diff --git a/.gitignore b/.gitignore index 1cdc9f7..ea76a74 100644 --- a/.gitignore +++ b/.gitignore @@ -1,9 +1,3 @@ target/ -pom.xml.tag -pom.xml.releaseBackup -pom.xml.versionsBackup -pom.xml.next -release.properties -dependency-reduced-pom.xml -buildNumber.properties -.mvn/timing.properties +.idea/ + diff --git a/pom.xml b/pom.xml new file mode 100644 index 0000000..845f5de --- /dev/null +++ b/pom.xml @@ -0,0 +1,29 @@ + + + 4.0.0 + + com.chaoticity.citationsentiment + citationsentimentclassifier + 1.0-SNAPSHOT + + + + maven + https://repo1.maven.org/maven2 + + + + + nz.ac.waikato.cms.weka + weka-stable + 3.6.6 + + + nz.ac.waikato.cms.weka + LibSVM + 1.0.3 + + + \ No newline at end of file diff --git a/src/main/java/com/chaoticity/citationsentiment/CitationSentimentClassifier.java b/src/main/java/com/chaoticity/citationsentiment/CitationSentimentClassifier.java new file mode 100644 index 0000000..f32a544 --- /dev/null +++ b/src/main/java/com/chaoticity/citationsentiment/CitationSentimentClassifier.java @@ -0,0 +1,75 @@ +/* + * To change this template, choose Tools | Templates + * and open the template in the editor. + */ +package com.chaoticity.citationsentiment; + +import weka.classifiers.Evaluation; +import weka.classifiers.functions.LibSVM; +import weka.core.Instances; +import weka.core.converters.ConverterUtils; +import weka.core.tokenizers.NGramTokenizer; +import weka.core.tokenizers.WordTokenizer; +import weka.filters.Filter; +import weka.filters.unsupervised.attribute.StringToWordVector; + +import java.util.Random; + +/** + * Code and data for citation sentiment classification reported in http://www.aclweb.org/anthology/P11-3015 + * The file test.arff contains only the test set with dependency triplets generated with Stanford CoreNLP + * Full corpus available at http://www.cl.cam.ac.uk/~aa496/citation-sentiment-corpus + * + * @author Awais Athar + */ +public class CitationSentimentClassifier { + + + /** + * @param args the command line arguments + */ + public static void main(String[] args) throws Exception { + ConverterUtils.DataSource source = new ConverterUtils.DataSource("test.arff"); + Instances data = source.getDataSet(); + + // Set class attribute + data.setClassIndex(data.attribute("@@class@@").index()); + + // delete unused attributes + data.deleteAttributeAt(1); + data.deleteAttributeAt(2); + + // split dependencies on space + StringToWordVector unigramFilter = new StringToWordVector(); + unigramFilter.setInputFormat(data); + unigramFilter.setIDFTransform(true); + unigramFilter.setAttributeIndices("3"); + WordTokenizer whitespaceTokenizer = new WordTokenizer(); + whitespaceTokenizer.setDelimiters(" "); + unigramFilter.setTokenizer(whitespaceTokenizer); + data = Filter.useFilter(data,unigramFilter); + + // make trigrams from citation sentences + StringToWordVector trigramFilter = new StringToWordVector(); + trigramFilter.setInputFormat(data); + trigramFilter.setIDFTransform(true); + trigramFilter.setAttributeIndices("2"); + NGramTokenizer tokenizer = new NGramTokenizer(); + tokenizer.setNGramMinSize(1); + tokenizer.setNGramMaxSize(3); + trigramFilter.setTokenizer(tokenizer); + data = Filter.useFilter(data,trigramFilter); + + // Train and test 10x cross-validation + int folds = 10; + LibSVM svm = new LibSVM(); + svm.setCost(1000); + Evaluation eval = new Evaluation(data); + eval.crossValidateModel(svm, data, folds, new Random(1)); + System.out.println(eval.toMatrixString()); + System.out.println(eval.toSummaryString()); + System.out.println(eval.toClassDetailsString()); + } + + +} diff --git a/test.arff b/test.arff new file mode 100644 index 0000000..95400e9 --- /dev/null +++ b/test.arff @@ -0,0 +1,7273 @@ +@relation 'C:\\work\\data\\arff\\postwriteup\\aan3withauth.txt' + +@attribute @@class@@ {o,n,p} +@attribute @@id@@ string +@attribute @@sentence@@ string +@attribute @@author@@ string +@attribute @@dependencies@@ string + +@data +o,736,'In it was observed that a significant percent of the queries made by a user in a search engine are associated to a repeated search ',Berger,'prep_observed_In pobj_In_ nsubjpass_observed_it auxpass_observed_was complm_associated_that det_percent_a amod_percent_significant nsubjpass_associated_percent prep_percent_of det_queries_the pobj_of_queries partmod_queries_made prep_made_by det_user_a pobj_by_user prep_user_in det_engine_a nn_engine_search pobj_in_engine auxpass_associated_are ccomp_observed_associated prep_associated_to det_search_a amod_search_repeated pobj_to_search ' +o,737,'Output sequence optimization Rather than basing classifications only on model parameters estimated from co-occurrences between input and output symbols employed for maximizing the likelihood of point-wise single-label predictions at the output level , classifier output may be augmented by an optimization over the output sequence as a whole using optimization techniques such as beam searching in the space of a conditional markov models output or hidden markov models ',Ratnaparkhi,'nn_optimization_Output nn_optimization_sequence dep_than_Rather advcl_augmented_than pcomp_than_basing dobj_basing_classifications dep_on_only prep_basing_on nn_parameters_model pobj_on_parameters partmod_parameters_estimated prep_estimated_from pobj_from_co-occurrences prep_co-occurrences_between nn_symbols_input cc_input_and conj_input_output pobj_between_symbols partmod_symbols_employed prep_employed_for pcomp_for_maximizing det_likelihood_the dobj_maximizing_likelihood prep_likelihood_of amod_predictions_point-wise amod_predictions_single-label pobj_of_predictions prep_maximizing_at det_level_the nn_level_output pobj_at_level nn_output_classifier nsubjpass_augmented_output aux_augmented_may auxpass_augmented_be dep_optimization_augmented prep_augmented_by det_optimization_an pobj_by_optimization prep_optimization_over det_sequence_the nn_sequence_output pobj_over_sequence prep_augmented_as det_whole_a pobj_as_whole partmod_whole_using nn_techniques_optimization dobj_using_techniques dep_as_such prep_techniques_as pobj_as_beam partmod_beam_searching prep_searching_in det_space_the pobj_in_space prep_space_of det__a amod__conditional amod__markov nn__models nn__output pobj_of_ cc__or amod_models_hidden amod_models_markov conj__models advmod_searching_ ' +o,738,'Dredze et al yielded the second highest score1 in the domain adaptation track ',Dredze,'nsubj_yielded_Dredze cc_Dredze_et conj_Dredze_al det_score1_the amod_score1_second amod_score1_highest dobj_yielded_score1 prep_yielded_in det_track_the nn_track_domain nn_track_adaptation pobj_in_track ' +o,739,'The IBM models search a version of permutation space with a one-to-many constraint ',Brown,'det_search_The nn_search_IBM nn_search_models nn_search_ det_version_a dep_search_version prep_version_of nn_space_permutation pobj_of_space prep_space_with det_constraint_a amod_constraint_one-to-many pobj_with_constraint ' +o,740,' propose the use of language models for sentiment analysis task and subjectivity extraction ',Pang,'advmod_propose_ det_use_the dobj_propose_use prep_use_of nn_models_language pobj_of_models prep_models_for nn_task_sentiment nn_task_analysis pobj_for_task cc_task_and amod_extraction_subjectivity conj_task_extraction ' +o,741,'In training process , we use GIZA + + 4 toolkit for word alignment in both translation directions , and apply grow-diag-final method to refine it ',Koehn,'prep_use_In nn_process_training pobj_In_process nsubj_use_we advmod_+_GIZA acomp_use_+ cc_+_+ dep_toolkit_4 conj_+_toolkit prep_+_for nn_alignment_word pobj_for_alignment prep_alignment_in det_directions_both nn_directions_translation pobj_in_directions cc_use_and conj_use_apply amod_method_grow-diag-final dobj_use_method aux_refine_to infmod_method_refine dobj_refine_it ' +o,742,'The models in the comparative study by did not include such features , and so , again for consistency of comparison , we experimentally verified that our maximum entropy model -LRB- a -RRB- consistently yielded higher scores than when the features were not used , and -LRB- b -RRB- consistently yielded higher scores than nave Bayes using the same features , in agreement with ',Klein,'det_models_The nsubj_include_models prep_models_in det_study_the amod_study_comparative pobj_in_study prep_study_by pobj_by_ aux_include_did neg_include_not amod_features_such dobj_include_features cc_include_and advmod_include_so advmod_verified_again prep_verified_for pobj_for_consistency prep_consistency_of pobj_of_comparison nsubj_verified_we advmod_verified_experimentally ccomp_include_verified complm_yielded_that poss_model_our amod_model_maximum amod_model_entropy nsubj_yielded_model dep_-LRB-_a advmod_yielded_consistently ccomp_verified_yielded amod_scores_higher dobj_yielded_scores prep_yielded_than advmod_used_when det_features_the nsubjpass_used_features auxpass_used_were neg_used_not pcomp_than_used cc_include_and dep_-RRB-_b nsubj_yielded_consistently conj_include_yielded amod_scores_higher dobj_yielded_scores prep_yielded_than amod_Bayes_nave pobj_than_Bayes dep_yielded_using det_features_the amod_features_same dobj_using_features prep_yielded_in pobj_in_agreement prep_yielded_with ' +o,743,' and et al ',Rosti,'cc__and conj__ cc__et conj__al ' +o,744,'We used the WordNet : : Similarity package to compute baseline scores for several existing measures , noting that one word pair was not processed in WS-353 because one of the words was missing from WordNet ',Pedersen,'nsubj_used_We det_WordNet_the dobj_used_WordNet nn_package_Similarity nsubj__package parataxis_used_ aux_compute_to xcomp__compute nn_scores_baseline dobj_compute_scores prep_compute_for amod_measures_several amod_measures_existing pobj_for_measures xcomp__noting complm_processed_that num_pair_one nn_pair_word nsubjpass_processed_pair auxpass_processed_was neg_processed_not ccomp_noting_processed prep_processed_in pobj_in_WS-353 mark_missing_because nsubj_missing_one prep_one_of det_words_the pobj_of_words aux_missing_was advcl_processed_missing prep_missing_from pobj_from_WordNet ' +o,745,'We use MER to tune the decoders parameters using a development data set ',Och,'nsubj_use_We nn__MER nsubj_tune_ aux_tune_to xcomp_use_tune det_parameters_the amod_parameters_decoders dobj_tune_parameters partmod_parameters_using det_set_a nn_set_development nn_set_data dobj_using_set ' +o,746,'The training set is extracted from TreeBank section 1518 , the development set , used in tuning parameters of the system , from section 20 , and the test set from section 21 ',Marcus,'det_set_The nn_set_training nsubjpass_extracted_set auxpass_extracted_is prep_extracted_from pobj_from_TreeBank num_TreeBank_ punct_TreeBank_section num_TreeBank_1518 det_development_the nsubj_set_development ccomp_extracted_set ccomp_set_used prep_used_in amod_parameters_tuning pobj_in_parameters prep_parameters_of det_system_the pobj_of_system prep_used_from pobj_from_section num_section_20 cc_extracted_and det_test_the nsubj_set_test conj_extracted_set prep_set_from pobj_from_section num_section_21 ' +o,747,'For non-local features , we adapt cube pruning from forest rescoring , since the situation here is analogous to machine translation decoding with integrated language models : we can view the scores of unit nonlocal features as the language model cost , computed on-the-fly when combining sub-constituents ',Huang,'prep_adapt_For amod_features_non-local pobj_For_features nsubj_adapt_we nn_pruning_cube dobj_adapt_pruning prep_adapt_from nn__forest amod__rescoring pobj_from_ mark_analogous_since det_situation_the nsubj_analogous_situation advmod_analogous_here cop_analogous_is dep__analogous prep_analogous_to nn_translation_machine pobj_to_translation partmod_translation_decoding prep_decoding_with amod_models_integrated nn_models_language pobj_with_models nsubj_view_we aux_view_can parataxis_adapt_view det_scores_the dobj_view_scores prep_scores_of nn_features_unit nn_features_nonlocal pobj_of_features prep_view_as det_cost_the nn_cost_language nn_cost_model pobj_as_cost amod_on-the-fly_computed appos_cost_on-the-fly advmod_combining_when advcl_view_combining dobj_combining_sub-constituents ' +o,748,'31 Agreement for Emotion Classes The kappa coefficient of agreement is a statistic adopted by the Computational Linguistics community as a standard measure for this purpose ',Carletta,'num_Agreement_31 nsubj_Classes_Agreement prep_Agreement_for pobj_for_Emotion det_coefficient_The nn_coefficient_kappa nsubj_statistic_coefficient prep_coefficient_of pobj_of_agreement cop_statistic_is det_statistic_a ccomp_Classes_statistic partmod_statistic_adopted prep_adopted_by det_community_the nn_community_Computational nn_community_Linguistics pobj_by_community prep_adopted_as det_measure_a amod_measure_standard pobj_as_measure prep_measure_for det_purpose_this pobj_for_purpose ' +p,749,'The averaged version of the perceptron , like the voted perceptron , reduces the effect of over-training ',Collins,'det_version_The amod_version_averaged nsubj_reduces_version prep_version_of det__the nn__perceptron pobj_of_ prep_version_like det__the amod__voted nn__perceptron pobj_like_ det_effect_the dobj_reduces_effect prep_effect_of pobj_of_over-training ' +o,750,'ITGs translate into simple -LRB- 2,2 -RRB- - BRCGs in the following way ; see for a definition of ITGs ',Wu,'dep_ITGs_translate prep_translate_into pobj_into_simple appos_simple_2,2 dep_simple_BRCGs prep_BRCGs_in det_way_the amod_way_following pobj_in_way parataxis_translate_see dobj_see_ prep_see_for det_definition_a pobj_for_definition prep_definition_of pobj_of_ITGs ' +o,751,'This may be because their system was not tuned using minimum error rate training ',Och,'nsubj_be_This aux_be_may mark_tuned_because poss_system_their nsubjpass_tuned_system auxpass_tuned_was neg_tuned_not advcl_be_tuned dep_tuned_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +o,752,'However , most of the existing models have been developed for English and trained on the Penn Treebank , which raises the question whether these models generalize to other languages , and to annotation schemes that differ from the Penn Treebank markup ',Marcus,'advmod_developed_However nsubjpass_developed_most prep_most_of det_models_the amod_models_existing pobj_of_models aux_developed_have auxpass_developed_been prep_developed_for pobj_for_English cc_developed_and conj_developed_trained prep_trained_on det__the nn__Penn nn__Treebank pobj_on_ nsubj_raises_which rcmod__raises det_question_the dobj_raises_question complm_generalize_whether det_models_these nsubj_generalize_models ccomp_raises_generalize prep_generalize_to amod_languages_other pobj_to_languages cc_on_and conj_on_to nn_schemes_annotation pobj_to_schemes nsubj_differ_that rcmod_schemes_differ prep_differ_from det_markup_the nn_markup_Penn nn_markup_Treebank pobj_from_markup ' +o,753,'Following , we used sections 0-18 of the Wall Street Journal -LRB- WSJ -RRB- corpus for training , sections 19-21 for development , and sections 22-24 for final evaluation ',Collins,'prep_used_Following pobj_Following_ nsubj_used_we dobj_used_sections num_sections_0-18 prep_sections_of det_corpus_the nn_Journal_Wall nn_Journal_Street nn_corpus_Journal abbrev_Journal_WSJ pobj_of_corpus prep_corpus_for pobj_for_training conj_sections_sections num_sections_19-21 prep_sections_for pobj_for_development cc_sections_and conj_sections_sections num_sections_22-24 prep_sections_for amod_evaluation_final pobj_for_evaluation ' +o,754,'In , the authors provide some sample subtrees resulting from such a 1,000-word clustering ',Brown,'prep_provide_In pobj_In_ det_authors_the nsubj_provide_authors det_sample_some nsubj_subtrees_sample ccomp_provide_subtrees xcomp_subtrees_resulting prep_resulting_from predet_clustering_such det_clustering_a amod_clustering_1,000-word pobj_from_clustering ' +o,755,'We took part the Multilingual Track of all ten languages provided by the CoNLL-2007 shared task organizer ',Marcus,'nsubj_took_We iobj_took_part det_Track_the nn_Track_Multilingual dobj_took_Track prep_Track_of quantmod_ten_all num_languages_ten pobj_of_languages partmod_languages_provided prep_provided_by det_organizer_the amod_organizer_CoNLL-2007 amod_organizer_shared nn_organizer_task pobj_by_organizer ' +o,756,'To set the weights , m , we carried out minimum error rate training using BLEU as the objective function ',Och,'aux_set_To dep_carried_set det_weights_the dobj_set_weights appos_weights_m nsubj_carried_we prt_carried_out amod__minimum nn__error nn__rate nn__training dobj_carried_ partmod__using nn__BLEU dobj_using_ prep__as det_function_the amod_function_objective pobj_as_function ' +o,757,'Use of global features for structured prediction problem has been explored by several NLP applications such as sequential labeling and dependency parsing with a great deal of success ',Finkel,'nsubjpass_explored_Use prep_Use_of amod_features_global pobj_of_features prep_features_for amod_problem_structured nn_problem_prediction pobj_for_problem aux_explored_has auxpass_explored_been prep_explored_by amod_applications_several nn_applications_NLP pobj_by_applications dep_as_such prep_applications_as amod__sequential nn__labeling pobj_as_ cc__and amod__dependency nn__parsing conj__ prep_explored_with det_deal_a amod_deal_great pobj_with_deal prep_deal_of pobj_of_success ' +o,758,'It is important to realize that the output of all mentioned processing steps is noisy and contains plenty of mistakes , since the data has huge variability in terms of quality , style , genres , domains etc , and domain adaptation for the NLP tasks involved is still an open problem ',Dredze,'nsubj_important_It cop_important_is aux_realize_to xcomp_important_realize complm_mentioned_that det_output_the nsubj_mentioned_output prep_output_of pobj_of_all ccomp_realize_mentioned nn_steps_processing nsubj_noisy_steps cop_noisy_is ccomp_mentioned_noisy cc_mentioned_and conj_mentioned_contains dobj_contains_plenty prep_plenty_of pobj_of_mistakes mark_has_since det_data_the nsubj_has_data advcl_important_has amod_variability_huge dobj_has_variability prep_variability_in pobj_in_terms prep_terms_of pobj_of_quality conj_quality_style conj_quality_genres nn_etc_domains conj_quality_etc cc_quality_and nn_adaptation_domain conj_quality_adaptation prep_adaptation_for det_NLP_the pobj_for_NLP nsubj_problem_tasks partmod_tasks_involved cop_problem_is advmod_problem_still det_problem_an amod_problem_open dep_important_problem ' +o,759,'They are also used for inducing alignments ',Wu,'nsubjpass_used_They auxpass_used_are advmod_used_also prep_used_for amod_alignments_inducing pobj_for_alignments ' +o,760,'In recent work , proposed a general framework for including morphological features in a phrase-based SMT system by factoring the representation of words into a vector of morphological features and allowing a phrase-based MT system to work on any of the factored representations , which is implemented in the Moses system ',Koehn,'prep_proposed_In amod_work_recent pobj_In_work nsubj_proposed_ det_framework_a amod_framework_general dobj_proposed_framework prep_proposed_for pcomp_for_including amod_features_morphological dobj_including_features prep_features_in det_system_a amod_system_phrase-based nn_system_SMT pobj_in_system prep_including_by pcomp_by_factoring det_representation_the dobj_factoring_representation prep_representation_of pobj_of_words prep_factoring_into det_vector_a pobj_into_vector prep_vector_of amod_features_morphological pobj_of_features cc_including_and conj_including_allowing det_system_a amod_system_phrase-based nn_system_MT dobj_allowing_system aux_work_to xcomp_allowing_work prep_work_on pobj_on_any prep_any_of det_representations_the amod_representations_factored pobj_of_representations nsubjpass_implemented_which auxpass_implemented_is rcmod_representations_implemented prep_implemented_in det_system_the nn_system_Moses pobj_in_system ' +o,761,'2 Architecture of the system The goal of statistical machine translation -LRB- SMT -RRB- is to produce a target sentence e from a source sentence f It is today common practice to use phrases as translation units and a log linear framework in order to introduce several models explaining the translation process : e ? ? = argmaxp -LRB- e f -RRB- = argmaxe -LCB- exp -LRB- summationdisplay i ihi -LRB- e , f -RRB- -RRB- -RCB- -LRB- 1 -RRB- The feature functions hi are the system models and the i weights are typically optimized to maximize a scoring function on a development set ',Koehn,'number_Architecture_2 nsubj_is_Architecture prep_Architecture_of det_system_the pobj_of_system det_goal_The dep_system_goal prep_goal_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_Architecture_SMT aux_produce_to xcomp_is_produce det_sentence_a nn_sentence_target dobj_produce_sentence dep_from_e dep_practice_from det_sentence_a nn_sentence_source pobj_from_sentence dep_practice_f nsubj_practice_It cop_practice_is advmod_practice_today amod_practice_common dep_is_practice aux_use_to infmod_practice_use dobj_use_phrases mark__as nn_units_translation nsubj__units advcl_use_ cc__and det_log_a conj__log amod_framework_linear dobj_log_framework prep_framework_in pobj_in_order aux_introduce_to xcomp_log_introduce amod_models_several dobj_introduce_models dep_e_explaining det_process_the nn_process_translation dobj_explaining_process dep_use_e dep_use_= nsubj_=_argmaxp dep_f_e dep_-LRB-_f ccomp_=_= dobj_=_argmaxe dep_argmaxe_exp dep_argmaxe_summationdisplay dep_summationdisplay_i dep_summationdisplay_ihi dep_ihi_e appos_e_f dep_models_1 det_hi_The nn_hi_feature nn_hi_functions nsubj_models_hi cop_models_are det_models_the nn_models_system rcmod_argmaxe_models cc_models_and det_weights_the amod_weights_i nsubjpass_optimized_weights auxpass_optimized_are advmod_optimized_typically conj_models_optimized aux_maximize_to xcomp_optimized_maximize det_function_a amod_function_scoring dobj_maximize_function prep_=_on det__a nn__development nn__set pobj_on_ ' +o,762,'1 Introduction Sentiment analysis have been widely conducted in several domains such as movie reviews , product reviews , news and blog reviews ',Turney,'num_analysis_1 nn_analysis_Introduction nn_analysis_Sentiment nsubjpass_conducted_analysis aux_conducted_have auxpass_conducted_been advmod_conducted_widely prep_conducted_in amod_domains_several pobj_in_domains dep_as_such prep_domains_as nn_reviews_movie pobj_as_reviews nn_reviews_product conj_reviews_reviews conj_reviews_news cc_reviews_and amod_reviews_blog conj_reviews_reviews ' +o,763,'Their approaches include the use of a vector-based information retrieval technique \\/ bin\\/bash : line 1 : a : command not found Our do - mains are more varied , which may results in more recognition errors ',Chu-Carroll,'poss_approaches_Their nsubj_found_approaches rcmod_approaches_include det_use_the dobj_include_use prep_use_of det_information_a amod_information_vector-based pobj_of_information nn_\\/_retrieval nn_\\/_technique nn_\\/_ dep_include_\\/ dep_\\/_bin\\/bash dep_bin\\/bash_line num_line_1 dep_\\/_a dep_a_command neg_command_not nsubj_do_Our ccomp_found_do nsubjpass_varied_mains auxpass_varied_are advmod_varied_more parataxis_do_varied nsubj_results_which aux_results_may ccomp_do_results prep_results_in amod_errors_more nn_errors_recognition pobj_in_errors ' +o,764,'The corpus was aligned with GIZA + + and symmetrized with the grow-diag-finaland heuristic ',Koehn,'det_corpus_The nsubjpass_aligned_corpus auxpass_aligned_was prep_aligned_with nn_+_GIZA pobj_with_+ cc__+ amod_+_ cc__and conj__symmetrized prep_aligned_with det_heuristic_the amod_heuristic_grow-diag-finaland pobj_with_heuristic ' +p,765,'Since the use of cluster of machines is not always practical , showed a randomized data structure called Bloom filter , that can be used to construct space efficient language models 513 for SMT ',Talbot,'mark_practical_Since det_use_the nsubj_practical_use prep_use_of pobj_of_cluster prep_cluster_of pobj_of_machines cop_practical_is neg_practical_not advmod_practical_always advcl_showed_practical nsubj_showed_ det_structure_a amod_structure_randomized nn_structure_data nsubj_called_structure ccomp_showed_called nn_filter_Bloom dobj_called_filter nsubjpass_used_that aux_used_can auxpass_used_be rcmod_filter_used aux_construct_to purpcl_used_construct nn_models_space amod_models_efficient nn_models_language nsubj_513_models xcomp_construct_513 prep_513_for pobj_for_SMT ' +o,766,'BLEU was devised to provide automatic evaluation of MT output ',Papineni,'nn__BLEU nsubjpass_devised_ auxpass_devised_was aux_provide_to xcomp_devised_provide amod_evaluation_automatic dobj_provide_evaluation prep_evaluation_of nn_output_MT pobj_of_output ' +o,767,'Statistics in linguistics , Oxford : Basil Blackwell N Chinchor Evaluating message understanding systems : an analysis of the third Message Understanding Conference -LRB- MUC-3 1993 Computational Linguistics 19 409 -- 449 Chinchor , 1993 Chinchor , N , et al , 1993 ',Marcus,'dep_Chinchor_Statistics prep_Statistics_in pobj_in_linguistics appos_linguistics_Oxford nn_systems_Basil nn_systems_Blackwell nn_systems_ nn_systems_ nn_systems_ nn_systems_ nn_systems_ nn_systems_N nn_systems_Chinchor nn_systems_ nn_systems_ nn_systems_ nn_systems_Evaluating nn_systems_message nn_systems_understanding dep_Oxford_systems det_analysis_an dep_Oxford_analysis prep_analysis_of det_</volume>_the nn_</volume>_third nn_Conference_Message nn_Conference_Understanding dep_third_Conference nn_</volume>_MUC-3 nn_</volume>_ nn__ num__1993 nn__ nn__ nn__Computational nn__Linguistics nn__ nn__ num__19 pobj_of_ number_409_ num__409 number__449 dep_Oxford_ nn_Chinchor_ dep__Chinchor dep_Chinchor_1993 nn_Chinchor_ nn_Chinchor_ appos_Chinchor_N nn_al_et appos_Chinchor_al appos_Chinchor_1993 ' +o,768,'Note that it is straightforward to calculate these expected counts using a variant of the inside-outside algorithm applied to the dependency-parsing data structures for projective dependency structures , or the matrix-tree theorem for nonprojective dependency structures ',Smith,'complm_straightforward_that nsubj_straightforward_it cop_straightforward_is ccomp_Note_straightforward aux_calculate_to xcomp_straightforward_calculate det_counts_these amod_counts_expected dobj_calculate_counts partmod_counts_using det_variant_a dobj_using_variant prep_variant_of det__the amod__inside-outside nn__algorithm pobj_of_ partmod__applied prep_applied_to det__the amod__ amod__dependency-parsing nn__data nn__structures pobj_to_ prep__for amod_structures_projective amod_structures_dependency pobj_for_structures cc_counts_or det__the amod__matrix-tree nn__theorem conj_counts_ prep__for amod_structures_nonprojective amod_structures_dependency pobj_for_structures ' +o,769,'Following , we consider an anaphoric reference , NPi , correctly resolved if NPi and its closest antecedent are in the same coreference chain in the resulting partition ',Ponzetto,'prep_consider_Following pobj_Following_ nsubj_consider_we det_reference_an amod_reference_anaphoric nsubj_resolved_reference appos_reference_NPi advmod_resolved_correctly ccomp_consider_resolved mark_are_if nsubj_are_NPi cc_NPi_and poss_antecedent_its amod_antecedent_closest conj_NPi_antecedent advcl_resolved_are prep_are_in det_chain_the amod_chain_same nn_chain_coreference pobj_in_chain prep_chain_in det_partition_the amod_partition_resulting pobj_in_partition ' +o,770,'This algorithm adjusts the log-linear weights so that BLEU is maximized over a given development set ',Papineni,'det_algorithm_This nsubj_weights_algorithm dep_weights_adjusts det_weights_the amod_weights_log-linear advmod_maximized_so dep_maximized_that nn__BLEU nsubjpass_maximized_ auxpass_maximized_is dep_weights_maximized prep_maximized_over det_set_a amod_set_given nn_set_development pobj_over_set ' +o,771,'We discriminatively trained our parser in an on-line fashion using a variant of the voted perceptron ',Collins,'nsubj_trained_We advmod_trained_discriminatively poss_parser_our dobj_trained_parser prep_parser_in det_fashion_an amod_fashion_on-line pobj_in_fashion partmod_fashion_using det_variant_a dobj_using_variant prep_variant_of det_perceptron_the amod_perceptron_voted pobj_of_perceptron ' +o,772,'In fact , we found that it doesnt do so badly at all : the bitag HMM estimated by EM achieves a mean 1-to1 tagging accuracy of 40 \% , which is approximately the same as the 413 \% reported by for their sophisticated MRF model ',Haghighi,'prep_found_In pobj_In_fact nsubj_found_we complm_do_that nsubj_do_it aux_do_doesnt ccomp_found_do advmod_badly_so advmod_do_badly advmod_do_at dep_at_all det_HMM_the amod_HMM_bitag nsubj_achieves_HMM partmod_HMM_estimated prep_estimated_by pobj_by_EM parataxis_found_achieves det_accuracy_a amod_accuracy_mean amod_accuracy_1-to1 nn_accuracy_tagging dobj_achieves_accuracy prep_accuracy_of num_\%_40 pobj_of_\% nsubj_same_which cop_same_is advmod_same_approximately det_same_the rcmod_\%_same prep_same_as det_\%_the num_\%_413 pobj_as_\% partmod_\%_reported prep_reported_by pobj_by_ prep__for poss_model_their amod_model_sophisticated nn_model_MRF pobj_for_model ' +o,773,'2 Motivation and Prior Work While several authors have looked at the supervised adaptation case , there are less -LRB- and especially less successful -RRB- studies on semi-supervised domain adaptation ',Dredze,'dep_Motivation_2 cc_Motivation_and amod_Work_Prior conj_Motivation_Work mark_looked_While amod_authors_several nsubj_looked_authors aux_looked_have advcl_are_looked prep_looked_at det_case_the amod_case_supervised nn_case_adaptation pobj_at_case expl_are_there dep_Motivation_are acomp_are_less dep_-LRB-_and advmod_less_especially dep_-LRB-_less dep_less_successful dep_on_studies prep_are_on amod_adaptation_semi-supervised nn_adaptation_domain pobj_on_adaptation ' +o,774,'There are other types of variations for phrases ; for example , insertion , deletion or substitution of words , and permutation of words such as view point and point of view are such variations ',Berger,'expl_are_There amod_types_other nsubj_are_types prep_types_of pobj_of_variations prep_variations_for pobj_for_phrases prep_phrases_for pobj_for_example conj_example_insertion conj_example_deletion cc_example_or conj_example_substitution prep_phrases_of pobj_of_words cc_are_and nsubj_variations_permutation prep_permutation_of pobj_of_words dep_as_such prep_words_as nn_point_view pobj_as_point cc_point_and conj_point_point prep_point_of pobj_of_view cop_variations_are amod_variations_such conj_are_variations advmod_variations_ ' +o,775,'? ? search engines : uses the Altavista web browser , while we consider and combine the frequency information acquired from three web search engines ',Turney,'nn_engines_search nsubj_uses_engines nsubj_uses_ det_browser_the nn_browser_Altavista nn_browser_web dobj_uses_browser mark_consider_while nsubj_consider_we advcl_uses_consider cc_consider_and conj_consider_combine det_information_the nn_information_frequency dobj_consider_information partmod_information_acquired prep_acquired_from num_engines_three nn_engines_web nn_engines_search pobj_from_engines ' +o,776,'Many methods have been proposed to measure the co-occurrence relation between two words such as 2 , mutual information , t-test , and loglikelihood ',Dunning,'amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been aux_measure_to xcomp_proposed_measure det_relation_the amod_relation_co-occurrence dobj_measure_relation prep_relation_between num_words_two pobj_between_words dep_as_such prep_words_as num__2 pobj_as_ amod__mutual nn__information conj__ amod__t-test conj__ cc__and conj__loglikelihood ' +p,777,'1 Introduction The Maximum Entropy -LRB- ME -RRB- statistical framework has been successfully deployed in several NLP tasks ',Berger,'num__1 nn_Entropy_Introduction nn_Entropy_The nn_Entropy_Maximum measure_statistical_Entropy abbrev_Entropy_ME amod__statistical nn__framework nsubjpass_deployed_ aux_deployed_has auxpass_deployed_been advmod_deployed_successfully prep_deployed_in amod_tasks_several nn_tasks_NLP pobj_in_tasks ' +n,778,'Unfortunately , this is not the case for such widely used MT evaluation metrics as BLEU and NIST ',Papineni,'advmod_case_Unfortunately nsubj_case_this cop_case_is neg_case_not det_case_the prep_case_for amod_metrics_such advmod_used_widely amod_metrics_used nn_metrics_MT nn_metrics_evaluation pobj_for_metrics prep_case_as nn__BLEU pobj_as_ cc__and nn__NIST conj__ ' +o,779,'It has been argued that the reliability of a coding schema can be assessed only on the basis of judgments made by naive coders ',Carletta,'nsubjpass_argued_It aux_argued_has auxpass_argued_been complm_assessed_that det_reliability_the nsubjpass_assessed_reliability prep_reliability_of det_schema_a amod_schema_coding pobj_of_schema aux_assessed_can auxpass_assessed_be ccomp_argued_assessed advmod_assessed_only prep_assessed_on det_basis_the pobj_on_basis prep_basis_of pobj_of_judgments partmod_judgments_made prep_made_by amod_coders_naive pobj_by_coders ' +o,780,'to the pair-wise TER alignment described in ',Rosti,'dep_alignment_to det_alignment_the amod_alignment_pair-wise nn_alignment_TER partmod_alignment_described prep_described_in ' +o,781,'We obtain aligned parallel sentences and the phrase table after the training of Moses , which includes running GIZA + + , grow-diagonal-final symmetrization and phrase extraction ',Koehn,'nsubj_obtain_We ccomp_obtain_aligned amod_sentences_parallel dobj_aligned_sentences cc_sentences_and det_table_the nn_table_phrase conj_sentences_table prep_aligned_after det_training_the pobj_after_training prep_training_of pobj_of_Moses nsubj_includes_which rcmod_training_includes xcomp_includes_running nn__GIZA nn__+ nn__+ dobj_running_ amod_symmetrization_grow-diagonal-final conj__symmetrization cc__and nn_extraction_phrase conj__extraction ' +o,782,'From the above discussion , we can see that traditional tree sequence-based method uses single tree as translation input while the forestbased model uses single sub-tree as the basic translation unit that can only learn tree-to-string rules ',Galley,'prep_see_From det_discussion_the amod_discussion_above pobj_From_discussion nsubj_see_we aux_see_can complm_uses_that amod_method_traditional nn_method_tree nn_method_sequence-based nsubj_uses_method ccomp_see_uses amod_tree_single dobj_uses_tree prep_uses_as nn_input_translation pobj_as_input mark_uses_while det_model_the amod_model_forestbased nsubj_uses_model advcl_uses_uses amod_sub-tree_single dobj_uses_sub-tree prep_uses_as det_unit_the amod_unit_basic nn_unit_translation pobj_as_unit nsubj_learn_that aux_learn_can advmod_learn_only rcmod_unit_learn amod_rules_tree-to-string amod_rules_ dobj_learn_rules ' +o,783,'Our baseline method for ambiguity resolution is the Collins parser as implemented by Bikel ',Collins,'poss_method_Our amod_method_baseline nsubj_parser_method prep_method_for amod_resolution_ambiguity pobj_for_resolution cop_parser_is det_parser_the nn_parser_Collins mark_implemented_as advcl_parser_implemented prep_implemented_by pobj_by_Bikel ' +o,784,'We then built separate English-to-Spanish and Spanish-to-English directed word alignments using IBM model 4 , combined them using the intersect + grow heuristic , and extracted phrase-level translation pairs of maximum length 7 using the alignment template approach ',Och,'nsubj_built_We advmod_built_then dobj_built_separate dep_separate_English-to-Spanish cc_separate_and amod_alignments_Spanish-to-English amod_alignments_directed nn_alignments_word conj_separate_alignments xcomp_built_using nn_model_IBM nsubj__model num__4 xcomp_using_ partmod__combined nsubj_using_them dep_combined_using det__the amod__intersect cc_intersect_+ conj_intersect_grow amod__heuristic dobj_using_ cc_built_and conj_built_extracted amod_pairs_phrase-level nn_pairs_translation dobj_extracted_pairs prep_pairs_of nn_length_maximum pobj_of_length num_length_7 xcomp_built_using det_approach_the amod_approach_alignment nn_approach_template dobj_using_approach ' +o,785,'For instance , both Pang and Lee and consider the thumbs up\\/thumbs down decision : is a film review positive or negative ? ',Turney,'prep_consider_For pobj_For_instance preconj__both nn__Pang cc_Pang_and conj_Pang_Lee nsubj_consider_ cc__and nn__ conj__ det_thumbs_the nsubj_up\\/thumbs_thumbs ccomp_consider_up\\/thumbs prt_up\\/thumbs_down dobj_up\\/thumbs_decision cop_review_is det_review_a nn_review_film dep_decision_review amod_review_positive cc_positive_or conj_positive_negative ' +o,786,'Binarizing the syntax trees for syntax-based machine translation is similar in spirit to generalizing parsing models via markovization ',Collins,'det_trees_the nn_trees_syntax nsubj_similar_trees prep_trees_for amod_translation_syntax-based nn_translation_machine pobj_for_translation cop_similar_is ccomp_Binarizing_similar prep_similar_in pobj_in_spirit aux_generalizing_to xcomp_similar_generalizing amod_models_parsing dobj_generalizing_models prep_models_via pobj_via_markovization ' +o,787,'4 Options from the Translation Table Phrase-based statistical machine translation methods acquire their translation knowledge in form of large phrase translation tables automatically from large amounts of translated texts ',Koehn,'num_Options_4 nsubj_Table_Options prep_Options_from det_Translation_the pobj_from_Translation amod_methods_Phrase-based amod_methods_statistical nn_methods_machine nn_methods_translation nsubj_acquire_methods ccomp_Table_acquire poss_knowledge_their nn_knowledge_translation dobj_acquire_knowledge prep_acquire_in pobj_in_form prep_form_of amod_tables_large nn_tables_phrase nn_tables_translation pobj_of_tables advmod_acquire_automatically prep_acquire_from amod_amounts_large pobj_from_amounts prep_amounts_of amod_texts_translated pobj_of_texts ' +o,788,'For example , collected reviews from a movie database and rated them as positive , negative , or neutral based on the rating -LRB- eg , number of stars -RRB- given by the reviewer ',Pang,'prep_collected_For pobj_For_example nsubj_collected_ dobj_collected_reviews prep_collected_from det_database_a nn_database_movie pobj_from_database cc_collected_and conj_collected_rated dobj_rated_them prep_rated_as pobj_as_positive conj_positive_negative cc_positive_or conj_positive_neutral prep_rated_based dep_based_on det_rating_the pobj_on_rating dep_rating_eg appos_eg_number prep_number_of pobj_of_stars prep_rating_given dep_given_by det_reviewer_the pobj_by_reviewer ' +o,789,'The earliest work in this direction are those of , , , , and ',Weeds,'det_work_The amod_work_earliest nsubj_those_work prep_work_in det_direction_this pobj_in_direction cop_those_are prep_those_of nn_and_ appos_and_ appos_and_ appos_and_ appos_and_ pobj_of_and ' +o,790,' established that it is important to tune -LRB- the trade-off between Precision and Recall -RRB- to maximize performance ',Fraser,'nsubj_established_ complm_important_that nsubj_important_it cop_important_is ccomp_established_important prep_important_to pobj_to_tune det_trade-off_the dep_tune_trade-off prep_trade-off_between pobj_between_Precision cc_Precision_and conj_Precision_Recall aux_maximize_to infmod_tune_maximize dobj_maximize_performance ' +o,791,'Moreover , under this view , SMT becomes quite similar to sequential natural language annotation problems such as part-of-speech tagging and shallow parsing , and the novel training algorithm presented in this paper is actually most similar to work on training algorithms presented for these task , eg the on-line training algorithm presented in and the perceptron training algorithm presented in ',Collins,'advmod_similar_Moreover prep_similar_under det_view_this pobj_under_view nsubj_similar_SMT cop_similar_becomes advmod_similar_quite aux_sequential_to xcomp_similar_sequential amod_problems_natural nn_problems_language nn_problems_annotation dobj_sequential_problems dep_as_such prep_problems_as amod_tagging_part-of-speech pobj_as_tagging cc_tagging_and nn_parsing_shallow conj_tagging_parsing cc_similar_and det_algorithm_the amod_algorithm_novel nn_algorithm_training nsubj_similar_algorithm partmod_algorithm_presented prep_presented_in det_paper_this pobj_in_paper cop_similar_is advmod_similar_actually advmod_similar_most conj_similar_similar aux_work_to xcomp_similar_work prep_work_on nn_algorithms_training pobj_on_algorithms partmod_algorithms_presented prep_presented_for det_task_these pobj_for_task partmod_task_eg det_algorithm_the amod_algorithm_on-line nn_algorithm_training dobj_eg_algorithm partmod_algorithm_presented prep_presented_in pobj_in_ cc_similar_and det_algorithm_the nn_algorithm_perceptron nn_algorithm_training nsubj_presented_algorithm conj_similar_presented prt_presented_in ' +o,792,'Feature weights vector are trained discriminatively in concert with the language model weight to maximize the BLEU automatic evaluation metric via Minimum Error Rate Training -LRB- MERT -RRB- ',Och,'amod_vector_Feature nn_vector_weights nsubjpass_trained_vector auxpass_trained_are advmod_trained_discriminatively prep_trained_in pobj_in_concert prep_trained_with det_weight_the nn_weight_language nn_weight_model pobj_with_weight aux_maximize_to xcomp_trained_maximize det_metric_the amod_metric_BLEU amod_metric_ amod_metric_automatic nn_metric_evaluation dobj_maximize_metric prep_metric_via nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_via_Training abbrev_Training_MERT ' +o,793,' and Wiebe -LRB- 2000 -RRB- focused on learning adjectives and adjectival phrases and Wiebe et al ',Turney,'nsubj_focused_ cc__and conj__Wiebe appos__2000 prep_focused_on pcomp_on_learning dobj_learning_adjectives cc_adjectives_and amod_phrases_adjectival conj_adjectives_phrases cc_phrases_and conj_phrases_Wiebe cc_Wiebe_et conj_Wiebe_al ' +p,794,'To achieve efficient parsing , we use a beam search strategy like the previous methods ',Collins,'aux_achieve_To dep_use_achieve amod_parsing_efficient dobj_achieve_parsing nsubj_use_we det_strategy_a nn_strategy_beam nn_strategy_search dobj_use_strategy prep_strategy_like det_methods_the amod_methods_previous pobj_like_methods ' +o,795,'A number of alignment techniques have been proposed , varying from statistical methods to lexical methods ',Brown,'det_number_A nsubjpass_proposed_number prep_number_of amod_techniques_alignment pobj_of_techniques aux_proposed_have auxpass_proposed_been dobj_proposed_varying prep_varying_from amod_methods_statistical pobj_from_methods partmod_methods_ prep__to amod__lexical nn__methods pobj_to_ ' +p,796,'Successflfl examples of reuse of data resources include : the WordNet thesaurus ; the Penn Tree Bank ; the Longmans Dictionary of Contemporary English ',Marcus,'amod_examples_Successflfl nsubj_include_examples prep_examples_of pobj_of_reuse prep_reuse_of nn_resources_data pobj_of_resources det__the nn__WordNet nn__thesaurus dobj_include_ det__the nn__Penn nn__Tree nn__Bank dep__ det_Dictionary_the nn_Dictionary_Longmans dep__Dictionary prep_Dictionary_of nn__Contemporary nn__English pobj_of_ ' +o,797,'Even though there are some studies that compare the results from statistically computed association measures with word association norms from psycholinguistic experiments there has not been any research on the usage of a digital , network-based dictionary reflecting the organization of the mental lexicon to our knowledge ',Rapp,'advmod_are_Even dep_are_though expl_are_there det_studies_some nsubj_are_studies nsubj_compare_that rcmod_studies_compare det_results_the nsubj__results prep_results_from advmod_computed_statistically amod_measures_computed nn_measures_association pobj_from_measures prep_measures_with nn_norms_word nn_norms_association pobj_with_norms prep_norms_from amod_experiments_psycholinguistic pobj_from_experiments ccomp_compare_ expl_been_there aux_been_has neg_been_not ccomp__been det_research_any dobj_been_research prep_research_on det_usage_the pobj_on_usage prep_usage_of det_dictionary_a amod_dictionary_digital amod_dictionary_network-based pobj_of_dictionary partmod_dictionary_reflecting det_organization_the dobj_reflecting_organization prep_organization_of det_lexicon_the amod_lexicon_mental pobj_of_lexicon prep_reflecting_to poss_knowledge_our pobj_to_knowledge ' +o,798,'Finally , we use as a feature the mappings produced in of WordNet senses to Oxford English Dictionary senses ',Navigli,'advmod_use_Finally nsubj_use_we prep_use_as det_feature_a pobj_as_feature det_mappings_the nsubj_produced_mappings dep_use_produced prep_produced_in pobj_in_ prep__of nn_senses_WordNet pobj_of_senses prep_produced_to nn_senses_Oxford nn_senses_English nn_senses_Dictionary pobj_to_senses ' +o,799,'3 Semantic Representation 31 The Need for Dependencies Perhaps the most common representation of text for assessing content is Bag-Of-Words or Bag-of-NGrams ',Papineni,'dep_Need_3 nn_Representation_Semantic dep_3_Representation dep_Need_31 det_Need_The dep_representation_Need prep_Need_for pobj_for_Dependencies advmod_representation_Perhaps det_representation_the advmod_common_most amod_representation_common prep_representation_of pobj_of_text prep_text_for pcomp_for_assessing nsubj_Bag-Of-Words_content cop_Bag-Of-Words_is ccomp_assessing_Bag-Of-Words cc_Bag-Of-Words_or conj_Bag-Of-Words_Bag-of-NGrams ' +o,800,'First , we adopt an ONTOLOGICALLY PROMISCUOUS representation that includes a wide variety of types of entities ',Hobbs,'advmod_adopt_First nsubj_adopt_we det__an nn__ONTOLOGICALLY nn__PROMISCUOUS nn__representation dobj_adopt_ nsubj_includes_that rcmod__includes det_variety_a amod_variety_wide dobj_includes_variety prep_variety_of pobj_of_types prep_types_of pobj_of_entities ' +o,801,'For each training direction , we run GIZA + + , specifying 5 iterations of Model 1 , 4 iterations of the HMM model , and 4 iterations of Model 4 ',Och,'prep_run_For det_direction_each nn_direction_training pobj_For_direction nsubj_run_we nsubj_+_GIZA xcomp_run_+ cc_+_+ conj_+_ xcomp_run_specifying num_iterations_5 dobj_specifying_iterations prep_iterations_of pobj_of_Model num_Model_1 num_iterations_4 conj_iterations_iterations prep_iterations_of det__the nn__HMM nn__model pobj_of_ cc_iterations_and num_iterations_4 conj_iterations_iterations prep_iterations_of pobj_of_Model num_Model_4 ' +o,802,'To set the weight vector w , we train twenty averaged perceptrons on different shuffles of data drawn from sections 0221 of the Penn Treebank ',Collins,'aux_set_To dep_train_set det_w_the nn_w_weight nn_w_vector dobj_set_w nsubj_train_we nsubj_averaged_twenty ccomp_train_averaged amod__perceptrons dobj_averaged_ prep_averaged_on amod_shuffles_different pobj_on_shuffles prep_shuffles_of pobj_of_data partmod_data_drawn prep_drawn_from pobj_from_sections num_sections_0221 prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,803,'In the refined model 2 alignment probabilities a -LRB- ilj , l , m -RRB- are included to model the effect that the position of a word influences the position of its translation ',Brown,'det_model_the amod_model_refined pobj_In_model number__2 num_probabilities_ amod_probabilities_alignment nsubjpass_included_probabilities amod_probabilities_a nn_m_ilj dep_m_l dep_a_m auxpass_included_are rcmod_model_included aux_model_to xcomp_included_model det_effect_the dobj_model_effect complm_influences_that det_position_the nsubj_influences_position prep_position_of det_word_a pobj_of_word ccomp_model_influences det_position_the dobj_influences_position prep_position_of poss_translation_its pobj_of_translation ' +o,804,'6 Conclusions and Future Directions In previous work , statistical NLP computation over large corpora has been a slow , of ine process , as in KNOWITALL and also in PMI-IR applications such as sentiment classi cation ',Turney,'num_Conclusions_6 tmod_been_Conclusions dep_Conclusions_and nn_Directions_Future dep_Conclusions_Directions prep_been_In amod_work_previous pobj_In_work amod_computation_statistical nn_computation_NLP nsubj_been_computation prep_computation_over amod_corpora_large pobj_over_corpora aux_been_has det_slow_a dobj_been_slow prep_slow_of nn_process_ine pobj_of_process dep_in_as prep_been_in nn__KNOWITALL pobj_in_ cc_in_and conj_in_also conj_in_in nn_applications_PMI-IR pobj_in_applications dep_as_such prep_applications_as nn_cation_sentiment nn_cation_classi pobj_as_cation ' +p,805,'Averaging has been shown to help reduce overfitting ',Collins,'nsubjpass_shown_Averaging aux_shown_has auxpass_shown_been aux_help_to xcomp_shown_help xcomp_help_reduce dobj_reduce_overfitting ' +o,806,'The first solution might also introduce errors elsewhere As already noted : ` While this automatic derivation process introduced a small percentage of errors on its own , it was the only practical way both to provide the amount of training data required and to allow for fully-automatic testing \' ',Ramshaw,'det_solution_The amod_solution_first nsubj_introduce_solution aux_introduce_might advmod_introduce_also dobj_introduce_errors advmod_As_elsewhere advmod_introduce_As nsubj_noted_ advmod_noted_already dep_introduce_noted mark_introduced_While det_process_this amod_process_automatic nn_process_derivation nsubj_introduced_process advcl_way_introduced det_percentage_a amod_percentage_small dobj_introduced_percentage prep_percentage_of pobj_of_errors prep_introduced_on poss_own_its pobj_on_own nsubj_way_it cop_way_was det_way_the amod_way_only amod_way_practical ccomp_noted_way preconj_provide_both aux_provide_to dep_way_provide det_amount_the dobj_provide_amount prep_amount_of nn_data_training pobj_of_data partmod_data_required cc_provide_and aux_allow_to conj_provide_allow prep_allow_for amod_testing_fully-automatic pobj_for_testing ' +p,807,'Hypergraphs have been successfully used in parsing and machine translation ',Huang,'nsubjpass_used_Hypergraphs aux_used_have auxpass_used_been advmod_used_successfully prep_used_in pcomp_in_parsing nn_translation_ cc__and conj__machine dobj_parsing_translation ' +o,808,'1 To train their system , R&M used a 200k-word chunk of the Penn Treebank Parsed Wall Street Journal tagged using a transformation-based tagger and extracted base noun phrases from its parses by selecting noun phrases that contained no nested noun phrases and further processing the data with some heuristics -LRB- like treating the possessive marker as the first word of a new base noun phrase -RRB- to flatten the recursive structure of the parse ',Ramshaw,'aux_train_To dep_used_train poss_system_their dobj_train_system nsubj_used_R&M dep_1_used det_chunk_a amod_chunk_200k-word dobj_used_chunk prep_used_of det__the nn__Penn nn__Treebank nn__Parsed nn__Wall nn__Street nn__Journal pobj_of_ dep__tagged xcomp_tagged_using det__a amod__transformation-based nn__tagger dobj_using_ cc_tagged_and conj_tagged_extracted amod_phrases_base nn_phrases_noun dobj_extracted_phrases prep_extracted_from poss_parses_its pobj_from_parses prep_tagged_by pcomp_by_selecting nn_phrases_noun dobj_selecting_phrases nsubj_contained_that rcmod_phrases_contained det_phrases_no amod_phrases_nested nn_phrases_noun dobj_contained_phrases cc_selecting_and advmod_selecting_further conj_selecting_processing det_data_the dobj_processing_data prep_processing_with det_heuristics_some pobj_with_heuristics dep_used_like pcomp_like_treating det_marker_the amod_marker_possessive dobj_treating_marker prep_treating_as det_word_the amod_word_first pobj_as_word prep_word_of det_phrase_a amod_phrase_new nn_phrase_base nn_phrase_noun pobj_of_phrase aux_flatten_to xcomp_used_flatten det_structure_the amod_structure_recursive dobj_flatten_structure prep_structure_of det_parse_the pobj_of_parse ' +o,809,'This can either be semi-supervised parsing , using both annotated and unannotated data or unsupervised parsing , training entirely on unannotated text ',McClosky,'nsubjpass_semi-supervised_This aux_semi-supervised_can advmod_semi-supervised_either auxpass_semi-supervised_be purpcl_semi-supervised_parsing dep_parsing_using preconj_annotated_both amod_data_annotated cc_annotated_and conj_annotated_unannotated nsubj_parsing_data amod_data_ cc__or conj__unsupervised ccomp_parsing_parsing dep_parsing_training advmod_training_entirely prep_training_on amod_text_unannotated pobj_on_text ' +o,810,'In , as well as other similar works , only left-toright search was employed ',Collins,'prep_employed_In pobj_In_ dep_well_as cc__well dep_well_as amod_works_other amod_works_similar conj__works partmod_works_ advmod_search_only amod_search_left-toright nsubjpass_employed_search auxpass_employed_was ' +o,811,'This iterative optimiser , derived from a word disambiguation technique , finds the nearest local maximum in the lexical cooccurrence network from each concept seed ',Yarowsky,'det_optimiser_This amod_optimiser_iterative nsubj_finds_optimiser partmod_optimiser_derived prep_derived_from det__a nn__word nn__disambiguation nn__technique pobj_from_ det_maximum_the amod_maximum_nearest amod_maximum_local dobj_finds_maximum prep_maximum_in det_network_the amod_network_lexical nn_network_cooccurrence pobj_in_network prep_network_from det_seed_each nn_seed_concept pobj_from_seed ' +p,812,'1 Introduction introduced minimum error rate training -LRB- MERT -RRB- for optimizing feature weights in statistical machine translation -LRB- SMT -RRB- models , and demonstrated that it produced higher translation quality scores than maximizing the conditional likelihood of a maximum entropy model using the same features ',Och,'num__1 nn__Introduction nsubj_introduced_ amod_training_minimum nn_training_error nn_training_rate dobj_introduced_training abbrev_training_MERT prep_training_for pcomp_for_optimizing nn_weights_feature dobj_optimizing_weights prep_optimizing_in amod_models_statistical nn_models_machine nn_models_translation nn_models_SMT pobj_in_models cc_introduced_and conj_introduced_demonstrated complm_produced_that nsubj_produced_it ccomp_demonstrated_produced amod_scores_higher nn_scores_translation nn_scores_quality dobj_produced_scores prep_scores_than pcomp_than_maximizing det_likelihood_the amod_likelihood_conditional dobj_maximizing_likelihood prep_likelihood_of det_model_a amod_model_maximum nn_model_entropy pobj_of_model partmod_model_using det_features_the amod_features_same dobj_using_features ' +n,813,' applied the parser of developed for English , to Czech , and found thatthe performance wassubstantially lower when compared to the results for English ',Collins,'nsubj_applied_ det_parser_the dobj_applied_parser prep_parser_of pobj_of_ partmod__developed prep_developed_for pobj_for_English prep_applied_to pobj_to_Czech cc_applied_and conj_applied_found amod_performance_thatthe nsubj_compared_performance advmod_lower_wassubstantially dep_compared_lower advmod_compared_when xcomp_found_compared prep_compared_to det_results_the pobj_to_results prep_results_for pobj_for_English ' +o,814,'This model is related to the averaged perceptron algorithm of ',Collins,'det_model_This nsubjpass_related_model auxpass_related_is prep_related_to det_algorithm_the amod_algorithm_averaged nn_algorithm_perceptron pobj_to_algorithm prep_algorithm_of ' +o,815,'a22 a14 is the sufficient statistic of a16 a14 Then , we can rewrite a2a24a3 a10a27 a42a7 a25 as : a5a7a6a9a8a11a10 a23 a3 a10 a7 a15 a27 a25a18a17a26a25 a12a28a27 a5a7a6a29a8a30a10 a23 a3 a10 a7 a15 a27 a25a18a17 3 Loss Functions for Label Sequences Given the theoretical advantages of discriminative models over generative models and the empirical support by , and that CRFs are the state-of-the-art among discriminative models for label sequences , we chose CRFs as our model , and trained by optimizing various objective functions a31 a3 a10a36 a25 with respect to the corpus a36 The application of these models to the label sequence problems vary widely ',Klein,'amod_a14_a22 nsubj_statistic_a14 cop_statistic_is det_statistic_the amod_statistic_sufficient prep_statistic_of amod_a14_a16 pobj_of_a14 advmod_statistic_Then nsubj_rewrite_we aux_rewrite_can ccomp_statistic_rewrite amod_a25_a2a24a3 amod_a25_a10a27 nn_a25_a42a7 dobj_rewrite_a25 prep_rewrite_as amod_Functions_a5a7a6a9a8a11a10 nn_Functions_a23 amod_Functions_a3 amod_Functions_a10 nn_Functions_a7 amod_Functions_a15 amod_Functions_a27 nn_Functions_a25a18a17a26a25 amod_Functions_a12a28a27 amod_Functions_a5a7a6a29a8a30a10 nn_Functions_a23 amod_Functions_a3 amod_Functions_a10 nn_Functions_a7 amod_Functions_a15 amod_Functions_a27 nn_Functions_a25a18a17 nn_Functions_3 nn_Functions_Loss dep_as_Functions dep_Functions_for nn_Sequences_Label pobj_for_Sequences partmod_Sequences_Given det_advantages_the amod_advantages_theoretical dobj_Given_advantages prep_advantages_of amod_models_discriminative pobj_of_models prep_Given_over amod_models_generative pobj_over_models cc_models_and det_support_the amod_support_empirical conj_models_support prep_Given_by pobj_by_ cc_statistic_and dep_state-of-the-art_that nsubj_state-of-the-art_CRFs cop_state-of-the-art_are det_state-of-the-art_the ccomp_chose_state-of-the-art prep_state-of-the-art_among amod_models_discriminative pobj_among_models prep_models_for nn_sequences_label pobj_for_sequences nsubj_chose_we conj_statistic_chose dobj_chose_CRFs prep_chose_as poss_model_our pobj_as_model cc_chose_and conj_chose_trained prep_trained_by pcomp_by_optimizing amod_functions_various nn_functions_objective nsubj_a25_functions amod_a25_a31 amod_a25_a3 amod_a25_a10a36 xcomp_optimizing_a25 prep_a25_with pobj_with_respect prep_optimizing_to det_a36_the nn_a36_corpus pobj_to_a36 det_application_The dobj_optimizing_application prep_application_of det_models_these pobj_of_models prep_models_to det_label_the pobj_to_label nn_problems_sequence nsubj_vary_problems rcmod_label_vary advmod_vary_widely ' +o,816,'Pustejovsky confronted with the problem of automatic acquisition more extensively in ',Dunning,'advmod_confronted_Pustejovsky prep_confronted_with det_problem_the pobj_with_problem prep_problem_of amod_acquisition_automatic pobj_of_acquisition advmod_extensively_more dep_confronted_extensively prep_extensively_in pobj_in_ ' +o,817,'PropBank encodes propositional information by adding a layer of argument structure annotation to the syntactic structures of the Penn Treebank ',Marcus,'nsubj_encodes_PropBank amod_information_propositional dobj_encodes_information prep_encodes_by pcomp_by_adding det_layer_a dobj_adding_layer prep_layer_of nn_annotation_argument nn_annotation_structure pobj_of_annotation prep_adding_to det_structures_the amod_structures_syntactic pobj_to_structures prep_structures_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,818,'However , while similarity measures -LRB- such as WordNet distance or Lins similarity metric -RRB- only detect cases of semantic similarity , association measures -LRB- such as the ones used by Poesio et al , or by Garera and Yarowsky -RRB- also find cases of associative bridg497 Lin98 RFF TheY TheY : G2 PL03 Land -LRB- country\\/state\\/land -RRB- Staat Staat Kemalismus Regierung Kontinent state state Kemalism government continent Stadt Stadt Bauernfamilie Prasident Region city city agricultural family president region Region Landesregierung Bankgesellschaft Dollar Stadt region country government banking corporation dollar city Bundesrepublik Bundesregierung Baht Albanien Staat federal republic federal government Baht Albania state Republik Gewerkschaft Gasag Hauptstadt Bundesland republic trade union -LRB- a gas company -RRB- capital state Medikament -LRB- medical drug -RRB- Arzneimittel Pille RU Patient Arzneimittel pharmaceutical pill -LRB- a drug -RRB- patient pharmaceutical Praparat Droge Abtreibungspille Arzt Lebensmittel preparation drug -LRB- non-medical -RRB- abortion pill doctor foodstuff Pille Praparat Viagra Pille Praparat pill preparation Viagra pill preparation Hormon Pestizid Pharmakonzern Behandlung Behandlung hormone pesticide pharmaceutical company treatment treatment Lebensmittel Lebensmittel Praparat Abtreibungspille Arznei foodstuff foodstuff preparation abortion pill drug highest ranked words , with very rare words removed : RU 486 , an abortifacient drug Lin98 : Lins distributional similarity measure RFF : Geffet and Dagans Relative Feature Focus measure TheY : association measure introduced by Garera and Yarowsky TheY : G2 : similar method using a log-likelihood-based statistic this statistic has a preference for higher-frequency terms PL03 : semantic space association measure proposed by Pado and Lapata Table 1 : Similarity and association measures : most similar items ing like 1a , b ; the result of this can be seen in table -LRB- 2 -RRB- : while the similarity measures -LRB- Lin98 , RFF -RRB- list substitutable terms -LRB- which behave like synonyms in many contexts -RRB- , the association measures -LRB- Garera and Yarowskys TheY measure , Pado and Lapatas association measure -RRB- also find non-compatible associations such as countrycapital or drugtreatment , which is why they are commonly called relationfree ',Dunning,'advmod_find_However mark_detect_while nn_measures_similarity nsubj_detect_measures dep_as_such dep_measures_as amod_distance_WordNet pobj_as_distance cc_distance_or nn_metric_Lins nn_metric_similarity conj_distance_metric advmod_detect_only dep_find_detect dobj_detect_cases prep_cases_of amod_similarity_semantic pobj_of_similarity nn_measures_association nsubj_find_measures dep_as_such dep_measures_as det_ones_the pobj_as_ones partmod_ones_used prep_used_by pobj_by_Poesio cc_Poesio_et conj_Poesio_al cc_by_or conj_by_by pobj_by_Garera cc_Garera_and conj_Garera_Yarowsky advmod_find_also dobj_find_cases prep_cases_of amod_TheY_associative amod_TheY_bridg497 nn_TheY_Lin98 nn_TheY_RFF nn_TheY_TheY pobj_of_TheY dep_list_G2 nn_Land_PL03 dobj_G2_Land dep_Land_country\\/state\\/land nn_Staat_Staat nn_Staat_Staat nn_Staat_Kemalismus nn_Staat_Regierung nn_Staat_Kontinent nn_Staat_state nn_Staat_state nn_Staat_Kemalism nn_Staat_government nn_Staat_continent nn_Staat_Stadt nn_Staat_Stadt nn_Staat_Bauernfamilie nn_Staat_Prasident nn_Staat_Region nn_Staat_city nn_Staat_city nn_Staat_agricultural nn_Staat_family nn_Staat_president nn_Staat_region nn_Staat_Region nn_Staat_Landesregierung nn_Staat_Bankgesellschaft nn_Staat_Dollar nn_Staat_Stadt nn_Staat_region nn_Staat_country nn_Staat_government nn_Staat_banking nn_Staat_corporation nn_Staat_dollar nn_Staat_city nn_Staat_Bundesrepublik nn_Staat_Bundesregierung nn_Staat_Baht nn_Staat_Albanien nsubj_G2_Staat amod_republic_federal dep_Bundesland_republic amod_government_federal dep_republic_government nn_Gewerkschaft_Baht nn_Gewerkschaft_Albania nn_Gewerkschaft_state nn_Gewerkschaft_Republik dep_Bundesland_Gewerkschaft nn_Bundesland_Gasag nn_Bundesland_Hauptstadt dep_seen_Bundesland nn_union_republic nn_union_trade dep_Bundesland_union det_company_a nn_company_gas appos_union_company nn_state_capital dep_union_state nn_patient_Medikament amod_drug_medical appos_patient_drug nn_patient_Arzneimittel nn_patient_Pille nn_patient_RU nn_patient_Patient nn_patient_Arzneimittel amod_patient_pharmaceutical nn_patient_pill det_drug_a appos_patient_drug nsubjpass_seen_patient amod_drug_pharmaceutical nn_drug_Praparat nn_drug_Droge nn_drug_Abtreibungspille nn_drug_Arzt nn_drug_Lebensmittel nn_drug_preparation dep_foodstuff_drug dep_drug_non-medical nn_Arznei_abortion nn_Arznei_pill nn_Arznei_doctor nn_Arznei_foodstuff nn_Arznei_Pille nn_Arznei_Praparat nn_Arznei_Viagra nn_Arznei_Pille nn_Arznei_Praparat nn_Arznei_pill nn_Arznei_preparation nn_Arznei_Viagra nn_Arznei_pill nn_Arznei_preparation nn_Arznei_Hormon nn_Arznei_Pestizid nn_Arznei_Pharmakonzern nn_Arznei_Behandlung nn_Arznei_Behandlung nn_Arznei_hormone nn_Arznei_pesticide amod_Arznei_pharmaceutical nn_Arznei_company nn_Arznei_treatment nn_Arznei_treatment nn_Arznei_Lebensmittel nn_Arznei_Lebensmittel nn_Arznei_Praparat nn_Arznei_Abtreibungspille dep_foodstuff_Arznei rcmod_patient_foodstuff nn_drug_foodstuff nn_drug_preparation nn_drug_abortion nn_drug_pill iobj_foodstuff_drug dobj_foodstuff_highest partmod_highest_ranked dobj_ranked_words prep_ranked_with advmod_rare_very amod_words_rare pobj_with_words partmod_words_removed dep_highest_RU num_RU_486 det_Lin98_an amod_Lin98_abortifacient nn_Lin98_drug conj_RU_Lin98 nn_RFF_Lins amod_RFF_distributional nn_RFF_similarity nn_RFF_measure nn_RFF_ dep_RU_RFF dep_RU_Geffet cc_RU_and nn_TheY_Dagans nn_TheY_Relative nn_TheY_Feature nn_TheY_Focus nn_TheY_measure nn_TheY_ conj_RU_TheY nn_measure_association dep_highest_measure partmod_measure_introduced prep_introduced_by pobj_by_Garera cc_Garera_and nn_TheY_Yarowsky nn_TheY_ conj_Garera_TheY dep_highest_G2 amod_method_similar dep_G2_method partmod_method_using det__a amod__log-likelihood-based nn__statistic dobj_using_ det_statistic_this nsubj_has_statistic rcmod__has det_preference_a dobj_has_preference prep_preference_for amod_PL03_higher-frequency nn_PL03_terms pobj_for_PL03 amod_measure_semantic nn_measure_space nn_measure_association dep_highest_measure partmod_measure_proposed prep_proposed_by pobj_by_Pado cc_highest_and nn_Table_Lapata nn_Table_ conj_highest_Table dep_Table_1 dep_Table_Similarity cc_Similarity_and nn_measures_association conj_Similarity_measures advmod_similar_most amod_items_similar nn_ing_items dep_Table_ing prep_ing_like pobj_like_1a appos_1a_b det_result_the dep_Table_result prep_result_of pobj_of_this aux_seen_can auxpass_seen_be dep_G2_seen prep_seen_in pobj_in_table appos_table_2 mark_list_while det_measures_the nn_measures_similarity nsubj_list_measures appos_measures_Lin98 dep_Lin98_RFF parataxis_find_list amod_terms_substitutable dobj_list_terms nsubj_behave_which parataxis_list_behave prep_behave_like pobj_like_synonyms prep_synonyms_in amod_contexts_many pobj_in_contexts det_measures_the nn_measures_association nsubj_find_measures nn_measure_Garera cc_Garera_and conj_Garera_Yarowskys nn_measure_TheY dep_find_measure conj_measure_Pado cc_measure_and nn_measure_Lapatas nn_measure_association conj_measure_measure advmod_find_also dep_list_find amod_associations_non-compatible dobj_find_associations dep_as_such prep_associations_as pobj_as_countrycapital cc_countrycapital_or conj_countrycapital_drugtreatment nsubj_is_which rcmod_countrycapital_is advmod_called_why nsubjpass_called_they auxpass_called_are advmod_called_commonly advcl_is_called dep_called_relationfree ' +o,819,'Jiao et al propose semi-supervised conditional random fields that try to maximize the conditional log-likelihood on the training data and simultaneously minimize the conditional entropy of the class labels on the unlabeled data ',Jiao,'nsubj_propose_Jiao cc_Jiao_et conj_Jiao_al amod_fields_semi-supervised amod_fields_conditional amod_fields_random nsubj__fields ccomp_propose_ complm_try_that ccomp__try aux_maximize_to xcomp_try_maximize det_log-likelihood_the amod_log-likelihood_conditional dobj_maximize_log-likelihood prep_maximize_on det_data_the nn_data_training pobj_on_data cc_maximize_and advmod_maximize_simultaneously conj_maximize_minimize det_entropy_the amod_entropy_conditional dobj_minimize_entropy prep_entropy_of det_labels_the nn_labels_class pobj_of_labels prep_minimize_on det_data_the amod_data_unlabeled pobj_on_data ' +o,820,' report extracting database records by learning record field compatibility ',Wick,'nsubj_report_ xcomp_report_extracting nn_records_database dobj_extracting_records prep_extracting_by pcomp_by_learning amod_compatibility_record nn_compatibility_field dobj_learning_compatibility ' +o,821,'Unfortunately , a counterexample illustrated in shows that the max function does not produce valid kernels in general ',Pedersen,'advmod_illustrated_Unfortunately det_counterexample_a nsubj_illustrated_counterexample prep_illustrated_in amod_shows_ pobj_in_shows complm_produce_that det_function_the nn_function_max nsubj_produce_function aux_produce_does neg_produce_not ccomp_illustrated_produce amod_kernels_valid dobj_produce_kernels prep_produce_in pobj_in_general ' +p,822,'1 Introduction When data have distinct sub-structures , models exploiting latent variables are advantageous in learning ',Matsuzaki,'num_Introduction_1 nsubj_advantageous_Introduction advmod_have_When nsubj_have_data dep_Introduction_have amod_sub-structures_distinct dobj_have_sub-structures nsubj_exploiting_models dep_have_exploiting amod_variables_latent dobj_exploiting_variables cop_advantageous_are prep_advantageous_in pcomp_in_learning ' +o,823,'2 Detecting Discourse-New Definite Descriptions 21 Vieira and Poesio Poesio and Vieira carried out corpus studies indicating that in corpora like the Wall Street Journal portion of the Penn Treebank , around 52 \% of DDs are discourse-new , and another 15 \% or so are bridging references , for a total of about 66-67 \% firstmention ',Marcus,'num_Vieira_2 nn_Vieira_Detecting nn_Vieira_Discourse-New nn_Vieira_Definite nn_Vieira_Descriptions num_Vieira_21 nsubj_carried_Vieira cc_Vieira_and nn_Poesio_Poesio conj_Vieira_Poesio cc_Vieira_and nn__Vieira conj_Vieira_ prt_carried_out nn_studies_corpus dobj_carried_studies xcomp_carried_indicating dobj_indicating_that prep_that_in pobj_in_corpora prep_indicating_like det_portion_the nn_portion_Wall nn_portion_Street nn_portion_Journal pobj_like_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ prep_carried_around num_\%_52 pobj_around_\% prep_\%_of nn_discourse-new_DDs dep_discourse-new_are pobj_of_discourse-new num_discourse-new_ cc_carried_and dep_\%_another number_\%_15 nsubj_bridging_\% cc_\%_or conj_\%_so aux_bridging_are conj_carried_bridging dobj_bridging_references prep_bridging_for det_total_a pobj_for_total prep_total_of quantmod_66-67_about num_\%_66-67 pobj_of_\% partmod_\%_firstmention ' +o,824,'The distinction between lexical and relational similarity for word pair comparison is recognized by -LRB- hecallstheformer attributional similarity -RRB- , though the methods he presents focus on relational similarity ',Turney,'det_distinction_The nsubjpass_recognized_distinction prep_distinction_between amod_similarity_lexical cc_lexical_and conj_lexical_relational pobj_between_similarity prep_similarity_for nn_comparison_word nn_comparison_pair pobj_for_comparison auxpass_recognized_is prep_recognized_by pobj_by_ amod_similarity_hecallstheformer amod_similarity_attributional appos__similarity advmod__though det_methods_the dep__methods nsubj_presents_he dep_recognized_presents dobj_presents_focus prep_focus_on amod_similarity_relational pobj_on_similarity ' +o,825,'The POS disambiguation has usually been performed by statistical approaches mainly using hidden markov model -LRB- HMM -RRB- -LRB- et al , 1992 ; Kupiec ',Cutting,'det_disambiguation_The dep_disambiguation_POS nsubjpass_performed_disambiguation aux_performed_has advmod_performed_usually auxpass_performed_been prep_performed_by amod_approaches_statistical pobj_by_approaches advmod_using_mainly partmod_approaches_using amod_model_hidden amod_model_markov dobj_using_model abbrev_model_HMM dep_model_ cc__et conj__al dep__1992 dep_approaches_Kupiec ' +o,826,'As a baseline model we used a maximum entropy tagger , very similar to the one described in ',Ratnaparkhi,'prep_used_As det_model_a nn_model_baseline pobj_As_model nsubj_used_we det_tagger_a amod_tagger_maximum nn_tagger_entropy dobj_used_tagger advmod_similar_very dep_used_similar prep_similar_to det_one_the pobj_to_one partmod_one_described prep_described_in ' +o,827,'We assign tags of part-of-speech -LRB- POS -RRB- to the words with MXPOST that adopts the Penn Treebank tag set ',Ratnaparkhi,'nsubj_assign_We dobj_assign_tags prep_tags_of pobj_of_part-of-speech appos_part-of-speech_POS prep_assign_to det_words_the pobj_to_words prep_words_with pobj_with_MXPOST nsubj_adopts_that rcmod_words_adopts det_tag_the nn_tag_Penn nn_tag_Treebank nsubj_set_tag ccomp_adopts_set ' +o,828,'Such coarse-grained inventories can be produced manually from scratch or by automatically relating or clustering existing word senses ',Navigli,'amod_inventories_Such amod_inventories_coarse-grained nsubjpass_produced_inventories aux_produced_can auxpass_produced_be advmod_produced_manually prep_produced_from nn__scratch pobj_from_ cc_from_or conj_from_by advmod_relating_automatically pcomp_by_relating dobj_relating_ cc__or amod_senses_clustering nn_senses_ amod_senses_existing nn_senses_word conj__senses ' +o,829,'Given a contextual word cw that occurs in the paragraphs of bc , a log-likelihood ratio -LRB- G2 -RRB- test is employed , which checks if the distribution of cw in bc is similar to the distribution of cw in rc ; p -LRB- cw bc -RRB- = p -LRB- cw rc -RRB- -LRB- null hypothesis -RRB- ',Dunning,'prep_employed_Given det_cw_a amod_cw_contextual nn_cw_word dep_Given_cw nsubj_occurs_that rcmod_cw_occurs prep_occurs_in det_paragraphs_the pobj_in_paragraphs prep_paragraphs_of pobj_of_bc det_ratio_a amod_ratio_log-likelihood appos_bc_ratio appos_bc_G2 dep_bc_test auxpass_employed_is nsubjpass_employed_ nsubj_checks_which rcmod__checks mark_similar_if det_distribution_the nsubj_similar_distribution prep_distribution_of pobj_of_cw prep_cw_in pobj_in_bc cop_similar_is advcl_checks_similar prep_similar_to det_distribution_the pobj_to_distribution prep_distribution_of pobj_of_cw prep_cw_in pobj_in_rc dep_cw_p nn_bc_cw nsubj_p_bc dep_p_= ccomp_p_p nn_rc_cw appos_cw_rc amod_hypothesis_null appos_cw_hypothesis ' +o,830,'In this paper we use the so-called Model 4 from ',Brown,'prep_use_In det_paper_this pobj_In_paper nsubj_use_we det_4_the amod_4_so-called number_4_Model dobj_use_4 prep_use_from ' +o,831,'We would expect the opposite effect with hand-aligned data ',Galley,'nsubj_expect_We aux_expect_would det_effect_the amod_effect_opposite dobj_expect_effect prep_expect_with amod_data_hand-aligned pobj_with_data ' +o,832,'Extensions to Hiero Several authors describe extensions to Hiero , to incorporate additional syntactic information , or to combine it with discriminative latent models ',Zhang,'nsubj_describe_Extensions prep_Extensions_to nn_authors_Hiero amod_authors_Several pobj_to_authors dobj_describe_extensions prep_describe_to pobj_to_Hiero aux_incorporate_to dep_describe_incorporate amod__additional amod__syntactic nn__information dobj_incorporate_ cc_incorporate_or aux_combine_to conj_incorporate_combine dobj_combine_it prep_combine_with amod__discriminative amod__latent nn__models pobj_with_ ' +o,833,'The other form of hybridization ? ? a statistical MT model that is based on a deeper analysis of the syntactic 33 structure of a sentence ? ? has also long been identified as a desirable objective in principle -LRB- consider -RRB- ',Wu,'det_form_The amod_form_other dep_?_form prep_form_of pobj_of_hybridization det_MT_a amod_MT_statistical nsubjpass_identified_model nsubjpass_based_that auxpass_based_is dep_model_based prep_based_on det_analysis_a amod_analysis_deeper pobj_on_analysis prep_analysis_of det_structure_the amod_structure_syntactic tmod_syntactic_33 pobj_of_structure prep_structure_of det_sentence_a pobj_of_sentence aux_identified_has advmod_identified_also advmod_identified_long auxpass_identified_been rcmod_MT_identified prep_identified_as det_objective_a amod_objective_desirable pobj_as_objective prep_MT_in pobj_in_principle dep_principle_consider acomp_consider_ ' +o,834,'272 Similarity-based estimation was first used for language modeling in the cooccurrence smoothing method of Essen and Steinbiss , derived from work on acoustic model smoothing by Sugawara et al ',Brown,'amod_estimation_272 amod_estimation_Similarity-based nsubjpass_used_estimation auxpass_used_was advmod_used_first prep_used_for nn_modeling_language pobj_for_modeling prep_used_in det_method_the amod_method_cooccurrence amod_method_smoothing pobj_in_method prep_method_of nn__Essen cc_Essen_and conj_Essen_Steinbiss pobj_of_ partmod_method_derived prep_derived_from pobj_from_work prep_derived_on amod_model_acoustic pobj_on_model partmod_model_smoothing prep_smoothing_by pobj_by_Sugawara cc_Sugawara_et conj_Sugawara_al ' +o,835,'Following the setup in , we initialize the transition and emission distributions to be uniform with a small amount of noise , and run EM and VB for 1000 iterations ',Johnson,'prep_initialize_Following det_setup_the pobj_Following_setup prep_setup_in pobj_in_ nsubj_initialize_we det_distributions_the nn_distributions_transition cc_transition_and conj_transition_emission nsubj_uniform_distributions aux_uniform_to cop_uniform_be xcomp_initialize_uniform prep_uniform_with det_amount_a amod_amount_small pobj_with_amount prep_amount_of pobj_of_noise cc_uniform_and conj_uniform_run dobj_run_EM cc_EM_and conj_EM_VB prep_run_for num_iterations_1000 pobj_for_iterations ' +o,836,'Our method uses assumptions similar to et al 1996 but is naturally suitable for distributed parallel computations ',Berger,'poss_method_Our nsubj_uses_method dobj_uses_assumptions amod_assumptions_similar prep_similar_to pobj_to_ cc__et conj__al tmod_similar_1996 cc_uses_but cop_suitable_is advmod_suitable_naturally conj_uses_suitable prep_suitable_for amod_computations_distributed amod_computations_parallel pobj_for_computations ' +o,837,'The agreement on identifying the boundaries of units , using the statistic discussed in , was = 9 -LRB- for two annotators and 500 units -RRB- ; the agreement on features -LRB- 2 annotators and at least 200 units -RRB- was as follows : UTYPE : = 76 ; VERBED : = 9 ; FINITE : = 81 ',Carletta,'det_agreement_The prep_agreement_on pcomp_on_identifying det_boundaries_the dobj_identifying_boundaries prep_boundaries_of pobj_of_units dep_identifying_using det_statistic_the dobj_using_statistic dep_agreement_discussed prep_discussed_in pobj_in_ aux_=_was dep__= dobj_=_9 dep_discussed_for num_annotators_two pobj_for_annotators cc_annotators_and num_units_500 conj_annotators_units det_agreement_the nsubj_was_agreement prep_agreement_on pobj_on_features dep_features_2 dep_2_annotators cc_annotators_and quantmod_200_at dep_at_least num_units_200 conj_annotators_units parataxis_discussed_was mark_follows_as advcl_was_follows parataxis_discussed_UTYPE dep_UTYPE_= dobj_=_76 parataxis_discussed_VERBED parataxis_discussed_= dobj_=_9 parataxis_discussed_FINITE dep_FINITE_= dobj_=_81 ' +n,838,'Hanks and proposed using pointwise mutual information to identify collocations in lexicography ; however , the method may result in unacceptable collocations for low-count pairs ',Church,'nsubj_proposed_Hanks cc_Hanks_and conj_Hanks_ xcomp_proposed_using amod_information_pointwise amod_information_mutual dobj_using_information aux_identify_to xcomp_using_identify dobj_identify_collocations prep_identify_in pobj_in_lexicography advmod_result_however det_method_the nsubj_result_method aux_result_may parataxis_proposed_result prep_result_in amod_collocations_unacceptable pobj_in_collocations prep_collocations_for amod_pairs_low-count pobj_for_pairs ' +o,839,'In comparison we introduce 28 several metrics coefficients reported in Albrecht and Hwa including smoothed BLEU , METEOR , HWCM , and the metric proposed in Albrecht and Hwa using the full feature set ',Liu,'prep_reported_In pobj_In_comparison nsubj_introduce_we rcmod_comparison_introduce num_coefficients_28 amod_coefficients_several nn_coefficients_metrics dobj_introduce_coefficients prep_reported_in pobj_in_Albrecht cc_Albrecht_and conj_Albrecht_Hwa nsubj_reported_ prep__including nn__smoothed nn__BLEU pobj_including_ nn__METEOR conj__ nn__HWCM conj__ cc__and det_metric_the conj__metric amod_metric_proposed prep_metric_in pobj_in_Albrecht cc__and nn__Hwa conj__ partmod__using det_set_the amod_set_full nn_set_feature dobj_using_set ' +o,840,'This is one manifestation of what is commonly referred to as the data sparseness problem , and was discussed by as a side-effect of specificity ',Rapp,'nsubj_manifestation_This cop_manifestation_is num_manifestation_one prep_manifestation_of nsubjpass_referred_what auxpass_referred_is advmod_referred_commonly pcomp_of_referred prep_referred_to mark_sparseness_as det_data_the nsubjpass_sparseness_data pcomp_to_sparseness dobj_sparseness_problem cc_sparseness_and auxpass_discussed_was conj_sparseness_discussed prep_discussed_by pobj_by_ prep_discussed_as det_side-effect_a pobj_as_side-effect prep_side-effect_of pobj_of_specificity ' +o,841,'Techniques for weakening the independence assumptions made by the IBM models 1 and 2 have been proposed in recent work ',Berger,'nsubj_made_Techniques prep_Techniques_for pcomp_for_weakening det_assumptions_the nn_assumptions_independence dobj_weakening_assumptions prep_made_by det_models_the nn_models_IBM pobj_by_models nsubjpass_proposed_1 cc_1_and conj_1_2 aux_proposed_have auxpass_proposed_been dep_made_proposed prep_proposed_in amod_work_recent pobj_in_work ' +o,842,'C3BTC5 and CCCDCA were used in and , respectively ',Turney,'nsubjpass_used_C3BTC5 cc_C3BTC5_and conj_C3BTC5_CCCDCA auxpass_used_were prep_used_in pobj_in_ cc__and conj__ advmod_used_respectively ' +o,843,'In this work we will use structured linear classifiers ',Collins,'prep_use_In det_work_this pobj_In_work nsubj_use_we aux_use_will amod_classifiers_structured amod_classifiers_linear dobj_use_classifiers ' +o,844,'This is the best automatically learned part-of-speech tagging result known to us , representing an error reduction of 44 \% on the model presented in , using the same data splits , and a larger error reduction of 121 \% from the more similar best previous loglinear model in Toutanova and Manning ',Collins,'nsubj_best_This cop_best_is det_best_the advmod_learned_automatically partmod_best_learned amod_result_part-of-speech amod_result_tagging dobj_learned_result partmod_result_known prep_known_to pobj_to_us csubj_splits_representing det_reduction_an nn_reduction_error dobj_representing_reduction prep_reduction_of num_\%_44 pobj_of_\% prep_representing_on det_model_the pobj_on_model partmod_model_presented prep_presented_in pobj_in_ dep_representing_using det_data_the amod_data_same dobj_using_data ccomp_best_splits cc_best_and det_reduction_a amod_reduction_larger nn_reduction_error nsubj__reduction prep_reduction_of num_\%_121 pobj_of_\% prep_\%_from det_model_the amod_model_more amod_model_similar dep_similar_best amod_model_previous nn_model_loglinear pobj_from_model prep_\%_in pobj_in_Toutanova cc_Toutanova_and conj_Toutanova_Manning conj_best_ ' +o,845,'The template we use here is similar to , but we have added extra context words before the X and after the Y Our morphological processing also differs from ',Turney,'det_template_The nsubj_use_we dep_template_use nsubj_similar_here cop_similar_is ccomp_use_similar aux__to xcomp_similar_ cc_use_but nsubj_added_we aux_added_have conj_use_added amod_words_extra nn_words_context dobj_added_words prep_added_before det_X_the pobj_before_X cc_before_and conj_before_after det_Y_the pobj_after_Y poss_processing_Our amod_processing_morphological nsubj_differs_processing advmod_differs_also dep_template_differs prep_differs_from ' +o,846,'The algorithm employs the OpenNLP MaxEnt implementation of the maximum entropy classification algorithm to develop word sense recognition signatures for each lemma which predicts the most likely sense for the lemma according to the context in which the lemma occurs ',Berger,'det_algorithm_The nsubj_employs_algorithm det_implementation_the nn_implementation_OpenNLP nn_implementation_MaxEnt dobj_employs_implementation prep_implementation_of det_algorithm_the amod_algorithm_maximum amod_algorithm_entropy nn_algorithm_classification pobj_of_algorithm partmod_algorithm_ aux_develop_to xcomp__develop nn_signatures_word nn_signatures_sense nn_signatures_recognition dobj_develop_signatures prep_signatures_for det_lemma_each pobj_for_lemma nsubj_predicts_which rcmod_lemma_predicts det_sense_the advmod_likely_most amod_sense_likely dobj_predicts_sense prep_sense_for det_lemma_the pobj_for_lemma prep_predicts_according dep_according_to det_context_the pobj_to_context rel_occurs_in pobj_in_which det_lemma_the nsubj_occurs_lemma rcmod_context_occurs ' +o,847,'Alternatively , order is modelled in terms of movement of automatically induced hierarchical structure of sentences ',Wu,'advmod_modelled_Alternatively nsubjpass_modelled_order auxpass_modelled_is prep_modelled_in pobj_in_terms prep_terms_of pobj_of_movement prep_movement_of advmod_induced_automatically amod_structure_induced amod_structure_hierarchical pobj_of_structure prep_structure_of pobj_of_sentences ' +o,848,'Parameters used to calculate P -LRB- D -RRB- are trained using MER training on development data ',Och,'nsubjpass_trained_Parameters partmod_Parameters_used aux_calculate_to xcomp_used_calculate dobj_calculate_P appos_Parameters_D auxpass_trained_are xcomp_trained_using nn__MER nn__training dobj_using_ prep_using_on nn_data_development pobj_on_data ' +o,849,'Chiang distinguishes statistical MT approaches that are syntactic in a formal sense , going beyond the nite-state underpinnings of phrasebased models , from approaches that are syntactic in a linguistic sense , ie taking advantage of a priori language knowledge in the form of annotations derived from human linguistic analysis or treebanking1 The two forms of syntactic modeling are doubly dissociable : current research frameworks include systems that are nite state but informed by linguistic annotation prior to training -LRB- eg , -RRB- , and also include systems employing contextfree models trained on parallel text without bene t of any prior linguistic analysis -LRB- eg ',Koehn,'nn__Chiang nsubj_distinguishes_ amod_approaches_statistical nn_approaches_MT dobj_distinguishes_approaches nsubj_going_that cop_going_are acomp_going_syntactic prep_going_in det_sense_a amod_sense_formal pobj_in_sense rcmod_approaches_going prep_going_beyond det_underpinnings_the amod_underpinnings_nite-state pobj_beyond_underpinnings prep_underpinnings_of amod_models_phrasebased pobj_of_models prep_going_from pobj_from_approaches nsubj_syntactic_that cop_syntactic_are rcmod_approaches_syntactic prep_syntactic_in det_sense_a amod_sense_linguistic pobj_in_sense appos_sense_ie partmod_ie_taking dobj_taking_advantage prep_advantage_of det_knowledge_a amod_knowledge_priori nn_knowledge_language pobj_of_knowledge prep_taking_in det_form_the pobj_in_form prep_form_of pobj_of_annotations partmod_annotations_derived prep_derived_from amod_analysis_human amod_analysis_linguistic pobj_from_analysis cc_analysis_or conj_analysis_treebanking1 det_forms_The num_forms_two nsubj_dissociable_forms prep_forms_of amod_modeling_syntactic pobj_of_modeling cop_dissociable_are advmod_dissociable_doubly dep_distinguishes_dissociable amod_frameworks_current nn_frameworks_research nsubj_include_frameworks ccomp_dissociable_include dobj_include_systems nsubj_state_that cop_state_are amod_state_nite rcmod_systems_state cc_state_but conj_state_informed prep_informed_by amod_annotation_linguistic pobj_by_annotation dep_to_prior prep_informed_to pobj_to_training appos_training_eg dep_eg_ cc_include_and advmod_include_also conj_include_include dobj_include_systems partmod_systems_employing amod_models_contextfree dobj_employing_models partmod_models_trained prep_trained_on amod_text_parallel pobj_on_text prep_trained_without amod_t_bene pobj_without_t prep_t_of det_linguistic_any amod_linguistic_prior pobj_of_linguistic nn_eg_analysis dep_linguistic_eg ' +o,850,'This can be done in a supervised , a semi-supervised or a fully unsupervised way ',Yarowsky,'nsubjpass_done_This aux_done_can auxpass_done_be prep_done_in det__a amod__supervised pobj_in_ det__a amod__semi-supervised conj__ cc__or det__a advmod_unsupervised_fully amod__unsupervised nn__way conj__ ' +o,851,'Equation -LRB- 3 -RRB- reads If the target noun appears , then it is distinguished by the majority The log-likelihood ratio decides in which order rules are applied to the target noun in novel context ',Yarowsky,'nsubj_reads_Equation appos_Equation_3 mark_appears_If det_noun_the nn_noun_target nsubj_appears_noun advcl_distinguished_appears advmod_distinguished_then nsubj_distinguished_it cop_distinguished_is ccomp_reads_distinguished prep_distinguished_by det_majority_the pobj_by_majority det__The amod__log-likelihood nn__ratio nsubj_decides_ dep_reads_decides dep_applied_in pobj_in_which nn_rules_order nsubjpass_applied_rules auxpass_applied_are ccomp_decides_applied prep_applied_to det_noun_the nn_noun_target pobj_to_noun prep_noun_in amod_context_novel pobj_in_context ' +p,852,' and shows how some of the methods which have been used in the past -LRB- particularly mutual information scores -RRB- are invalid for rare events , and introduce accurate measures of how ` surprising \' rare events are ',Dunning,'amod_shows_ cc__and conj__ advmod_invalid_how nsubj_invalid_some prep_some_of det_methods_the pobj_of_methods nsubjpass_used_which aux_used_have auxpass_used_been rcmod_methods_used prep_used_in det_past_the pobj_in_past dep_scores_particularly amod_scores_mutual nn_scores_information appos_past_scores cop_invalid_are dep_shows_invalid prep_invalid_for amod_events_rare pobj_for_events cc_invalid_and dep_measures_introduce acomp_measures_accurate conj_invalid_measures prep_shows_of advmod_are_how amod_events_surprising dep_surprising_rare nsubj_are_events pcomp_of_are ' +n,853,'Unlike , Smadja -LRB- 1993 -RRB- goes beyond the ` two-word \' limitation and deals with ` collocations of arbitrary length \' ',Church,'prep_goes_Unlike pobj_Unlike_ nsubj_goes_Smadja appos_Smadja_1993 prep_goes_beyond det_limitation_the amod_limitation_two-word pobj_beyond_limitation cc_limitation_and conj_limitation_deals prep_goes_with pobj_with_collocations prep_collocations_of amod_length_arbitrary pobj_of_length ' +o,854,'Statistic-based algorithms based on Belief Network such as Hidden-MarkovModel -LRB- HMM -RRB- , Lexicalized HMM and Maximal-Entropy model use the statistical information of a manually tagged corpus as background knowledge to tag new sentences ',Ratnaparkhi,'amod_algorithms_Statistic-based nsubj_use_algorithms prep_algorithms_based dep_based_on nn__Belief nn__Network pobj_on_ dep_as_such prep__as amod__Hidden-MarkovModel abbrev__HMM amod__ nn__ dep__Lexicalized nn__HMM pobj_as_ cc__and amod__Maximal-Entropy nn__model conj__ det_information_the amod_information_statistical dobj_use_information prep_information_of det_manually_a pobj_of_manually partmod_manually_tagged dobj_tagged_corpus prep_tagged_as nn_knowledge_background pobj_as_knowledge aux_tag_to xcomp_tagged_tag amod_sentences_new dobj_tag_sentences ' +o,855,'One way of resolving query ambiguities is to use the statistics , such as mutual information , to measure associations of query terms , on the basis of existing corpora ',Church,'num_way_One nsubj_is_way prep_way_of pcomp_of_resolving amod_ambiguities_query dobj_resolving_ambiguities aux_use_to xcomp_is_use det_statistics_the dobj_use_statistics dep_as_such prep_statistics_as amod__mutual nn__information pobj_as_ aux_measure_to xcomp_use_measure dobj_measure_associations prep_associations_of amod_terms_query pobj_of_terms prep_use_on det_basis_the pobj_on_basis prep_basis_of amod__existing nn__corpora pobj_of_ ' +p,856,'Nowadays , most of the state-of-the-art SMT systems are based on bilingual phrases ',Koehn,'nsubjpass_based_Nowadays nsubjpass_based_most prep_most_of det_systems_the amod_systems_state-of-the-art nn_systems_SMT pobj_of_systems auxpass_based_are prep_based_on amod_phrases_bilingual pobj_on_phrases ' +o,857,'The wn : : similarity package to compute the Jiang & Conrath -LRB- J&C -RRB- distance as in ',Pedersen,'nsubj_wn_The nn__similarity nn__package nsubj_compute_ aux_compute_to parataxis_wn_compute det_Jiang_the dobj_compute_Jiang cc_Jiang_& nn__Conrath appos__J&C nn__distance conj_Jiang_ dep_in_as prep_Jiang_in pobj_in_ ' +o,858,' propose using a statistical word alignment algorithm as a more robust way of aligning -LRB- monolingual -RRB- outputs into a confusion network for system com2 construct lattices over paraphrases using an iterative pairwise multiple sequence alignment -LRB- MSA -RRB- algorithm ',Barzilay,'num_lattices_ amod_lattices_propose dep_propose_using det_algorithm_a amod_algorithm_statistical nn_algorithm_word nn_algorithm_alignment dobj_using_algorithm prep_using_as det_way_a advmod_robust_more amod_way_robust pobj_as_way prep_way_of pcomp_of_aligning appos_outputs_monolingual dobj_aligning_outputs prep_aligning_into det_network_a nn_network_confusion pobj_into_network prep_network_for pobj_for_system amod_lattices_com2 amod_lattices_ nn_lattices_construct prep_lattices_over pobj_over_paraphrases partmod_paraphrases_using det_algorithm_an amod_algorithm_iterative dep_iterative_pairwise amod_algorithm_multiple nn_algorithm_sequence nn_algorithm_alignment abbrev_algorithm_MSA dobj_using_algorithm ' +o,859,'One of the first large scale hand tagging efforts is reported in , where a subset of the Brown corpus was tagged with WordNet July 2002 , pp ',Marcus,'nsubjpass_reported_One prep_One_of det_hand_the amod_hand_first amod_hand_large nn_hand_scale pobj_of_hand partmod_hand_tagging dobj_tagging_efforts auxpass_reported_is prep_reported_in pobj_in_ advmod_tagged_where det_subset_a nsubjpass_tagged_subset prep_subset_of det_corpus_the nn_corpus_Brown pobj_of_corpus auxpass_tagged_was rcmod__tagged prep_tagged_with pobj_with_WordNet tmod_tagged_July num_July_2002 num_July_pp ' +o,860,'42 Experiments To build all alignment systems , we start with 5 iterations of Model 1 followed by 4 iterations of HMM , as implemented in GIZA + + ',Och,'num_Experiments_42 aux_build_To dep_start_build det_systems_all amod_systems_alignment dobj_build_systems nsubj_start_we dep_Experiments_start prep_start_with num_iterations_5 pobj_with_iterations prep_iterations_of pobj_of_Model num_Model_1 partmod_iterations_followed prep_followed_by num_iterations_4 pobj_by_iterations prep_iterations_of nn__HMM pobj_of_ mark_implemented_as dep__implemented prep_implemented_in nn_+_GIZA nn_+_+ pobj_in_+ ' +o,861,'The usual recall and precision metrics -LRB- eg , how many of the interesting bits of information were detected , and how many of the found bits were actually correct -RRB- require either a test corpus previously annotated with the required information , or manual evaluation ',Fleischman,'det_recall_The amod_recall_usual nsubj_require_recall cc_recall_and nn_metrics_precision conj_recall_metrics dep_recall_eg advmod_many_how dep_detected_many dep_many_of det_bits_the amod_bits_interesting pobj_of_bits prep_bits_of pobj_of_information auxpass_detected_were dep_eg_detected cc_detected_and advmod_many_how dep_found_many dep_the_of nsubj_found_the conj_detected_found nsubj_correct_bits cop_correct_were advmod_correct_actually ccomp_found_correct advmod_corpus_either det_corpus_a nn_corpus_test dobj_require_corpus advmod_annotated_previously partmod_corpus_annotated prep_annotated_with det_information_the amod_information_required pobj_with_information cc_corpus_or amod_evaluation_manual conj_corpus_evaluation ' +p,862,'In this paper , we show that a noisy channel model instantiated within the paradigm of Statistical Machine Translation -LRB- SMT -RRB- can successfully provide editorial assistance for non-native writers ',Brown,'prep_show_In det_paper_this pobj_In_paper nsubj_show_we complm_provide_that det_model_a amod_model_noisy nn_model_channel nsubj_provide_model partmod_model_instantiated prep_instantiated_within det_paradigm_the pobj_within_paradigm prep_paradigm_of nn__Statistical nn__Machine nn__Translation abbrev__SMT pobj_of_ aux_provide_can advmod_provide_successfully ccomp_show_provide amod_assistance_editorial dobj_provide_assistance prep_provide_for amod_writers_non-native pobj_for_writers ' +o,863,'The CRF tagger was implemented in MALLET using the original feature templates from ',Ratnaparkhi,'det_tagger_The nn_tagger_CRF nsubjpass_implemented_tagger auxpass_implemented_was prep_implemented_in nn__MALLET nsubj_using_ pcomp_in_using det_templates_the amod_templates_original nn_templates_feature dobj_using_templates prep_using_from ' +p,864,'However , formally syntax-based methods propose simple but efficient ways to parse and translate sentences ',Wu,'advmod_propose_However amod_methods_formally amod_methods_syntax-based nsubj_propose_methods amod_ways_simple cc_simple_but conj_simple_efficient dobj_propose_ways aux_parse_to xcomp_propose_parse cc_parse_and conj_parse_translate dobj_parse_sentences ' +o,865,'Given a set of terms with unknown sentiment orientation , then uses the PMI-IR algorithm to issue queries to the web and determine , for each of these terms , its pointwise mutual information -LRB- PMI -RRB- with the two seed words across a large set of documents ',Turney,'prep_uses_Given det_set_a dep_Given_set prep_set_of pobj_of_terms prep_terms_with amod_orientation_unknown nn_orientation_sentiment pobj_with_orientation nsubj_uses_ advmod_uses_then det__the nn__PMI-IR nn__algorithm dobj_uses_ aux_issue_to xcomp_uses_issue dobj_issue_queries prep_issue_to det_web_the pobj_to_web cc_issue_and conj_issue_determine prep_determine_for pobj_for_each prep_each_of det_terms_these pobj_of_terms poss_information_its amod_information_pointwise amod_information_mutual appos_terms_information abbrev_information_PMI prep_determine_with det_words_the num_words_two nn_words_seed pobj_with_words prep_words_across det_set_a amod_set_large pobj_across_set prep_set_of pobj_of_documents ' +o,866,'This is similartothegraphconstructionmethodof and Rao et al ',Hughes,'nsubj__This cop__is nn__similartothegraphconstructionmethodof cc__and conj__Rao cc_Rao_et conj_Rao_al ' +p,867,'441 N-gram Co-Occurrence Statistics for Answer Extraction N-gram co-occurrence statistics have been successfully used in automatic evaluation , and more recently as training criteria in statistical machine translation ',Och,'num_Statistics_441 amod_Statistics_N-gram nn_Statistics_Co-Occurrence nsubjpass_used_Statistics prep_Statistics_for nn_statistics_Answer nn_statistics_Extraction nn_statistics_N-gram amod_statistics_co-occurrence pobj_for_statistics aux_used_have auxpass_used_been advmod_used_successfully prep_used_in amod__automatic nn__evaluation pobj_in_ cc_in_and advmod_recently_more conj_in_recently conj_in_as nn_criteria_training pobj_as_criteria prep_criteria_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,868,'We distinguish two main approaches to domain adaptation that have been addressed in the literature : supervised and semi-supervised ','Daume III','nsubj_distinguish_We num_approaches_two amod_approaches_main dobj_distinguish_approaches prep_distinguish_to nn_adaptation_domain pobj_to_adaptation nsubjpass_addressed_that aux_addressed_have auxpass_addressed_been rcmod_adaptation_addressed prep_addressed_in det__the nn__literature pobj_in_ amod__supervised cc_supervised_and conj_supervised_semi-supervised ' +o,869,'Smadja,Frank.(1993).',Smadja,'' +o,870,'The experimental results in show a negative impact on the parsing accuracy from too long dependency relation ',McDonald,'det_results_The amod_results_experimental nsubj_show_results prep_results_in pobj_in_ det_impact_a amod_impact_negative dobj_show_impact prep_show_on det_accuracy_the amod_accuracy_parsing pobj_on_accuracy prep_accuracy_from advmod_long_too amod_relation_long amod_relation_dependency pobj_from_relation ' +p,871,'3 Experimental Results Whereas stochastic modelling is widely used in speech recognition , there are so far only a few research groups that apply stochastic modelling to language translation ',Brown,'num_Results_3 amod_Results_Experimental mark_used_Whereas amod_modelling_stochastic nsubjpass_used_modelling auxpass_used_is advmod_used_widely advcl_are_used prep_used_in nn_recognition_speech pobj_in_recognition expl_are_there dep_Results_are advmod_far_so advmod_are_far quantmod_few_only quantmod_few_a num_groups_few nn_groups_research nsubj_are_groups nsubj_apply_that rcmod_groups_apply amod_modelling_stochastic dobj_apply_modelling prep_apply_to nn_translation_language pobj_to_translation ' +o,872,'k - ~ P -LRB- A -RRB- P -LRB- E -RRB- -LRB- 3 -RRB- 1P -LRB- E -RRB- suggests that the units over which the kappa statistic is computed affects the outcome ',Carletta,'dep_suggests_k nn_P_~ nn_P_P appos_P_A nsubj_suggests_P appos_P_E appos_P_3 measure__1P appos_1P_E amod_P_ complm_affects_that det_units_the nsubj_affects_units prep_units_over dobj_computed_which det_statistic_the nn_statistic_kappa nsubjpass_computed_statistic auxpass_computed_is pcomp_over_computed ccomp_suggests_affects det_outcome_the dobj_affects_outcome ' +o,873,'Among the chunk types , NP chunking is the first to receive the attention , than other chunk types , such as VP and PP chunking ',Ramshaw,'prep_first_Among det_types_the nn_types_chunk pobj_Among_types nn_chunking_NP nsubj_first_chunking cop_first_is det_first_the aux_receive_to dep_first_receive det__the nn__attention dobj_receive_ dep_as_than amod_types_other nn_types_chunk pobj_than_types dep_as_such rcmod__as pobj_as_VP cc_VP_and nn__PP amod__chunking conj_VP_ ' +o,874,'This model shares some similarities with the stochastic inversion transduction grammars -LRB- SITG -RRB- presented by Wu in ',Wu,'det_model_This dep_similarities_model dep_model_shares dep_similarities_some dep_grammars_similarities prep_similarities_with det_inversion_the amod_inversion_stochastic pobj_with_inversion nn_grammars_transduction abbrev_grammars_SITG partmod_grammars_presented prep_presented_by pobj_by_Wu prep_presented_in ' +o,875,'-LRB- makes a similar point , noting that for reviews , the whole is not necessarily the sum of the parts \' -RRB- ',Turney,'nsubj_makes_ det_point_a amod_point_similar dobj_makes_point xcomp_makes_noting complm_sum_that prep_sum_for pobj_for_reviews det_whole_the nsubj_sum_whole cop_sum_is neg_sum_not advmod_sum_necessarily det_sum_the ccomp_noting_sum prep_sum_of det_parts_the pobj_of_parts ' +o,876,'Identifying transliteration pairs is an important component in many linguistic applications which require identifying out-of-vocabulary words , such as machine translation and multilingual information retrieval ',Hermjakob,'amod_pairs_Identifying nn_pairs_transliteration nsubj_component_pairs cop_component_is det_component_an amod_component_important prep_component_in amod_applications_many amod_applications_linguistic pobj_in_applications nsubj_require_which rcmod_applications_require xcomp_require_identifying amod_words_out-of-vocabulary dobj_identifying_words dep_as_such prep_applications_as nn_translation_machine pobj_as_translation cc_translation_and amod_retrieval_multilingual nn_retrieval_information conj_translation_retrieval ' +o,877,' and Chan et al ',Carpuat,'cc__and conj__Chan cc_Chan_et conj_Chan_al ' +o,878,'Based on these grammars , a great number of SMT models have been recently proposed , including string-to-string model -LRB- Synchronous FSG -RRB- , tree-to-string model -LRB- TSG-string -RRB- , string-totree model -LRB- string-CFG\\/TSG -RRB- , tree-to-tree model -LRB- Synchronous CFG\\/TSG , Data-Oriented Translation -RRB- and so on ',Brown,'prep_proposed_Based dep_Based_on det_grammars_these pobj_on_grammars det_number_a amod_number_great appos_grammars_number prep_number_of nn_models_SMT pobj_of_models aux_proposed_have auxpass_proposed_been advmod_proposed_recently prep_proposed_including amod_model_string-to-string pobj_including_model nn_FSG_Synchronous abbrev_model_FSG nsubjpass_proposed_ amod__tree-to-string nn__model abbrev__TSG-string appos__ amod__string-totree nn__model appos__string-CFG\\/TSG appos__ amod_model_tree-to-tree appos__model nn_CFG\\/TSG_Synchronous appos_model_CFG\\/TSG amod_Translation_Data-Oriented dep_CFG\\/TSG_Translation partmod_model_ cc_so_and advmod__so dep_so_on ' +o,879,'We use a standard maximum entropy classifier implemented as part of MALLET ',Berger,'nsubj_use_We det__a amod__standard amod__maximum nn__entropy nn__classifier dobj_use_ partmod__implemented prep_implemented_as pobj_as_part prep_part_of nn__MALLET pobj_of_ ' +o,880,'8412 only PTB -LRB- baseline -RRB- 8358 1st 8342 2nd 8338 3rd 8308 third row lists the three highest scores of the domain adaptation track of the CoNLL 2007 shared task ',Dredze,'num_PTB_8412 advmod_PTB_only appos_PTB_baseline number_1st_8358 num_row_1st nn_row_ num_row_8342 amod_row_2nd number_8338_ num_row_8338 amod_row_3rd number_8308_ num_row_8308 amod_row_third nsubj_lists_row dep_PTB_lists det_scores_the num_scores_three amod_scores_highest dobj_lists_scores prep_scores_of det_track_the nn_track_domain nn_track_adaptation pobj_of_track prep_track_of det_task_the nn_task_CoNLL num_task_2007 amod_task_shared pobj_of_task ' +o,881,'(Johnson [1997] notes that this structure has a higher probability than the correct, flat structure, given counts taken from the treebank for a standard PCFG).',Collins,'' +o,882,'We used a loglinear model with no Markov dependency between adjacent tags ,3 and trained the parameters of the model with the perceptron algorithm , with averaging to control for over-training ',Collins,'nsubj_used_We det_model_a nn_model_loglinear dobj_used_model prep_model_with det_dependency_no nn_dependency_Markov pobj_with_dependency prep_dependency_between amod_tags_adjacent pobj_between_tags dep_model_,3 cc_,3_and conj_,3_trained det_parameters_the dobj_trained_parameters prep_parameters_of det_model_the pobj_of_model prep_trained_with det_algorithm_the nn_algorithm_perceptron pobj_with_algorithm prep_trained_with pcomp_with_averaging aux_control_to xcomp_averaging_control prep_control_for pobj_for_over-training ' +o,883,'In Turneys work , the co-occurrence is considered as the appearance in the same window ',Turney,'prep_considered_In nn_work_Turneys pobj_In_work det_co-occurrence_the nsubjpass_considered_co-occurrence auxpass_considered_is prep_considered_as det_appearance_the pobj_as_appearance prep_appearance_in det_window_the amod_window_same pobj_in_window ' +o,884,'Named entities also pose another problem with the coreference model ; since it models only the heads of NPs , it will fail to resolve some references to named entities : -LRB- Ford Motor Co , Ford -RRB- , while erroneously merging others : -LRB- Ford Motor Co , Lockheed Martin Co -RRB- ',Haghighi,'amod_entities_Named nsubj_pose_entities advmod_pose_also det_problem_another dobj_pose_problem prep_problem_with det_model_the amod_model_ nn_model_coreference pobj_with_model mark_models_since nsubj_models_it advcl_fail_models advmod_heads_only det_heads_the dobj_models_heads prep_heads_of pobj_of_NPs nsubj_fail_it aux_fail_will parataxis_pose_fail aux_resolve_to xcomp_fail_resolve det_references_some dobj_resolve_references prep_references_to amod_entities_named pobj_to_entities nn_Co_Ford nn_Co_Motor appos_references_Co dep_Co_Ford mark_merging_while advmod_merging_erroneously dep_references_merging dobj_merging_others nn_Co_Ford nn_Co_Motor dep_others_Co nn_Co_Lockheed nn_Co_Martin appos_Co_Co ' +o,885,'Translation rules can : look like phrase pairs with syntax decoration : NPB -LRB- NNP -LRB- prime -RRB- NNP -LRB- minister -RRB- NNP -LRB- keizo -RRB- NNP -LRB- obuchi -RRB- -RRB- BUFDFKEUBWAZ carry extra contextual constraints : VP -LRB- VBD -LRB- said -RRB- x0 : SBAR-C -RRB- DKx0 -LRB- according to this rule , DK can translate to said only if some Chinese sequence to the right ofDK is translated into an SBAR-C -RRB- be non-constituent phrases : VP -LRB- VBD -LRB- said -RRB- SBAR-C -LRB- IN -LRB- that -RRB- x0 : S-C -RRB- -RRB- DKx0 VP -LRB- VBD -LRB- pointed -RRB- PRT -LRB- RP -LRB- out -RRB- -RRB- x0 : SBAR-C -RRB- DXGPx0 contain non-contiguous phrases , effectively phrases with holes : PP -LRB- IN -LRB- on -RRB- NP-C -LRB- NPB -LRB- DT -LRB- the -RRB- x0 : NNP -RRB- -RRB- NN -LRB- issue -RRB- -RRB- -RRB- -RRB- GRx0 EVABG6 PP -LRB- IN -LRB- on -RRB- NP-C -LRB- NPB -LRB- DT -LRB- the -RRB- NN -LRB- issue -RRB- -RRB- x0 : PP -RRB- -RRB- GRx0 EVEVABABG6 be purely structural -LRB- no words -RRB- : S -LRB- x0 : NP-C x1 : VP -RRB- x0 x1 re-order their children : NP-C -LRB- NPB -LRB- DT -LRB- the -RRB- x0 : NN -RRB- PP -LRB- IN -LRB- of -RRB- x1 : NP-C -RRB- -RRB- x1 DFx0 Decoding with this model produces a tree in the target language , bottom-up , by parsing the foreign string using a CYK parser and a binarized rule set ',Zhang,'nn_rules_Translation nsubj_look_rules aux_look_can prep_look_like nn_pairs_phrase pobj_like_pairs prep_look_with nn_decoration_syntax pobj_with_decoration nsubj_carry_NPB nn_NNP_NNP dep_NNP_prime abbrev_NPB_NNP appos_NNP_minister nn_keizo_NNP dep_NNP_keizo nn_obuchi_NNP dep_NNP_obuchi advmod_carry_BUFDFKEUBWAZ parataxis_look_carry amod_constraints_extra amod_constraints_contextual dobj_carry_constraints nsubj_DKx0_VP nsubj_said_VBD parataxis_DKx0_said dobj_said_x0 dep_x0_SBAR-C parataxis_look_DKx0 dep_DKx0_according dep_according_to det_rule_this pobj_to_rule appos_rule_DK aux_translate_can rcmod_DK_translate aux_said_to xcomp_translate_said advmod_translated_only dep_translated_if det_sequence_some amod_sequence_Chinese nsubjpass_translated_sequence prep_sequence_to det_ofDK_the amod_ofDK_right pobj_to_ofDK auxpass_translated_is ccomp_said_translated prep_translated_into det_SBAR-C_an pobj_into_SBAR-C cop_phrases_be amod_phrases_non-constituent dep_DKx0_phrases advmod_contain_VP nsubj_said_VBD parataxis_contain_said dobj_said_SBAR-C dep_SBAR-C_IN dep_-LRB-_that pobj_IN_x0 dep_x0_S-C nn_PRT_DKx0 nn_PRT_VP nn_PRT_VBD nn_PRT_pointed nn_x0_PRT dep_PRT_RP dep_PRT_out dep_SBAR-C_x0 dep_SBAR-C_SBAR-C nsubj_contain_DXGPx0 parataxis_look_contain amod_phrases_non-contiguous nsubj_structural_phrases dep_phrases_effectively dep_phrases_phrases prep_phrases_with pobj_with_holes dep_holes_PP dep_PP_IN dep_NP-C_on pobj_IN_NP-C nn_x0_NPB nn_x0_DT nn_x0_the abbrev_NP-C_x0 dep_x0_NNP dep_PP_NN appos_NN_issue nn_PP_GRx0 nn_PP_EVABG6 dep_phrases_PP dep_phrases_IN dep_NP-C_on pobj_IN_NP-C abbrev_NP-C_NPB nn_NN_DT dep_-LRB-_the nsubj_x0_NN appos_NN_issue rcmod_NPB_x0 dobj_x0_PP nn_EVEVABABG6_GRx0 dep_phrases_EVEVABABG6 cop_structural_be advmod_structural_purely ccomp_contain_structural det_words_no dep_-LRB-_words dobj_structural_S abbrev_S_x0 amod_x1_NP-C dep_x0_x1 dep_x1_VP amod_x1_x0 nsubj_re-order_x1 rcmod_S_re-order poss_children_their dobj_re-order_children nn_Decoding_NP-C nn_PP_NPB nn_x0_DT nn_x0_the abbrev_PP_x0 dep_x0_NN dep_NP-C_PP dep_NP-C_IN dep_x1_of pobj_IN_x1 dep_-LRB-_NP-C nn_Decoding_x1 nn_Decoding_DFx0 nsubj_produces_Decoding prep_Decoding_with det_model_this pobj_with_model parataxis_look_produces det_tree_a dobj_produces_tree prep_tree_in det_language_the nn_language_target pobj_in_language appos_language_bottom-up prep_produces_by pcomp_by_parsing det_string_the amod_string_foreign dobj_parsing_string xcomp_parsing_using det_parser_a nn_parser_CYK dobj_using_parser cc_look_and det_rule_a amod_rule_binarized nsubj_set_rule conj_look_set ' +o,886,'7 Model Structure In our statistical model , trees are generated according to a process similar to that described in ',Collins,'num_Structure_7 nn_Structure_Model prep_generated_In poss_model_our amod_model_statistical pobj_In_model nsubjpass_generated_trees auxpass_generated_are dep_Structure_generated prep_generated_according dep_according_to det_process_a pobj_to_process amod_process_similar prep_similar_to pobj_to_that partmod_that_described prep_described_in ' +o,887,'For the constituent-based models , constituent information was obtained from the output of for English and Dubeys parser -LRB- 2004 -RRB- for German ',Collins,'prep_obtained_For det_models_the amod_models_constituent-based pobj_For_models amod_information_constituent nsubjpass_obtained_information auxpass_obtained_was prep_obtained_from det_output_the pobj_from_output prep_output_of amod_parser_ prep__for pobj_for_English cc_English_and conj_English_Dubeys pobj_of_parser appos_parser_2004 prep_obtained_for pobj_for_German ' +o,888,'This finding has been previously reported , among others , in ',Liu,'det_finding_This nsubjpass_reported_finding aux_reported_has auxpass_reported_been advmod_reported_previously prep_reported_among pobj_among_others prep_reported_in ' +p,889,'1 Introduction By exploiting information encoded in human-produced syntactic trees , research on probabilistic models of syntax has driven the performance of syntactic parsers to about 90 \% accuracy ',Marcus,'num_Introduction_1 prep_driven_By pcomp_By_exploiting dobj_exploiting_information partmod_information_encoded prep_encoded_in amod__human-produced amod__syntactic nn__trees pobj_in_ nsubj_driven_research prep_research_on amod_models_probabilistic pobj_on_models prep_models_of pobj_of_syntax aux_driven_has rcmod_Introduction_driven det_performance_the dobj_driven_performance prep_performance_of amod_parsers_syntactic pobj_of_parsers prep_driven_to quantmod_90_about dep_\%_90 dep__\% nn__accuracy pobj_to_ ' +o,890,'In order increase the likelihood that 909 only true paraphrases were considered as phraselevel alternations for an example , extracted sentences were clustered using complete-link clustering using a technique proposed in ',Barzilay,'mark_likelihood_In dep_likelihood_order dep_likelihood_increase det_likelihood_the dep_clustered_likelihood complm_considered_that amod_paraphrases_909 amod_paraphrases_only amod_paraphrases_true nsubjpass_considered_paraphrases auxpass_considered_were ccomp_likelihood_considered prep_considered_as amod_alternations_phraselevel pobj_as_alternations prep_alternations_for det_example_an pobj_for_example amod_sentences_extracted nsubjpass_clustered_sentences auxpass_clustered_were xcomp_clustered_using amod_clustering_complete-link dobj_using_clustering xcomp_using_using det_technique_a dobj_using_technique partmod_technique_proposed prep_proposed_in ' +o,891,'As reported in , parameter averaging can effectively avoid overfitting ',Collins,'mark_reported_As advcl_avoid_reported prep_reported_in pobj_in_ nn_averaging_parameter nsubj_avoid_averaging aux_avoid_can advmod_avoid_effectively dobj_avoid_overfitting ' +p,892,'Unigram models have been previously shown to give good results in sentiment classification tasks : unigram representations can capture a variety of lexical combinations and distributions , including those of emotion words ',Pang,'nn_models_Unigram nsubjpass_shown_models aux_shown_have auxpass_shown_been advmod_shown_previously aux_give_to xcomp_shown_give amod_results_good iobj_give_results prep_results_in nn_tasks_sentiment nn_tasks_classification pobj_in_tasks dobj_give_ nn_representations_unigram nsubj_capture_representations aux_capture_can parataxis_shown_capture det_variety_a dobj_capture_variety prep_variety_of amod_combinations_lexical pobj_of_combinations cc_combinations_and conj_combinations_distributions prep_variety_including pobj_including_those prep_those_of nn_words_emotion pobj_of_words ' +o,893,'Several representations to encode region information are proposed and examined ',Ramshaw,'amod_representations_Several nsubjpass_proposed_representations aux_encode_to infmod_representations_encode nn_information_region dobj_encode_information auxpass_proposed_are cc_proposed_and conj_proposed_examined ' +o,894,'One important application of bitext maps is the construction of translation lexicons and , as discussed , translation lexicons are an important information source for bitext mapping ',Brown,'num_application_One amod_application_important nsubj_construction_application prep_application_of amod_maps_bitext pobj_of_maps cop_construction_is det_construction_the prep_construction_of nn_lexicons_translation pobj_of_lexicons dep_lexicons_ cc__and mark_source_as ccomp_source_discussed nn_lexicons_translation nsubj_source_lexicons cop_source_are det_source_an amod_source_important nn_source_information advcl_,_source prep_source_for amod_mapping_bitext pobj_for_mapping ' +o,895,'This method is described hereafter , while the subsequent steps , that use deeper -LRB- rulebased -RRB- levels of knowledge , are implemented into the ARIOSTO_LEX lexical learning system , described in ',Marcus,'det_method_This nsubjpass_described_method auxpass_described_is advmod_described_hereafter mark_implemented_while det_steps_the amod_steps_subsequent nsubjpass_implemented_steps nsubj_use_that rcmod_steps_use amod_levels_deeper dep_deeper_rulebased dobj_use_levels prep_levels_of pobj_of_knowledge auxpass_implemented_are advcl_described_implemented prep_implemented_into det_system_the nn_system_ARIOSTO_LEX amod_system_lexical nn_system_learning pobj_into_system partmod_system_described prep_described_in pobj_in_ ' +o,896,'This can be the base of a principled method for detecting structural contradictions ',Marneffe,'nsubj_base_This aux_base_can cop_base_be det_base_the prep_base_of det_method_a amod_method_principled pobj_of_method prep_method_for pcomp_for_detecting amod_contradictions_structural dobj_detecting_contradictions ' +o,897,'6 Related Work Several works attempt to extend WordNet with additional lexical semantic information ',Snow,'num_works_6 nn_works_Related nn_works_Work amod_works_Several nsubj_attempt_works aux_extend_to xcomp_attempt_extend dobj_extend_WordNet prep_WordNet_with amod_information_additional amod_information_lexical amod_information_semantic pobj_with_information ' +o,898,'In order to improve sentence-level evaluation performance , several metrics have been proposed , including ROUGE-W , ROUGE-S and METEOR ',Banerjee,'mark_improve_In dep_improve_order aux_improve_to dep_proposed_improve amod_performance_sentence-level nn_performance_evaluation dobj_improve_performance amod_metrics_several nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been prep_proposed_including nn__ROUGE-W conj__ROUGE-S pobj_including_ cc__and conj__METEOR ' +o,899,'Besides the the case-sensitive BLEU-4 used in the two experiments , we design another evaluation metrics Reordering Accuracy -LRB- RAcc -RRB- for forced decoding evaluation ',Papineni,'prep_design_Besides det__the dep__the amod__case-sensitive nn__BLEU-4 pobj_Besides_ partmod__used prep_used_in det_experiments_the num_experiments_two pobj_in_experiments nsubj_design_we det_metrics_another nn_metrics_evaluation iobj_design_metrics nn_Accuracy_Reordering dobj_design_Accuracy abbrev_Accuracy_RAcc prep_Accuracy_for amod_evaluation_forced nn_evaluation_decoding pobj_for_evaluation ' +o,900,'2 Syntactic-oriented evaluation metrics We investigated the following metrics oriented on the syntactic structure of a translation output : POSBLEU The standard BLEU score calculated on POS tags instead of words ; POSP POS n-gram precision : percentage of POS ngrams in the hypothesis which have a counterpart in the reference ; POSR Recall measure based on POS n-grams : percentage of POS n-grams in the reference which are also present in the hypothesis ; POSF POS n-gram based F-measure : takes into account all POS n-grams which have a counter29 part , both in the reference and in the hypothesis ',Papineni,'num_metrics_2 amod_metrics_Syntactic-oriented nn_metrics_evaluation nsubj_investigated_We dep_metrics_investigated det_metrics_the amod_metrics_following dobj_investigated_metrics amod_metrics_oriented prep_oriented_on det_structure_the amod_structure_syntactic pobj_on_structure prep_structure_of det_output_a nn_output_translation pobj_of_output csubj_takes_POSBLEU det__The amod__standard amod__BLEU nn__score dobj_POSBLEU_ partmod__calculated prep_calculated_on dep_tags_POS pobj_on_tags dep_of_instead prep_tags_of pobj_of_words nn_POS_POSP dep_tags_POS amod_precision_n-gram dep_POS_precision dep_tags_percentage prep_percentage_of dep_ngrams_POS pobj_of_ngrams prep_ngrams_in det_hypothesis_the pobj_in_hypothesis nsubj_have_which rcmod_tags_have det_counterpart_a dobj_have_counterpart prep_counterpart_in det_reference_the pobj_in_reference amod_measure_POSR amod_measure_Recall dep_reference_measure partmod_measure_based prep_based_on dep_n-grams_POS pobj_on_n-grams dep_reference_percentage prep_percentage_of dep_n-grams_POS pobj_of_n-grams prep_n-grams_in det_reference_the pobj_in_reference nsubj_present_which cop_present_are advmod_present_also rcmod_reference_present prep_present_in det_hypothesis_the pobj_in_hypothesis nn_n-gram_POSF dep_n-gram_POS dep_reference_n-gram partmod_n-gram_based dep_based_F-measure parataxis_investigated_takes prep_takes_into pobj_into_account det_n-grams_all dep_n-grams_POS dobj_takes_n-grams nsubj_have_which rcmod_n-grams_have det_part_a amod_part_counter29 dobj_have_part preconj_in_both prep_have_in det_reference_the pobj_in_reference cc_in_and conj_in_in det_hypothesis_the pobj_in_hypothesis ' +o,901,'4 Building Noun Similarity Lists A lot of work has been done in the NLP community on clustering words according to their meaning in text ',Hindle,'num_lot_4 nn_lot_Building nn_lot_Noun nn_lot_Similarity nn_lot_Lists nn_lot_A nsubjpass_done_lot prep_lot_of pobj_of_work aux_done_has auxpass_done_been prep_done_in det_community_the nn_community_NLP pobj_in_community prep_done_on pcomp_on_clustering dobj_clustering_words prep_clustering_according dep_according_to poss_meaning_their pobj_to_meaning prep_meaning_in pobj_in_text ' +o,902,'In other words , -LRB- 4b -RRB- can be used in substitution of -LRB- 4a -RRB- , whereas -LRB- 5b -RRB- can not , so easily 41n , a value of K between 8 and I indicates good agreement ; a value between 6 and 8 indicates some agreement ',Carletta,'prep_used_In amod_words_other pobj_In_words nsubjpass_used_4b aux_used_can auxpass_used_be prep_used_in pobj_in_substitution prep_substitution_of pobj_of_4a dep_used_whereas nsubj_can_5b dep_used_can neg_can_not advmod_can_so advmod__easily amod__41n dep_can_ det_value_a appos__value prep_value_of pobj_of_K prep_K_between pobj_between_8 cc_used_and nsubj_indicates_I conj_used_indicates amod_agreement_good dobj_indicates_agreement det_value_a nsubj_indicates_value prep_value_between pobj_between_6 cc_6_and conj_6_8 parataxis_indicates_indicates det_agreement_some dobj_indicates_agreement ' +o,903,'For example , have studied synchronous context free grammar ',Wu,'prep_studied_For pobj_For_example nsubj_studied_ aux_studied_have dep_studied_synchronous prep_studied_context amod_grammar_free dep_context_grammar ' +o,904,' Peter F Brown , Vincent J Della Pietra , Petere V deSouza , Jenifer C Lai , and Robert L Mercer ',Brown,'nn_Brown_Peter nn_Brown_F nsubj__Brown nn_Pietra_Vincent nn_Pietra_J nn_Pietra_Della appos_Brown_Pietra nn_deSouza_Petere nn_deSouza_V conj_Pietra_deSouza nn_Lai_Jenifer nn_Lai_C conj_Pietra_Lai cc_Pietra_and nn_Mercer_Robert nn_Mercer_L conj_Pietra_Mercer ' +o,905,'Table 1 reports values for the Kappa -LRB- K -RRB- coefficient of agreement for Forward and Backward Functions 6 The columns in the tables read as follows : if utterance Ui has tag X , do coders agree on the subtag ? ',Carletta,'nn_reports_Table num_reports_1 nsubj_values_reports dep_agree_values prep_values_for det_coefficient_the nn_coefficient_Kappa appos_coefficient_K pobj_for_coefficient prep_coefficient_of nn__agreement pobj_of_ prep__for nn_Functions_Forward cc_Forward_and conj_Forward_Backward pobj_for_Functions dobj_values_6 det_columns_The nsubj_read_columns prep_columns_in det_tables_the pobj_in_tables rcmod_6_read mark_follows_as advcl_read_follows mark_has_if amod_Ui_utterance nsubj_has_Ui advcl_follows_has nn_X_tag dobj_has_X aux_agree_do nsubj_agree_coders prep_agree_on det_subtag_the pobj_on_subtag ' +p,906,'The state-of-the art taggers are using feature sets discribed in the corresponding articles -LRB- , , and -RRB- , Morce supervised and Morce semi-supervised are using feature set desribed in section 4 ',Shen,'det_taggers_The amod_taggers_state-of-the nn_taggers_art nsubj_using_taggers aux_using_are nn_sets_feature dobj_using_sets partmod_sets_discribed prep_discribed_in det_articles_the amod_articles_corresponding pobj_in_articles appos_articles_ dep__ dep__ cc__and conj__ nsubj_supervised_Morce conj_using_supervised cc_using_and nsubj_using_Morce amod_Morce_semi-supervised aux_using_are conj_using_using nn_set_feature dobj_using_set partmod_set_desribed prep_desribed_in pobj_in_section num_section_4 ' +o,907,'When we have a junction tree for each document , we can efficiently perform belief propagation in order to compute argmax in Equation -LRB- 1 -RRB- , or the marginal probabilities of cliques and labels , necessary for the parameter estimation of machine learning classifiers , including perceptrons , and maximum entropy models ',Collins,'advmod_have_When nsubj_have_we dep_perform_have det_tree_a nn_tree_junction dobj_have_tree prep_tree_for det_document_each pobj_for_document nsubj_perform_we aux_perform_can advmod_perform_efficiently nn_propagation_belief dobj_perform_propagation mark_compute_in dep_compute_order aux_compute_to purpcl_perform_compute dobj_compute_argmax prep_compute_in pobj_in_Equation appos_Equation_1 cc_Equation_or det_probabilities_the amod_probabilities_marginal conj_Equation_probabilities prep_probabilities_of pobj_of_cliques cc_cliques_and conj_cliques_labels amod_Equation_necessary prep_necessary_for det_estimation_the nn_estimation_parameter pobj_for_estimation prep_estimation_of nn_classifiers_machine nn_classifiers_learning pobj_of_classifiers prep_Equation_including amod__perceptrons pobj_including_ cc_perform_and amod_models_maximum nn_models_entropy nsubj__models conj_perform_ ' +o,908,' -RRB- , concordancing for bilingual lexicography , computerassisted language learning , corpus linguistics -LRB- Melby ',Brown,'dep__concordancing prep_concordancing_for pobj_for_bilingual nn__lexicography dobj__ amod_learning_computerassisted nn_learning_language appos__learning advmod_linguistics_corpus dep__linguistics dobj_linguistics_Melby ' +p,909,'It performed slightly worse on baseNP recognition than the experiments -LRB- Fz = 1 = 916 -RRB- ',Ramshaw,'nsubj_performed_It advmod_worse_slightly acomp_performed_worse prep_worse_on nn_recognition_baseNP pobj_on_recognition prep_recognition_than det_experiments_the amod_experiments_ pobj_than_experiments nsubj_916_Fz dep_916_= num_916_1 amod_916_= dep_worse_916 ' +p,910,'21 The averaged perceptron The averaged perceptron algorithm was proposed as a way of reducing overfitting on the training data ',Collins,'det_perceptron_The amod_perceptron_averaged nsubjpass_proposed_perceptron det__The amod__averaged nn__perceptron nn__algorithm dep_perceptron_ auxpass_proposed_was dep_21_proposed prep_proposed_as det_way_a pobj_as_way prep_way_of pcomp_of_reducing xcomp_reducing_overfitting prep_overfitting_on det_data_the nn_data_training pobj_on_data ' +o,911,'In particular , previous work has investigated the use of Markov random fields -LRB- MRFs -RRB- or log-linear models as probabilistic models with global features for parsing and other NLP tasks ',Collins,'prep_investigated_In pobj_In_particular amod__previous nn__work nsubj_investigated_ aux_investigated_has det_use_the dobj_investigated_use prep_use_of nn_fields_Markov amod_fields_random pobj_of_fields abbrev_fields_MRFs cc_use_or amod_models_log-linear conj_use_models prep_models_as amod_models_probabilistic pobj_as_models prep_investigated_with amod_features_global pobj_with_features prep_features_for nn_tasks_parsing cc_parsing_and conj_parsing_other nn_tasks_NLP pobj_for_tasks ' +o,912,'The model consists of a set of word-pair parameters p -LRB- t -LRB- s -RRB- and position parameters p -LRB- j -LRB- i , \\/ -RRB- ; in model 1 -LRB- IBM1 -RRB- the latter are fixed at 1 \\/ -LRB- 1 + 1 -RRB- , as each position , including the empty position 0 , is considered equally likely to contain a translation for w Maximum likelihood estimates for these parameters can be obtained with the EM algorithm over a bilingual training corpus , as described in ',Brown,'det_model_The nsubj_consists_model prep_consists_of det_set_a pobj_of_set prep_set_of amod_p_word-pair nn_p_parameters pobj_of_p dep_p_t dep_-LRB-_s cc_set_and nn_parameters_position conj_set_parameters partmod_parameters_p dobj_p_j appos_j_i dep_i_\\/ rel_fixed_in pobj_in_model num_model_1 dep_model_IBM1 det_latter_the nsubjpass_fixed_latter auxpass_fixed_are rcmod_j_fixed prep_fixed_at number_\\/_1 pobj_at_\\/ dep_\\/_1 cc_1_+ conj_1_1 mark_considered_as det_position_each nsubjpass_considered_position prep_position_including det_0_the amod_0_empty nn_0_position pobj_including_0 auxpass_considered_is advcl_fixed_considered advmod_likely_equally acomp_considered_likely aux_contain_to xcomp_likely_contain det_translation_a dobj_contain_translation prep_translation_for amod_estimates_w amod_estimates_Maximum nn_estimates_likelihood pobj_for_estimates mark_obtained_for det_parameters_these nsubjpass_obtained_parameters aux_obtained_can auxpass_obtained_be advcl_p_obtained prep_obtained_with det_algorithm_the nn_algorithm_EM pobj_with_algorithm prep_obtained_over det_corpus_a amod_corpus_bilingual nn_corpus_training pobj_over_corpus mark_described_as dep_corpus_described prt_described_in ' +o,913,'We perform minimum error rate training to tune the feature weights for the log-linear modeltomaximizethesystemssBLEUscoreonthe development set ',Och,'nsubj_perform_We amod__minimum nn__error nn__rate nn__training dobj_perform_ prep_perform_to pobj_to_tune det_weights_the nn_weights_feature nsubj_set_weights prep_weights_for det_development_the amod_development_log-linear nn_development_modeltomaximizethesystemssBLEUscoreonthe pobj_for_development dep_perform_set ' +o,914,'73 ID Participant BBN-COMBO BBN system combination (Rosti et al., 2008) CMU-COMBO Carnegie Mellon University system combination (Jayaraman and Lavie, 2005) CMU-GIMPEL Carnegie Mellon University Gimpel (Gimpel and Smith, 2008) CMU-SMT Carnegie Mellon University SMT (Bach et al., 2008) CMU-STATXFER Carnegie Mellon University Stat-XFER (Hanneman et al., 2008) CU-TECTOMT Charles University TectoMT (Zabokrtsky et al., 2008) CU-BOJAR Charles University Bojar (Bojar and Hajic, 2008) CUED Cambridge University (Blackwood et al., 2008) DCU Dublin City University (Tinsley et al., 2008) LIMSI LIMSI (Dechelotte et al., 2008) LIU Linkoping University (Stymne et al., 2008) LIUM-SYSTRAN LIUM / Systran (Schwenk et al., 2008) MLOGIC Morphologic (Novak et al., 2008) PCT a commercial MT provider from the Czech Republic RBMT16 Babelfish, Lingenio, Lucy, OpenLogos, ProMT, SDL (ordering anonymized) SAAR University of Saarbruecken (Eisele et al., 2008) SYSTRAN Systran (Dugast et al., 2008) UCB University of California at Berkeley (Nakov, 2008) UCL University College London (Wang and Shawe-Taylor, 2008) UEDIN University of Edinburgh (Koehn et al., 2008) UEDIN-COMBO University of Edinburgh system combination (Josh Schroeder) UMD University of Maryland (Dyer, 2007) UPC Universitat Politecnica de Catalunya, Barcelona (Khalilov et al., 2008) UW University of Washington (Axelrod et al., 2008) XEROX Xerox Research Centre Europe (Nikoulina and Dymetman, 2008) Table 2: Participants in the shared translation task.',Rosti,'' +o,915,'The candidates of unknown words can be generated by heuristic rules or statistical word models which predict the probabilities for any strings to be unknown words ',Berger,'det_candidates_The nsubjpass_generated_candidates prep_candidates_of amod_words_unknown pobj_of_words aux_generated_can auxpass_generated_be prep_generated_by amod__heuristic nn__rules pobj_by_ cc__or amod_models_statistical nn_models_word conj__models nsubj_predict_which rcmod_models_predict det_probabilities_the nsubj_words_probabilities prep_probabilities_for det_strings_any pobj_for_strings aux_words_to cop_words_be amod_words_unknown xcomp_predict_words advmod_words_ ' +o,916,'For each differently tokenized corpus , we computed word alignments by a HMM translation model and by a word alignment refinement heuristic of grow-diagfinal ',Koehn,'prep_computed_For det_corpus_each advmod_tokenized_differently amod_corpus_tokenized pobj_For_corpus nsubj_computed_we nn_alignments_word dobj_computed_alignments prep_computed_by det__a nn__HMM nn__translation nn__model pobj_by_ cc_by_and conj_by_by det_heuristic_a nn_heuristic_word nn_heuristic_alignment nn_heuristic_refinement pobj_by_heuristic prep_heuristic_of pobj_of_grow-diagfinal ' +o,917,'On the other hand , purely statistical systems extract discriminating MWUs from text corpora by means of association measure regularities ',Dunning,'prep__On det_hand_the amod_hand_other pobj_On_hand advmod_systems_purely amod_systems_statistical nsubj__systems dep__extract xcomp_extract_discriminating dobj_discriminating_MWUs prep_MWUs_from nn_corpora_text pobj_from_corpora prep_discriminating_by pobj_by_means prep_means_of nn_regularities_association nn_regularities_measure pobj_of_regularities ' +o,918,'The chunker is trained on the answer side of the Training corpus in order to learn 2 and 3word collocations , defined using the likelihood ratio of ',Dunning,'det_chunker_The nsubjpass_trained_chunker auxpass_trained_is prep_trained_on det_side_the nn_side_answer pobj_on_side prep_side_of det_corpus_the nn_corpus_Training pobj_of_corpus mark_learn_in dep_learn_order aux_learn_to purpcl_trained_learn dobj_learn_2 cc_2_and amod_collocations_3word conj_2_collocations partmod_2_defined xcomp_defined_using det_ratio_the nn_ratio_likelihood dobj_using_ratio prep_using_of ' +o,919,'We then tested the best models for each vocabulary size on the testing set4 Standard measures of performance are shown in table 15 3We used a publicly available tagger to provide the tags used in these experiments , rather than the handcorrected tags which come with the corpus ',Ratnaparkhi,'nsubj_tested_We advmod_tested_then det_models_the amod_models_best dobj_tested_models prep_models_for det_size_each nn_size_vocabulary pobj_for_size prep_tested_on det_testing_the pobj_on_testing amod_measures_set4 nn_measures_Standard nsubjpass_shown_measures prep_measures_of pobj_of_performance auxpass_shown_are dep_tested_shown prep_shown_in nn_3We_table num_3We_15 pobj_in_3We partmod_3We_used det__a advmod_available_publicly amod__available nn__tagger dobj_used_ aux_provide_to xcomp_used_provide det_tags_the dobj_provide_tags partmod_tags_used prep_used_in det_experiments_these pobj_in_experiments cc_tags_rather dep_rather_than det_tags_the amod_tags_handcorrected conj_tags_tags nsubj_come_which rcmod_tags_come prep_come_with det_corpus_the pobj_with_corpus ' +o,920,'First , we can let the number of nonterminals grow unboundedly , as in the Infinite PCFG , where the nonterminals of the grammar can be indefinitely refined versions of a base PCFG ',Liang,'advmod_let_First nsubj_let_we aux_let_can det_number_the nsubj_grow_number prep_number_of pobj_of_nonterminals ccomp_let_grow advmod_grow_unboundedly dep_in_as prep_grow_in det_PCFG_the nn_PCFG_Infinite pobj_in_PCFG advmod_refined_where det_nonterminals_the nsubjpass_refined_nonterminals prep_nonterminals_of det_grammar_the pobj_of_grammar aux_refined_can auxpass_refined_be advmod_refined_indefinitely rcmod_PCFG_refined dobj_refined_versions prep_versions_of det_PCFG_a nn_PCFG_base pobj_of_PCFG ' +o,921,'The tree-to-string model views the translation as a structure mapping process , which first breaks the source syntax tree into many tree fragments and then maps each tree fragment into its corresponding target translation using translation rules , finally combines these target translations into a complete sentence ',Galley,'det__The amod__tree-to-string nn__model nsubj_views_ det_translation_the dobj_views_translation mark_combines_as det_process_a nn_process_structure nn_process_mapping nsubj_combines_process nsubj_breaks_which advmod_breaks_first rcmod_process_breaks det_tree_the nn_tree_source nn_tree_syntax dobj_breaks_tree prep_breaks_into amod_fragments_many nn_fragments_tree pobj_into_fragments cc_breaks_and advmod_maps_then conj_breaks_maps det_fragment_each nn_fragment_tree dobj_maps_fragment prep_maps_into poss_translation_its amod_translation_corresponding nn_translation_target pobj_into_translation xcomp_maps_using nn_rules_translation dobj_using_rules advmod_combines_finally advcl_views_combines det_translations_these nn_translations_target dobj_combines_translations prep_combines_into det_sentence_a amod_sentence_complete pobj_into_sentence ' +o,922,'It is important because a wordaligned corpus is typically used as a first step in order to identify phrases or templates in phrase-based Machine Translation , , -LRB- et al , 2003 , sec ',Koehn,'nsubj_important_It cop_important_is mark_used_because det_corpus_a amod_corpus_wordaligned nsubjpass_used_corpus auxpass_used_is advmod_used_typically advcl_important_used prep_used_as det_step_a amod_step_first pobj_as_step prep_step_in pobj_in_order aux_identify_to xcomp_used_identify dobj_identify_phrases cc_phrases_or conj_phrases_templates prep_identify_in amod__phrase-based nn__Machine nn__Translation pobj_in_ appos__ dep__ cc__et conj__al dep__2003 appos__sec ' +o,923,'Parse Parse score from Model 2 of the statistical parser , normalized by the number of words ',Collins,'nn_Parse_Parse nsubj_score_Parse prep_score_from pobj_from_Model num_Model_2 prep_Model_of det__the amod__statistical nn__parser pobj_of_ partmod__normalized prep_normalized_by det_number_the pobj_by_number prep_number_of pobj_of_words ' +o,924,'But Koehn , find that phrases longer than three words improve performance little for training corpora of up to 20 million words , suggesting that the data may be too sparse to learn longer phrases ',Och,'cc_find_But nn__Koehn nsubj_find_ complm_improve_that nsubj_improve_phrases advmod_improve_longer dep_longer_than num_words_three pobj_than_words ccomp_find_improve dobj_improve_performance amod_performance_little prep_improve_for nn_corpora_training pobj_for_corpora prep_corpora_of quantmod_million_up dep_million_to number_million_20 num_words_million pobj_of_words xcomp_improve_suggesting complm_sparse_that det_data_the nsubj_sparse_data aux_sparse_may cop_sparse_be advmod_sparse_too ccomp_suggesting_sparse aux_learn_to xcomp_sparse_learn amod_phrases_longer dobj_learn_phrases ' +o,925,'To do this , we first identify initial phrase pairs using the same criterion as previous systems : Definition 1 ',Och,'aux_do_To dep_identify_do dobj_do_this nsubj_identify_we advmod_identify_first amod_pairs_initial nn_pairs_phrase dobj_identify_pairs partmod_pairs_using det_criterion_the amod_criterion_same dobj_using_criterion prep_criterion_as amod__previous nn__systems pobj_as_ dep_criterion_Definition num_Definition_1 ' +o,926,'The first is a novel stochastic search strategy that appears to make better use of s algorithm for finding the global minimum along any given search direction than either coordinate descent or Powells method ',Och,'det_first_The nsubj_strategy_first cop_strategy_is det_strategy_a amod_strategy_novel amod_strategy_stochastic nn_strategy_search nsubj_appears_that rcmod_strategy_appears aux_make_to xcomp_appears_make amod_use_better dobj_make_use prep_use_of amod_algorithm_ nn_algorithm_s pobj_of_algorithm prep_make_for pcomp_for_finding det_minimum_the amod_minimum_global dobj_finding_minimum prep_finding_along det_direction_any amod_direction_given nn_direction_search pobj_along_direction prep_direction_than preconj_descent_either amod_descent_coordinate pobj_than_descent cc_descent_or nn_method_Powells conj_descent_method ' +o,927,'However , as do not propose any evaluation of which clustering algorithm should be used , we experiment a set of clustering algorithms and present the comparative results ',Barzilay,'dep_propose_However mark_propose_as nsubj_propose_ aux_propose_do neg_propose_not det_evaluation_any dobj_propose_evaluation prep_evaluation_of dobj_used_which amod_algorithm_clustering nsubjpass_used_algorithm aux_used_should auxpass_used_be pcomp_of_used nsubj_experiment_we ccomp_used_experiment det_set_a dobj_experiment_set prep_set_of amod_algorithms_clustering pobj_of_algorithms cc_used_and conj_used_present det_results_the amod_results_comparative dobj_present_results ' +o,928,'Transformation-based error-driven learning has been applied to a number of natural language problems , including part of speech tagging , prepositional phrase attachment disambiguation , speech generation and syntactic parsing ',Marcus,'amod_learning_Transformation-based amod_learning_error-driven nsubjpass_applied_learning aux_applied_has auxpass_applied_been prep_applied_to det_number_a pobj_to_number prep_number_of amod_problems_natural nn_problems_language pobj_of_problems prep_number_including pobj_including_part prep_part_of nn_tagging_speech pobj_of_tagging amod_disambiguation_prepositional nn_disambiguation_phrase nn_disambiguation_attachment conj_tagging_disambiguation nn_generation_speech conj_tagging_generation cc_tagging_and amod_parsing_syntactic conj_tagging_parsing ' +o,929,'To measure interannotator agreement , we compute Cohens Kappa from the two sets of annotations , obtaining a Kappa value of only 043 ',Carletta,'aux_measure_To dep_compute_measure nn_agreement_interannotator dobj_measure_agreement nsubj_compute_we nn__Cohens nn__Kappa dobj_compute_ prep_compute_from det_sets_the num_sets_two pobj_from_sets prep_sets_of pobj_of_annotations xcomp_compute_obtaining det_value_a nn_value_Kappa dobj_obtaining_value prep_value_of quantmod_043_only pobj_of_043 ' +o,930,'Many 412 Similarity of Semantic Relations researchers have argued that metaphor is the heart of human thinking ',Turney,'amod_Similarity_Many nn_Similarity_412 nn_Similarity_ nsubj_argued_Similarity prep_Similarity_of nn_researchers_Semantic nn_researchers_Relations pobj_of_researchers aux_argued_have complm_heart_that nsubj_heart_metaphor cop_heart_is det_heart_the ccomp_argued_heart prep_heart_of amod__human nn__thinking pobj_of_ ' +o,931,'We use Entropy Regularization -LRB- ER -RRB- to leverage unlabeled instances7 We weight the ER term by choosing the best8 weight in -LCB- 103,102,101,1,10 -RCB- multiplied by # labeled # unlabeled for each data set and query selection method ',Jiao,'nsubj_use_We nn__Entropy nn__Regularization abbrev__ER dobj_use_ prep__to nn_instances7_leverage amod_instances7_unlabeled pobj_to_instances7 nsubj_weight_We dep_use_weight det_term_the nn_term_ER dobj_weight_term prep_weight_by pcomp_by_choosing det_weight_the amod_weight_best8 dobj_choosing_weight prep_weight_in pobj_in_103,102,101,1,10 partmod_weight_multiplied prep_multiplied_by dep_labeled_# pobj_by_labeled dep_unlabeled_# dep_use_unlabeled prep_unlabeled_for det_set_each nn_set_data pobj_for_set cc_set_and nn_method_query nn_method_selection conj_set_method ' +p,932,'In the statistical NLP community , the most widely used grammatical resource is the Penn Treebank ',Marcus,'prep_Treebank_In det_community_the amod_community_statistical nn_community_NLP pobj_In_community det_resource_the advmod_used_most advmod_used_widely amod_resource_used nn_resource_grammatical nsubj_Treebank_resource cop_Treebank_is det_Treebank_the nn_Treebank_Penn ' +o,933,' applied to the output of the reranking parser of Charniak and Johnson , whereas in BE -LRB- in the version presented here -RRB- dependencies are generated by the Minipar parser ',Lin,'nsubj_applied_ prep_applied_to det_output_the pobj_to_output prep_output_of det_parser_the amod_parser_reranking pobj_of_parser prep_parser_of nn__Charniak cc_Charniak_and conj_Charniak_Johnson pobj_of_ dep_applied_whereas prep_generated_in pobj_in_BE dep_BE_in det_version_the pobj_in_version partmod_version_presented advmod_presented_here nsubjpass_generated_dependencies auxpass_generated_are dep_applied_generated prep_generated_by det_parser_the nn_parser_Minipar pobj_by_parser ' +o,934,'Traditionally , generative word alignment models have been trained on massive parallel corpora ',Brown,'advmod_trained_Traditionally amod_models_generative nn_models_word nn_models_alignment nsubjpass_trained_models aux_trained_have auxpass_trained_been prep_trained_on amod_corpora_massive amod_corpora_parallel pobj_on_corpora ' +o,935,'There are other approaches in which the generation grammars are extracted semiautomatically or automatically -LRB- such as HPSG , LFG and CCG -RRB- ',Cahill,'expl_are_There amod_approaches_other nsubj_are_approaches rel_extracted_in pobj_in_which det_grammars_the nn_grammars_generation nsubjpass_extracted_grammars auxpass_extracted_are rcmod_approaches_extracted advmod__semiautomatically acomp_extracted_ cc__or advmod__automatically dep_as_such dep_approaches_as nn__HPSG pobj_as_ nn__LFG conj__ cc__and nn__CCG conj__ ' +o,936,' 1993 -RRB- or else -LRB- as with mutual information -RRB- eschew significance testing in favor of a generic information-theoretic approach ',Dunning,'nsubj_1993_ cc_significance_or amod_significance_else dep_with_as dep_significance_with amod_information_mutual pobj_with_information nn_significance_eschew dobj_1993_significance dep_significance_testing prep_1993_in pobj_in_favor prep_favor_of det_approach_a amod_approach_generic amod_approach_information-theoretic pobj_of_approach ' +o,937,'We automatically converted the phrase structure output of the parser into the syntactic dependency representation used by our syntactic realizer , RealPro ',Collins,'nsubj_converted_We advmod_converted_automatically det_output_the nn_output_phrase nn_output_structure dobj_converted_output prep_output_of det_parser_the amod_parser_ pobj_of_parser prep_converted_into det_representation_the amod_representation_syntactic amod_representation_dependency pobj_into_representation partmod_representation_used prep_used_by poss_realizer_our amod_realizer_syntactic pobj_by_realizer nn__RealPro appos_realizer_ ' +p,938,' , Pedersen , Yarowsky and Florian -RRB- as well as maximum entropy models -LRB- eg , Dang and Palmer , and Manning -RRB- in particular have shown a large degree of success for WSD , and have established challenging state-of-the-art benchmarks ',Klein,'advmod_shown_ nn__Pedersen nn__ conj__Yarowsky cc__and conj__Florian nsubj_shown_ dep_well_as cc__well dep_well_as amod_models_maximum nn_models_entropy conj__models dep_models_eg conj_eg_Dang cc_eg_and nn__Palmer nn__ conj__ cc__and conj__Manning conj_eg_ prep_models_in pobj_in_particular aux_shown_have det_degree_a amod_degree_large dobj_shown_degree prep_degree_of pobj_of_success prep_shown_for pobj_for_WSD cc_shown_and aux_established_have conj_shown_established xcomp_established_challenging amod_benchmarks_state-of-the-art dobj_challenging_benchmarks ' +o,939,'3 We then run , using just the sentence pairs where parsing succeeds with a negative log likelihood below 200 ',Collins,'nsubj_run_We advmod_run_then dep_3_run acomp_run_ xcomp_run_using advmod_pairs_just det_pairs_the nn_pairs_sentence dobj_using_pairs advmod_succeeds_where nsubj_succeeds_parsing rcmod_pairs_succeeds prep_succeeds_with det_likelihood_a amod_likelihood_negative nn_likelihood_log pobj_with_likelihood prep_succeeds_below pobj_below_200 ' +p,940,'1 Introduction Large scale annotated corpora such as the Penn TreeBank have played a central role in speech and natural language research ',Marcus,'num_scale_1 nn_scale_Introduction amod_scale_Large nsubj_annotated_scale nsubj_played_corpora dep_as_such prep_corpora_as det__the nn__Penn nn__TreeBank pobj_as_ aux_played_have ccomp_annotated_played det_role_a amod_role_central dobj_played_role prep_role_in nn_research_speech cc_speech_and conj_speech_natural nn_research_language pobj_in_research ' +o,941,'On the one hand using 1 human reference with uniform results is essential for our methodology , since it means that there is no more trouble with Recall a systems ability to avoid under-generation of N-grams can now be reliably measured ',Papineni,'prep_essential_On det_hand_the num_hand_one pobj_On_hand partmod_hand_using num_reference_1 amod_reference_human dobj_using_reference prep_using_with amod_results_uniform pobj_with_results cop_essential_is prep_essential_for poss_methodology_our pobj_for_methodology mark_means_since nsubj_means_it advcl_essential_means complm_is_that expl_is_there ccomp_means_is det_trouble_no amod_trouble_more nsubj_is_trouble prep_trouble_with nn__Recall pobj_with_ det_ability_a nn_ability_systems nsubj_essential_ability aux_avoid_to infmod_ability_avoid nsubjpass_measured_under-generation prep_under-generation_of pobj_of_N-grams aux_measured_can advmod_measured_now auxpass_measured_be advmod_measured_reliably ccomp_avoid_measured ' +o,942,'2 Linguistic and Context Features 21 Non-terminal Labels In the original string-to-dependency model , a translation rule is composed of a string of words and non-terminals on the source side and a well-formed dependency structure on the target side ',He,'dep_Features_2 nsubj_Features_Linguistic cc_Linguistic_and conj_Linguistic_Context num_Labels_21 amod_Labels_Non-terminal dobj_Features_Labels prep_composed_In det__the amod__original amod__string-to-dependency nn__model pobj_In_ det_rule_a nn_rule_translation nsubjpass_composed_rule auxpass_composed_is dep_Features_composed prep_composed_of det_string_a pobj_of_string prep_string_of pobj_of_words cc_words_and conj_words_non-terminals prep_composed_on det_side_the nn_side_source pobj_on_side cc_side_and det_structure_a amod_structure_well-formed amod_structure_dependency conj_side_structure prep_composed_on det_side_the nn_side_target pobj_on_side ' +o,943,'Results in terms of word-error-rate -LRB- WER -RRB- and BLEU score are reported in Table 4 for those sentences that contain at least one unknown word ',Papineni,'nsubjpass_reported_Results prep_Results_in pobj_in_terms prep_terms_of pobj_of_word-error-rate abbrev_word-error-rate_WER cc_word-error-rate_and amod__BLEU nn__score conj_word-error-rate_ auxpass_reported_are prep_reported_in pobj_in_Table num_Table_4 prep_reported_for det_sentences_those pobj_for_sentences nsubj_contain_that rcmod_sentences_contain quantmod_one_at dep_at_least num_word_one amod_word_unknown dobj_contain_word ' +o,944,'This was used , for example , by in information extraction , and by in POS tagging ',Collins,'nsubjpass_used_This auxpass_used_was prep_used_for pobj_for_example prep_used_by pcomp_by_ prep__in nn_extraction_information pobj_in_extraction cc_by_and conj_by_by pcomp_by_ prep__in dep_tagging_POS pobj_in_tagging ' +n,945,'2This can explain why previous attempts to use WordNet for generating sentence-level paraphrases were unsuccessful ',Barzilay,'nsubj_explain_2This aux_explain_can advmod_use_why amod_attempts_previous nsubj_use_attempts aux_use_to xcomp_explain_use dobj_use_WordNet prep_WordNet_for amod_paraphrases_generating amod_paraphrases_sentence-level pobj_for_paraphrases nsubj_unsuccessful_ cop_unsuccessful_were dep_explain_unsuccessful ' +o,946,'5 Combining In-Domain and Out-of-Domain Data for Training In this section , we will first introduce the AUGMENT technique of , before showing the performance of our WSD system with and without using this technique ','Daume III','num_In-Domain_5 nn_In-Domain_Combining nsubj_introduce_In-Domain cc_In-Domain_and nn_Data_Out-of-Domain conj_In-Domain_Data prep_In-Domain_for pobj_for_Training prep_introduce_In det_section_this pobj_In_section nsubj_introduce_we aux_introduce_will advmod_introduce_first det_technique_the amod_technique_AUGMENT dobj_introduce_technique prep_technique_of pobj_of_ prep_introduce_before pcomp_before_showing det_performance_the dobj_showing_performance prep_performance_of poss_system_our nn_system_WSD pobj_of_system prep_showing_with cc_with_and conj_with_without pcomp_with_using det_technique_this dobj_using_technique ' +o,947,' use hand-coded slot-filling rules to determine the semantic roles of the arguments of a nominalization ',Berger,'nsubj_use_ amod_rules_hand-coded amod_rules_slot-filling nsubj_determine_rules aux_determine_to xcomp_use_determine det_roles_the amod_roles_semantic dobj_determine_roles prep_roles_of det_arguments_the pobj_of_arguments prep_arguments_of det_nominalization_a pobj_of_nominalization ' +p,948,' report an improvement in MT grammaticality on a very restricted test set : short sentences parsable by an LFG grammar without back-off rules ',Riezler,'nsubj_report_ det_improvement_an dobj_report_improvement prep_improvement_in pobj_in_MT dep_on_grammaticality prep_report_on det_set_a advmod_restricted_very amod_set_restricted nn_set_test pobj_on_set amod_sentences_short nsubj_parsable_sentences parataxis_report_parsable prep_parsable_by det_grammar_an nn_grammar_LFG pobj_by_grammar prep_parsable_without amod_rules_back-off pobj_without_rules ' +o,949,'Previous work on linguistic annotation pipelines has enforced consistency from one stage to the next ',Finkel,'amod_work_Previous nsubj_enforced_work prep_work_on amod__linguistic nn__annotation nn__pipelines pobj_on_ aux_enforced_has dobj_enforced_consistency prep_enforced_from num_stage_one pobj_from_stage prep_enforced_to det_next_the pobj_to_next ' +o,950,' suggested comparing the frequency of phrase co-occurrences with words predetermined by the sentiment lexicon ',Turney,'nsubj_suggested_ xcomp_suggested_comparing det_frequency_the dobj_comparing_frequency prep_frequency_of nn_co-occurrences_phrase pobj_of_co-occurrences prep_comparing_with pobj_with_words partmod_words_predetermined prep_predetermined_by det_lexicon_the nn_lexicon_sentiment pobj_by_lexicon ' +o,951,' has called attention to the log-likelihood ratio , G 2 , as appropriate for the analysis of such contingency tables , especially when such contingency tables concern very low frequency words ',Dunning,'nsubj_called_ aux_called_has dobj_called_attention prep_called_to det_ratio_the amod_ratio_log-likelihood pobj_to_ratio appos_ratio_G num_G_2 advmod_appropriate_as amod_ratio_appropriate prep_appropriate_for det_analysis_the pobj_for_analysis prep_analysis_of amod_tables_such nn_tables_contingency pobj_of_tables advmod_when_especially advmod_concern_when amod_tables_such nn_tables_contingency nsubj_concern_tables dep_ratio_concern advmod_low_very amod_words_low nn_words_frequency dobj_concern_words ' +o,952,'The original publication evaluated their NP chunker on two data sets , the second holding a larger amount of training data -LRB- Penn Treebank sections 02-21 -RRB- while using 00 as test data ',Ramshaw,'det_publication_The amod_publication_original amod_publication_ partmod_publication_evaluated poss_chunker_their nn_chunker_NP dobj_evaluated_chunker prep_evaluated_on num_sets_two nn_sets_data pobj_on_sets det_second_the appos_sets_second partmod_second_holding det_amount_a amod_amount_larger dobj_holding_amount prep_amount_of nn_data_training pobj_of_data nn_02-21_Penn nn_02-21_Treebank nn_02-21_sections dep_publication_02-21 prep_publication_while pcomp_while_using dobj_using_00 prep_using_as nn_data_test pobj_as_data ' +o,953,'In addition to precision and recall , we also evaluate the Bleu score changes before and after applying our measure word generation method to the SMT output ',Papineni,'prep_evaluate_In pobj_In_addition prep_addition_to pobj_to_precision cc_precision_and conj_precision_recall nsubj_evaluate_we advmod_evaluate_also det_Bleu_the nsubj_score_Bleu ccomp_evaluate_score amod_changes_ dobj_score_changes prep_score_before cc_before_and conj_before_after pcomp_before_applying poss_method_our nn_method_measure nn_method_word nn_method_generation dobj_applying_method prep_applying_to det_output_the nn_output_SMT pobj_to_output ' +o,954,'We used a bottom-up , CKY-style decoder that works with binary xRs rules obtained via a synchronous binarization procedure ',Zhang,'nsubj_used_We det_decoder_a amod_decoder_bottom-up amod_decoder_CKY-style dobj_used_decoder nsubj_works_that rcmod_decoder_works prep_works_with amod_rules_binary nn_rules_xRs pobj_with_rules partmod_rules_obtained prep_obtained_via det_procedure_a amod_procedure_synchronous nn_procedure_binarization pobj_via_procedure ' +o,955,'Alternatively , one can train them with respect to the final translation quality measured by an error criterion ',Och,'advmod_train_Alternatively nsubj_train_one aux_train_can dobj_train_them prep_train_with pobj_with_respect prep_train_to det_quality_the amod_quality_final nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_by det_criterion_an nn_criterion_error pobj_by_criterion ' +o,956,'53 Performance of Taxonomy Induction In this section , we compare the following automatic taxonomy induction systems : HE , the system by Hearst with 6 hypernym patterns ; GI , the system by Girju et al ',Brown,'num_Performance_53 nsubj_compare_Performance prep_Performance_of nn_Induction_Taxonomy pobj_of_Induction prep_compare_In det_section_this pobj_In_section nsubj_compare_we det_taxonomy_the amod_taxonomy_following amod_taxonomy_automatic dobj_compare_taxonomy nn_systems_induction dep_taxonomy_systems dep_taxonomy_HE det_system_the appos_HE_system prep_system_by nn__Hearst pobj_by_ prep__with num_patterns_6 nn_patterns_hypernym pobj_with_patterns dep_taxonomy_GI det_system_the appos_GI_system prep_taxonomy_by pobj_by_Girju cc_Girju_et conj_Girju_al ' +o,957,'Examples are Andersen , , Sun et al ',Okanohara,'advmod_are_Examples nn__Andersen nsubj_are_ appos__ conj__Sun cc__et conj__al ' +o,958,'To address this , standard measures like precision and recall could be used , as in some previous research ',Banerjee,'aux_address_To dep_used_address dobj_address_this amod_measures_standard nsubjpass_used_measures prep_measures_like pobj_like_precision cc_precision_and conj_precision_recall aux_used_could auxpass_used_be advmod_used_as dep_as_in det_research_some amod_research_previous pobj_in_research ' +p,959,'We carried out automatic evaluation of our summaries using ROUGE toolkit , which has been widely adopted by DUC for automatic summarization evaluation ',Lin,'nsubj_carried_We prt_carried_out amod_evaluation_automatic dobj_carried_evaluation prep_evaluation_of poss_summaries_our pobj_of_summaries xcomp_carried_using amod_toolkit_ROUGE amod_toolkit_ dobj_using_toolkit nsubjpass_adopted_which aux_adopted_has auxpass_adopted_been advmod_adopted_widely rcmod_toolkit_adopted prep_adopted_by pobj_by_DUC prep_DUC_for amod_evaluation_automatic nn_evaluation_summarization pobj_for_evaluation ' +o,960,'Many statistical translation models try to model word-to-word correspondences between source and target words ',Brown,'amod__Many amod__statistical nn__translation nn__models nsubj_try_ aux_model_to xcomp_try_model amod_correspondences_word-to-word dobj_model_correspondences prep_correspondences_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,961,'The models are trained using the Margin Infused Relaxed Algorithm or MIRA instead of the standard minimum-error-rate training or MERT algorithm ',Och,'det_models_The nsubjpass_trained_models auxpass_trained_are xcomp_trained_using det_Margin_the nsubj_Algorithm_Margin nn_Algorithm_Infused nn_Algorithm_Relaxed xcomp_using_Algorithm cc_Algorithm_or nn__MIRA conj_Algorithm_ dep_of_instead prep_Algorithm_of det_training_the amod_training_standard amod_training_minimum-error-rate pobj_of_training cc_training_or nn_algorithm_MERT conj_training_algorithm ' +o,962,'Phrases are then extracted from the word alignments using the method described in ',Och,'nsubjpass_extracted_Phrases auxpass_extracted_are advmod_extracted_then prep_extracted_from det_alignments_the nn_alignments_word pobj_from_alignments partmod_alignments_using det_method_the dobj_using_method partmod_method_described prep_described_in ' +o,963,' is the first , to the best of our knowledge , to raise the issue of a unified approach ',Turney,'nsubj_first_ cop_first_is det_first_the prep_first_to det_best_the pobj_to_best prep_best_of poss_knowledge_our pobj_of_knowledge aux_raise_to infmod_first_raise det_issue_the dobj_raise_issue prep_issue_of det_approach_a amod_approach_unified pobj_of_approach ' +o,964,'IBM constraints , the lexical word reordering model , and inversion transduction grammar -LRB- ITG -RRB- constraints belong to this type of approach ',Wu,'nn_constraints_IBM nsubj__constraints det_model_the amod_model_lexical nn_model_word nn_model_reordering nsubj__model ccomp__ cc__and nn_grammar_inversion nn_grammar_transduction nn_constraints_grammar abbrev_constraints_ITG nsubj_belong_constraints advmod_belong_ conj__belong prep_belong_to det_type_this pobj_to_type prep_type_of pobj_of_approach ' +o,965,'The effectiveness of these features for recognition of discourse relations has been previously shown by ',Marcu,'det_effectiveness_The nsubjpass_shown_effectiveness prep_effectiveness_of det_features_these pobj_of_features prep_features_for pobj_for_recognition prep_recognition_of nn_relations_discourse pobj_of_relations aux_shown_has auxpass_shown_been advmod_shown_previously prep_shown_by ' +o,966,'However , such methods require the existence of either a parallel corpus\\/machine translation engine for projecting\\/translating annotations\\/lexica from a resource-rich language to the target language , or a domain that is similar enough to the target domain ',Blitzer,'advmod_require_However amod_methods_such nsubj_require_methods det_existence_the dobj_require_existence prep_existence_of dep_engine_either det_engine_a amod_engine_parallel amod_engine_corpus\\/machine nn_engine_translation pobj_of_engine prep_require_for amod_annotations\\/lexica_projecting\\/translating pobj_for_annotations\\/lexica prep_annotations\\/lexica_from det_language_a amod_language_resource-rich pobj_from_language prep_language_to det__the nn__target nn__language pobj_to_ cc_annotations\\/lexica_or det_domain_a conj_annotations\\/lexica_domain nsubj_similar_that cop_similar_is rcmod_domain_similar advmod_similar_enough prep_similar_to det_domain_the nn_domain_target pobj_to_domain ' +o,967,'Given an input sentence x , the correct output segmentation F -LRB- x -RRB- satisfies : F -LRB- x -RRB- = argmax yGEN -LRB- x -RRB- Score -LRB- y -RRB- where GEN -LRB- x -RRB- denotes the set of possible segmentations for an input sentence x , consistent with notation from ',Collins,'det_sentence_an nn_sentence_input dep_Given_sentence nsubj_satisfies_x det_F_the amod_F_correct nn_F_output nn_F_segmentation appos_x_F dep_F_x rcmod_sentence_satisfies dep_Given_F dep_argmax_x amod_argmax_= dobj_F_argmax dep_argmax_yGEN dep_yGEN_x dep_argmax_Score appos_Score_y advmod_denotes_where nsubj_denotes_GEN dep_GEN_x rcmod_Score_denotes det_set_the dobj_denotes_set prep_set_of amod_segmentations_possible pobj_of_segmentations prep_segmentations_for det_x_an nn_x_input nn_x_sentence pobj_for_x amod_argmax_consistent prep_consistent_with pobj_with_notation prep_consistent_from ' +p,968,'In machine translation , confusion-network based combination techniques -LRB- eg , -RRB- have achieved the state-of-theart performance in MT evaluations ',Rosti,'prep_achieved_In nn_translation_machine pobj_In_translation nsubj_achieved_confusion-network partmod_confusion-network_based nn_techniques_combination dobj_based_techniques appos_confusion-network_eg dep_eg_ aux_achieved_have det_performance_the amod_performance_state-of-theart dobj_achieved_performance prep_performance_in nn_evaluations_MT pobj_in_evaluations ' +o,969,'For the statistics-based approaches , developed a statistics-based method for automatically identifying existential definite NPs which are non-anaphoric ',Bean,'prep_developed_For det_approaches_the amod_approaches_statistics-based pobj_For_approaches nsubj_developed_ det_method_a amod_method_statistics-based dobj_developed_method prep_developed_for advmod_identifying_automatically pcomp_for_identifying amod_NPs_existential amod_NPs_definite dobj_identifying_NPs nsubj_non-anaphoric_which cop_non-anaphoric_are rcmod_NPs_non-anaphoric ' +o,970,'A large database of human judgments might also be useful as an objective function for minimum error rate training or in other system development tasks ',Och,'det_database_A amod_database_large nsubj_useful_database prep_database_of amod_judgments_human pobj_of_judgments aux_useful_might advmod_useful_also cop_useful_be prep_useful_as det_function_an amod_function_objective pobj_as_function prep_function_for amod__minimum nn__error nn__rate amod__training pobj_for_ cc_as_or conj_as_in amod_tasks_other nn_tasks_system nn_tasks_development pobj_in_tasks ' +o,971,'During evaluation two performance metrics , BLEU and NIST , were computed ',Papineni,'pobj_During_evaluation num_metrics_two nn_metrics_performance nsubjpass_computed_metrics nn__BLEU appos_metrics_ cc__and conj__NIST auxpass_computed_were rcmod_evaluation_computed ' +n,972,'We have also illustrated that ASIA outperforms three other English systems , even though many of these use more input than just a semantic class name ',Snow,'nsubj_illustrated_We aux_illustrated_have advmod_illustrated_also complm_outperforms_that nsubj_outperforms_ASIA ccomp_illustrated_outperforms num_systems_three amod_systems_other amod_systems_English nsubj__systems xcomp_outperforms_ advmod_use_even dep_use_though nsubj_use_many prep_many_of pobj_of_these ccomp_outperforms_use amod_input_more dobj_use_input dep_just_than cc_input_just det_name_a amod_name_semantic nn_name_class conj_input_name ' +o,973,'-LRB- 1 -RRB- Here , the candidate generator gen -LRB- s -RRB- enumerates candidates of destination -LRB- correct -RRB- strings , and the scorer P -LRB- t s -RRB- denotes the conditional probability of the string t for the given s The scorer was modeled by a noisy-channel model and maximum entropy framework ',Berger,'dep_enumerates_1 nsubj_enumerates_Here det_gen_the nn_gen_candidate nn_gen_generator appos_Here_gen dep_-LRB-_s dobj_enumerates_candidates prep_candidates_of amod_strings_destination dep_strings_correct pobj_of_strings cc_enumerates_and det_P_the nn_P_scorer nsubj_denotes_P nn_s_t appos_P_s conj_enumerates_denotes det_probability_the amod_probability_conditional dobj_denotes_probability prep_probability_of det_t_the nn_t_string pobj_of_t prep_t_for det_s_the amod_s_given pobj_for_s det_scorer_The nsubjpass_modeled_scorer auxpass_modeled_was dep_enumerates_modeled prep_modeled_by det__a amod__noisy-channel nn__model pobj_by_ cc__and amod_framework_maximum nn_framework_entropy conj__framework ' +o,974,'Meanwhile , translation grammars have grown in complexity from simple inversion transduction grammars to general tree-to-string transducers and have increased in size by including more synchronous tree fragments ',Galley,'advmod_grown_Meanwhile nn_grammars_translation nsubj_grown_grammars aux_grown_have prep_grown_in pobj_in_complexity prep_grown_from amod_grammars_simple nn_grammars_inversion nn_grammars_transduction pobj_from_grammars partmod_grammars_ prep__to amod__general amod__tree-to-string nn__transducers pobj_to_ cc_grown_and aux_increased_have conj_grown_increased prep_increased_in pobj_in_size prep_increased_by pcomp_by_including advmod_synchronous_more amod_fragments_synchronous nn_fragments_tree dobj_including_fragments ' +p,975,'However , except for , none of these advances in alignment quality has improved translation quality of a state-of-the-art system ',Fraser,'advmod_has_However prep_has_except dep_except_for pobj_for_ nsubj_has_none prep_none_of det_advances_these pobj_of_advances prep_advances_in amod_quality_alignment pobj_in_quality amod_quality_improved nn_quality_translation dobj_has_quality prep_quality_of det_system_a amod_system_state-of-the-art pobj_of_system ' +o,976,'220 ; they can overlap5 Additionally , since phrase features can be any function of words and alignments , we permit features that consider phrase pairs in which a target word outside the target phrase aligns to a source word inside the source phrase , as well as phrase pairs with gaps ',Koehn,'num__220 nsubj_overlap5_they aux_overlap5_can dep__overlap5 advmod_permit_Additionally mark_function_since nn_features_phrase nsubj_function_features aux_function_can cop_function_be det_function_any dep_permit_function prep_function_of pobj_of_words cc_words_and conj_words_alignments nsubj_permit_we ccomp_overlap5_permit dobj_permit_features nsubj_consider_that rcmod_features_consider nn_pairs_phrase dobj_consider_pairs rel_aligns_in pobj_in_which det_word_a nn_word_target nsubj_aligns_word prep_word_outside det_phrase_the nn_phrase_target pobj_outside_phrase rcmod_pairs_aligns prep_aligns_to det_word_a nn_word_source pobj_to_word prep_word_inside det_phrase_the nn_phrase_source pobj_inside_phrase dep_well_as cc_word_well dep_well_as nn_pairs_phrase conj_word_pairs prep_pairs_with amod__gaps pobj_with_ ' +o,977,'There are three major types of models : Heuristic models as in , generative models as the IBM models and discriminative models ',Brown,'expl_are_There num_types_three amod_types_major nsubj_are_types prep_types_of pobj_of_models amod_models_Heuristic dep_types_models mark__as prep__in pobj_in_ amod_models_generative nsubj__models prep_models_as det__the nn__IBM nn__models pobj_as_ cc__and amod_models_discriminative conj__models dep_models_ ' +o,978,' invented heuristic symmetriza57 FRENCH\\/ENGLISH ARABIC\\/ENGLISH SYSTEM F-MEASURE -LRB- = 04 -RRB- BLEU F-MEASURE -LRB- = 01 -RRB- BLEU GIZA + + 735 3063 758 5155 741 3140 791 5289 LEAF UNSUPERVISED 745 723 LEAF SEMI-SUPERVISED 763 3186 845 5434 Table 3 : Experimental Results tion of the output of a 1-to-N model and a M-to-1 model resulting in a M-to-N alignment , this was extended in ',Fraser,'advmod_invented_ amod_symmetriza57_heuristic dobj_invented_symmetriza57 nn_F-MEASURE_FRENCH\\/ENGLISH nn_F-MEASURE_ARABIC\\/ENGLISH nn_F-MEASURE_SYSTEM nsubj_invented_F-MEASURE number_04_= dep_F-MEASURE_04 amod_F-MEASURE_BLEU dep_F-MEASURE_F-MEASURE number_01_= dep_F-MEASURE_01 nn_GIZA_BLEU nsubjpass_extended_GIZA amod_GIZA_+ number_5434_+ cc_5434_735 conj_5434_3063 number_5434_758 number_5434_5155 number_5434_ number_5434_741 number_5434_3140 number_5434_791 number_5434_5289 number_5434_LEAF number_5434_UNSUPERVISED number_5434_745 cc_5434_723 conj_5434_LEAF number_5434_SEMI-SUPERVISED number_5434_763 number_5434_3186 number_5434_845 dep_+_5434 cc_5434_Table conj_5434_3 nn_tion_Experimental nn_tion_Results dep_5434_tion prep_tion_of det_output_the pobj_of_output prep_output_of det_model_a amod_model_1-to-N pobj_of_model cc_output_and det_model_a amod_model_M-to-1 conj_output_model partmod_model_resulting prep_resulting_in det_alignment_a amod_alignment_M-to-N pobj_in_alignment appos_alignment_this auxpass_extended_was rcmod_F-MEASURE_extended prep_extended_in pobj_in_ ' +p,979,'In this paper , we build on recent work that demonstrated how the Bloom filter -LRB- Bloom -LRB- 1970 -RRB- ; BF -RRB- , a space-efficient randomised data structure for representing sets , could be used to store corpus statistics efficiently ',Talbot,'prep_build_In det_paper_this pobj_In_paper nsubj_build_we prep_build_on amod__recent nn__work pobj_on_ nsubj_demonstrated_that rcmod__demonstrated advmod_used_how det_filter_the nn_filter_Bloom nsubjpass_used_filter nn_1970_Bloom abbrev_filter_1970 dep_1970_BF det_structure_a amod_structure_space-efficient amod_structure_randomised nn_structure_data appos_filter_structure prep_structure_for pcomp_for_representing dobj_representing_sets aux_used_could auxpass_used_be ccomp_demonstrated_used aux_store_to purpcl_used_store nn_statistics_corpus dobj_store_statistics advmod_store_efficiently ' +o,980,'The limited contexts used in this model are similar to the previous methods ',Collins,'det_contexts_The amod_contexts_limited nsubj_similar_contexts partmod_contexts_used prep_used_in det_model_this pobj_in_model cop_similar_are prep_similar_to det_methods_the amod_methods_previous pobj_to_methods ' +p,981,'Among all the automatic MT evaluation metrics , BLEU is the most widely used ',Papineni,'prep_most_Among predet_metrics_all det_metrics_the nn_metrics_automatic nn_metrics_MT nn_metrics_evaluation pobj_Among_metrics nn__BLEU nsubj_most_ cop_most_is det_most_the advmod_used_widely amod_most_used ' +o,982,'Distributional measures of distance , such as those proposed by , quantify how similar the two sets of contexts of a target word pair are ',Lin,'amod_measures_Distributional nsubj_quantify_measures prep_measures_of pobj_of_distance dep_as_such prep_measures_as pobj_as_those partmod_those_proposed prep_proposed_by pobj_by_ advmod_are_how dep_how_similar det_sets_the num_sets_two nsubj_are_sets prep_sets_of pobj_of_contexts prep_contexts_of det_pair_a nn_pair_target nn_pair_word pobj_of_pair ccomp_quantify_are ' +p,983,'To simulate real world scenario , we use n-best lists from ISIs state-of-the-art statistical machine translation system , AlTemp , and the 2002 NIST Chinese-English evaluation corpus as the test corpus ',Och,'aux_simulate_To dep_use_simulate amod_scenario_real nn_scenario_world dobj_simulate_scenario nsubj_use_we amod_lists_n-best dobj_use_lists prep_use_from nn_system_ISIs amod_system_state-of-the-art amod_system_statistical nn_system_machine nn_system_translation pobj_from_system nn__AlTemp conj_system_ cc_system_and det_corpus_the num_corpus_2002 nn_corpus_NIST nn_corpus_Chinese-English nn_corpus_evaluation conj_system_corpus prep_corpus_as det_corpus_the nn_corpus_test pobj_as_corpus ' +o,984,'GIZA + + , an implementation of the IBM and HMM -LRB- ? -RRB- ',Och,'nn_+_GIZA nn__+ dep_+_ det_implementation_an conj__implementation prep_implementation_of det__the nn__IBM pobj_of_ cc__and conj__HMM ' +o,985,'2 Block Orientation Bigrams This section describes a phrase-based model for SMT similar to the models presented in ',Och,'num_Bigrams_2 nn_Bigrams_Block nn_Bigrams_Orientation det_section_This nsubj_describes_section dep_Bigrams_describes det_model_a amod_model_phrase-based nsubj_similar_model prep_model_for pobj_for_SMT xcomp_describes_similar prep_similar_to det_models_the pobj_to_models partmod_models_presented prep_presented_in ' +o,986,'A perceptron algorithm gives 9711 \% ',Collins,'det_algorithm_A nn_algorithm_perceptron nsubj_gives_algorithm num_\%_9711 dobj_gives_\% ' +n,987,'For unknown words , SCL gives a relative reduction in error of 195 \% over , even with 40,000 sentences of source domain training data ',Ratnaparkhi,'prep_gives_For amod_words_unknown pobj_For_words nsubj_gives_SCL det_reduction_a amod_reduction_relative dobj_gives_reduction prep_gives_in pobj_in_error prep_error_of num_\%_195 pobj_of_\% prep_gives_over pobj_over_ dep_with_even prep_gives_with num_sentences_40,000 pobj_with_sentences prep_sentences_of nn_data_source nn_data_domain nn_data_training pobj_of_data ' +o,988,'The f are trained using a held-out corpus using maximum BLEU training ',Och,'det_f_The nsubjpass_trained_f auxpass_trained_are xcomp_trained_using det_corpus_a amod_corpus_held-out dobj_using_corpus xcomp_using_using amod_training_maximum amod_training_BLEU dobj_using_training ' +o,989,'Standard CI Model 1 training , initialised with a uniform translation table so that t -LRB- ejf -RRB- is constant for all source\\/target word pairs -LRB- f , e -RRB- , was run on untagged data for 10 iterations in each direction ',Brown,'nn_Model_Standard nn_Model_CI num_training_1 nsubjpass_run_training partmod_training_initialised prep_initialised_with det_table_a amod_table_uniform nn_table_translation pobj_with_table advmod_constant_so dep_constant_that nsubj_constant_t appos_t_ejf cop_constant_is ccomp_initialised_constant prep_constant_for det_pairs_all amod_pairs_source\\/target nn_pairs_word pobj_for_pairs dep_pairs_f appos_f_e auxpass_run_was dep_Model_run prep_run_on amod_data_untagged pobj_on_data prep_data_for num_iterations_10 pobj_for_iterations prep_iterations_in det_direction_each pobj_in_direction ' +o,990,'According to the document , it is the output of Ratnaparkhis tagger ',Ratnaparkhi,'prep_output_According dep_According_to det_document_the pobj_to_document nsubj_output_it cop_output_is det_output_the prep_output_of nn_tagger_Ratnaparkhis pobj_of_tagger ' +o,991,'1 Introduction Current methods for large-scale information extraction take advantage of unstructured text available from either Web documents or , more recently , logs of Web search queries to acquire useful knowledge with minimal supervision ',Snow,'num_methods_1 nn_methods_Introduction amod_methods_Current nsubj_take_methods prep_methods_for amod_extraction_large-scale nn_extraction_information pobj_for_extraction dobj_take_advantage prep_advantage_of amod_text_unstructured pobj_of_text amod_text_available prep_available_from det__either nn__Web nn__documents pobj_from_ cc_available_or advmod_recently_more dep_available_recently conj_available_logs prep_text_of nn_queries_Web nn_queries_search pobj_of_queries partmod_queries_ aux_acquire_to xcomp__acquire amod_knowledge_useful dobj_acquire_knowledge prep_acquire_with amod_supervision_minimal pobj_with_supervision ' +o,992,'6 The Experimental Results We used the Penn Treebank to perform empirical experiments on this parsing model ',Marcus,'num_Results_6 nn_Results_The nn_Results_Experimental nsubj_used_We dep_Results_used det__the nn__Penn nn__Treebank nsubj_perform_ aux_perform_to xcomp_used_perform amod_experiments_empirical dobj_perform_experiments prep_perform_on det_model_this amod_model_parsing pobj_on_model ' +o,993,'Given a source sentence f , the preferred translation output is determined by computing the lowest-cost derivation -LRB- combination of hierarchical and glue rules -RRB- yielding f as its source side , where the cost of a derivation R1 Rn with respective feature vectors v1 , , vn Rm is given by msummationdisplay i = 1 i nsummationdisplay j = 1 -LRB- vj -RRB- i Here , 1 , , m are the parameters of the loglinear model , which we optimize on a held-out portion of the training set using minimum-error-rate training ',Och,'prep_determined_Given det_sentence_a nn_sentence_source dep_Given_sentence dep_sentence_f det_output_the amod_output_preferred nn_output_translation nsubjpass_determined_output auxpass_determined_is prep_determined_by pcomp_by_computing det_derivation_the amod_derivation_lowest-cost dobj_computing_derivation dep_derivation_combination prep_combination_of nn_rules_hierarchical cc_hierarchical_and conj_hierarchical_glue pobj_of_rules partmod_derivation_yielding dep_1_f dep_1_as poss_side_its nn_side_source nsubj_1_side advmod_j_where det_cost_the nsubj_j_cost prep_cost_of det_R1_a nn_R1_derivation pobj_of_R1 partmod_R1_Rn prep_Rn_with amod_v1_respective nn_v1_feature nn_v1_vectors pobj_with_v1 appos_R1_vn nsubjpass_given_Rm auxpass_given_is rcmod_vn_given prep_given_by pobj_by_msummationdisplay dep_msummationdisplay_i dep_j_= num_j_1 amod_j_i nn_j_nsummationdisplay rcmod_side_j dep_1_= ccomp_yielding_1 appos_derivation_vj dep_parameters_i nsubj_parameters_Here appos_Here_1 appos_Here_m cop_parameters_are det_parameters_the dep_determined_parameters prep_parameters_of det_model_the nn_model_loglinear pobj_of_model rel_optimize_which nsubj_optimize_we rcmod_model_optimize prep_optimize_on det_portion_a amod_portion_held-out nsubj_using_portion prep_portion_of det__the nn__training nn__set pobj_of_ pcomp_on_using amod_training_minimum-error-rate dobj_using_training ' +o,994,'The dependency trees induced when each rewrite rule in an i-th order LCFRS distinguish a unique head can similarly be characterized by being of gap-degree i , so that i is the maximum number of gaps that may appear between contiguous substrings of any subtree in the dependency tree ',Kuhlmann,'det_trees_The amod_trees_dependency nsubj_induced_trees advmod_LCFRS_when det_rule_each amod_rule_rewrite nsubj_LCFRS_rule prep_rule_in det_order_an amod_order_i-th pobj_in_order advcl_induced_LCFRS dep_LCFRS_distinguish det_head_a amod_head_unique nsubjpass_characterized_head aux_characterized_can advmod_characterized_similarly auxpass_characterized_be dep_distinguish_characterized prep_characterized_by pcomp_by_being prep_being_of pobj_of_gap-degree dep_gap-degree_i cc_characterized_so dep_number_that nsubj_number_i cop_number_is det_number_the amod_number_maximum conj_characterized_number prep_number_of pobj_of_gaps nsubj_appear_that aux_appear_may rcmod_gaps_appear prep_appear_between amod_substrings_contiguous pobj_between_substrings prep_substrings_of det_subtree_any pobj_of_subtree prep_subtree_in det_tree_the amod_tree_dependency pobj_in_tree ' +o,995,'This improvement is close to that of one sense per discourse -LRB- improvement ranging from 13 \% to 17 \% -RRB- , which seems to be a sensible upper bound of the proposed method ',Yarowsky,'det_improvement_This nsubj_close_improvement cop_close_is prep_close_to pobj_to_that prep_that_of num_sense_one pobj_of_sense prep_sense_per nn__discourse pobj_per_ dep__improvement partmod_improvement_ranging prep_ranging_from number_\%_13 dep_\%_\% dep_\%_to number_\%_17 pobj_from_\% nsubj_seems_which rcmod__seems aux_bound_to cop_bound_be det_bound_a amod_bound_sensible amod_bound_upper xcomp_seems_bound prep_bound_of det_method_the amod_method_proposed pobj_of_method ' +o,996,'Besides precision , recall and -LRB- balanced -RRB- F-measure , we also include an F-measure variant strongly biased towards recall -LRB- # 0B = 01 -RRB- , which found to be best to tune their LEAF aligner for maximum MT accuracy ',Fraser,'prep_include_Besides pobj_Besides_precision conj_precision_recall cc_precision_and dep_precision_balanced conj_precision_F-measure nsubj_include_we advmod_include_also det_variant_an amod_variant_F-measure dobj_include_variant advmod_biased_strongly partmod_variant_biased prep_biased_towards pobj_towards_recall dep_0B_# nsubj_01_0B dep_01_= dep_recall_01 nsubj_found_which advmod_found_ rcmod_recall_found aux_best_to cop_best_be xcomp_found_best prep_best_to pcomp_to_tune poss_aligner_their nn_aligner_LEAF dobj_tune_aligner prep_tune_for amod_accuracy_maximum nn_accuracy_MT pobj_for_accuracy ' +o,997,'1 Introduction Sentiment detection and classification has received considerable attention recently ',Turney,'num_detection_1 nn_detection_Introduction nn_detection_Sentiment nsubj_received_detection cc_detection_and conj_detection_classification aux_received_has amod_attention_considerable dobj_received_attention advmod_received_recently ' +o,998,'4 -RRB- , it constitutes a bijection between source and target sentence positions , since the intersecting alignments are functions according to their definition in 3 ',Brown,'nsubj_constitutes_it dep_4_constitutes det_bijection_a dobj_constitutes_bijection prep_bijection_between nn_positions_source cc_source_and conj_source_target nn_positions_sentence pobj_between_positions prep_constitutes_since det_intersecting_the pobj_since_intersecting nsubj_functions_alignments cop_functions_are dep_4_functions prep_functions_according dep_according_to poss_definition_their pobj_to_definition prep_definition_in pobj_in_ num__3 ' +o,999,'In fact , in it was shown that this neural network can be viewed as a coarse approximation to the corresponding ISBN model ',Titov,'prep_shown_In pobj_In_fact prep_fact_in pobj_in_ nsubjpass_shown_it auxpass_shown_was complm_viewed_that det_network_this amod_network_neural nsubjpass_viewed_network aux_viewed_can auxpass_viewed_be ccomp_shown_viewed prep_viewed_as det_approximation_a amod_approximation_coarse pobj_as_approximation prep_viewed_to det_model_the amod_model_corresponding nn_model_ISBN pobj_to_model ' +o,1000,'The evaluation metric is case-sensitive BLEU-4 ',Papineni,'det_metric_The nn_metric_evaluation nsubj_BLEU-4_metric cop_BLEU-4_is amod_BLEU-4_case-sensitive ' +o,1001,'To this end , we adopt techniques from statistical machine translation and use statistical alignment to learn the edit patterns ',Brown,'prep_adopt_To det_end_this pobj_To_end nsubj_adopt_we dobj_adopt_techniques prep_adopt_from amod__statistical nn__machine nn__translation pobj_from_ cc_adopt_and conj_adopt_use amod_alignment_statistical dobj_use_alignment aux_learn_to xcomp_use_learn det_patterns_the amod_patterns_edit dobj_learn_patterns ' +o,1002,'Indeed , in the II scenario , reported no improvement of the base parser for small -LRB- 500 sentences , in the first paper -RRB- and large -LRB- 40K sentences , in the last two papers -RRB- seed datasets respectively ',McClosky,'advmod_reported_Indeed prep_reported_in det_scenario_the nn_scenario_II pobj_in_scenario nsubj_reported_ det_improvement_no dobj_reported_improvement prep_improvement_of det_parser_the nn_parser_base pobj_of_parser prep_reported_for amod_datasets_small num_sentences_500 dep_small_sentences prep_sentences_in det_paper_the amod_paper_first pobj_in_paper cc_small_and conj_small_large num_sentences_40K dep_large_sentences prep_sentences_in det_papers_the amod_papers_last num_papers_two pobj_in_papers nn_datasets_seed pobj_for_datasets dep_for_respectively ' +o,1003,'The basic phrase-based model is an instance of the noisy-channel approach ',Brown,'det_model_The amod_model_basic amod_model_phrase-based nsubj_instance_model cop_instance_is det_instance_an prep_instance_of det_approach_the amod_approach_noisy-channel pobj_of_approach ' +o,1004,'Here , ppicker shows the accuracy when phrases are extracted by using the N-best phrase alignment method described in Section 41 , while growdiag-final shows the accuracy when phrases are extracted using the standard phrase extraction algorithm described in ',Koehn,'advmod_shows_Here nsubj_shows_ppicker det_accuracy_the dobj_shows_accuracy advmod_extracted_when nsubjpass_extracted_phrases auxpass_extracted_are dep_accuracy_extracted prep_extracted_by pcomp_by_using det_method_the amod_method_N-best nn_method_phrase nn_method_alignment dobj_using_method partmod_method_described prep_described_in pobj_in_Section num_Section_41 mark_shows_while nsubj_shows_growdiag-final advcl_shows_shows det_accuracy_the dobj_shows_accuracy advmod_extracted_when nsubjpass_extracted_phrases auxpass_extracted_are dep_accuracy_extracted xcomp_extracted_using det_algorithm_the amod_algorithm_standard nn_algorithm_phrase nn_algorithm_extraction dobj_using_algorithm partmod_algorithm_described prep_described_in ' +o,1005,'The word sense disambiguation method proposed in can also be viewed as a kind of co-training ',Yarowsky,'det_method_The nn_method_word nn_method_sense nn_method_disambiguation nsubjpass_viewed_method partmod_method_proposed prep_proposed_in pobj_in_ aux_viewed_can advmod_viewed_also auxpass_viewed_be prep_viewed_as det_kind_a pobj_as_kind prep_kind_of pobj_of_co-training ' +p,1006,'A quite different approach from our hypotheses testing implemented in the TREQ-AL aligner is taken by the model-estimating aligners , most of them relying on the IBM models -LRB- 1 to 5 -RRB- described in the seminal paper ',Brown,'det_approach_A advmod_different_quite amod_approach_different nsubjpass_taken_approach prep_approach_from poss_testing_our amod_testing_hypotheses pobj_from_testing partmod_testing_implemented prep_implemented_in det_aligner_the amod_aligner_TREQ-AL pobj_in_aligner auxpass_taken_is prep_taken_by det_aligners_the amod_aligners_model-estimating pobj_by_aligners appos_aligners_most prep_most_of nsubj_relying_them pcomp_of_relying prep_relying_on det_models_the nn_models_IBM pobj_on_models number_5_1 dep_5_to dep_models_5 partmod_models_described prep_described_in det_paper_the amod_paper_ amod_paper_seminal pobj_in_paper ' +p,1007,'In our future work we plan to investigate the effect of more sophisticated and , probably , more accurate filtering methods on the QA results ',Fleischman,'prep_plan_In poss_work_our amod_work_future pobj_In_work nsubj_plan_we aux_investigate_to xcomp_plan_investigate det_effect_the dobj_investigate_effect prep_effect_of advmod_sophisticated_more amod_methods_sophisticated cc_sophisticated_and dep_sophisticated_probably advmod_accurate_more conj_sophisticated_accurate amod_methods_filtering pobj_of_methods partmod_methods_ prep__on det_results_the nn_results_QA pobj_on_results ' +o,1008,'These joint counts are estimated using the phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Brown,'det_counts_These amod_counts_joint nsubjpass_estimated_counts auxpass_estimated_are xcomp_estimated_using det_algorithm_the nn_algorithm_phrase nn_algorithm_induction dobj_using_algorithm partmod_algorithm_described prep_described_in pobj_in_ prep_estimated_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments partmod_alignments_generated xcomp_generated_using nsubj_model_IBM ccomp_using_model dobj_model_2 ' +o,1009,'have been proposed ',Hindle,'aux_proposed_have auxpass_proposed_been ' +o,1010,'Also related are the areas of word alignment for machine translation , induction of translation lexicons , and cross-language annotation projections to a second language ',Mohammad,'advmod_related_Also dep_are_related det_areas_the nsubj_are_areas prep_areas_of nn_alignment_word pobj_of_alignment prep_areas_for nn__machine nn__translation pobj_for_ conj_areas_induction prep_induction_of nn__translation nn__lexicons pobj_of_ cc_areas_and amod_projections_cross-language nn_projections_annotation conj_areas_projections prep_projections_to det_language_a amod_language_second pobj_to_language ' +o,1011,'In Table 6 we report our results , together with the state-of-the-art from the ACL wiki5 and the scores of -LRB- PairClass -RRB- and from Amac Herdagdelens PairSpace system , that was trained on ukWaC ',Turney,'prep_report_In pobj_In_Table num_Table_6 nsubj_report_we poss_results_our dobj_report_results advmod_report_together prep_report_with det_state-of-the-art_the pobj_with_state-of-the-art prep_state-of-the-art_from det_wiki5_the nn_wiki5_ACL pobj_from_wiki5 cc_state-of-the-art_and det_scores_the conj_state-of-the-art_scores prep_scores_of pobj_of_ appos__PairClass cc_with_and conj_with_from nn_system_Amac nn_system_Herdagdelens nn_system_PairSpace pobj_from_system nsubjpass_trained_that auxpass_trained_was rcmod_system_trained prep_trained_on pobj_on_ukWaC ' +o,1012,'The features used in this study are : the length of t ; a single-parameter distortion penalty on phrase reordering in a , as described in ; phrase translation model probabilities ; and 4-gram language model probabilities logp -LRB- t -RRB- , using Kneser-Ney smoothing as implemented in the SRILM toolkit ',Koehn,'det_features_The nsubj_length_features partmod_features_used prep_used_in det_study_this pobj_in_study cop_length_are det_length_the prep_length_of pobj_of_t det_penalty_a amod_penalty_single-parameter nn_penalty_distortion dep_length_penalty prep_penalty_on nn_reordering_phrase pobj_on_reordering prep_reordering_in det_logp_a advmod_described_as amod_logp_described prep_described_in pobj_in_ nn_probabilities_phrase nn_probabilities_translation nn_probabilities_model dep__probabilities cc__and amod_probabilities_4-gram nn_probabilities_language nn_probabilities_model conj__probabilities pobj_in_logp appos_logp_t partmod_reordering_using dobj_using_Kneser-Ney partmod_Kneser-Ney_smoothing mark_implemented_as advcl_smoothing_implemented prep_implemented_in det_toolkit_the nn_toolkit_SRILM pobj_in_toolkit ' +o,1013,'We have computed the BLEU score -LRB- accumulated up to 4-grams -RRB- , the NIST score -LRB- accumulated up to 5-grams -RRB- , the General Text Matching -LRB- GTM -RRB- F-measure -LRB- e = 1,2 -RRB- , and the METEOR measure ',Banerjee,'nsubj_computed_We aux_computed_have det_score_the amod_score_BLEU dobj_computed_score dep_score_accumulated prt_accumulated_up prep_accumulated_to pobj_to_4-grams nsubj__ det_score_the amod_score_NIST appos__score dep__accumulated prt_accumulated_up prep_accumulated_to pobj_to_5-grams dep_computed_ det__the nn_Matching_General nn_Matching_Text measure_F-measure_Matching abbrev_Matching_GTM amod__F-measure nsubj_1,2_e dep_1,2_= dep_F-measure_1,2 dobj__ cc__and det_measure_the amod_measure_METEOR conj__measure ' +o,1014,'As , we adopted an evaluation of mutual information as a cohesion measure of each cooccurrence ',Church,'prep_adopted_As pobj_As_ nsubj_adopted_we det_evaluation_an dobj_adopted_evaluation prep_evaluation_of amod_information_mutual pobj_of_information prep_adopted_as det_measure_a nn_measure_cohesion pobj_as_measure prep_measure_of det_cooccurrence_each pobj_of_cooccurrence ' +o,1015,'The last two counts -LRB- CAUS and ANIM -RRB- were performed on a 29-million word parsed corpus -LRB- gall Street Journal 1988 , provided by Michael Collins -RRB- ',Collins,'det_counts_The amod_counts_last num_counts_two nsubjpass_performed_counts appos_counts_CAUS cc_CAUS_and conj_CAUS_ANIM auxpass_performed_were prep_performed_on det_word_a amod_word_29-million pobj_on_word partmod_word_parsed dobj_parsed_corpus nn_Journal_gall nn_Journal_Street dep_parsed_Journal dep_Journal_1988 dep_Journal_provided prep_provided_by nn__Michael nn__Collins pobj_by_ ' +o,1016,'The tagger used is thus one that does not need tagged and disambiguated material to be trained on , namely the XPOST originally constructed at Xerox Parc ',Cutting,'det_tagger_The nsubj_one_tagger partmod_tagger_used cop_one_is advmod_one_thus nsubj_need_that aux_need_does neg_need_not rcmod_one_need amod_material_tagged cc_tagged_and conj_tagged_disambiguated dobj_need_material aux_trained_to auxpass_trained_be xcomp_need_trained prt_trained_on advmod_one_namely det_XPOST_the dep_one_XPOST advmod_constructed_originally partmod_XPOST_constructed prep_constructed_at nn_Parc_Xerox pobj_at_Parc ' +o,1017,'We perform word alignment using GIZA + + , symmetrize the alignments using the grow-diag-final-and heuristic , and extract phrases up to length 3 ',Och,'nsubj_perform_We nn_alignment_word dobj_perform_alignment xcomp_perform_using nsubj_+_GIZA xcomp_using_+ cc_+_+ conj_+_ conj_perform_symmetrize det_alignments_the nsubj_using_alignments dep_symmetrize_using det_heuristic_the amod_heuristic_grow-diag-final-and dobj_using_heuristic cc_perform_and conj_perform_extract dobj_extract_phrases prt_extract_up prep_extract_to pobj_to_length num_length_3 ' +o,1018,'Congress of the Italian Association for Artificial Intelligence , Palermo , 1991 B Boguraev , Building a Lexicon : the Contribution of Computers , IBM Report , TJ Watson Research Center , 1991 M Brent , Automatic Aquisition of Subcategorization frames from Untagged Texts , in N Calzolari , R Bindi , Acquisition of Lexical Information from Corpus , in K W , P Hanks , Word Association Norms , Mutual Information , and Lexicography , Computational Linguistics , vol ',Church,'nsubj_vol_Congress prep_Congress_of det_Association_the amod_Association_Italian pobj_of_Association prep_Association_for nn_Intelligence_Artificial pobj_for_Intelligence appos_Congress_Palermo num_Boguraev_1991 nn_Boguraev_B appos_Congress_Boguraev appos_Congress_Building det_Lexicon_a dep_Building_Lexicon det_Contribution_the dep_Building_Contribution prep_Contribution_of pobj_of_Computers nn_Report_IBM appos_Congress_Report nn_Center_TJ nn_Center_Watson nn_Center_Research appos_Congress_Center num_Brent_1991 nn_Brent_M appos_Congress_Brent nn_Aquisition_Automatic appos_Congress_Aquisition prep_Aquisition_of amod_frames_Subcategorization pobj_of_frames prep_frames_from nn_Texts_Untagged pobj_from_Texts prep_frames_in nn_Calzolari_ nn_Calzolari_N pobj_in_Calzolari nn_Bindi_R appos_Congress_Bindi appos_Congress_Acquisition prep_Acquisition_of nn_Information_Lexical pobj_of_Information prep_Information_from pobj_from_Corpus prep_Information_in nn__ nn__K nn__W pobj_in_ nn_Hanks_P conj__Hanks nn_Norms_Word nn_Norms_Association conj__Norms nn_Information_Mutual conj__Information cc__and conj__Lexicography nn_Linguistics_Computational appos_Congress_Linguistics ' +o,1019,'4 Maximum Entropy To explain our method , we l -RRB- riefly des -LRB- : ribe the con -LRB- : ept of maximum entrol -RRB- y Recently , many al -RRB- lnoaches l -RRB- ased on the maximum entroi -RRB- y lnodel have t -RRB- een applied to natural language processing ',Berger,'num_Maximum_4 measure_Entropy_Maximum aux_explain_To dep_Entropy_explain poss_method_our dobj_explain_method measure_l_we amod_method_l advmod_des_riefly amod_een_des dep_-LRB-_ribe det_con_the dep_ribe_con dep_con_ept prep_ept_of amod_y_maximum amod_y_entrol pobj_of_y advmod_y_Recently amod_lnoaches_many amod_lnoaches_al appos_ept_lnoaches dep_lnoaches_l amod_con_ased prep_con_on det_entroi_the amod_entroi_maximum pobj_on_entroi nn_lnodel_y nsubj_t_lnodel aux_t_have dep_des_t dep_method_een partmod_een_applied prep_applied_to amod_processing_natural nn_processing_language pobj_to_processing ' +o,1020,'These are most directly presented in ',Ostler,'nsubjpass_presented_These auxpass_presented_are advmod_directly_most advmod_presented_directly prep_presented_in ' +o,1021,'Modeling reordering as the inversion in order of two adjacent blocks is similar to the approach taken by the Inverse Transduction Model -LRB- ITG -RRB- , except that here we are not limited to a binary tree ',Wu,'dobj_Modeling_reordering prep_Modeling_as det_inversion_the pobj_as_inversion prep_inversion_in pobj_in_order prep_order_of num_blocks_two amod_blocks_adjacent pobj_of_blocks cop_similar_is dep_Modeling_similar prep_similar_to det_approach_the pobj_to_approach partmod_approach_taken prep_taken_by det_Inverse_the pobj_by_Inverse nn__Transduction nn__Model abbrev__ITG nsubj_Modeling_ mark_limited_except dep_limited_that advmod_limited_here nsubjpass_limited_we auxpass_limited_are neg_limited_not dep__limited prep_limited_to det_tree_a amod_tree_binary pobj_to_tree ' +o,1022,'EM-HMM tagger provided with good initial conditions 914 \\* -LRB- \\* uses linguistic constraints and manual adjustments to the dictionary -RRB- Figure 1 : Previous results on unsupervised POS tagging using a dictionary on the full 45-tag set ',Goldberg,'nn_tagger_EM-HMM nsubj_provided_tagger prep_provided_with amod_conditions_good amod_conditions_initial pobj_with_conditions number_914_ dobj_provided_914 prep_914_\\* dep_-LRB-_\\* dep_\\*_uses amod_constraints_linguistic dobj_uses_constraints cc_constraints_and amod_adjustments_manual conj_constraints_adjustments prep_uses_to det_dictionary_the pobj_to_dictionary pobj_\\*_Figure num_Figure_1 amod_results_Previous dep_Figure_results prep_results_on amod_POS_unsupervised pobj_on_POS partmod_POS_tagging xcomp_tagging_using det__a nn__dictionary dobj_using_ prep_using_on det_set_the amod_set_full amod_set_45-tag pobj_on_set ' +p,1023,'In Statistical Machine Translation -LRB- SMT -RRB- , recent work shows that WSD helps translation quality when the WSD system directly uses translation candidates as sense inventories ',Carpuat,'prep_shows_In nn_Translation_Statistical nn_Translation_Machine pobj_In_Translation abbrev_Translation_SMT amod_work_recent nsubj_shows_work complm_helps_that nsubj_helps_WSD ccomp_shows_helps nn_quality_translation dobj_helps_quality advmod_uses_when det_system_the nn_system_WSD nsubj_uses_system advmod_uses_directly dep_quality_uses nn_candidates_translation dobj_uses_candidates prep_uses_as nn_inventories_sense pobj_as_inventories ' +o,1024,'Dialogs Speakers Turns Words Fragments Distinct Words Distinct Words\\/POS Singleton Words Singleton Words\\/POS Intonational Phrases Speech Repairs 98 34 6163 58298 756 859 1101 252 350 1094 7 2396 Table 1 : Size of the Trains Corpus 21 POS Annotations Our POS tagset is based on the Penn Treebank tagset , but modified to include tags for discourse markers and end-of-turns , and to provide richer syntactic information ',Marcus,'amod_Speakers_Dialogs nsubj_Turns_Speakers nn_Repairs_Words nn_Repairs_Fragments nn_Repairs_Distinct nn_Repairs_Words nn_Repairs_Distinct nn_Repairs_Words\\/POS nn_Repairs_Singleton nn_Repairs_Words nn_Repairs_Singleton nn_Repairs_Words\\/POS nn_Repairs_Intonational nn_Repairs_Phrases nn_Repairs_Speech dobj_Turns_Repairs num_Table_98 num_Table_34 num_Table_6163 num_Table_58298 num_Table_756 859 1101 num_Table_252 350 1094 num_Table_7 nn_Table_2396 dep_Turns_Table dep_Table_1 dep_Table_Size prep_Size_of det_Annotations_the nn_Annotations_Trains nn_Annotations_Corpus num_Annotations_21 dep_Annotations_POS pobj_of_Annotations poss_tagset_Our dep_tagset_POS nsubjpass_based_tagset auxpass_based_is dep_Turns_based prep_based_on det__the nn__Penn nn__Treebank amod__tagset pobj_on_ cc_based_but conj_based_modified aux_include_to ccomp_modified_include dobj_include_tags prep_tags_for nn_markers_discourse pobj_for_markers cc_markers_and conj_markers_end-of-turns cc_include_and aux_provide_to conj_include_provide amod__richer amod__syntactic nn__information dobj_provide_ ' +p,1025,'Most semi-automated approaches have met with limited success and supervised learning models have tended to outperform dictionary-based classi cation schemes ',Pang,'amod_approaches_Most amod_approaches_semi-automated nsubj_met_approaches aux_met_have prep_met_with amod__limited nn__success pobj_with_ cc_met_and conj_met_supervised xcomp_supervised_learning nsubj_tended_models aux_tended_have ccomp_learning_tended aux_outperform_to xcomp_tended_outperform amod_schemes_dictionary-based nn_schemes_classi nn_schemes_cation dobj_outperform_schemes ' +n,1026,'For comparison purposes , we revisit a fullygenerative Bayesian model for unsupervised coreference resolution recently introduced by , discuss its potential weaknesses and consequently propose three modifications to their model -LRB- Section 3 -RRB- ',Haghighi,'prep_revisit_For nn_purposes_comparison pobj_For_purposes nsubj_revisit_we det_model_a amod_model_fullygenerative nn_model_Bayesian dobj_revisit_model prep_model_for amod_resolution_unsupervised nn_resolution_coreference pobj_for_resolution advmod_introduced_recently partmod_resolution_introduced prep_introduced_by pobj_by_ dep_revisit_discuss poss_weaknesses_its amod_weaknesses_potential dobj_discuss_weaknesses cc_discuss_and advmod_propose_consequently conj_discuss_propose num_modifications_three dobj_propose_modifications prep_propose_to poss_model_their pobj_to_model appos_model_Section num_Section_3 ' +o,1027,'html 162 311 Penn Treebank 3 The Penn Treebank 3 corpus consists of hand-coded parses of the Wall Street Journal -LRB- test , development and training -RRB- and a small subset of the Brown corpus -LRB- test only -RRB- ',Marcus,'amod__html num__162 nn__311 nn__Penn nn__Treebank num__3 nn__The nn__Penn nn__Treebank num__3 nn__corpus nsubj_consists_ prep_consists_of amod_parses_hand-coded pobj_of_parses prep_parses_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal dep_Journal_test appos_test_development cc_development_and conj_development_training cc_parses_and det_subset_a amod_subset_small conj_parses_subset prep_subset_of det__the nn__Brown nn__corpus pobj_of_ dep__test advmod_test_only ' +o,1028,'Default parameters were used for all experiments except for the numberofiterationsforGIZA + ',Och,'amod_parameters_Default nsubjpass_used_parameters auxpass_used_were prep_used_for det_experiments_all pobj_for_experiments prep_used_except dep_except_for det_+_the nn_+_numberofiterationsforGIZA pobj_for_+ ' +o,1029,'The baseline we measure against in all of these experiments is the state-of-the-art grow-diag-final -LRB- gdf -RRB- alignment refinement heuristic commonly used in phrase-based SMT ',Koehn,'det_baseline_The nsubj_heuristic_baseline nsubj_measure_we rcmod_baseline_measure dep_in_against prep_measure_in pobj_in_all prep_all_of det_experiments_these pobj_of_experiments cop_heuristic_is det_heuristic_the amod_grow-diag-final_state-of-the-art nn_heuristic_grow-diag-final appos_grow-diag-final_gdf nn_heuristic_alignment nn_heuristic_refinement advmod_used_commonly partmod_heuristic_used prep_used_in amod_SMT_phrase-based pobj_in_SMT ' +o,1030,' , 1993 -RRB- : 1 ',Smadja,'amod__1993 dep__1 ' +o,1031,'This decomposition applies both to discriminative linear models and to generative models such as HMMs and CRFs , in which case the linear sum corresponds to log likelihood assigned to the input\\/output pair by the model -LRB- for details see for the classi cation case and for the structured case -RRB- ',Collins,'det_decomposition_This nsubj_applies_decomposition preconj_discriminative_both aux_discriminative_to ccomp_applies_discriminative amod_models_linear dobj_discriminative_models cc_discriminative_and aux_generative_to conj_discriminative_generative dobj_generative_models dep_as_such prep_models_as pobj_as_HMMs cc_HMMs_and conj_HMMs_CRFs rel_corresponds_in det_case_which pobj_in_case det_sum_the amod_sum_linear nsubj_corresponds_sum rcmod_HMMs_corresponds aux_log_to xcomp_corresponds_log dobj_log_likelihood partmod_likelihood_assigned prep_assigned_to det_pair_the amod_pair_input\\/output pobj_to_pair prep_assigned_by det_model_the pobj_by_model mark_see_for nsubj_see_details dep_model_see ccomp_see_ prep__for det_case_the nn_case_classi nn_case_cation pobj_for_case cc__and conj__ prep__for det_case_the amod_case_structured pobj_for_case ' +o,1032,'So far , pivot features on the word level were used , eg Does the bigram not buy occur in this document ? ',Blitzer,'dep_far_So advmod_used_far amod_features_pivot nsubjpass_used_features prep_features_on det_level_the nn_level_word pobj_on_level auxpass_used_were ccomp_Does_used dep_used_ nsubj_Does_eg det_bigram_the nsubj_buy_bigram neg_buy_not ccomp_Does_buy dep_buy_occur prep_occur_in det_document_this pobj_in_document ' +o,1033,'In order to get a better understanding of these matters , we replicate parts of the error analysis presented by , where parsing errors are related to different structural properties of sentences and their dependency graphs ',McDonald,'mark_get_In dep_get_order aux_get_to dep_replicate_get det_understanding_a amod_understanding_better dobj_get_understanding prep_understanding_of det_matters_these pobj_of_matters nsubj_replicate_we dobj_replicate_parts prep_parts_of det_analysis_the nn_analysis_error pobj_of_analysis partmod_analysis_presented prep_presented_by pobj_by_ advmod_related_where amod_errors_parsing nsubjpass_related_errors auxpass_related_are rcmod__related prep_related_to amod_properties_different amod_properties_structural pobj_to_properties prep_properties_of pobj_of_sentences cc_sentences_and poss_graphs_their amod_graphs_dependency conj_sentences_graphs ' +o,1034,'Dubey et al proposed an unlexicalized PCFG parser that modied PCFG probabilities to condition the existence of syntactic parallelism ',Dubey,'nsubj_proposed_Dubey cc_Dubey_et conj_Dubey_al det_parser_an amod_parser_unlexicalized nn_parser_PCFG dobj_proposed_parser nsubj_modied_that rcmod_parser_modied xcomp_modied_PCFG dobj_PCFG_probabilities prep_PCFG_to pobj_to_condition det_existence_the tmod_PCFG_existence prep_existence_of amod_parallelism_syntactic pobj_of_parallelism ' +p,1035,'While studies have shown that ratings of MT systems by BLEU and similar metrics correlate well with human judgments , we are not aware of any studies that have shown that corpus-based evaluation metrics of NLG systems are correlated with human judgments ; correlation studies have been made of individual components , but not of systems ',Papineni,'mark_shown_While nsubj_shown_studies aux_shown_have advcl_aware_shown complm_correlate_that nsubj_correlate_ratings prep_ratings_of nn_systems_MT pobj_of_systems prep_systems_by amod_metrics_BLEU cc_BLEU_and conj_BLEU_similar pobj_by_metrics ccomp_shown_correlate advmod_correlate_well prep_correlate_with amod__human nn__judgments pobj_with_ nsubj_aware_we cop_aware_are neg_aware_not prep_aware_of det_studies_any pobj_of_studies nsubj_shown_that aux_shown_have rcmod_studies_shown complm_correlated_that amod_metrics_corpus-based nn_metrics_evaluation nsubjpass_correlated_metrics prep_metrics_of nn_systems_NLG pobj_of_systems auxpass_correlated_are ccomp_shown_correlated prep_correlated_with amod_judgments_human pobj_with_judgments nn_studies_correlation nsubjpass_made_studies aux_made_have auxpass_made_been parataxis_aware_made prep_made_of amod__individual nn__components pobj_of_ cc_of_but conj_of_not dep_of_of pobj_of_systems ' +o,1036,'We use the cosine similarity measure for windowbased contexts and the following commonly used similarity measures for the syntactic vector space : measure , the weighted Lin measure , the - Skew divergence measure , the Jensen-Shannon -LRB- JS -RRB- divergence measure , Jaccards coef cient and the Confusion probability ',Hindle,'nsubj_use_We det_measure_the amod_measure_cosine nn_measure_similarity dobj_use_measure prep_measure_for amod_contexts_windowbased pobj_for_contexts cc_measure_and det_following_the conj_measure_following advmod_used_commonly partmod_following_used nn_measures_similarity dobj_used_measures prep_measures_for det_space_the amod_space_syntactic nn_space_vector pobj_for_space amod_measure_ dep_following_measure det__the amod__weighted nn__Lin nn__measure conj_measure_ det__the nn__Skew nn__divergence nn__measure conj_measure_ det_measure_the amod_measure_Jensen-Shannon abbrev_measure_JS nn_measure_divergence conj_measure_measure amod__ amod__Jaccards nn__coef nn__cient dep_measure_ cc_measure_and det__the nn__Confusion nn__probability conj_measure_ ' +o,1037,'Note that this early discarding is related to ideas behind cube pruning , which generates the top n most promising hypotheses , but in our method the decision not to generate hypotheses is guided by the quality of hypotheses on the result stack ',Huang,'complm_related_that det_discarding_this amod_discarding_early nsubjpass_related_discarding auxpass_related_is ccomp_Note_related prep_related_to pobj_to_ideas prep_ideas_behind nn__cube nn__pruning pobj_behind_ nsubj_generates_which rcmod__generates det_n_the amod_n_top nsubj_hypotheses_n advmod_hypotheses_most amod_hypotheses_promising xcomp_generates_hypotheses cc_generates_but prep_decision_in poss_method_our pobj_in_method det_decision_the conj_generates_decision neg_generate_not aux_generate_to infmod_decision_generate nsubjpass_guided_hypotheses auxpass_guided_is ccomp_generate_guided prep_guided_by det_quality_the pobj_by_quality prep_quality_of pobj_of_hypotheses prep_hypotheses_on det_stack_the nn_stack_result pobj_on_stack ' +o,1038,'Examples of such methods are the introduction of information weights as in the NIST measure or the comparison of stems or synonyms , as in METEOR ',Banerjee,'nsubj_introduction_Examples prep_Examples_of amod_methods_such pobj_of_methods cop_introduction_are det_introduction_the prep_introduction_of nn_weights_information pobj_of_weights dep_in_as prep_introduction_in det_measure_the amod_measure_NIST pobj_in_measure cc_measure_or det_comparison_the conj_measure_comparison prep_comparison_of pobj_of_stems cc_stems_or conj_stems_synonyms dep_in_as dep_in_in pobj_in_METEOR ' +o,1039,'Supervision for simple features has been explored in the literature ',Haghighi,'nsubjpass_explored_Supervision prep_Supervision_for amod_features_simple pobj_for_features aux_explored_has auxpass_explored_been prep_explored_in det_literature_the pobj_in_literature ' +o,1040,'In general , they can be divided into two major categories , namely lexicalized models and un-lexicalized models ',Matsuzaki,'prep_divided_In pobj_In_general nsubjpass_divided_they aux_divided_can auxpass_divided_be prep_divided_into num_categories_two amod_categories_major pobj_into_categories advmod_models_namely amod_models_lexicalized conj_categories_models amod_models_ cc_categories_and amod_models_un-lexicalized conj_categories_models ' +o,1041,'1 Introduction Many state-of-the-art machine translation -LRB- MT -RRB- systems over the past few years rely on several models to evaluate the goodness of a given candidate translation in the target language ',Och,'num_translation_1 amod_translation_Introduction amod_translation_Many amod_translation_state-of-the-art nn_translation_machine abbrev_translation_MT nsubj_rely_systems prep_systems_over det__the amod__past amod__few nn__years pobj_over_ dep_translation_rely prep_rely_on amod_models_several pobj_on_models aux_evaluate_to xcomp_rely_evaluate det_goodness_the dobj_evaluate_goodness prep_goodness_of det_translation_a amod_translation_given nn_translation_candidate pobj_of_translation prep_translation_in det_language_the nn_language_target pobj_in_language ' +o,1042,'Initial results show the potential benefit of factors for statistical machine translation , and ',Koehn,'amod_results_Initial nsubj_show_results det_benefit_the amod_benefit_potential dobj_show_benefit prep_benefit_of pobj_of_factors prep_factors_for amod_translation_statistical nn_translation_machine pobj_for_translation nn_and_ appos_translation_and ' +p,1043,'While EM has worked quite well for a few tasks , notably machine translations -LRB- starting with the IBM models 1-5 , it has not had success in most others , such as part-of-speech tagging , named-entity recognition and context-free-grammar induction -LRB- numerous attempts , too many to mention -RRB- ',Brown,'mark_worked_While nsubj_worked_EM aux_worked_has advmod_well_quite advmod_worked_well dep_well_for det_tasks_a amod_tasks_few pobj_for_tasks advmod_tasks_notably nn_translations_machine dep_tasks_translations parataxis_worked_starting prep_starting_with det_models_the nn_models_IBM pobj_with_models num__1-5 dobj_starting_ nsubj_had_it aux_had_has neg_had_not dep__had dobj_had_success prep_success_in amod_others_most pobj_in_others dep_as_such prep__as amod__part-of-speech amod__tagging pobj_as_ amod__named-entity nn__recognition conj__ cc__and amod_induction_context-free-grammar conj__induction amod_attempts_numerous dep_induction_attempts advmod_many_too amod_induction_many prep_induction_to pobj_to_mention ' +o,1044,'Part-of-speech tagging is an active area of research ; a great deal of work has been done in this area over the past few years ',Cutting,'nn_tagging_Part-of-speech nsubj_area_tagging cop_area_is det_area_an amod_area_active prep_area_of pobj_of_research det_deal_a amod_deal_great nsubjpass_done_deal prep_deal_of pobj_of_work aux_done_has auxpass_done_been parataxis_area_done prep_done_in det_area_this pobj_in_area prep_area_over det_years_the amod_years_past amod_years_few pobj_over_years ' +n,1045,'Unfortunately , there is no straightforward generalization of the method of to the two edge marginal problem ',Smith,'advmod_is_Unfortunately expl_is_there det_generalization_no amod_generalization_straightforward nsubj_is_generalization prep_generalization_of det_method_the pobj_of_method prep_method_of amod_problem_ prep__to det_edge_the num_edge_two pobj_to_edge amod_problem_marginal pobj_of_problem ' +o,1046,'Evaluation We evaluate translation output using three automatic evaluation measures : BLEU , NIST , and METEOR 5 All measures used were the case-sensitive , corpuslevel versions ',Banerjee,'aux_evaluate_Evaluation nsubj_evaluate_We nn_output_translation dobj_evaluate_output partmod_output_using num_measures_three amod_measures_automatic nn_measures_evaluation dobj_using_measures nn__BLEU dep_output_ nn__NIST appos__ cc_output_and conj_output_METEOR number_5_ num_measures_5 det_measures_All nsubj_versions_measures partmod_measures_used cop_versions_were det_versions_the amod_versions_case-sensitive amod_versions_corpuslevel rcmod_METEOR_versions ' +o,1047,'In the early statistical translation model work at IBM , these representations were called cepts , short for concepts ',Brown,'prep_called_In det_work_the amod_work_early amod_work_statistical nn_work_translation nn_work_model pobj_In_work prep_work_at pobj_at_IBM det_representations_these nsubjpass_called_representations auxpass_called_were dep_called_cepts acomp_called_short prep_short_for pobj_for_concepts ' +o,1048,'We adopt an approach , similar to , in which the meaning representation , in our case XML , is transformed into a sorted flat list of attribute-value pairs indicating the core contentful concepts of each command ',Brown,'nsubj_adopt_We det_approach_an dobj_adopt_approach amod_approach_similar dep_similar_to dep_similar_ rel_transformed_in pobj_in_which det_representation_the amod_representation_meaning nsubjpass_transformed_representation prep_representation_in poss_XML_our nn_XML_case pobj_in_XML auxpass_transformed_is rcmod_approach_transformed prep_transformed_into det_list_a amod_list_sorted amod_list_flat pobj_into_list prep_list_of amod_pairs_attribute-value pobj_of_pairs xcomp_transformed_indicating det_concepts_the amod_concepts_core amod_concepts_contentful dobj_indicating_concepts prep_concepts_of det_command_each pobj_of_command ' +o,1049,'An alternative to tercom , considered in this paper , is to use the Inversion Transduction Grammar -LRB- ITG -RRB- formalism which allows one to view the problem of alignment as a problem of bilingual parsing ',Wu,'det_alternative_An nsubj_is_alternative prep_alternative_to pobj_to_tercom partmod_alternative_considered prep_considered_in det_paper_this pobj_in_paper aux_use_to xcomp_is_use det__the nn_Grammar_Inversion nn_Grammar_Transduction nn__Grammar abbrev_Grammar_ITG nn__formalism dobj_use_ nsubj_allows_which rcmod__allows dobj_allows_one aux_view_to xcomp_allows_view det_problem_the dobj_view_problem prep_problem_of pobj_of_alignment prep_view_as det_problem_a pobj_as_problem prep_problem_of amod_parsing_bilingual pobj_of_parsing ' +o,1050,'We obtain weights for the combinations of the features by performing minimum error rate training on held-out data ',Och,'nsubj_obtain_We dobj_obtain_weights prep_weights_for det_combinations_the pobj_for_combinations prep_combinations_of det_features_the pobj_of_features prep_obtain_by pcomp_by_performing amod__minimum nn__error nn__rate nn__training dobj_performing_ prep_performing_on amod_data_held-out pobj_on_data ' +o,1051,'Actually , it is defined similarly to the translation model in SMT ',Koehn,'advmod_defined_Actually nsubjpass_defined_it auxpass_defined_is advmod_defined_similarly prep_defined_to det_model_the nn_model_translation pobj_to_model prep_model_in pobj_in_SMT ' +o,1052,'2 Related Work There has been extensive research in opinion mining at the document level , for example on product and movie reviews ',Pang,'num_Work_2 amod_Work_Related expl_been_There aux_been_has dep_Work_been amod_research_extensive dobj_been_research prep_research_in nn_mining_opinion pobj_in_mining prep_mining_at det_level_the nn_level_document pobj_at_level prep_been_for pobj_for_example prep_been_on nn_reviews_product cc_product_and conj_product_movie pobj_on_reviews ' +o,1053,'2 Experimental System and Data HMIHY is a spoken dialogue system based on the notion of call routing ',Chu-Carroll,'num_System_2 nn_System_Experimental nsubj_system_System cc_System_and nn_HMIHY_Data conj_System_HMIHY cop_system_is det_system_a amod_system_spoken nn_system_dialogue prep_system_based dep_based_on det_notion_the pobj_on_notion prep_notion_of nn_routing_call pobj_of_routing ' +o,1054,'Our results on Chinese data confirm previous findings on English data shown in ',McClosky,'poss_results_Our nsubj_confirm_results prep_results_on amod_data_Chinese pobj_on_data amod_findings_previous dobj_confirm_findings prep_confirm_on amod_data_English pobj_on_data partmod_data_shown prep_shown_in ' +n,1055,'Previous literature on GB parsing \\/ Wehrli , 1984 ; Sharp , 1985 ; , 1986 ; Kuhns , 1986 ; Abney , 1986\\/has not addressed the issue of implementation of the Binding theory -RRB- The present paper intends in part to fill this gap ',Kashket,'amod_literature_Previous nsubj_parsing_literature prep_literature_on pobj_on_GB nn_Wehrli_\\/ dobj_parsing_Wehrli appos_Wehrli_1984 dep_Wehrli_Sharp appos_Sharp_1985 dep_Wehrli_ appos__1986 dep_Wehrli_Kuhns appos_Kuhns_1986 dep_Wehrli_Abney appos_Abney_1986\\/has neg_1986\\/has_not partmod_1986\\/has_addressed det_issue_the dobj_addressed_issue prep_issue_of pobj_of_implementation prep_implementation_of det_theory_the nn_theory_Binding pobj_of_theory det_paper_The amod_paper_present nsubj_intends_paper dep_parsing_intends prep_intends_in pobj_in_part aux_fill_to xcomp_intends_fill det_gap_this dobj_fill_gap ' +o,1056,'We use the log-likelihood ratio for determining significance as in , but other measures are possible as well ',Dunning,'nsubj_use_We det_ratio_the amod_ratio_log-likelihood nsubj_possible_ratio prep_ratio_for pcomp_for_determining dobj_determining_significance dep_in_as prep_determining_in pobj_in_ cc_in_but amod_measures_other conj_in_measures cop_possible_are ccomp_use_possible advmod_well_as advmod_possible_well ' +o,1057,'First , word frequencies , context word frequencies in surrounding positions -LRB- here three-words window -RRB- are computed following a statistics-based metrics , the log-likelihood ratio ',Dunning,'advmod_computed_First nn_frequencies_word nsubjpass_computed_frequencies nn_frequencies_context nn_frequencies_word nsubjpass_computed_frequencies prep_frequencies_in amod_positions_surrounding pobj_in_positions advmod_window_here amod_window_three-words appos_positions_window auxpass_computed_are prep_computed_following det_metrics_a amod_metrics_statistics-based pobj_following_metrics det_ratio_the amod_ratio_log-likelihood appos_metrics_ratio ' +o,1058,'The two annotators agreed on the annotations of 385\\/453 turns , achieving 8499 \% agreement , with Kappa = 0682 This inter-annotator agreement exceeds that of prior studies of emotion annotation in naturally occurring speech 2a3a5a4a7a6a8a6a9a4a11a10a13a12a15a14a17a16a19a18a21a20a22a12a23a14a25a24a26a18 a27 a20a22a12a23a14a25a24a26a18 ',Carletta,'det_annotators_The num_annotators_two nsubj_agreed_annotators prep_agreed_on det_annotations_the pobj_on_annotations prep_annotations_of num_turns_385\\/453 pobj_of_turns xcomp_agreed_achieving num_\%_8499 dobj_achieving_\% nsubj_exceeds_agreement mark_=_with nsubj_=_Kappa dep_agreement_= iobj_=_0682 det_agreement_This amod_agreement_inter-annotator dobj_=_agreement dep_agreed_exceeds dobj_exceeds_that prep_that_of amod_studies_prior pobj_of_studies prep_studies_of nn_annotation_emotion pobj_of_annotation prep_exceeds_in advmod_occurring_naturally pcomp_in_occurring nn_a20a22a12a23a14a25a24a26a18_speech nn_a20a22a12a23a14a25a24a26a18_2a3a5a4a7a6a8a6a9a4a11a10a13a12a15a14a17a16a19a18a21a20a22a12a23a14a25a24a26a18 nn_a20a22a12a23a14a25a24a26a18_a27 dobj_occurring_a20a22a12a23a14a25a24a26a18 ' +o,1059,',.~.eqmvalent ot duty in a parallel French text, the correct sense of the Enghsh word is identified These studies exploit th~s lnformatmn m order to gather co-occurrence data for the different senses, which ts then used to dtsamb~guate new texts In related work, Dywk (1998) used patterns of translational relatmns in an EnghshNorwegian paralle ! corpus (ENPC, Oslo Umverslty) to define semantic propemes such as synonymy, ambtgmty, vagueness, and semantic helds and suggested a derivation otsemantic representations for signs (eg, lexemes), captunng semantm relatmnshlps such as hyponymy etc, fiom such translatmnal relatmns Recently, Resnlk and Yarowsky (1997) suggested that fol the purposes ot WSD, the different senses of a wo~d could be detelmlned by considering only sense d~stmctmns that are lextcahzed cross-hngmstlcally In particular, they propose that some set of target languages be ~dent~fied, and that the sense d~stmctmns to be considered for language processing appllcatmns and evaluatmn be restricted to those that are reahzed lexlcally in some minimum subset of those languages This idea would seem to p~ovtde an answer, at least m part, to the problem of determining different senses of a word mtumvely, one assumes that ff another language lexlcahzes a word m two or more ways, there must be a conceptual monvatmn If we look at enough languages, we would be likely to fred the s~gmficant lexlcal differences that dehmtt different senses of a word However, th~s suggestmn raises several questions Fo~ instance, ~t ~s well known that many amb~gumes are preserved across languages (for example, the French tntdrYt and the Enghsh interest), especmlly languages that are relatively closely related Assuming this problem can be overcome, should differences found m closely related languages be given lesser (or greater) weight than those found m more distantly related languages 9 More generally, which languages should be considered for this exermse 9 All languages 9 Closely related languages9 Languages from different language famlhes \'~ A mixture of the two 9 How many languages, and of which types, would be \'enough\' to provide adequate lnfotmanon tot this purpose~ There ts also the questmn ot the crlterm that would be used to estabhsh that a sense distinction is \'lexlcahzed cross-hngu~stmally\' How consistent must the d~stlnCtlOn be 9 Does it mean that two concepts are expressed by mutually non-lntetchangeable lexmal items in some slgmficant number ot other languages, or need tt only be the case that the option ot a different lexlcahzatlon exists m a certain percentage of cases 9 Another conslderatmn ts where the cross-hngual mformatlon to answer these questmns would come from Using bdmgual dictionaries would be extremely tedmus and error-prone, g~ven the substantial d~vergence among d~ctlonanes in terms of the kinds and degree of sense dlstmctmns they make Resmk and Yalowsky (1997) suggest EutoWordNet (Vossen, 1998) as a possible somce of mformatmn, but, given that EuroWordNet ts pttmatdy a lexmon and not a corpus, ~t is subject to many of the same objections as for bl-hngual dictionaries An alternative would be to gather the reformation from parallel, ahgned corpma Unlike bilingual and muttt-hngual dictionaries, translatmn eqmvalents xn parallel texts a~e determined by experienced translatols, who evaluate each instance ot a word\'s use m context rather than as a part of the meta-hngmst~c actlvlty of classifying senses for mclusmn in a dictionary However, at present very few parallel ahgned corpora exist The vast majority ot these are bl-texts, mvolwng only two languages, one of which is very often English Ideally, a serious 53 evaluation of Resnik and Yarowsky\'s proposal would include parallel texts m languages from several different language families, and, to maximally ensure that the word m question is used in the exact same sense across languages, ~t would be preferable that the same text were used over all languages in the study The only currently avadable parallel corpora for more than two languages are Olwell\'s Nmeteen Eighty-Four (Erjavec and Ide, 1998), Plato\'s Repubhc (Erjavec, et al, 1998), the MULTEXT Journal .o/ the Commt.~ston corpus (Ide and V6roms, 1994), and the Bible (Resnlk, et al, m press) It is likely that these corpora do not provide enough appropriate data to reliably determine sense distinctions Also, ~t Is not clear how the lexlcahzatlon of sense distractions across languages Is affected by genre, domain, style, etc Thls paper attempts to provide some prehmlnary answers to the questions outhned above, In order to eventually determine the degree to which the use of parallel data ts vmble to determine sense distinctions, and, ff so, the ways in which th~s reformation might be used Given the lack of lalge parallel texts across multiple languages, the study is necessarily hmlted, however, close exammanon of a small sample of parallel data can, as a first step, provide the basis and dlrectmn for more extensive studies 1 Methodology I have conducted a small study using parallel, aligned versmns ot George Orwell\'s Nineteen Etghtv-Fo,lr (Euavec and Ide, 1998)m five languages Enghsh, Slovene, Estonian, Romanlan, and Czech I The study therefole Involves languages from four language families The O~well parallel corpus also includes vers|ons o) Ntneteen-E~gho Four m Hungarian, Bulgarmn, Latwan, Llthuaman, Se~bmn, and Russmn (Germanic, Slavic, Fmno-Ugrec, and Romance), two languages from the same family (Czech and Slovene), as well as one non-Indo-European language (Estoman) Nmeteen Eighty-Four Is a text of about 100,000 words, translated directly from the original English m each of the other languages The parallel versions of the text are sentence-aligned to the English and tagged for part of speech Although Nineteen Eighty-Four is a work of fiction, Orwell\'s prose IS not highly stylized and, as such, it provides a reasonable sample ot modern, ordinary language that ~s not tied to a given topic or sub-domain (such as newspapers, technical reports, etc ) Furthermore, the translations of the text seem to be relatively faithful to the original for instance, over 95\% ot the sentence alignments in the full pmallel corpus of seven languages are one-to-one (Prlest-Dorman, et al, 1997) Nine ambiguous English words were considered hard, head, country, hne, promise, shght, seize, scrap, float The first four were chosen because they have been used in other dlsamb~guatlon studies, the latter five were chosen from among the words used m the Senseval dlsamblguatlon exercise (Kllgamff and Palmer, forthcoming) In all cases, the study was necessarily hmlted to words that occurred frequently enough in the Orwell text to warrant consideration F~ve hundred forty-two sentences conta|nmg an occurrence or occurrences (Including morphological variants) of each of the nine words were extracted from the Enghsh text, together w~th the parallel sentences m which they occur m the texts ot the four comparison languages (Czech, Estonian, Romantan, Slovene) As Walks and Stevenson (1998) have pointed out, pa~t-of-speech tagging accomplishes a good portion of the work ot semantic dlsamb~guatmn, therefore occmrences of wolds that appemed in the data in more than 54 one part of speech were grouped separately 2 The Enghsh occurrences were then grouped usmg the sense distinctions m WordNet, (version 1 6) [Miller et al, 1990, Fellbaum, 1998]) The sense categonzatmn was performed by the author and two student assistants, results from the three were compared and a final, mutually agreeable set of sense assignments was estabhshed For each of the four comparison languages, the corpus of sense-grouped parallel sentences were sent to a llngmst and natl,ve speaker of the comparison language The hngmsts were asked to provide the lexlcal item m each parallel sentence that corresponds to the ambiguous Enghsh word If inflected, they were asked to provide both the inflected form and the root form In addttmn, the lmgmsts were asked to indicate the type of translatmn, according to the dtstmctmns given m Table 1 For over 85\% of the Enghsh word occurrences (corresponding to types 1 and 2 m Table 1), a specific lexlcal item or items could be identified as the translation equivalent for the corresponding Enghsh word For comparison purposes, each translanon equivalent was represented by ~ts lemma (or the lemma of the toot form in the case of derivatives) and associated w~th the WordNet sense to which it corresponds In order to determine the degree to which the assigned sense dlstlncttons correspond to translation eqmvalents, a coherence index ( Cl) was computed that measures how often each pmr of senses is translated usmg the same word as well as the consistency with which a g~ven se,ls,z ~s translated with the same word ~ Note that the z The adJective and adverb senses of hard are consadeied together because the distinction is not consistent across the translations used m the study Note that the CI ~s similar to semanuc entropy (Melamed, 1997) However, Melamed computes CIs do not determine whether or not a sense dtstmctton can be lextcahzed in the target language, but only the degree to whmh they are lexicahzed differently m the translated text However, tt can be assumed that the CIs provide a measure of the tendency to lex~cahze different WordNet senses differently, which can m turn be seen as an mdtcatmn of the degree to which the distraction ts vahd For each ambiguous word, the CI Is computed for each pair of senses, as follows S Cl(sqS, ) = \'=1 m rnrt where @ n ~s the number of comparison languages under consideration, nl~q and m,, are the nt~mber of occurrences olsense sqand sense s~ m the Enghsh corpus, respectively, including occurrences that have no idenufiable translation, s<~ ~>m ts the number of times that senses q and r are translated by the same lex~cal Item m language t, i e, x=y t ~tJan ~( q ), r~oan~( r ) The CI ts a value between 0 and 1, computed by examining clusters of occurrences translated by the same word In the othel languages If sense and sense ) are consistently translated w~th the same wo~d in each comparison language, then Cl(s, s~) = 1, if they are translated with a different word m every occurrence, Cl(s, ~) = 0 In general, the CI for pans of different senses provides an index of thmr relatedness, t e, the greater the value of Cl(s, sj), the more frequently occurrences of-sense t and sense j are translated with the same lextcal item When t = j, we entropy tOl wold types, lather than word senses 55 obtain a measure of the coherence of a ~lven sense Type Meaning 1 A slngle lexlcal Item is used to translate the En@izsh equivalent (possibly a 2 The English word is translated by a phrase of two or more words or a compound, meaning as the slngle English word 3 The En@izsh word is not lexzcalized in the translation 4 A pronoun is substituted for the English word In the translation An English phrase contalnmng the ambiguous word Is translated by a single language which has a broader or more specific meanlng, or by a phrase in whl corresponding to the English word Is not explicltl~ lexlcallzed Table 1 Translation types and their trequencles \% dizen whl\%h h 6\% 6\% 6\% of s p same Word # Description hard 1 1 difficult 2 head i i i 1 Table 2 1 2 _meta~horlcally hard _] 3 not yielding to pressure, 1 4 very strong or ~lgorous, ar 2 I wlth force or vigor (adv) 3 earnestly, intently (adv) i_ ~art of the body 3 intellect 4 _r~le_!r, ch,\%ef 7 front, front part WoldNet senses ot hard and head CIs were also computed for each language individually as well as for different language groupings Romaman, Czech, and Estonian (three different language families) Czech and Slovene (same family), Romaman, Czech, Slovene (Indo-European, and Estonian (nonIndo-European) To better visualize the relationship between senses, a hierarchical clustering algorithm was applied to the CI data to generate trees reflecting sense proximity 4 Finally, in order to determine the degree to which the linguistic relaUon between languages may affect coherence, a correlation was run among CIs for all pairs of the four target languages Fol example, Table 2 gives the senses of hard and head that occurred in the data s The CI data .s \'sobS\' hard and head are given in Tables 3 and 4 ~uous CIs measuring the aff, mty of a sense with itself--that is, the tendency for all occurrences of that sense to be translated wlth the same word--show that all of the s,x senses of ha,d have greatel internal consistency tfian athmty with other senses, with senses 1 1 (\'dlff|cult\' CI = 56) and 13 (,\'not soft,, ci = 63) registenng the h,ghest internal consistency 6 The same holds true for three of the four senses of head, while the CI for senses 1 3 (\'Intellect\') and 1 1 (\'part of the body\') is higher than the CI for 1 3/1 3 WordNet Sense 2 1 2 3 1 4 1 3 1 1 1 2 21 23 1 4 13 0 50 o 13 i ool 0 O0 0 25 i O0 0 04 0 50 0 17 0 56 0 19 0 00 0 00 0 00 0 00 0 00 0 25 0 21 Table 3 CIs for hard I i 12 0,,63 0 00 0 50 2 Results Although the data sample is small, It gives some insight into ways m which a larger sample might contribute to sense discrimination 4 Developed by Andleas Stolcke Results tor all words m the study are avadable at http//www cs vassar edu/~~de/wsd/cross-hng html 6 Senses 2 3 and 1 4 have CIs ot 1 because each ot these senses exists m a single occurrence m the corpus, and have theretote been dlscarded horn consideration ot CIs to~ individual senses We a~e currently mvesugatmg the use oI the Kappa staUst~c (Carletta, 1996) to normahze these sparse data 56 WordNet Sense 1 1 1 3 1 4 1 7 1 1 0 69 1 3 0 53 0 45 1 4 0 12 0 07, 0 50 1 7 0 40 0 001 0 00 1 00 Table 4 CIs for head Figure 2 shows the sense clusters for hard generated from the CI data 7 The senses fall into two mare clusters, w~th the two most internally consistent senses (1 1 and 1 3) at the deepest level of each ot the respecuve groups The two adverbml forms 8 are placed in separate groups, leflectmg thmr semantic proximity to the different adjecuval meanings of hard The clusters for head (Figure 2) stmdarly show two dlstmct groupings, each anchored in the two senses with the h~ghest internal consistency and the lowest mutual CI (\'part of the body\' (1 1) and \'ruler, chief\' (1 4)) The h~erarchtes apparent m the cluster graphs make intuitive sense Structured hke dictmnary enmes, the clusters for hard and head might appeal as m F~gure 1 This ts not dissimilar to actual dlctLonary entries for hard and head, for example, the enmes for hard in four differently constructed dlctmnanes ( Colhns Enghsh (CED), Longman\'s (LDOCE), OxJotd Advanced Learner\'s (OALD), and COBUILD) all hst the \'\'d~fficult\' and \'not soft\' senses first and second, whmh, since most dictionaries hst the most common Ol frequently used senses hrst, reflects the gross dlwslon apparent m the clusters Beyond this, ~t ~s difficult to assess the 7 Foi the purposes ot the cluster analys~s, CIs of l 00 resulting from a single occurrrence were normahzed to 5 8 Because ~oot to, ms were used m the analysis, no dzstlncUon m UanslaUon eqmvalents was made tor part ot speech correspondence between the senses In the dictionary entries and the clusters The remamlng WordNet senses are scattered at various places within the entries or, m some cases, split across various senses The h~erarchlcal relatmns apparent m the clusters are not reflected m the d~cttonary enmes, smce the senses are for the most part presented in flat, hnear hsts However, It is interesting to note that the first five senses of hard In the COBUILD d~cuonary, which is the only d~cttonary in the group constructed on the bas~s of colpus examples 9 and presents senses m ruder of frequency, correspond to hve of the six WordNet senses in thls study WordNet\'s \'metaphorically hard\' is spread over multiple senses in the COB UILD, as it.is In the other d~ctlonarles HARD HEAD I 1 dlfflcult 2 vlgorously II 1 a not soft b strong 2 a earnestly b metaphorlcally hard I 1 a part of the body b zntellect 2 front, front part II ruler, chlef Flgme 1 Clusteis tol hard and head suuctured as dlcuonary entt ~es The results tor dlftment language groupings show that the tendency to lextcahze senses differently is not aftected by language d~stance (Table 5) In fact, the mean CI fol Estonian, the only non-Indo-European language m the study, ~s lower than that for any other group, mdmatmg that WordNet sense dtstmctmns are slightly less hkely to be lexlcahzed differently m Estonian 9 Edmons ot the LDOCE (1987 vexsmn) and OALD (1985 version) dictlonalles consulted m this study ple-date edmons ol those same d~ctlonanes based on colpus evidence 57 Correlations of CIs for each language pair (Table 5) also show no relationship between the degree to which sense d~stmcuons are lexlcahzed differently and language distance This is contrary to results obtained by Resmk and Yarowsky (subm,tted), who, using a memc slmdar to the one used in this study, found that that non-Indo-European languages tended to lexlcallze English sense d~stmctlons more than Indo-European languages, especially at finergrained levels However, their translation data was generated by native speakers presented with Isolated sentences in English, who were asked to provide the translation for a given word In the sentence It is not clear how this data compares to translations generated by trained translators working with full context Lanquaqe qroup Averaqe CI ALL 0 27 RO/ES/SL 0 28 SL/CS 0 28 RO/SL/CS 0 27 ES 0 26 Table 5 Average CI values Lanqs Hard Country Llne Head Ave ES/CS 0 86 0 72 0 68 0 69 0 74 RO/SL 0 73 0 78 0 68 1 00 0 80 RO/CS 0 83 0 66 0 67 0 72 0 72 SL/CS 0 88 0 51 0 72 0 71 0 71 RO/ES 0 97 0 26 0 70 0 98 0 73 ES/SL 0 73 0 59 0 90 0 99 0 80 Table 6 CI correlauon tor the tour target languages I -I I I I m~nlmum dlstance = 0 249399 m~nlmum d~stance = 0 434856 mlnlmum dlstance = 0 555158 mlnlmum dlstance = 0 602972 m~nlmum dlstance = 0 761327 I >21 I >ii I >23 l >13 l >14 I >12 (13) (23) (12) (1,4) (ii) (21) (1412) (2313) ( 2 3 1 3 1 4 1 2 ) ( 2 111 ) Figure 2 Cluster tree and distance measures tor the sm senses of hard I >14 -i I > i i I--- 1 J > i 3 I >17 mlnlmum dlstance = 0 441022 mlnlmum dlstance = 0 619052 mln~mum dlstance = 0 723157 (13) (ll) (17) (1113) (111317) (14) F,gure 3 Cluster tree and dmtance measures tot the tout senses ot head 58 Conclusion The small sample m this study suggests that cross-hngual lexlcahzat~on can be used to define and structure sense d~stmct~ons The cluster graphs above provide mformat~on about relations among WordNet senses that could be used, for example, to determine the granularity of sense differences, whtch m turn could be used in tasks such as machine translatton, mtormaUon retrieval, etc For example, it is hkely that as sense dtstmcttons become finer, the degree of error ~s less severe Resmk and Yarowsky (1997) suggest that confusing freer-grained sense dtstmctlons should be penahzed less severely than confusing grosser d~stmct~ons when evaluatmg the performance of sense dtsambtguatt0n systems The clusters also provide insight into the lexlcallzatlon of sense dtstmcttons related by various semantic relations (metonymy, meronymy, etc ) across languages, for instance, the \'part of the body\' and \'intellect\' senses of head are lex~cahzed with the same ~tem a s~gnlficant portion of the t~me across all languages, reformation that could be used m machine translatton In addtt~on, cluster data such as that presented here could be used m lexicography, to determine a mole detaded hierarchy of relations among senses in dtct~onary entries It is less clear how cross-hngual reformation can be used to determine sense d~st~nctlons independent of a pre-deflned set, such as the WordNet senses used here In an effort to explore how thts mlght be done, I have used the small sample from thts study to create word groupmgs from \'back translations\' (l e, additional translations m the original language ot the translations m the target language) and developed a metric that uses th~s mformatton to determine relatedness between occurrences, whtch ~s m turn used to cluster occurrences into sense groups I have also compared sets of back translations for words representing the various WordNet senses, which provtde word groups s~mdar to WordNet synsets Interestingly, there ts virtually no overlap between the WordNet synsets and word groups generated from back translations The results show, however, that sense dlstmctlons useful for natural language processing tasks such as machme translanon could potentsally be determined, ot at least influenced, by constdeHng this mformatton The automatically generated synsets themselves may also be useful m the same apphcatlons; where WordNet synsets (and ontologtes) have been used tn the past More work needs to be done on the topic of cross-hngual sense determination, utthzmg substantially larger parallel corpora that include a variety ot language types as well as texts fiom several genres This small study explores a possible methodology to apply when such resources become avatlable Acknowledgements The author would hke to gratefully acknowledge the contrtbut~on of those who provided the translatton mfotmat~on Tomaz Eua~ec (Slovene), Kadrt Muxschnek (Estonian), Vladtmlr Petkevtc (Czech), and Dan Tubs (Romanlan), as well as Dana Fleut and Darnel Khne, who helped to transcrtbe and evaluate the data Special thanks to Dan Melamed and Hlnrtch Schutze for their helpful comments 59 [] [] in [] in i i Hg nn i an i am References Ca~letta, Jean (1996) Assessing Agreement on Classthcatton Tasks The Kappa Stat~st~t. Computational Lmgulstlcs, 22(2), 249-254 Dagan, Ido and Ita~, Alon (1994) Wo~d sense dlsambxguat~on using a second language monohngual corpus Computattonal Ltngmsttcs, 20(4), 563-596 Dagan, Ido, Ital, Alon, and Schwall, Ulnke (1991) Two languages a~e more mformattve than one Proceedings of the 29th Annual Meettng of the Assoctatton for Computattonal Ltngutsttcs, 18-21 June 1991, Berkeley, Cahfornm, 130-137 Dyvtk, Helge (1998) Translations as Semantic Mirrors Proceedmgs of Workshop W13 Multzlmguahty in the Lextcon II, The 13th Biennial European Conference on Arttftctal lntelhgence (ECA198), Brighton, UK, 24-44 Eqavec, Tomaz and Ide, Nancy (1998) The MULTEXT-EAST Corpus Proceedlng~ of the Fltst International Conference on Language Resources and Evaluatton, 27-30 May 1998, Granada, 971-74 Erjavec, Tomaz, Lawson, Ann, and Romary, Laurent (1998) East meets West Producing Multflmgual Resources m a European Context Pioceedtngs of the Ftrst Internattonal Conference on Language Resources and Evaluation, 27-30 May 1998, Gtanada, 981-86 Fellbaum, Chttstmne (ed) (1998) WordNet An Electrontc Lexlcal Database MIT Press, Cambridge, Massachusetts Gale, Wdham A, Church, Kenneth W and Yatowsky, Davtd (1993) A method tor dlsamblguatmg word senses m a large cmpus Computers and the Humamtles, 26, 415-439, Hearst, M\'attl A (1991) Noun homograph \' dlsamblguatlon using local:\'~.\'0ntext m large corpora Proceedtngs of the 7th Annual Conference of the Umver~lt~ of Waterloo Centre for the New OED and Text ReaeaJch, Oxford, Umted Kingdom, 1-19 Ide, Nancy and V61oms, Jean (1998) Word sense d~samb~guat~on The state of the alt Computational Lmgut~ttc~, 24 1, 1-40 Kdgar~ttt, Adam and Palmer, Ma~tha, Eds (forthcoming) Proceedmgs ot the Senseval Word Sense D~samb~guatlon Workshop, Specml double ~ssue otComputer~ and the Humamttes, 33 4-5 Leacock, Claudia, Towell, Geoffrey and Voorhees, Ellen (1993) Corpus-based stattstlcal sense resolution Proceedtng~ of the ARPA Human Language Technology Worsl~shop, San Francisco, Morgan Kautman Melamed, I Dan (1997) Measuring Semantic Entropy ACL-SIGLEX Workshop Taggmg Tert wtth Lextcal Semanttcs Why, What, and How ~ April 4-5, 1997, Washington, D C, 41-46 Mtllet, George A, Beckwlth, Richard T Fellbaum.',Carletta,'' +o,1060,'We build a subset S C ~ \' incrementally by iterating to adjoin a feature f E ~ \' which maximizes loglikelihood of the model to S This algorithm is called the Basic Feature Selection ',Berger,'nsubj_build_We det_~_a nn_~_subset nn_~_S nn_~_C poss_~_~ amod_~_incrementally prep_incrementally_by pcomp_by_iterating aux_adjoin_to xcomp_iterating_adjoin det_feature_a dobj_adjoin_feature dep_adjoin_f dep_f_E dobj_build_~ nsubj_maximizes_which rcmod_~_maximizes dobj_maximizes_loglikelihood prep_loglikelihood_of det_model_the pobj_of_model prep_maximizes_to pobj_to_S det_algorithm_This nsubjpass_called_algorithm auxpass_called_is dep_build_called det_Selection_the nn_Selection_Basic nn_Selection_Feature dobj_called_Selection ' +o,1061,'From this point of view , some of the measures used in the evaluation of Machine Translation systems , such as BLEU , have been imported into the summarization task ',Papineni,'prep_imported_From det_point_this pobj_From_point prep_point_of pobj_of_view nsubjpass_imported_some prep_some_of det_measures_the pobj_of_measures partmod_measures_used prep_used_in det_evaluation_the pobj_in_evaluation prep_evaluation_of nn_systems_Machine nn_systems_Translation pobj_of_systems dep_as_such prep_some_as nn__BLEU pobj_as_ aux_imported_have auxpass_imported_been prep_imported_into det_task_the nn_task_summarization pobj_into_task ' +o,1062,'The linear kernel derived from the L1 distance is the same as the difference-weighted token-based similarity measure of ',Weeds,'det_kernel_The amod_kernel_linear nsubj_measure_kernel partmod_kernel_derived prep_derived_from det_distance_the nn_distance_L1 pobj_from_distance cop_measure_is det_measure_the amod_measure_same prep_same_as det_similarity_the amod_similarity_difference-weighted amod_similarity_token-based pobj_as_similarity prep_measure_of ' +o,1063,'Thus , it may not suffer from the issues of non-isomorphic structure alignment and non-syntactic phrase usage heavily ',Wellington,'advmod_suffer_Thus nsubj_suffer_it aux_suffer_may neg_suffer_not prep_suffer_from det_issues_the pobj_from_issues prep_issues_of amod_alignment_non-isomorphic nn_alignment_structure pobj_of_alignment cc_alignment_and amod_usage_non-syntactic nn_usage_phrase conj_alignment_usage advmod_suffer_heavily ' +o,1064,'For example , applied multiple-sequence alignment -LRB- MSA -RRB- to parallel news sentences and induced paraphrasing patterns for generating new sentences ',Barzilay,'prep_applied_For pobj_For_example nsubj_applied_ amod_alignment_multiple-sequence dobj_applied_alignment abbrev_alignment_MSA prep_applied_to amod_sentences_parallel nn_sentences_news pobj_to_sentences cc_applied_and conj_applied_induced xcomp_induced_paraphrasing dobj_paraphrasing_patterns prep_patterns_for amod_sentences_generating amod_sentences_new pobj_for_sentences ' +o,1065,'Many authors claim that class-based methods are more robust against data sparseness problems , , ',Brown,'amod_authors_Many nsubj_claim_authors complm_robust_that amod_methods_class-based nsubj_robust_methods cop_robust_are advmod_robust_more ccomp_claim_robust prep_robust_against nn__data nn__sparseness nn__problems pobj_against_ advmod_robust_ ' +o,1066,'It achieves 901 \% average precision\\/recall for sentences with maximum length 40 and 895 \% for sentences with maximum length 100 when trained and tested on the standard sections of the Wall Street Journal Treebank ',Marcus,'nsubj_achieves_It num_precision\\/recall_901 nn_precision\\/recall_\% nn_precision\\/recall_average dobj_achieves_precision\\/recall prep_precision\\/recall_for pobj_for_sentences prep_sentences_with nn_length_maximum pobj_with_length num_length_40 cc_sentences_and num_\%_895 conj_sentences_\% prep_\%_for pobj_for_sentences prep_sentences_with nn_length_maximum pobj_with_length num_length_100 advmod_trained_when amod_\%_trained cc_trained_and conj_trained_tested prep_tested_on det_sections_the amod_sections_standard pobj_on_sections prep_sections_of det_Treebank_the nn_Treebank_Wall nn_Treebank_Street nn_Treebank_Journal pobj_of_Treebank ' +o,1067,'The usefulness of prosody was found to be very limited by itself , if the effect of utterance length is not considered ',Penn,'det_usefulness_The nsubjpass_found_usefulness prep_usefulness_of pobj_of_prosody auxpass_found_was aux_limited_to cop_limited_be advmod_limited_very xcomp_found_limited prep_limited_by pobj_by_itself mark_considered_if det_effect_the nsubjpass_considered_effect prep_effect_of amod_length_utterance pobj_of_length auxpass_considered_is neg_considered_not advcl_limited_considered ' +p,1068,'Incremental top-down and left-corner parsers have been shown to effectively -LRB- and efficiently -RRB- make use of non-local features from the left-context to yield very high accuracy syntactic parses , and we will use such rich models to derive our scores ',Collins,'nsubjpass_shown_Incremental dep_Incremental_top-down cc_Incremental_and amod_parsers_left-corner conj_Incremental_parsers aux_shown_have auxpass_shown_been aux_make_to advmod_make_effectively dep_-LRB-_and dep_-LRB-_efficiently xcomp_shown_make dobj_make_use prep_make_of amod_features_non-local pobj_of_features prep_shown_from det_left-context_the pobj_from_left-context aux_yield_to infmod_left-context_yield advmod_high_very amod_accuracy_high dobj_yield_accuracy amod_parses_syntactic nsubj__parses dep_shown_ cc__and nsubj_use_we aux_use_will conj__use amod_models_such amod_models_rich dobj_use_models aux_derive_to xcomp_use_derive poss_scores_our dobj_derive_scores ' +o,1069,'In our search procedure , we use a mixture-based alignment model that slightly differs from the model introduced as Model 2 in ',Brown,'prep_use_In poss_procedure_our nn_procedure_search pobj_In_procedure nsubj_use_we det_model_a amod_model_mixture-based nn_model_alignment dobj_use_model nsubj_differs_that advmod_differs_slightly rcmod_model_differs prep_differs_from det_model_the pobj_from_model partmod_model_introduced prep_introduced_as pobj_as_Model num_Model_2 prep_introduced_in ' +o,1070,'One can imagine the same techniques coupled with more informative probability distributions , such as lexicalized PCFGs , or even grammars not based upon literal rules , but probability distributions that describe how rules are built up from smaller components ',Collins,'nsubj_imagine_One aux_imagine_can det_techniques_the amod_techniques_same dobj_imagine_techniques partmod_techniques_coupled prep_coupled_with advmod_informative_more amod_distributions_informative nn_distributions_probability pobj_with_distributions dep_as_such prep_coupled_as amod__lexicalized nn__PCFGs pobj_as_ cc_techniques_or advmod_grammars_even conj_techniques_grammars neg_based_not dep_grammars_based prep_based_upon amod_rules_literal pobj_upon_rules cc_based_but nn_distributions_probability conj_based_distributions nsubj_describe_that rcmod_distributions_describe advmod_built_how nsubjpass_built_rules auxpass_built_are ccomp_describe_built prt_built_up prep_built_from amod_components_smaller pobj_from_components ' +o,1071,'4 Related Work 41 Acquisition of Classes of Instances Although some researchers focus on re-organizing or extending classes of instances already available explicitly within manually-built resources such as Wikipedia or WordNet or both , a large body of previous work focuses on compiling sets of instances , not necessarily labeled , from unstructured text ',Snow,'dep_focuses_4 nn_Acquisition_Related nn_Acquisition_Work num_Acquisition_41 dep_focus_Acquisition prep_Acquisition_of pobj_of_Classes prep_Classes_of pobj_of_Instances dep_focus_Although det_researchers_some nsubj_focus_researchers dep_focuses_focus prep_focus_on pcomp_on_re-organizing cc_re-organizing_or conj_re-organizing_extending dobj_re-organizing_classes prep_classes_of pobj_of_instances advmod_available_already amod_instances_available advmod_instances_explicitly prep_instances_within amod_resources_manually-built pobj_within_resources dep_as_such prep_resources_as nn__Wikipedia pobj_as_ cc_instances_or nn__WordNet conj_instances_ cc__or det__both conj__ det_body_a amod_body_large nsubj_focuses_body prep_body_of amod_work_previous pobj_of_work prep_focuses_on pcomp_on_compiling dobj_compiling_sets prep_sets_of pobj_of_instances neg_labeled_not advmod_labeled_necessarily partmod_instances_labeled prep_labeled_from amod_text_unstructured pobj_from_text ' +o,1072,'An alternative method we considered was to estimate certain conditional probabilities , similarly to the formula used in : SW -LRB- t -RRB- log P -LRB- p C A\\/t -RRB- f -LRB- t , A -RRB- f -LRB- A -RRB- = ~ log -LRB- 2 -RRB- P -LRB- p C R\\/t -RRB- f -LRB- t , R -RRB- f -LRB- l ~ -RRB- Here f -LRB- A -RRB- is -LRB- an estimate of -RRB- the probability that any given candidate phrase will be accepted by the spotter , and f -LRB- R -RRB- is the probability that this phrase is rejected , ie , f -LRB- R -RRB- = l-f -LRB- A -RRB- ',Yarowsky,'det_method_An amod_method_alternative nsubj_was_method nsubj_considered_we rcmod_method_considered aux_estimate_to xcomp_was_estimate amod_probabilities_certain amod_probabilities_conditional dobj_estimate_probabilities advmod_estimate_similarly prep_estimate_to det_formula_the pobj_to_formula partmod_formula_used prep_used_in pobj_in_ nsubj_log_SW appos_SW_t parataxis_was_log dobj_log_P nn_A\\/t_p nn_A\\/t_C appos_P_A\\/t dobj_=_f dep_f_t dep_t_A nsubj_=_f appos_f_A rcmod_P_= number_log_~ num_f_log appos_f_2 nn_f_P nn_R\\/t_p nn_R\\/t_C appos_f_R\\/t dobj_=_f appos_f_t dep_t_R dobj_is_f number_~_l dep_f_~ advmod_is_Here nsubj_is_f appos_f_A rcmod_f_is det_estimate_an dep_probability_estimate dep_estimate_of det_probability_the dep_is_probability complm_accepted_that det_phrase_any amod_phrase_given nn_phrase_candidate nsubjpass_accepted_phrase aux_accepted_will auxpass_accepted_be dep_probability_accepted prep_accepted_by det_spotter_the pobj_by_spotter cc_was_and nsubj_probability_f appos_f_R cop_probability_is det_probability_the conj_was_probability complm_rejected_that det_phrase_this nsubjpass_rejected_phrase auxpass_rejected_is ccomp_probability_rejected dep_rejected_ie xcomp_rejected_f nsubj_A_R dep_A_= nn_A_l-f ccomp_f_A ' +o,1073,'3 Automatic Evaluation of MT Quality We utilize BLEU for the automatic evaluation of MT quality in this paper ',Papineni,'dep_Evaluation_3 amod_Evaluation_Automatic prep_Evaluation_of nn_Quality_MT pobj_of_Quality nsubj_utilize_We rcmod_Quality_utilize nn__BLEU dobj_utilize_ prep__for det_evaluation_the amod_evaluation_automatic pobj_for_evaluation prep_evaluation_of nn_quality_MT pobj_of_quality prep_quality_in det_paper_this pobj_in_paper ' +o,1074,'Since the texts in the RST Treebank are taken from the syntactically annotated Penn Treebank , it is natural to ask what the relation is between the discourse structures in the RST Treebank and the syntactic structures of the Penn Treebank ',Marcus,'mark_taken_Since det_texts_the nsubjpass_taken_texts prep_texts_in det_Treebank_the nn_Treebank_RST pobj_in_Treebank auxpass_taken_are advcl_natural_taken prep_taken_from det__the advmod_annotated_syntactically amod__annotated nn__Penn nn__Treebank pobj_from_ nsubj_natural_it cop_natural_is aux_ask_to xcomp_natural_ask dep_is_what det_relation_the nsubj_is_relation ccomp_ask_is prep_is_between det_structures_the nn_structures_discourse pobj_between_structures prep_structures_in det_Treebank_the nn_Treebank_RST pobj_in_Treebank cc_structures_and det_structures_the amod_structures_syntactic conj_structures_structures prep_structures_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,1075,'Syntactic context information is used to compute term similarities , based on which similar words to a particular word can directly be returned ',Lin,'amod_information_Syntactic nn_information_context nsubjpass_used_information auxpass_used_is advmod_used_ aux_compute_to xcomp_used_compute nn_similarities_term dobj_compute_similarities prep_used_based dep_based_on dobj_returned_which amod_words_similar nsubjpass_returned_words prep_words_to det_word_a amod_word_particular pobj_to_word aux_returned_can advmod_returned_directly auxpass_returned_be pcomp_on_returned ' +o,1076,'B-X I-X 0 first word of a chunk of type X non-initial word in an X chunk word outside of any chunk This representation type is based on a representation proposed by for noun phrase chunks ',Ramshaw,'nn_I-X_B-X nsubj_0_I-X amod_word_first dobj_0_word prep_word_of det_chunk_a pobj_of_chunk prep_chunk_of nn_word_type nn_word_X amod_word_non-initial pobj_of_word prep_0_in det_outside_an nn_outside_X nn_outside_chunk nn_outside_word pobj_in_outside prep_outside_of det_chunk_any pobj_of_chunk det_type_This nn_type_representation nsubjpass_based_type auxpass_based_is dep_0_based prep_based_on det_representation_a pobj_on_representation partmod_representation_proposed prep_proposed_by pobj_by_ prep__for nn_chunks_noun nn_chunks_phrase pobj_for_chunks ' +o,1077,'The approach is evaluated by cross-validation on the WSJ treebank corpus ',Marcus,'det_approach_The nsubjpass_evaluated_approach auxpass_evaluated_is prep_evaluated_by pobj_by_cross-validation prep_cross-validation_on det_corpus_the nn_corpus_WSJ nn_corpus_treebank pobj_on_corpus ' +o,1078,'The block set is generated using a phrase-pair selection algorithm similar to , which includes some heuristic filtering to mal statement here ',Koehn,'det_set_The nn_set_block nsubjpass_generated_set auxpass_generated_is xcomp_generated_using det_algorithm_a amod_algorithm_phrase-pair nn_algorithm_selection nsubj_similar_algorithm xcomp_using_similar prep_similar_to pobj_to_ nsubj_includes_which rcmod__includes det_heuristic_some nsubj_mal_heuristic amod_heuristic_filtering aux_mal_to xcomp_includes_mal dobj_mal_statement advmod_mal_here ' +o,1079,'For details please refer to ',Wu,'prep_refer_For pobj_For_details nsubj_refer_please prep_refer_to ' +o,1080,'5 Translation performance was measured using the automatic BLEU evaluation metric on four reference translations ',Papineni,'num_performance_5 nn_performance_Translation nsubjpass_measured_performance auxpass_measured_was xcomp_measured_using det__the amod__automatic amod__BLEU nn__evaluation amod__metric dobj_using_ prep_using_on num_translations_four nn_translations_reference pobj_on_translations ' +o,1081,'The elementary trees were extracted from the parse trees in sections 02-21 of the Wall Street Journal in Penn Treebank , which is transformed by using parent-child annotation and left factoring ',Marcus,'det_trees_The amod_trees_elementary nsubjpass_extracted_trees auxpass_extracted_were prep_extracted_from det_trees_the amod_trees_parse pobj_from_trees prep_trees_in pobj_in_sections num_sections_02-21 prep_sections_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal prep_extracted_in nn__Penn nn__Treebank pobj_in_ nsubjpass_transformed_which auxpass_transformed_is rcmod__transformed prep_transformed_by pcomp_by_using amod_annotation_parent-child dobj_using_annotation cc_annotation_and amod__left nn__factoring conj_annotation_ ' +o,1082,'The probabilities of derivation decisions are modelled using the neural network approximation to a type of dynamic Bayesian Network called an Incremental Sigmoid Belief Network -LRB- ISBN -RRB- ',Titov,'det_probabilities_The nsubjpass_modelled_probabilities prep_probabilities_of nn_decisions_derivation pobj_of_decisions auxpass_modelled_are xcomp_modelled_using det__the amod__neural nn__network nn__approximation dobj_using_ prep_using_to det_type_a pobj_to_type prep_type_of pobj_of_dynamic nn_Network_Bayesian nsubj_called_Network dep_modelled_called det_Network_an nn_Network_Incremental nn_Network_Sigmoid nn_Network_Belief dobj_called_Network abbrev_Network_ISBN ' +o,1083,'To test whether a better set of initial parameter estimates can improve Model 1 alignment accuracy , we use a heuristic model based on the loglikelihood-ratio -LRB- LLR -RRB- statistic recommended by ',Dunning,'aux_test_To dep_use_test complm_improve_whether det_set_a amod_set_better nsubj_improve_set prep_set_of amod_estimates_initial nn_estimates_parameter pobj_of_estimates aux_improve_can ccomp_test_improve number_1_Model num_accuracy_1 nn_accuracy_alignment dobj_improve_accuracy nsubj_use_we det_model_a amod_model_heuristic nsubj_recommended_model prep_model_based dep_based_on det_statistic_the amod_statistic_loglikelihood-ratio abbrev_statistic_LLR pobj_on_statistic ccomp_use_recommended prep_recommended_by ' +o,1084,'The intuition is that the produced clusters will be less sense-conflating than those produced by other graph-based approaches , since collocations provide strong and consistent clues to the senses of a target word ',Yarowsky,'det_intuition_The nsubj_is_intuition complm_sense-conflating_that det_clusters_the amod_clusters_produced nsubj_sense-conflating_clusters aux_sense-conflating_will cop_sense-conflating_be advmod_sense-conflating_less ccomp_is_sense-conflating mark_produced_than nsubj_produced_those ccomp_sense-conflating_produced prep_produced_by amod_approaches_other amod_approaches_graph-based pobj_by_approaches mark_provide_since nsubj_provide_collocations advcl_produced_provide amod_clues_strong cc_strong_and conj_strong_consistent dobj_provide_clues prep_provide_to det_senses_the pobj_to_senses prep_senses_of det_word_a nn_word_target pobj_of_word ' +o,1085,'At the sentence level , employed an unsupervised learning approach to cluster sentences and extract lattice pairs from comparable monolingual corpora ',Barzilay,'prep__At det_level_the nn_level_sentence pobj_At_level dep__employed det_approach_an amod_approach_unsupervised nn_approach_learning dobj_employed_approach prep_employed_to nn_sentences_cluster pobj_to_sentences cc__and conj__extract nn_pairs_lattice dobj_extract_pairs prep_extract_from amod_corpora_comparable amod_corpora_monolingual pobj_from_corpora ' +o,1086,'Similarly , if the task is to distinguish between binary , coarse sense distinction , then current WSD techniques can achieve very high accuracy -LRB- in excess of 96 \% when tested on a dozen words in -RRB- ',Yarowsky,'advmod_achieve_Similarly mark_is_if det_task_the nsubj_is_task dep_achieve_is aux_distinguish_to xcomp_is_distinguish prep_distinguish_between amod_distinction_binary amod_distinction_coarse nn_distinction_sense pobj_between_distinction advmod_techniques_then amod_techniques_current nn_techniques_WSD nsubj_achieve_techniques aux_achieve_can advmod_high_very amod_accuracy_high dobj_achieve_accuracy dep_achieve_in pobj_in_excess prep_excess_of num_\%_96 pobj_of_\% advmod_tested_when dep_in_tested prep_tested_on quantmod_dozen_a num_words_dozen pobj_on_words prep_words_in pobj_in_ ' +o,1087,'These algorithms are usually applied to sequential labeling or chunking , but have also been applied to parsing , machine translation and summarization ',Liang,'det_algorithms_These nsubjpass_applied_algorithms auxpass_applied_are advmod_applied_usually aux_sequential_to xcomp_applied_sequential xcomp_sequential_labeling cc_labeling_or conj_labeling_chunking cc_applied_but aux_applied_have advmod_applied_also auxpass_applied_been conj_applied_applied prep_applied_to amod__parsing nn__ dep__machine nn__translation pobj_to_ cc__and nn__summarization conj__ ' +o,1088,'A variety of methods are used to account for the re-ordering stage : word-based , templatebased , and syntax-based , to name just a few ',Brown,'det_variety_A nsubjpass_used_variety prep_variety_of pobj_of_methods auxpass_used_are aux_account_to xcomp_used_account prep_account_for det_stage_the amod_stage_re-ordering pobj_for_stage amod__word-based dep_stage_ amod__templatebased conj__ cc__and amod__syntax-based conj__ aux_name_to xcomp_account_name quantmod_few_just quantmod_few_a dobj_name_few ' +o,1089,'For the multilingual dependency parsing track , which was the other track of the shared task , Nilsson et al achieved the best performance using an ensemble method ','Daume III','prep_achieved_For det_track_the amod_track_multilingual amod_track_dependency nn_track_parsing pobj_For_track nsubj_track_which cop_track_was det_track_the amod_track_other rcmod_track_track prep_track_of det_task_the amod_task_shared pobj_of_task nsubj_achieved_Nilsson cc_Nilsson_et conj_Nilsson_al det_performance_the amod_performance_best dobj_achieved_performance partmod_performance_using det__an nn__ensemble nn__method dobj_using_ ' +o,1090,'However , by exploiting the fact that the underlying scores assigned to competing hypotheses , w -LRB- e , h , f -RRB- , vary linearly wrt changes in the weight vector , w , proposed a strategy for finding the global minimum along any given search direction ',Och,'advmod_proposed_However prep_proposed_by pcomp_by_exploiting det_fact_the dobj_exploiting_fact complm_vary_that det_scores_the amod_scores_underlying nsubj_vary_scores partmod_scores_assigned prep_assigned_to amod_hypotheses_competing pobj_to_hypotheses appos_scores_w appos_w_e dep_e_h dep_e_f ccomp_exploiting_vary amod_changes_linearly nn_changes_wrt dobj_vary_changes prep_vary_in det_vector_the nn_vector_weight pobj_in_vector appos_vector_w nsubj_proposed_ det_strategy_a dobj_proposed_strategy prep_proposed_for pcomp_for_finding det_minimum_the amod_minimum_global dobj_finding_minimum prep_minimum_along det_direction_any amod_direction_given nn_direction_search pobj_along_direction ' +o,1091,'11 However , modeling word order under translation is notoriously difficult , and it is unclear how much improvement in accuracy a good model of word order would provide ',Brown,'advmod__However nn_order_modeling nn_order_word nsubj__order prep_order_under pobj_under_translation cop__is advmod_difficult_notoriously amod__difficult dep_11_ cc__and nsubj_unclear_it cop_unclear_is conj__unclear advmod_much_how dep_provide_much nsubj_provide_improvement prep_improvement_in pobj_in_accuracy det_model_a amod_model_good dep_accuracy_model prep_model_of nn_order_word pobj_of_order aux_provide_would ccomp_unclear_provide ' +o,1092,' present a system called BABAR that uses contextual role knowledge to do coreference resolution ',Bean,'nsubj_present_ det_system_a nsubj_called_system ccomp_present_called dobj_called_BABAR nsubj_uses_that rcmod_BABAR_uses amod_knowledge_contextual nn_knowledge_role dobj_uses_knowledge aux_do_to infmod_knowledge_do amod_resolution_coreference dobj_do_resolution ' +o,1093,'The system described in also makes use of syntactic heuristics ',Bean,'det_system_The nsubj_makes_system partmod_system_described prep_described_in pobj_in_ advmod_makes_also dobj_makes_use prep_use_of amod_heuristics_syntactic pobj_of_heuristics ' +o,1094,' , , Dave et al ',Turney,'nn__ appos__ conj__Dave cc__et conj__al ' +o,1095,'Our MT baseline system is based on Moses decoder with word alignment obtained from GIZA + + ',Koehn,'poss_system_Our nn_system_MT nn_system_baseline nsubjpass_based_system auxpass_based_is prep_based_on nn__Moses nn__decoder pobj_on_ prep_based_with nn_alignment_word pobj_with_alignment partmod_alignment_obtained prep_obtained_from pobj_from_GIZA amod_GIZA_+ cc_+_+ conj_+_ ' +o,1096,'1 Motivation A major component in phrase-based statistical Machine translation -LRB- PBSMT -RRB- is the table of conditional probabilities of phrase translation pairs ',Koehn,'nsubj_table_1 partmod_1_Motivation det_component_A amod_component_major dobj_Motivation_component prep_component_in amod_translation_phrase-based amod_translation_statistical nn_translation_Machine pobj_in_translation abbrev_translation_PBSMT advmod_Motivation_ cop_table_is det_table_the prep_table_of amod_probabilities_conditional pobj_of_probabilities prep_probabilities_of nn_pairs_phrase nn_pairs_translation pobj_of_pairs ' +o,1097,' , , Dave et al ',Turney,'nn__ appos__ conj__Dave cc__et conj__al ' +o,1098,'12 Related Work Recently , discriminative methods for alignment have rivaled the quality of IBM Model 4 alignments ',Fraser,'num_Work_12 amod_Work_Related nsubj_rivaled_Work advmod_Work_Recently amod_methods_discriminative nsubj_rivaled_methods prep_methods_for pobj_for_alignment aux_rivaled_have det_quality_the dobj_rivaled_quality prep_quality_of nn_alignments_IBM nn_alignments_Model num_alignments_4 pobj_of_alignments ' +o,1099,'The disambiguation algorithms also require that the semantic relatedness measures WordNet : : Similarity be installed ',Pedersen,'det_algorithms_The nn_algorithms_disambiguation nsubj_require_algorithms advmod_require_also complm_measures_that det_relatedness_the amod_relatedness_semantic nsubj_measures_relatedness ccomp_require_measures dobj_measures_WordNet nn__Similarity nsubjpass_installed_ auxpass_installed_be parataxis_require_installed ' +o,1100,'An analysis of the alignments shows that smoothing the fertility probabilities significantly reduces the frequently occurring problem of rare words forming garbage collectors in that they tend to align with too many words in the other language ',Brown,'det_analysis_An nsubj_shows_analysis prep_analysis_of det_alignments_the pobj_of_alignments complm_reduces_that csubj_reduces_smoothing det_probabilities_the nn_probabilities_fertility dobj_smoothing_probabilities advmod_reduces_significantly ccomp_shows_reduces det_problem_the advmod_occurring_frequently amod_problem_occurring dobj_reduces_problem prep_problem_of amod_words_rare pobj_of_words partmod_words_forming nn_collectors_garbage dobj_forming_collectors rel_tend_in pobj_in_that nsubj_tend_they rcmod_collectors_tend aux_align_to xcomp_tend_align prep_align_with advmod_many_too amod_words_many pobj_with_words prep_words_in det_language_the amod_language_other pobj_in_language ' +p,1101,'Erk compared a number of techniques for creating similar-word sets and found that both the Jaccard coefficient and s information-theoretic metric work best ',Lin,'nn__Erk nsubj_compared_ det_number_a dobj_compared_number prep_number_of pobj_of_techniques prep_compared_for pcomp_for_creating amod_sets_similar-word dobj_creating_sets cc_compared_and conj_compared_found complm_work_that preconj_coefficient_both det_coefficient_the nn_coefficient_Jaccard nsubj_work_coefficient cc_coefficient_and conj_coefficient_ dep_work_s amod_work_information-theoretic amod_work_metric ccomp_found_work advmod_work_best ' +p,1102,'4 Extended Minimum Error Rate Training Minimum error rate training is widely used to optimize feature weights for a linear model ',Och,'num__4 nn__Extended nn__Minimum nn__Error nn__Rate nn__Training nn__Minimum nn__error nn__rate amod__training nsubjpass_used_ auxpass_used_is advmod_used_widely aux_optimize_to xcomp_used_optimize nn_weights_feature dobj_optimize_weights prep_optimize_for det_model_a amod_model_linear pobj_for_model ' +o,1103,'Moses provides BLEU and NIST , but Meteor and TER can easily be used instead ',Banerjee,'nsubj_provides_Moses nn__BLEU dobj_provides_ cc__and nn__NIST conj__ cc_provides_but nn__Meteor nsubjpass_used_ cc__and nn__TER conj__ aux_used_can advmod_used_easily auxpass_used_be conj_provides_used advmod_used_instead ' +o,1104,'In an evaluation on the PENN treebank , the parser outperformed other unlexicalized PCFG parsers in terms of labeled bracketing fscore ',Marcus,'prep_outperformed_In det_evaluation_an pobj_In_evaluation prep_evaluation_on det__the nn__PENN nn__treebank pobj_on_ det_parser_the nsubj_outperformed_parser amod_parsers_other amod_parsers_unlexicalized nn_parsers_PCFG dobj_outperformed_parsers prep_outperformed_in pobj_in_terms prep_terms_of amod_fscore_labeled nn_fscore_bracketing pobj_of_fscore ' +o,1105,'Since the word support model and triple context matching model have been proposed in our previous work at the SIGHAN bakeoff 2005 and 2006 , the major descriptions of this paper is on the WBT model ',Levow,'mark_proposed_Since det_model_the nn_model_word nn_model_support nsubjpass_proposed_model cc_model_and amod_model_triple nn_model_context amod_model_matching conj_model_model aux_proposed_have auxpass_proposed_been advcl_is_proposed prep_proposed_in poss__our amod__previous nn__work pobj_in_ prep_proposed_at det__the nn__SIGHAN nn__bakeoff num__2005 pobj_at_ cc__and num__2006 conj__ det_descriptions_the amod_descriptions_major nsubj_is_descriptions prep_descriptions_of det_paper_this pobj_of_paper prep_is_on det_model_the nn_model_WBT pobj_on_model ' +o,1106,'612 ROUGE evaluation Table 4 presents ROUGE scores of each of human-generated 250-word surveys against each other ',Lin,'num_Table_612 amod_Table_ROUGE nn_Table_evaluation num_presents_4 measure_ROUGE_presents amod__ROUGE nn__scores dep_Table_ prep__of pobj_of_each prep_each_of amod_surveys_human-generated amod_surveys_250-word pobj_of_surveys prep_surveys_against det_other_each pobj_against_other ' +o,1107,'Two are conditionalized phrasal models , each EM trained until performance degrades : C-JPTM3 as described in Phrasal ITG as described in Section 41 Three provide alignments for the surface heuristic : GIZA + + with grow-diag-final -LRB- GDF -RRB- Viterbi Phrasal ITG with and without the noncompositional constraint We use the Pharaoh decoder with the SMT Shared Task baseline system ',Koehn,'dep_as_Two auxpass_conditionalized_are rcmod_Two_conditionalized amod_models_phrasal dobj_conditionalized_models det_EM_each appos_models_EM partmod_EM_trained prep_trained_until nn_degrades_performance pobj_until_degrades dep_as_C-JPTM3 dep_as_described prep_described_in nn_ITG_ nn_ITG_Phrasal pobj_in_ITG mark_provide_as amod_Three_described prep_described_in pobj_in_Section num_Section_41 nsubj_provide_Three advcl_described_provide dobj_provide_alignments prep_alignments_for det_heuristic_the nn_heuristic_surface pobj_for_heuristic nn_+_GIZA nn_+_+ dobj_provide_+ prep_+_with pobj_with_grow-diag-final abbrev_+_GDF nn_ITG_Viterbi nn_ITG_Phrasal dep_+_ITG prep_ITG_with cc_with_and conj_with_without det_constraint_the amod_constraint_noncompositional pobj_without_constraint nsubj_use_We rcmod_constraint_use det__the nn__Pharaoh nn__decoder dobj_use_ prep__with det_system_the nn_system_SMT nn_system_Shared nn_system_Task amod_system_baseline pobj_with_system ' +o,1108,'These words and phrases are usually compiled using different approaches -LRB- Hatzivassiloglou and McKeown , 1997 ; Kaji and Kitsuregawa , 2006 ; and Nasukawa , 2006 ; Esuli and Sebastiani , 2006 ; Breck et al , 2007 ; Ding , Liu and Yu ',Kanayama,'det_words_These cc_words_and conj_words_phrases auxpass_compiled_are advmod_compiled_usually dep_words_compiled xcomp_compiled_using amod_approaches_different dobj_using_approaches dep_approaches_Hatzivassiloglou cc_Hatzivassiloglou_and conj_Hatzivassiloglou_McKeown dep_Hatzivassiloglou_1997 dep_words_Kaji cc_Kaji_and conj_Kaji_Kitsuregawa appos_Kaji_2006 dep_Kaji_ cc__and conj__Nasukawa appos__2006 dep_Kaji_Esuli cc_Esuli_and conj_Esuli_Sebastiani appos_Esuli_2006 dep_Kaji_Breck cc_Breck_et conj_Breck_al appos_Breck_2007 dep_Kaji_Ding conj_Ding_Liu cc_Ding_and conj_Ding_Yu ' +o,1109,'3 Online Learning Again following , we have used the single best MIRA , which is a variant of the voted perceptron for structured prediction ',Collins,'num_Learning_3 nn_Learning_Online dep_following_Again prep_used_following pobj_following_ nsubj_used_we aux_used_have dep_Learning_used det__the amod__single amod__best nn__MIRA dobj_used_ nsubj_variant_which cop_variant_is det_variant_a rcmod__variant prep_variant_of det__the amod__voted nn__perceptron pobj_of_ prep__for amod_prediction_structured pobj_for_prediction ' +o,1110,'CP-STM -LRB- i -RRB- - l This metric corresponds to the STM metric presented by ',Liu,'nn_i_CP-STM aux_presented_l det_corresponds_This amod_corresponds_metric nsubj_presented_corresponds prep_corresponds_to det_metric_the nn_metric_STM pobj_to_metric dep_i_presented prep_presented_by ' +o,1111,'Our experiments created translation modules for two evaluation corpora : written news stories from the Penn Treebank corpus and spoken task-oriented dialogues from the TRAINS93 corpus ',Marcus,'poss_experiments_Our nsubj_created_experiments nn_modules_translation dobj_created_modules prep_created_for num_corpora_two nn_corpora_evaluation pobj_for_corpora amod_stories_written nn_stories_news dep_corpora_stories prep_stories_from det__the nn__Penn nn__Treebank nn__corpus pobj_from_ cc_created_and conj_created_spoken amod_dialogues_task-oriented dobj_spoken_dialogues prep_spoken_from det__the nn__TRAINS93 nn__corpus pobj_from_ ' +o,1112,'CIT -RRB- ',Turney,'nn_-RRB-_CIT ' +o,1113,'31 A Note on State-Splits Recent studies suggest that category-splits help in enhancing the performance of treebank grammars , and a previous study on MH outlines specific POS-tags splits that improve MH parsing accuracy ',Matsuzaki,'det_Note_A nsubj_suggest_Note prep_Note_on nn__State-Splits amod__Recent nn__studies pobj_on_ dep_31_suggest complm_help_that nsubj_help_category-splits ccomp_suggest_help prep_help_in pcomp_in_enhancing det_performance_the dobj_enhancing_performance prep_performance_of nn_grammars_treebank pobj_of_grammars cc_suggest_and det_study_a amod_study_previous nsubj_outlines_study prep_study_on nn__MH pobj_on_ conj_suggest_outlines amod_POS-tags_specific nsubj_splits_POS-tags ccomp_outlines_splits nsubj_improve_that ccomp_splits_improve nsubj_parsing_MH dep_improve_parsing dobj_parsing_accuracy ' +n,1114,'In general , these authors have found that existing lexicalized parsing models for English do not straightforwardly generalize to new languages ; this typically manifests itself in a severe reduction in parsing performance compared to the results for English ',Collins,'prep_found_In pobj_In_general det_authors_these nsubj_found_authors aux_found_have complm_generalize_that amod_models_existing amod_models_lexicalized nn_models_parsing nsubj_generalize_models prep_models_for nn__English pobj_for_ aux_generalize_do neg_generalize_not advmod_generalize_straightforwardly ccomp_found_generalize prep_generalize_to amod_languages_new pobj_to_languages nsubj_manifests_this advmod_manifests_typically parataxis_found_manifests dobj_manifests_itself prep_manifests_in det_reduction_a amod_reduction_severe pobj_in_reduction prep_reduction_in pcomp_in_parsing dobj_parsing_performance prep_parsing_compared dep_compared_to det_results_the pobj_to_results prep_results_for pobj_for_English ' +o,1115,'Given this , the mutual information ratio is expressed by Formula 1 ',Church,'prep_expressed_Given dep_Given_this det__the amod__mutual nn__information nn__ratio nsubjpass_expressed_ auxpass_expressed_is prep_expressed_by pobj_by_Formula num_Formula_1 ' +p,1116,'When we run our classifiers on resource-tight environments such as cell-phones , we can use a random feature mixing technique or a memory-efficient trie implementation based on a succinct data structure to reduce required memory usage ',Ganchev,'advmod_run_When nsubj_run_we dep_use_run poss_classifiers_our dobj_run_classifiers prep_run_on amod_environments_resource-tight pobj_on_environments dep_as_such prep_environments_as pobj_as_cell-phones nsubj_use_we aux_use_can det_feature_a amod_feature_random dobj_use_feature partmod_feature_mixing nn__technique dobj_mixing_ cc__or det_implementation_a amod_implementation_memory-efficient nn_implementation_trie conj__implementation prep_mixing_based dep_based_on det_structure_a amod_structure_succinct nn_structure_data pobj_on_structure partmod_structure_ aux_reduce_to xcomp__reduce amod_usage_required nn_usage_memory dobj_reduce_usage ' +o,1117,'In addition to sentence fusion , compression algorithms and methods for expansion of a multiparallel corpus are other instances of such methods ',Jing,'prep_instances_In pobj_In_addition prep_addition_to nn_fusion_sentence pobj_to_fusion nn__compression amod__algorithms nsubj_instances_ cc__and conj__methods prep__for pobj_for_expansion prep_expansion_of det__a amod__multiparallel nn__corpus pobj_of_ cop_instances_are amod_instances_other prep_instances_of amod_methods_such pobj_of_methods ' +o,1118,'by diag-and symmetrization ',Koehn,'amod_symmetrization_diag-and pobj_by_symmetrization ' +o,1119,'There are many research directions , eg , sentiment classification -LRB- classifying an opinion document as positive or negative -RRB- , subjectivity classification -LRB- determining whether a sentence is subjective or objective , and its associated opinion -RRB- , feature\\/topic-based sentiment analysis -LRB- assigning positive or negative sentiments to topics or product features -RRB- -LRB- Hu and Liu 2004 ; Popescu and Etzioni , 2005 ; Carenini et al , 2005 ; Ku et al , 2006 ; Kobayashi , Inui and Matsumoto , 2007 ; Titov and ',McDonald,'expl_directions_There cop_directions_are amod_directions_many nn_directions_research ccomp_classification_directions appos_directions_eg nn_classification_sentiment appos_eg_classification parataxis_directions_classifying det_document_an nn_document_opinion nsubj_positive_document advmod_positive_as xcomp_classifying_positive cc_positive_or conj_positive_negative advmod_directions_ advmod_classification_subjectivity parataxis_classification_determining complm_subjective_whether det_sentence_a nsubj_subjective_sentence cop_subjective_is ccomp_determining_subjective cc_subjective_or conj_subjective_objective cc_subjective_and poss_opinion_its amod_opinion_associated conj_subjective_opinion nsubj_classification_ amod_analysis_feature\\/topic-based nn_analysis_sentiment appos__analysis amod_sentiments_assigning amod_sentiments_positive cc_positive_or conj_positive_negative dep_analysis_sentiments prep_sentiments_to pobj_to_topics cc_topics_or nn_features_product conj_topics_features dep_analysis_Hu cc_Hu_and conj_Hu_Liu num_Liu_2004 dep_analysis_Popescu cc_Popescu_and conj_Popescu_Etzioni appos_Popescu_2005 dep_analysis_Carenini cc_Carenini_et conj_Carenini_al appos_Carenini_2005 dep_analysis_Ku cc_Ku_et conj_Ku_al appos_Ku_2006 dep_analysis_Kobayashi conj_Kobayashi_Inui cc_Kobayashi_and conj_Kobayashi_Matsumoto appos_Kobayashi_2007 dep_analysis_Titov dep_Titov_and ' +o,1120,'23 Online Learning Again following , we have used the single best MIRA , which is a margin aware variant of perceptron for structured prediction ',Collins,'num_Learning_23 nn_Learning_Online dep_following_Again prep_used_following pobj_following_ nsubj_used_we aux_used_have dep_Learning_used det__the amod__single amod__best nn__MIRA dobj_used_ nsubj_margin_which cop_margin_is det_margin_a rcmod__margin amod_margin_aware dep_aware_variant prep_margin_of nn__perceptron pobj_of_ prep__for amod_prediction_structured pobj_for_prediction ' +o,1121,'Reported work includes improved model variants and applications such as web data extraction , scientific citation extraction , word alignment , and discourselevel chunking ',Jiao,'dobj_Reported_work dep_Reported_includes amod__improved nn__model amod__variants dobj_includes_ cc__and conj__applications dep_as_such prep__as nn_data_web pobj_as_data nn__extraction nsubj_Reported_ amod__scientific nn__citation nn__extraction appos__ nn__word nn__alignment conj__ cc__and amod__discourselevel amod__chunking conj__ ' +o,1122,'They are not used in LN , but they are known to be useful for WSD ',Tanaka,'nsubjpass_used_They auxpass_used_are neg_used_not prep_used_in pobj_in_LN cc_used_but nsubjpass_known_they auxpass_known_are conj_used_known aux_useful_to cop_useful_be xcomp_known_useful prep_useful_for pobj_for_WSD ' +o,1123,'3 Building the CatVar The CatVar database was developed using a combination of resources and algorithms including the Lexical Conceptual Structure -LRB- LCS -RRB- Verb and Preposition Databases , the Brown Corpus section of the Penn Treebank , an English morphological analysis lexicon developed for PC-Kimmo -LRB- Englex -RRB- , NOMLEX , Longman Dictionary of Contemporary English 2For a deeper discussion and classification of Porter stemmers errors , see ',Marcus,'num_Building_3 det_database_the nn_database_CatVar nn_database_The nn_database_CatVar nsubjpass_developed_database auxpass_developed_was dep_Building_developed xcomp_developed_using det_combination_a dobj_using_combination prep_combination_of pobj_of_resources cc_resources_and conj_resources_algorithms prep_using_including det_Verb_the nn_Verb_Lexical nn_Verb_Conceptual nn_Verb_Structure nn_Verb_LCS pobj_including_Verb cc_Verb_and nn__Preposition nn__Databases conj_Verb_ det_section_the nn_section_Brown nn_section_Corpus appos_Verb_section prep_section_of det__the nn__Penn nn__Treebank pobj_of_ det_lexicon_an amod_lexicon_English amod_lexicon_morphological nn_lexicon_analysis appos_Verb_lexicon partmod_lexicon_developed prep_developed_for amod__PC-Kimmo appos__Englex pobj_for_ nn__NOMLEX appos__ nn_Dictionary_Longman dep_see_Dictionary prep_Dictionary_of nn_English_Contemporary pobj_of_English nn_errors_2For det_discussion_a amod_discussion_deeper nn_errors_discussion cc_discussion_and conj_discussion_classification prep_classification_of nn_stemmers_Porter pobj_of_stemmers dep_see_errors dep_Building_see acomp_see_ ' +o,1124,'POS tag the text using the tagger of ',Ratnaparkhi,'dep_tag_POS det_text_the nsubj_using_text dep_tag_using det_tagger_the dobj_using_tagger prep_tagger_of ' +o,1125,'The learning algorithm used for each stage of the classification task is a regularized variant of the structured Perceptron ',Collins,'det_algorithm_The nn_algorithm_learning nsubj_variant_algorithm partmod_algorithm_used prep_used_for det_stage_each pobj_for_stage prep_stage_of det_task_the nn_task_classification pobj_of_task cop_variant_is det_variant_a amod_variant_regularized prep_variant_of det_Perceptron_the amod_Perceptron_structured pobj_of_Perceptron ' +o,1126,'An alternative is to create an automatic system that uses a set of training question-answer pairs to learn the appropriate question-answer matching algorithm ',Chu-Carroll,'det_alternative_An nsubj_is_alternative aux_create_to xcomp_is_create det_system_an amod_system_automatic dobj_create_system nsubj_uses_that rcmod_system_uses det_set_a dobj_uses_set prep_set_of pcomp_of_training amod_pairs_question-answer dobj_training_pairs aux_learn_to xcomp_training_learn det_algorithm_the amod_algorithm_appropriate amod_algorithm_question-answer nn_algorithm_matching dobj_learn_algorithm ' +o,1127,' compared two Bayesian inference algorithms , Variational Bayes and what we call here a point-wise collapsed Gibbs sampler , and found that Variational Bayes produced the best solution , and that the Gibbs sampler was extremely slow to converge and produced a worse solution than EM ',Johnson,'dep__compared num_algorithms_two amod_algorithms_Bayesian nn_algorithms_inference dep_compared_algorithms nn_Bayes_Variational appos_algorithms_Bayes cc_compared_and dobj_call_what nsubj_call_we conj_compared_call advmod_sampler_here det_sampler_a amod_sampler_point-wise amod_sampler_collapsed nn_sampler_Gibbs dobj_call_sampler cc_call_and conj_call_found dep_produced_that nn_Bayes_Variational nsubj_produced_Bayes dep_found_produced det_solution_the amod_solution_best dobj_produced_solution cc_produced_and dep_slow_that det_sampler_the nn_sampler_Gibbs nsubj_slow_sampler cop_slow_was advmod_slow_extremely conj_produced_slow aux_converge_to xcomp_slow_converge cc_slow_and conj_slow_produced det_solution_a amod_solution_worse dobj_produced_solution prep_produced_than pobj_than_EM ' +o,1128,' , Pereira and Tishby , and Pereira , Tishby , and Lee propose methods that derive classes from the distributional properties of the corpus itself , while other authors use external information sources to define classes : Resnik uses the taxonomy of WordNet ; uses the categories of Roget \'s Thesaurus , Slator and Liddy and Paik use the subject codes in the LDOCE ; Luk uses conceptual sets built from the LDOCE definitions ',Yarowsky,'nsubj__ conj__Pereira cc__and conj__Tishby cc__and nn_Lee_Pereira conj_Lee_Tishby cc_Lee_and nsubj_propose_Lee advmod_propose_ conj__propose dobj_propose_methods nsubj_derive_that rcmod_methods_derive dobj_derive_classes prep_derive_from det_properties_the amod_properties_distributional pobj_from_properties prep_properties_of det_corpus_the pobj_of_corpus advmod_corpus_itself mark_use_while amod_authors_other nsubj_use_authors advcl_derive_use amod_sources_external nn_sources_information nsubj_define_sources aux_define_to xcomp_use_define dobj_define_classes nn__Resnik nsubj_uses_ parataxis__uses det_taxonomy_the dobj_uses_taxonomy prep_taxonomy_of pobj_of_WordNet nn__ nsubj_uses_ parataxis__uses det_categories_the dobj_uses_categories prep_categories_of poss_Thesaurus_Roget possessive_Roget_\'s pobj_of_Thesaurus nn__Slator conj_categories_ cc_categories_and conj_categories_Liddy cc__and nn__Paik nsubj_use_ conj__use det_codes_the nn_codes_subject dobj_use_codes prep_codes_in det_LDOCE_the pobj_in_LDOCE nn__Luk nsubj_uses_ parataxis_use_uses amod_sets_conceptual dobj_uses_sets partmod_sets_built prep_built_from det_definitions_the amod_definitions_LDOCE pobj_from_definitions ' +o,1129,'For the Brown corpus , we based our division on ',McClosky,'prep_based_For det_corpus_the nn_corpus_Brown pobj_For_corpus nsubj_based_we poss_division_our dobj_based_division prep_based_on ' +o,1130,'Much of the work in subjectivity analysis has been applied to English data , though work on other languages is growing : eg , Japanese data are used in , Chinese data are used in , and German data are used in ',Kanayama,'nsubjpass_applied_Much prep_Much_of det_work_the pobj_of_work prep_work_in amod_analysis_subjectivity pobj_in_analysis aux_applied_has auxpass_applied_been prep_applied_to amod_data_English pobj_to_data mark_growing_though nsubj_growing_work prep_work_on amod_languages_other pobj_on_languages cop_growing_is advcl_applied_growing nn_data_eg amod_data_Japanese nsubjpass_used_data auxpass_used_are parataxis_applied_used prep_used_in pobj_in_ amod_data_Chinese nsubjpass_used_data auxpass_used_are ccomp_applied_used prep_used_in pobj_in_ cc_applied_and amod_data_German nsubjpass_used_data auxpass_used_are conj_applied_used prep_used_in pobj_in_ ' +o,1131,'Morphosyntacticinformationhas in fact been shown to significantlyimprove the extractionresults ',Smadja,'prep_Morphosyntacticinformationhas_in pobj_in_fact auxpass_shown_been partmod_fact_shown aux_significantlyimprove_to xcomp_shown_significantlyimprove det_extractionresults_the dobj_significantlyimprove_extractionresults ' +p,1132,'32 Evaluation Criteria Well-established objective evaluation measures like the word error rate -LRB- WER -RRB- , positionindependent word error rate -LRB- PER -RRB- , and the BLEU score were used to assess the translation quality ',Papineni,'num_measures_32 nn_measures_Evaluation amod_measures_Criteria amod_measures_Well-established amod_measures_objective nn_measures_evaluation nsubjpass_used_measures prep_measures_like det_rate_the nn_rate_word nn_rate_error pobj_like_rate abbrev_rate_WER amod_rate_positionindependent nn_rate_word nn_rate_error conj_rate_rate abbrev_rate_PER cc_rate_and det__the amod__BLEU nn__score conj_rate_ auxpass_used_were aux_assess_to xcomp_used_assess det_quality_the nn_quality_translation dobj_assess_quality ' +p,1133,'In recent several years , the system combination methods based on confusion networks developed rapidly , which show state-of-the-art performance in benchmarks ',Rosti,'prep_developed_In amod_years_recent amod_years_several pobj_In_years det_methods_the nn_methods_system nn_methods_combination appos_years_methods partmod_methods_based prep_based_on nn_networks_confusion pobj_on_networks advmod_developed_rapidly nsubj_developed_ nsubj_show_which rcmod__show amod_performance_state-of-the-art dobj_show_performance prep_performance_in pobj_in_benchmarks ' +o,1134,'For the chunk part of the code , we adopt the Inside , Outside , and Between -LRB- IOB -RRB- encoding originating from ',Ramshaw,'prep_adopt_For det_part_the nn_part_chunk pobj_For_part prep_part_of det_code_the pobj_of_code nsubj_adopt_we det_Inside_the dobj_adopt_Inside dep_Inside_Outside cc_Inside_and conj_Inside_Between abbrev_Inside_IOB dep_adopt_encoding xcomp_encoding_originating prep_originating_from ' +o,1135,'Following , Iusevariational Bayes EM during the M-step for the transition distribution : l +1 j i = f -LRB- E -LRB- ni , j -RRB- + i -RRB- f -LRB- E -LRB- n i -RRB- + C i -RRB- -LRB- 3 -RRB- f -LRB- v -RRB- = exp -LRB- -LRB- v -RRB- -RRB- -LRB- 4 -RRB- 60 -LRB- v -RRB- = braceleftBigg g -LRB- v 1 2 -RRB- ifv -RRB- 7 -LRB- v + 1 -RRB- 1v ow ',Johnson,'nn__Following nn__Iusevariational nn__Bayes nn__EM appos__ prep__during det_M-step_the pobj_during_M-step prep_M-step_for det_distribution_the nn_distribution_transition pobj_for_distribution number_+1_l num_j_+1 dep__j dep_j_i rcmod_j_= dobj_=_f dep_f_E appos_E_ni dep_ni_j prep_E_+ pobj_+_i dep_3_f dep_-LRB-_E dep_E_n dep_n_i dep_n_+ dep_n_C dep_C_i dobj_=_3 nsubj_=_f appos_f_v rcmod_f_= dobj_=_exp dep_exp_v amod_exp_4 nsubj_=_60 appos_60_v rcmod_exp_= amod_ifv_braceleftBigg amod_ifv_g dep_ifv_v number_2_1 dobj_v_2 dobj_=_ifv nn_ow_7 dep_ow_v cc_v_+ conj_v_1 amod_ow_1v dep__ow ' +o,1136,'Our results are similar to those for conventional phrase-based models ',Koehn,'poss_results_Our nsubj_similar_results cop_similar_are prep_similar_to pobj_to_those prep_those_for amod_models_conventional amod_models_phrase-based pobj_for_models ' +o,1137,'While Kazama and Torisawa used a chunker , we parsed the definition sentence using Minipar ',Lin,'mark_used_While nsubj_used_Kazama cc_Kazama_and conj_Kazama_Torisawa advcl_parsed_used det_chunker_a dobj_used_chunker nsubj_parsed_we det_sentence_the nn_sentence_definition dobj_parsed_sentence xcomp_parsed_using dobj_using_Minipar ' +o,1138,'Perhaps the most well-known method is maximum marginal relevance -LRB- MMR -RRB- , as well as cross-sentence informational subsumption , mixture models , subtopic diversity , diversity penalty , and others ',Goldstein,'advmod_method_Perhaps det_method_the advmod_well-known_most amod_method_well-known nsubj_relevance_method cop_relevance_is amod_relevance_maximum amod_relevance_marginal abbrev_relevance_MMR dep_relevance_ dep_well_as cc_relevance_well dep_well_as amod__cross-sentence amod__informational nn__subsumption conj_relevance_ nn__mixture nn__models conj__ amod__subtopic nn__diversity conj__ nn__diversity nn__penalty conj__ cc__and conj__others ' +p,1139,'Although bi-alignments are known to exhibit high precision , in the face of sparse annotations we use unidirectional alignments as a fallback , as has been proposed in the context of phrase-based machine translation ',Koehn,'mark_known_Although nsubjpass_known_bi-alignments auxpass_known_are aux_exhibit_to xcomp_known_exhibit amod__high nn__precision dobj_exhibit_ prep_known_in det_face_the pobj_in_face prep_face_of amod_annotations_sparse pobj_of_annotations nsubj_use_we rcmod_annotations_use amod_alignments_unidirectional dobj_use_alignments prep_use_as det_fallback_a pobj_as_fallback mark_proposed_as aux_proposed_has auxpass_proposed_been dep_fallback_proposed prep_proposed_in det_context_the pobj_in_context prep_context_of amod_translation_phrase-based nn_translation_machine pobj_of_translation ' +o,1140,'Before training the classifiers , we perform feature ablation by imposing a count cutoff of 10 , and by limiting the number of features to the top 75K features in terms of log likelihood ratio ',Dunning,'prep_perform_Before pcomp_Before_training det_classifiers_the dobj_training_classifiers nsubj_perform_we nn_ablation_feature dobj_perform_ablation prep_perform_by pcomp_by_imposing det_cutoff_a nn_cutoff_count dobj_imposing_cutoff prep_cutoff_of pobj_of_10 cc_by_and conj_by_by pcomp_by_limiting det_number_the dobj_limiting_number prep_number_of pobj_of_features prep_limiting_to det_features_the amod_features_top num_features_75K pobj_to_features prep_limiting_in pobj_in_terms prep_terms_of nn_ratio_log nn_ratio_likelihood pobj_of_ratio ' +o,1141,'32 The parsers The parsers that we chose to evaluate are the C&C CCG parser , the Enju HPSG parser , the RASP parser , the Stanford parser , and the DCU postprocessor of PTB parsers , based on LFG and applied to the output of the Charniak and Johnson reranking parser ',Cahill,'det_parsers_The nsubj_parsers_parsers det_parsers_The xcomp_32_parsers complm_chose_that nsubj_chose_we ccomp_32_chose aux_evaluate_to xcomp_chose_evaluate cop__are det__the nn__C&C nn__CCG nn__parser dep_32_ det__the nn__Enju nn__HPSG nn__parser conj__ det__the nn__RASP nn__parser conj__ det__the nn__Stanford nn__parser conj__ cc__and det_postprocessor_the nn_postprocessor_DCU conj__postprocessor prep_postprocessor_of nn_parsers_PTB pobj_of_parsers nsubj_32_ dep__based prep_based_on pobj_on_LFG cc_based_and conj_based_applied prep_applied_to det_output_the pobj_to_output prep_output_of det_Charniak_the pobj_of_Charniak cc_Charniak_and conj_Charniak_Johnson xcomp_applied_reranking dobj_reranking_parser ' +p,1142,'It is based on Incremental Sigmoid Belief Networks -LRB- ISBNs -RRB- , a class of directed graphical model for structure prediction problems recently proposed in , where they were demonstrated to achieve competitive results on the constituent parsing task ',Titov,'nsubjpass_based_It auxpass_based_is prep_based_on nn_Networks_Incremental nn_Networks_Sigmoid nn_Networks_Belief pobj_on_Networks abbrev_Networks_ISBNs det_class_a appos_Networks_class prep_class_of amod_model_directed amod_model_graphical pobj_of_model prep_model_for nn_problems_structure nn_problems_prediction pobj_for_problems advmod_proposed_recently partmod_problems_proposed prep_proposed_in pobj_in_ advmod_demonstrated_where nsubjpass_demonstrated_they auxpass_demonstrated_were advcl_proposed_demonstrated aux_achieve_to xcomp_demonstrated_achieve amod_results_competitive dobj_achieve_results prep_achieve_on det_task_the nn_task_constituent amod_task_parsing pobj_on_task ' +o,1143,'Training via the voted perceptron algorithm or using a max-margin criterion also correspond to the first option -LRB- eg McCallum and Wellner , Finley and Joachims -RRB- ',Collins,'prep_Training_via det__the amod__voted nn__perceptron nn__algorithm pobj_via_ cc_Training_or conj_Training_using det_criterion_a amod_criterion_max-margin nsubj_correspond_criterion advmod_correspond_also ccomp_using_correspond prep_correspond_to det_option_the amod_option_first pobj_to_option nn_McCallum_eg dep_option_McCallum cc_McCallum_and nn__Wellner nn__ conj__Finley cc__and conj__Joachims conj_McCallum_ ' +o,1144,'Furthermore , early work on class-based language models was inconclusive ',Brown,'advmod_inconclusive_Furthermore amod_work_early nsubj_inconclusive_work prep_work_on amod_models_class-based nn_models_language pobj_on_models cop_inconclusive_was ' +p,1145,'In our experience , this approach is advantageous in terms of translation quality , eg by 07 \% in BLEU compared to a minimum Bayes risk primary ',Rosti,'prep_advantageous_In poss_experience_our pobj_In_experience det_approach_this nsubj_advantageous_approach cop_advantageous_is prep_advantageous_in pobj_in_terms prep_terms_of nn_quality_translation pobj_of_quality partmod_terms_eg prep_eg_by num_\%_07 pobj_by_\% prep_\%_in pobj_in_BLEU prep_eg_compared dep_compared_to det_primary_a amod_primary_minimum nn_primary_Bayes nn_primary_risk pobj_to_primary ' +o,1146,'Measures of cross-language relatedness are useful for a large number of applications , including cross-language information retrieval , cross-language text classification , lexical choice in machine translation , induction of translation lexicons , cross-language annotation and resource projections to a second language ',Mohammad,'nsubj_useful_Measures prep_Measures_of amod_relatedness_cross-language pobj_of_relatedness cop_useful_are ccomp_lexicons_useful prep_useful_for det_number_a amod_number_large pobj_for_number prep_number_of pobj_of_applications prep_number_including amod__cross-language nn__information nn__retrieval pobj_including_ amod__cross-language nn__text nn__classification appos__ amod_choice_lexical nsubj_lexicons_choice prep_choice_in nn__machine nn__translation pobj_in_ nsubj_lexicons_induction prep_induction_of pobj_of_translation amod_annotation_ amod_annotation_cross-language dobj_lexicons_annotation cc_annotation_and nn_projections_resource conj_annotation_projections prep_annotation_to det_language_a amod_language_second pobj_to_language ' +o,1147,'Related Work The recent availability of large amounts of bilingual data has attracted interest in several areas , including sentence alignment , word alignment , alignment of groups of words , and statistical translation ',Brown,'nn_Work_Related det_availability_The amod_availability_recent nsubj_attracted_availability prep_availability_of amod_amounts_large pobj_of_amounts prep_amounts_of amod_data_bilingual pobj_of_data aux_attracted_has dep_Work_attracted dobj_attracted_interest prep_attracted_in amod_areas_several pobj_in_areas prep_attracted_including nn__sentence nn__alignment pobj_including_ nn__word nn__alignment conj__ conj__alignment prep_alignment_of pobj_of_groups prep_groups_of nn__words pobj_of_ cc__and amod_translation_statistical conj__translation ' +p,1148,'To solve this problem , we adopt an idea one sense per collocation which was introduced in word sense disambiguation research ',Yarowsky,'aux_solve_To dep_adopt_solve det_problem_this dobj_solve_problem nsubj_adopt_we det_sense_an nn_sense_idea num_sense_one dobj_adopt_sense prep_sense_per pobj_per_collocation nsubjpass_introduced_which auxpass_introduced_was rcmod_sense_introduced prep_introduced_in nn_research_word nn_research_sense nn_research_disambiguation pobj_in_research ' +o,1149,'51 The baseline System used for comparison was Pharaoh , which uses a beam search algorithm for decoding ',Koehn,'det_System_The amod_System_baseline nsubj__System partmod_System_used prep_used_for pobj_for_comparison cop__was nn__Pharaoh dep_51_ nsubj_uses_which rcmod__uses det_algorithm_a nn_algorithm_beam nn_algorithm_search dobj_uses_algorithm prep_algorithm_for pobj_for_decoding ' +o,1150,'1 Introduction Base noun phrases -LRB- baseNPs -RRB- , broadly the initial portions of non-recursive noun phrases up to the head , are valuable pieces of linguistic structure which minimally extend beyond the scope of named entities ',Ramshaw,'num_phrases_1 nn_phrases_Introduction nn_phrases_Base nn_phrases_noun nsubj_pieces_phrases appos_phrases_baseNPs advmod_portions_broadly det_portions_the amod_portions_initial rcmod_phrases_portions prep_portions_of amod_phrases_non-recursive nn_phrases_noun pobj_of_phrases advmod_phrases_up dep_up_to det__the nn__head pobj_to_ cop_pieces_are amod_pieces_valuable prep_pieces_of amod_structure_linguistic pobj_of_structure nsubj_extend_which advmod_extend_minimally rcmod_pieces_extend prep_extend_beyond det_scope_the pobj_beyond_scope prep_scope_of amod_entities_named pobj_of_entities ' +p,1151,'The results show that , as compared to BLEU , several recently proposed metrics such as Semantic-role overlap , ParaEval-recall , and METEOR achieve higher correlation ',Banerjee,'det_results_The nsubj_show_results complm_proposed_that dep_compared_as prep_proposed_compared dep_compared_to pobj_to_BLEU nsubj_proposed_several advmod_proposed_recently ccomp_show_proposed nsubj_achieve_metrics dep_as_such prep_metrics_as nn__Semantic-role nn__overlap pobj_as_ nn__ParaEval-recall conj__ cc__and nn__METEOR conj__ ccomp_proposed_achieve amod_correlation_higher dobj_achieve_correlation ' +o,1152,'As modern systems move toward integrating many features , resources such as this will become increasingly important in improving translation quality ',Liang,'mark_move_As amod_systems_modern nsubj_move_systems advcl_important_move prep_move_toward pcomp_toward_integrating amod_features_many dobj_integrating_features advmod_integrating_ nsubj_important_resources dep_as_such prep_resources_as pobj_as_this aux_important_will cop_important_become advmod_important_increasingly prep_important_in pcomp_in_improving nn_quality_translation dobj_improving_quality ' +o,1153,'Metrics in the Rouge family allow for skip n-grams ; Kauchak and Barzilay take paraphrasing into account ; metrics such as METEOR and GTM calculate both recall and precision ; METEOR is also similar to SIA in that word class information is used ',Liu,'nsubj_allow_Metrics prep_Metrics_in det_family_the nn_family_Rouge pobj_in_family prep_allow_for amod__skip amod__n-grams pobj_for_ nn__Kauchak cc_Kauchak_and conj_Kauchak_Barzilay nsubj_take_ parataxis_allow_take xcomp_take_paraphrasing prep_paraphrasing_into pobj_into_account nsubj_calculate_metrics dep_as_such prep_metrics_as nn__METEOR pobj_as_ cc__and nn__GTM conj__ parataxis_take_calculate preconj_recall_both dobj_calculate_recall cc_recall_and conj_recall_precision nsubj_similar_METEOR cop_similar_is advmod_similar_also parataxis_take_similar prep_similar_to nn__SIA pobj_to_ rel_used_in pobj_in_that nn_information_word nn_information_class nsubjpass_used_information auxpass_used_is rcmod__used ' +o,1154,'We then piped the text through a maximum entropy sentence boundary detector and performed text normalization using NSW tools ',Ratnaparkhi,'nsubj_piped_We advmod_piped_then det_text_the dobj_piped_text prep_piped_through det__a amod__maximum nn__entropy nn__sentence nn__boundary nn__detector pobj_through_ cc_piped_and conj_piped_performed nn_normalization_text dobj_performed_normalization xcomp_performed_using nn_tools_NSW nsubj__tools xcomp_using_ ' +o,1155,'For English , we have used sections 03-06 of the WSJ portion of the Penn Treebank distributed by the Linguistic Data Consortium -LRB- LDC -RRB- , which have frequently been used to evaluate sentence boundary detection systems before ; compare Section 7 ',Dunning,'prep_used_For pobj_For_English nsubj_used_we aux_used_have dobj_used_sections nsubj_compare_03-06 prep_03-06_of det_portion_the nn_portion_WSJ pobj_of_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ partmod__distributed prep_distributed_by det_Consortium_the nn_Consortium_Linguistic nn_Consortium_Data pobj_by_Consortium abbrev_Consortium_LDC nsubjpass_used_which aux_used_have advmod_used_frequently auxpass_used_been rcmod_Consortium_used aux_evaluate_to xcomp_used_evaluate nn_systems_sentence nn_systems_boundary nn_systems_detection dobj_evaluate_systems advmod_evaluate_before rcmod_sections_compare dobj_compare_Section num_Section_7 ' +o,1156,'In the concept extension part of our algorithm we adapt our concept acquisition framework to suit diverse languages , including ones without explicit word segmentation ',Davidov,'prep_adapt_In det_part_the nn_part_concept nn_part_extension pobj_In_part prep_part_of poss_algorithm_our pobj_of_algorithm nsubj_adapt_we poss__our nn__concept nn__acquisition nn__framework dobj_adapt_ aux_suit_to xcomp_adapt_suit amod_languages_diverse dobj_suit_languages prep_languages_including pobj_including_ones prep_ones_without amod_segmentation_explicit nn_segmentation_word pobj_without_segmentation ' +o,1157,' , or -LRB- S+T - -RRB- , where no labeled target domain data is available , eg ','Daume III','cc__or nn_-_S+T advmod_available_where det_data_no amod_data_labeled nn_data_target nn_data_domain nsubj_available_data cop_available_is dep_eg_available dep__eg ' +o,1158,'Moreover , rather than predicting an intrinsic metric such as the PARSEVAL Fscore , the metric that the predictor learns to predict can be chosen to better fit the final metric on which an end-to-end system is measured , in the style of ',Och,'advmod_chosen_Moreover dep_than_rather ccomp_chosen_than pcomp_than_predicting det_metric_an amod_metric_intrinsic dobj_predicting_metric dep_as_such prep_metric_as det_Fscore_the nn_Fscore_PARSEVAL pobj_as_Fscore det_metric_the nsubjpass_chosen_metric complm_learns_that det_predictor_the nsubj_learns_predictor dep_metric_learns aux_predict_to xcomp_learns_predict aux_chosen_can auxpass_chosen_be aux_fit_to advmod_fit_better purpcl_chosen_fit det_metric_the amod_metric_final dobj_fit_metric rel_measured_on pobj_on_which det_system_an amod_system_end-to-end nsubjpass_measured_system auxpass_measured_is rcmod_metric_measured prep_measured_in det_style_the pobj_in_style prep_style_of ' +p,1159,'Parsing models have been developed for different languages and state-of-the-art results have been reported for , eg , English ',Collins,'amod_models_Parsing nsubjpass_developed_models aux_developed_have auxpass_developed_been mark_reported_for amod_languages_different nsubjpass_reported_languages cc_languages_and amod_results_state-of-the-art conj_languages_results aux_reported_have auxpass_reported_been advcl_developed_reported prep_reported_for dep_for_eg pobj_for_English ' +o,1160,'A possible solution to his problem might be the use of more general morphological rules like those used in part-of-speech tagging models -LRB- eg , 1 2 3 4 530 40 50 60 70 80 90 100 level error RAND BASE Boost_S NNtfidf NB Boost_M Figure 6 : Comparison of all models for a129 a48a51a95a66a97a98a97a180a222 -RRB- , where all suffixes up to a certain length are included ',Ratnaparkhi,'det_solution_A amod_solution_possible nsubj_use_solution prep_solution_to poss_problem_his pobj_to_problem aux_use_might cop_use_be det_use_the prep_use_of amod_rules_more amod_rules_general nn_rules_morphological pobj_of_rules prep_rules_like pobj_like_those partmod_those_used prep_used_in amod_models_part-of-speech amod_models_tagging pobj_in_models dep_models_eg number_100_1 punct_100_2 number_100_3 number_100_4 number_100_530 number_100_40 number_100_50 number_100_60 number_100_70 number_100_80 number_100_90 num_error_100 nn_error_level appos_eg_error nn_Figure_RAND nn_Figure_BASE nn_Figure_Boost_S nn_Figure_NNtfidf nn_Figure_NB nn_Figure_Boost_M dep_error_Figure num_Figure_6 dep_error_Comparison prep_Comparison_of det_models_all pobj_of_models prep_models_for amod__a129 amod__a48a51a95a66a97a98a97a180a222 pobj_for_ advmod_included_where det_suffixes_all nsubjpass_included_suffixes advmod_included_up dep_up_to det_length_a amod_length_certain pobj_to_length auxpass_included_are rcmod_models_included ' +o,1161,'4 Experiments The experiments described here were conducted using the Wall Street Journal Penn Treebank corpus ',Marcus,'num_Experiments_4 nsubjpass_conducted_Experiments det_experiments_The nsubj_described_experiments rcmod_Experiments_described advmod_described_here auxpass_conducted_were xcomp_conducted_using det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nn_corpus_Penn nn_corpus_Treebank dobj_using_corpus ' +o,1162,'Such studies follow the empiricist approach to word meaning summarized best in the famous dictum of the British 3 linguist J.R. Firth: You shall know a word by the company it keeps. (Firth, 1957, p. 11) Context similarity has been used as a means of extracting collocations from corpora, e.g. by Church & Hanks (1990) and by Dunning (1993), of identifying word senses, e.g. by Yarowski (1995) and by Schutze (1998), of clustering verb classes, e.g. by Schulte im Walde (2003), and of inducing selectional restrictions of verbs, e.g. by Resnik (1993), by Abe & Li (1996), by Rooth et al.',Church,'' +n,1163,'This method was shown to outperform the class based model proposed in and can thus be expected to discover better clusters of words ',Brown,'det_method_This nsubjpass_shown_method auxpass_shown_was aux_outperform_to xcomp_shown_outperform det_class_the dobj_outperform_class prep_outperform_based dep_based_model partmod_model_proposed prep_proposed_in pobj_in_ cc_shown_and aux_expected_can advmod_expected_thus auxpass_expected_be conj_shown_expected aux_discover_to purpcl_expected_discover amod_clusters_better dobj_discover_clusters prep_clusters_of pobj_of_words ' +o,1164,'For both experiments , we used dependency trees extracted from the Penn Treebank using the head rules and dependency extractor from Yamada and Matsumoto -LRB- 2003 -RRB- ',Marcus,'prep_used_For det_experiments_both pobj_For_experiments nsubj_used_we amod_trees_dependency nsubj_extracted_trees ccomp_used_extracted prep_extracted_from det__the nn__Penn nn__Treebank pobj_from_ partmod__using det_rules_the nn_rules_head dobj_using_rules cc_rules_and nn_extractor_dependency conj_rules_extractor prep_rules_from pobj_from_Yamada cc_Yamada_and conj_Yamada_Matsumoto dep_extracted_2003 ' +o,1165,'411 Lexical co-occurrences Lexical co-occurrences have previously been shown to be useful for discourse level learning tasks ',Marcu,'num_co-occurrences_411 amod_co-occurrences_Lexical amod_co-occurrences_co-occurrences nn_co-occurrences_Lexical nsubjpass_shown_co-occurrences aux_shown_have advmod_shown_previously auxpass_shown_been aux_useful_to cop_useful_be xcomp_shown_useful prep_useful_for nn_level_discourse pobj_for_level partmod_level_learning dobj_learning_tasks ' +o,1166,'Our evaluation metric is BLEU ',Papineni,'poss_metric_Our nn_metric_evaluation nsubj_BLEU_metric cop_BLEU_is ' +o,1167,'Similar to , eg , , we use a Naive Bayes algorithm trained on word features cooccurring with the subjective and the objective classifications ',Pang,'ccomp_use_Similar prep_Similar_to nsubj_use_eg advmod_use_ nsubj_use_we det_algorithm_a nn_algorithm_Naive nn_algorithm_Bayes dobj_use_algorithm partmod_algorithm_trained prep_trained_on nn_features_word pobj_on_features partmod_features_cooccurring prep_cooccurring_with det_subjective_the pobj_with_subjective cc_subjective_and det_classifications_the amod_classifications_objective conj_subjective_classifications ' +o,1168,'44 Experiment 2 : s Words We also conducted translation on seven of the twelve English words studied in ',Yarowsky,'num_Experiment_44 num_Experiment_2 amod_s_ nsubj_Words_s dep_Experiment_Words nsubj_conducted_We advmod_conducted_also ccomp_Words_conducted dobj_conducted_translation prep_conducted_on pobj_on_seven prep_seven_of det_words_the num_words_twelve amod_words_English pobj_of_words partmod_words_studied prep_studied_in nn__ pobj_in_ ' +o,1169,'Researchers extracted opinions from words , sentences , and documents , and both rule-based and statistical models are investigated ',Pang,'nsubj_extracted_Researchers dobj_extracted_opinions prep_extracted_from nn_documents_words conj_documents_sentences cc_documents_and pobj_from_documents cc_extracted_and det_models_both amod_models_rule-based cc_rule-based_and conj_rule-based_statistical nsubjpass_investigated_models auxpass_investigated_are conj_extracted_investigated ' +o,1170,'Since there is no well-agreed to definition of what an utterance is , we instead focus on intonational phrases , which end with an acoustically signaled boundary lone ',Brown,'mark_is_Since expl_is_there advcl_focus_is advmod_well-agreed_no acomp_is_well-agreed prep_well-agreed_to pobj_to_definition rel_is_of pobj_of_what det_utterance_an nsubj_is_utterance rcmod_definition_is nsubj_focus_we advmod_focus_instead prep_focus_on amod_phrases_intonational pobj_on_phrases advmod_focus_ dobj_signaled_which nsubj_signaled_end prep_end_with det_acoustically_an pobj_with_acoustically ccomp_focus_signaled dobj_signaled_boundary amod_boundary_lone ' +o,1171,'Running words 1,864 14,437 Vocabulary size 569 1,081 Table 2 : ChineseEnglish corpus statistics using Phramer , a 3-gram language model with Kneser-Ney smoothing trained with SRILM on the English side of the training data and Pharaoh with default settings to decode ',Koehn,'csubj_size_Running nsubj_Vocabulary_words number_14,437_1,864 measure_Vocabulary_14,437 xcomp_Running_Vocabulary number_1,081_569 num_Table_1,081 dobj_size_Table num_Table_2 nn_statistics_ChineseEnglish nn_statistics_corpus nsubj__statistics parataxis_size_ xcomp__using nn__Phramer dobj_using_ det_model_a amod_model_3-gram nn_model_language appos__model prep_model_with amod_smoothing_Kneser-Ney pobj_with_smoothing partmod_smoothing_trained prep_trained_with nn__SRILM pobj_with_ prep__on det_side_the amod_side_English pobj_on_side prep_side_of det_data_the nn_data_training pobj_of_data cc_side_and nn__Pharaoh conj_side_ prep__with nn_settings_default pobj_with_settings aux_decode_to xcomp_trained_decode ' +o,1172,'These tags are drawn from a tagset which is constructed by 363 extending each argument label by three additional symbols a80a44a81a83a82a84a81a86a85 , following ',Ramshaw,'det_tags_These nsubjpass_drawn_tags auxpass_drawn_are prep_drawn_from det_tagset_a pobj_from_tagset nsubjpass_constructed_which auxpass_constructed_is rcmod_tagset_constructed prep_constructed_by pobj_by_363 ccomp_constructed_extending det_label_each nn_label_argument dobj_extending_label prep_extending_by num_a80a44a81a83a82a84a81a86a85_three amod_a80a44a81a83a82a84a81a86a85_additional nn_a80a44a81a83a82a84a81a86a85_symbols pobj_by_a80a44a81a83a82a84a81a86a85 dep_extending_following ' +o,1173,'The first SMT systems were developed in the early nineties ',Brown,'det_systems_The amod_systems_first nn_systems_SMT nsubjpass_developed_systems auxpass_developed_were prep_developed_in det_nineties_the amod_nineties_early pobj_in_nineties ' +p,1174,'For English , after a relatively big jump achieved by , we have seen two significant improvements : and pushed the results by a significant amount each time1 1In our final comparison , we have also included the results of , because it has surpassed as well and we have used this tagger in the data preparation phase ',Collins,'dep_seen_For pobj_For_English mark_achieved_after det_jump_a advmod_big_relatively amod_jump_big nsubj_achieved_jump dep_seen_achieved prep_achieved_by pobj_by_ nsubj_seen_we aux_seen_have num_improvements_two amod_improvements_significant dobj_seen_improvements nsubj_pushed_ cc__and conj__ parataxis_seen_pushed det_results_the dobj_pushed_results prep_results_by det_amount_a amod_amount_significant pobj_by_amount det_1In_each amod_1In_time1 tmod_pushed_1In poss_comparison_our amod_comparison_final dep_1In_comparison nsubj_included_we aux_included_have advmod_included_also conj_seen_included det_results_the dobj_included_results prep_results_of pobj_of_ mark_surpassed_because nsubj_surpassed_it aux_surpassed_has advcl_included_surpassed xcomp_surpassed_ advmod_well_as advmod__well cc_seen_and nsubj_used_we aux_used_have conj_seen_used det_tagger_this dobj_used_tagger prep_tagger_in det_phase_the nn_phase_data nn_phase_preparation pobj_in_phase ' +o,1175,'Document level sentiment classification is mostly applied to reviews , where systems assign a positive or negative sentiment for a whole review document ',Pang,'nn_classification_Document nn_classification_level nn_classification_sentiment nsubjpass_applied_classification auxpass_applied_is advmod_applied_mostly prep_applied_to pobj_to_reviews advmod_assign_where nsubj_assign_systems rcmod_reviews_assign det_sentiment_a amod_sentiment_positive cc_positive_or conj_positive_negative dobj_assign_sentiment prep_sentiment_for det_document_a amod_document_whole nn_document_review pobj_for_document ' +o,1176,'A description of the flat featurized dependency-style syntactic representation we use is available in , which describes how the entire Penn Treebank was converted to this representation ',Marcus,'det_description_A nsubj_available_description prep_description_of det_representation_the amod_representation_flat dep_flat_featurized amod_representation_dependency-style amod_representation_syntactic pobj_of_representation nsubj_use_we rcmod_representation_use cop_available_is prep_available_in pobj_in_ nsubj_describes_which rcmod__describes advmod_converted_how det__the amod__entire nn__Penn nn__Treebank nsubjpass_converted_ auxpass_converted_was ccomp_describes_converted prep_converted_to det_representation_this pobj_to_representation ' +o,1177,'According to this model , when translating a stringf in the source language into the target language , a string e is chosen out of all target language strings e if it has the maximal probability given f : e = arg maxe -LCB- Pr -LRB- e f -RRB- -RCB- = arg maxe -LCB- Pr -LRB- f e -RRB- Pr -LRB- e -RRB- -RCB- where Pr -LRB- f e -RRB- is the translation model and Pr -LRB- e -RRB- is the target language model ',Brown,'prep_arg_According dep_According_to det_model_this pobj_to_model advmod_chosen_when dep_chosen_translating det_stringf_a dobj_translating_stringf prep_stringf_in det_language_the nn_language_source pobj_in_language prep_translating_into det_language_the nn_language_target pobj_into_language det_e_a nn_e_string nsubjpass_chosen_e auxpass_chosen_is dep_model_chosen prt_chosen_out prep_chosen_of det_strings_all nn_strings_target nn_strings_language pobj_of_strings dep_strings_e mark_has_if nsubj_has_it dep_strings_has det_probability_the amod_probability_maximal dobj_has_probability partmod_probability_given nsubj__f xcomp_given_ dep_chosen_e dep_e_= dobj_arg_maxe dep_maxe_Pr dep_f_e dep_-LRB-_f dep_maxe_= nn_maxe_arg nsubj_arg_maxe nn_Pr_Pr dep_-LRB-_f dep_f_e dep_maxe_Pr dep_-LRB-_e advmod_model_where nsubj_model_Pr dep_-LRB-_f dep_f_e cop_model_is det_model_the nn_model_translation rcmod_maxe_model cc_model_and nsubj_model_Pr dep_-LRB-_e cop_model_is det_model_the nn_model_target nn_model_language conj_model_model ' +o,1178,' evaluates both estimation techniques on the Bayesian bitag model ; Goldwater and Griffiths emphasize the advantage in the MCMC approach of integrating out the HMM parameters in a tritag model , yielding a tagging supported by many different parameter settings ',Johnson,'nsubj_evaluates_ det_techniques_both nn_techniques_estimation dobj_evaluates_techniques prep_evaluates_on det_model_the nn_model_Bayesian nn_model_bitag pobj_on_model nsubj__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths parataxis_evaluates_ xcomp__emphasize det_advantage_the dobj_emphasize_advantage prep_emphasize_in det_approach_the nn_approach_MCMC pobj_in_approach prep_approach_of pcomp_of_integrating prt_integrating_out det_parameters_the nn_parameters_HMM dobj_integrating_parameters prep_integrating_in det_model_a amod_model_tritag pobj_in_model dep_integrating_yielding det_tagging_a dobj_yielding_tagging partmod_tagging_supported prep_supported_by amod_settings_many amod_settings_different nn_settings_parameter pobj_by_settings ' +o,1179,'Conjunctions are a major source of errors for English chunking as well 9 , and we plan to address them in future work ',Ramshaw,'nsubj_source_Conjunctions cop_source_are det_source_a amod_source_major prep_source_of pobj_of_errors prep_errors_for pobj_for_English partmod_English_chunking prep_chunking_as quantmod_9_well number_9_ pobj_as_9 cc_source_and nsubj_plan_we conj_source_plan aux_address_to xcomp_plan_address dobj_address_them prep_address_in amod_work_future pobj_in_work ' +o,1180,'2.1 The Evaluator The evaluator is a function p(t[t\', s) which assigns to each target-text unit t an estimate of its probability given a source text s and the tokens t\' which precede t in the current translation of s. 1 Our approach to modeling this distribution is based to a large extent on that of the IBM group (Brown et al. , 1993), but it differs in one significant aspect: whereas the IBM model involves a \'noisy channel\' decomposition, we use a linear combination of separate predictions from a language model p(tlt ~) and a translation model p(tls ).',Brown,'' +o,1181,' shows that setting those weights should take into account the evaluation metric by which the MT system will eventually be judged ',Och,'advmod_shows_ dep_take_that csubj_take_setting det_weights_those dobj_setting_weights aux_take_should dep_shows_take prep_take_into pobj_into_account det_metric_the nn_metric_evaluation dobj_take_metric rel_judged_by pobj_by_which det_system_the nn_system_MT nsubjpass_judged_system aux_judged_will advmod_judged_eventually auxpass_judged_be rcmod_metric_judged ' +o,1182,'Two more recent investigations are by Yarowsky , , and later , Mihalcea , ',Yarowsky,'num_investigations_Two quantmod_Two_more amod_investigations_recent nsubj_are_investigations prep_are_by pobj_by_Yarowsky conj_Yarowsky_ cc_Yarowsky_and advmod_Yarowsky_later nn__Mihalcea conj_Yarowsky_ ' +o,1183,'F (Cahill et al. , 2004) overall 95.98 57.86 72.20 73.00 40.28 51.91 90.16 54.35 67.82 65.54 36.16 46.61 args only 98.64 42.03 58.94 82.69 30.54 44.60 86.36 36.80 51.61 66.08 24.40 35.64 Basic Model overall 92.44 91.28 91.85 63.87 62.15 63.00 63.12 62.33 62.72 42.69 41.54 42.10 args only 89.42 92.95 91.15 60.89 63.45 62.15 47.92 49.81 48.84 31.41 32.73 32.06 Basic Model with Subject Path Constraint overall 92.16 91.36 91.76 63.72 62.20 62.95 75.96 75.30 75.63 50.82 49.61 50.21 args only 89.04 93.08 91.02 60.69 63.52 62.07 66.15 69.15 67.62 42.77 44.76 44.76 Table 7: Evaluation of trace insertion and antecedent recovery for C04 algorithm, our basic algorithm and basic algorithm with the subject path constraint.',Cahill,'' +o,1184,'In general , Agold \\/ Acandidates ; following and for parse reranking and for translation reranking , we define Aoracle as alignment in Acandidates that is most similar to Agold8 We update each feature weight i as follows : i = i + hAoraclei hA1-besti 9 Following , after each training pass , we average all the feature weight vectors seen during the pass , and decode the discriminative training set using the vector of averaged feature weights ',Liang,'prep_define_In pobj_In_general nn_Acandidates_Agold nn_Acandidates_\\/ appos_general_Acandidates amod__following dep_Acandidates_ cc__and conj__ prep__for amod_reranking_parse pobj_for_reranking cc_reranking_and conj_reranking_ prep__for nn_reranking_translation pobj_for_reranking nsubj_define_we nsubj_alignment_Aoracle advmod_alignment_as xcomp_define_alignment prep_alignment_in pobj_in_Acandidates nsubj_similar_that cop_similar_is advmod_similar_most rcmod_Acandidates_similar aux_Agold8_to xcomp_similar_Agold8 nsubj_update_We ccomp_Agold8_update det_weight_each nn_weight_feature dobj_update_weight dep_average_i mark_follows_as dep_average_follows nn_hAoraclei_i nn_hAoraclei_= nn_hAoraclei_i nn_hAoraclei_+ dobj_follows_hAoraclei amod_hAoraclei_hA1-besti tmod_hA1-besti_9 prep_hAoraclei_Following pobj_Following_ mark_pass_after det_training_each nsubj_pass_training dep__pass nsubj_average_we dep_define_average predet_the_all nsubj_feature_the ccomp_average_feature nn_vectors_weight dobj_feature_vectors partmod_vectors_seen prep_seen_during det_pass_the pobj_during_pass cc_feature_and conj_feature_decode det_training_the amod_training_discriminative dobj_feature_training partmod_training_set xcomp_set_using det_vector_the dobj_using_vector prep_vector_of amod_weights_averaged nn_weights_feature pobj_of_weights ' +o,1185,'Here , under the ITG constraint , we need to consider just two kinds of reorderings , straight and inverted between two consecutive blocks ',Wu,'advmod_need_Here prep_need_under det__the nn__ITG nn__constraint pobj_under_ nsubj_need_we aux_consider_to xcomp_need_consider quantmod_two_just num_kinds_two dobj_consider_kinds prep_kinds_of pobj_of_reorderings amod_kinds_straight cc_straight_and conj_straight_inverted prep_inverted_between num_blocks_two amod_blocks_consecutive pobj_between_blocks ' +o,1186,'1 word w 2 word bigram w1w2 3 single-character word w 4 a word of length l with starting character c 5 a word of length l with ending character c 6 space-separated characters c1 and c2 7 character bigram c1c2 in any word 8 the first / last characters c1 / c2 of any word 9 word w immediately before character c 10 character c immediately before word w 11 the starting characters c1 and c2 of two consecutive words 12 the ending characters c1 and c2 of two consecutive words 13 a word of length l with previous word w 14 a word of length l with next word w Table 1: Feature templates for the baseline segmentor 2 The Baseline System We built a two-stage baseline system, using the perceptron segmentation model from our previous work (Zhang and Clark, 2007) and the perceptron POS tagging model from Collins (2002).',Collins,'' +o,1187,'A number of systems for automatically learning semantic parsers have been proposed ',Zettlemoyer,'det_number_A nsubjpass_proposed_number prep_number_of pobj_of_systems prep_systems_for advmod_learning_automatically pcomp_for_learning amod_parsers_semantic dobj_learning_parsers aux_proposed_have auxpass_proposed_been ' +o,1188,'791 and score the alignment template models phrases ',Koehn,'cc_791_and conj_791_score det_models_the amod_models_alignment nn_models_template dep_791_models dep_791_phrases ' +o,1189,'The pervading method for estimating these probabilities is a simple heuristic based on the relative frequency of the phrase pair in the multi-set of the phrase pairs extracted from the word-aligned corpus ',Koehn,'det_method_The amod_method_pervading nsubj_heuristic_method prep_method_for pcomp_for_estimating det_probabilities_these dobj_estimating_probabilities cop_heuristic_is det_heuristic_a amod_heuristic_simple partmod_heuristic_based prep_based_on det_frequency_the amod_frequency_relative pobj_on_frequency prep_frequency_of det_pair_the nn_pair_phrase pobj_of_pair prep_based_in det_multi-set_the pobj_in_multi-set prep_multi-set_of det_pairs_the nn_pairs_phrase pobj_of_pairs partmod_pairs_extracted prep_extracted_from det_corpus_the amod_corpus_word-aligned pobj_from_corpus ' +o,1190,'Metrics based on syntactic similarities such as the head-word chain metric -LRB- HWCM -RRB- ',Liu,'nsubj_based_Metrics prep_based_on amod_similarities_syntactic pobj_on_similarities dep_as_such prep_similarities_as det_metric_the amod_metric_head-word nn_metric_chain pobj_as_metric abbrev_metric_HWCM ' +o,1191,'4 Filtering with the CFG Rule Dictionary We use an idea that is similar to the method proposed by Ratnaparkhi for partof-speech tagging ',Ratnaparkhi,'partmod_4_Filtering prep_Filtering_with det_Dictionary_the nn_Dictionary_CFG nn_Dictionary_Rule pobj_with_Dictionary nsubj_use_We dep_4_use det_idea_an dobj_use_idea nsubj_similar_that cop_similar_is rcmod_idea_similar prep_similar_to det_method_the pobj_to_method partmod_method_proposed prep_proposed_by nn__Ratnaparkhi pobj_by_ prep__for amod_tagging_partof-speech pobj_for_tagging ' +o,1192,' used the Base-NP tag set as presented in : I for inside a Base-NP , O for outside a Base-NP , and B for the first word in a Base-NP following another Base-NP ',Ramshaw,'nsubj_used_ det_tag_the nn_tag_Base-NP nsubj_set_tag ccomp_used_set dep_presented_as advcl_set_presented prep_presented_in pobj_in_ dep_presented_I prep_used_for dep_for_inside det_Base-NP_a pobj_inside_Base-NP dep_outside_O dep_outside_for prep_used_outside det_B_a nn_B_Base-NP cc_B_and pobj_outside_B prep_B_for det_word_the amod_word_first pobj_for_word prep_word_in det_Base-NP_a pobj_in_Base-NP prep_Base-NP_following det_Base-NP_another pobj_following_Base-NP ' +o,1193,'After maximum BLEU tuning on a held-out tuning set , we evaluate translation quality on a held-out test set ',Och,'prep_set_After amod__maximum nn__BLEU amod__tuning pobj_After_ prep__on det_tuning_a amod_tuning_held-out pobj_on_tuning nsubj_evaluate_we ccomp_set_evaluate nn_quality_translation nsubj_set_quality prep_quality_on det_test_a amod_test_held-out pobj_on_test partmod_test_set ' +p,1194,'However , evaluations on the widely used WSJ corpus of the Penn Treebank show that the accuracy of these parsers still lags behind the state-of-theart ',Marcus,'advmod_show_However nsubj_show_evaluations prep_evaluations_on det_corpus_the advmod_used_widely amod_corpus_used nn_corpus_WSJ pobj_on_corpus prep_evaluations_of det__the nn__Penn nn__Treebank pobj_of_ complm_lags_that det_accuracy_the nsubj_lags_accuracy prep_accuracy_of det_parsers_these pobj_of_parsers advmod_lags_still ccomp_show_lags prep_lags_behind det_state-of-theart_the pobj_behind_state-of-theart ' +o,1195,'SEP\\/epsilon a\\/A # epsilon \\/ # a\\/epsilon a\\/epsilon b\\/epsilon b\\/B UNK\\/epsilon c\\/C b\\/epsilon c\\/BC e \\/ + E epsilon \\/ + d\\/epsilon d\\/epsilon epsilon\\/epsilon b\\/AB # b\\/A # B # e \\/ + DE c\\/epsilon d\\/BCD e \\/ + D+E Figure 1 : Illustration of dictionary based segmentation finite state transducer 31 Bootstrapping In addition to the model based upon a dictionary of stems and words , we also experimented with models based upon character n-grams , similar to those used for Chinese segmentation ',Berger,'prep_\\/_SEP\\/epsilon pobj_SEP\\/epsilon_a\\/A dep_epsilon_# nsubj_\\/_epsilon dep_a\\/epsilon_# num_b\\/epsilon_a\\/epsilon nn_b\\/epsilon_a\\/epsilon measure_b\\/B_b\\/epsilon amod_c\\/BC_b\\/B nn_c\\/BC_UNK\\/epsilon amod_c\\/BC_c\\/C nn_c\\/BC_b\\/epsilon dobj_\\/_c\\/BC dep_\\/_e dep_\\/_\\/ amod_epsilon_+ nn_epsilon_E nsubj_\\/_epsilon dep_\\/_\\/ cc_\\/_+ amod_B_d\\/epsilon nn_B_d\\/epsilon nn_B_epsilon\\/epsilon nn_B_b\\/AB dep_#_# number_#_b\\/A num_B_# measure_\\/_B dep_e_# num_B_e acomp_\\/_\\/ cc_\\/_+ conj_\\/_DE nsubj_d\\/BCD_c\\/epsilon ccomp_DE_d\\/BCD dep_based_e amod_Illustration_\\/ cc_\\/_+ conj_\\/_D+E dep_Illustration_Figure num_Figure_1 nsubj_based_Illustration prep_Illustration_of pobj_of_dictionary dep_experimented_based amod_state_segmentation amod_state_finite nsubj_transducer_state xcomp_based_transducer num_Bootstrapping_31 dep_transducer_Bootstrapping dep_transducer_In pobj_In_addition prep_based_to det_model_the pobj_to_model partmod_model_based prep_based_upon det_dictionary_a pobj_upon_dictionary prep_dictionary_of pobj_of_stems cc_stems_and conj_stems_words nsubj_experimented_we advmod_experimented_also ccomp_d\\/BCD_experimented prep_experimented_with pobj_with_models partmod_models_based prep_based_upon nn_n-grams_character pobj_upon_n-grams dep_experimented_similar prep_similar_to pobj_to_those partmod_those_used prep_used_for amod__Chinese nn__segmentation pobj_for_ ' +o,1196,'The results evaluated by BLEU score is shown in Table 2 ',Papineni,'det_results_The nsubjpass_shown_results partmod_results_evaluated prep_evaluated_by amod__BLEU nn__score pobj_by_ auxpass_shown_is prep_shown_in pobj_in_Table num_Table_2 ' +p,1197,'22 Maximum Entropy Models Maximum entropy -LRB- ME -RRB- models , also known as 928 log-linear and exponential learning models , provide a general purpose machine learning technique for classification and prediction which has been successfully applied to natural language processing including part of speech tagging , named entity recognition etc Maximum entropy models can integrate features from many heterogeneous information sources for classification ',Manning,'number_Maximum_22 num_models_Maximum nn_models_Entropy nn_models_Models nn_models_Maximum nn_models_entropy abbrev_models_ME nsubj_provide_ advmod_known_also partmod__known prep_known_as num_log-linear_928 pobj_as_log-linear cc_log-linear_and amod_models_exponential nn_models_learning conj_log-linear_models dep_models_provide det_machine_a amod_machine_general nn_machine_purpose dobj_provide_machine partmod_machine_learning dobj_learning_technique prep_technique_for pobj_for_classification cc_classification_and conj_classification_prediction nsubjpass_applied_which aux_applied_has auxpass_applied_been advmod_applied_successfully rcmod_technique_applied prep_applied_to amod_processing_natural nn_processing_language pobj_to_processing prep_processing_including pobj_including_part prep_part_of nn_tagging_speech pobj_of_tagging partmod_part_named nn_etc_entity nn_etc_recognition dobj_named_etc amod_models_Maximum nn_models_entropy nsubj_integrate_models aux_integrate_can rcmod_etc_integrate dobj_integrate_features prep_integrate_from amod_sources_many amod_sources_heterogeneous nn_sources_information pobj_from_sources prep_sources_for pobj_for_classification ' +o,1198,'2 Evaluating Heterogeneous Parser Output Two commonly reported shallow parsing tasks are Noun-Phrase -LRB- NP -RRB- Chunking and the CoNLL-2000 Chunking task , which extends the NPChunking task to recognition of 11 phrase types1 annotated in the Penn Treebank ',Ramshaw,'number_Evaluating_2 num_tasks_Evaluating nn_tasks_Heterogeneous nn_tasks_Parser nn_tasks_Output num_tasks_Two advmod_reported_commonly amod_tasks_reported nn_tasks_shallow nn_tasks_parsing nsubj_Noun-Phrase_tasks cop_Noun-Phrase_are abbrev_Noun-Phrase_NP nn__Chunking dep_Noun-Phrase_ cc_Noun-Phrase_and det__the amod__CoNLL-2000 nn__Chunking nn__task conj_Noun-Phrase_ nsubj_extends_which rcmod__extends det_task_the nn_task_NPChunking dobj_extends_task prep_extends_to pobj_to_recognition prep_recognition_of num_types1_11 nn_types1_phrase pobj_of_types1 partmod_types1_annotated prep_annotated_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +n,1199,'Although a rich literature covers bootstrapping methods applied to natural language problems several questions remain unanswered when these methods are applied to syntactic or semantic pattern acquisition ',Yarowsky,'mark_covers_Although det_literature_a amod_literature_rich nsubj_covers_literature csubj_applied_covers xcomp_covers_bootstrapping dobj_bootstrapping_methods prep_applied_to amod_problems_natural nn_problems_language pobj_to_problems amod_questions_ amod_questions_several nsubj_unanswered_questions cop_unanswered_remain dep_applied_unanswered advmod_applied_when det_methods_these nsubjpass_applied_methods auxpass_applied_are advcl_unanswered_applied prep_applied_to amod_acquisition_syntactic cc_syntactic_or conj_syntactic_semantic nn_acquisition_pattern pobj_to_acquisition ' +o,1200,'There are many possible methods for combining unlabeled and labeled data , but we simply concatenate unlabeled data with labeled data to see the effectiveness of the selected reliable parses ','Daume III','expl_are_There amod_methods_many amod_methods_possible nsubj_are_methods prep_methods_for pcomp_for_combining amod__unlabeled cc_unlabeled_and conj_unlabeled_labeled nn__data dobj_combining_ cc_are_but nsubj_concatenate_we advmod_concatenate_simply conj_are_concatenate amod_data_unlabeled dobj_concatenate_data prep_concatenate_with amod_data_labeled pobj_with_data aux_see_to xcomp_concatenate_see det_effectiveness_the dobj_see_effectiveness prep_effectiveness_of det_parses_the amod_parses_selected amod_parses_reliable pobj_of_parses ' +o,1201,'We ran the decoder with its default settings and then used Moses implementation of minimum error rate training to tune the feature weights on the development set ',Och,'nsubj_ran_We det_decoder_the dobj_ran_decoder prep_ran_with poss_settings_its nn_settings_default pobj_with_settings cc_ran_and advmod_used_then conj_ran_used nn_implementation_Moses dobj_used_implementation prep_implementation_of amod__minimum nn__error nn__rate nn__training pobj_of_ prep_used_to pobj_to_tune det_weights_the nn_weights_feature nsubj_set_weights prep_weights_on det_development_the pobj_on_development dep_ran_set ' +p,1202,'1 Introduction Many state-of-the-art machine translation -LRB- MT -RRB- systems over the past few years rely on several models to evaluate the goodness of a given candidate translation in the target language ',Koehn,'num_translation_1 amod_translation_Introduction amod_translation_Many amod_translation_state-of-the-art nn_translation_machine abbrev_translation_MT nsubj_rely_systems prep_systems_over det__the amod__past amod__few nn__years pobj_over_ dep_translation_rely prep_rely_on amod_models_several pobj_on_models aux_evaluate_to xcomp_rely_evaluate det_goodness_the dobj_evaluate_goodness prep_goodness_of det_translation_a amod_translation_given nn_translation_candidate pobj_of_translation prep_translation_in det_language_the nn_language_target pobj_in_language ' +o,1203,'55 Dependency validity features Like , we extract the dependency path from the question word to the common word -LRB- existing in both question and sentence -RRB- , and the path from candidate answer -LRB- such as CoNLL NE and numerical entity -RRB- to the common word for each pair of question and candidate sentence using Stanford dependency parser ',Manning,'num_validity_55 nn_validity_Dependency nsubj_features_validity prep_extract_Like pobj_Like_ nsubj_extract_we dep_features_extract det_path_the amod_path_dependency dobj_extract_path prep_path_from det_word_the nn_word_question pobj_from_word prep_word_to det_word_the amod_word_common pobj_to_word dep_path_existing prep_existing_in preconj_question_both pobj_in_question cc_question_and conj_question_sentence cc_path_and det_path_the conj_path_path prep_path_from nn_answer_candidate pobj_from_answer dep_as_such dep_answer_as nn_NE_CoNLL pobj_as_NE cc_NE_and amod_entity_numerical conj_NE_entity prep_answer_to det_word_the amod_word_common pobj_to_word prep_word_for det_pair_each pobj_for_pair prep_pair_of nn_sentence_question cc_question_and conj_question_candidate pobj_of_sentence partmod_path_using nn_parser_Stanford nn_parser_dependency dobj_using_parser ' +p,1204,'The use of dependencies in MT evaluation has not been extensively researched before -LRB- one exception here would be -RRB- , and requires more research to improve it , but the method shows potential to become an accurate evaluation metric ',Liu,'det_use_The nsubjpass_researched_use prep_use_of pobj_of_dependencies prep_dependencies_in nn_evaluation_MT pobj_in_evaluation aux_researched_has neg_researched_not auxpass_researched_been advmod_researched_extensively advmod_researched_before num_exception_one nsubj__exception advmod__here aux__would cop__be parataxis_researched_ cc_researched_and conj_researched_requires amod_research_more dobj_requires_research aux_improve_to infmod_research_improve dobj_improve_it cc_researched_but det_method_the nsubj_shows_method conj_researched_shows acomp_shows_potential aux_metric_to cop_metric_become det_metric_an amod_metric_accurate nn_metric_evaluation xcomp_potential_metric ' +o,1205,'Assuming that the parameters P -LRB- etk fsk -RRB- are known , the most likely alignment is computed by a simple dynamic-programming algorithm1 Instead of using an Expectation-Maximization algorithm to estimate these parameters , as commonly done when performing word alignment , we directly compute these parameters by relying on the information contained within the chunks ',Brown,'dep_compute_Assuming complm_known_that det_P_the nn_P_parameters nsubjpass_known_P nn_fsk_etk appos_P_fsk auxpass_known_are ccomp_Assuming_known det_alignment_the advmod_likely_most amod_alignment_likely nsubjpass_computed_alignment auxpass_computed_is parataxis_compute_computed prep_computed_by det_algorithm1_a amod_algorithm1_simple amod_algorithm1_dynamic-programming pobj_by_algorithm1 dep_of_Instead prep_algorithm1_of pcomp_of_using det_algorithm_an amod_algorithm_Expectation-Maximization dobj_using_algorithm aux_estimate_to xcomp_using_estimate det_parameters_these dobj_estimate_parameters advmod_commonly_as advmod_computed_commonly dep_computed_done advmod_performing_when advcl_done_performing nn__word nn__alignment dobj_performing_ nsubj_compute_we advmod_compute_directly det_parameters_these dobj_compute_parameters prep_compute_by pcomp_by_relying prep_relying_on det_information_the pobj_on_information partmod_information_contained prep_contained_within det_chunks_the pobj_within_chunks ' +o,1206,'We use binary Synchronous ContextFree Grammar -LRB- bSCFG -RRB- , based on Inversion Transduction Grammar -LRB- ITG -RRB- , to define the set of eligible segmentations for an aligned sentence pair ',Wu,'nsubj_use_We advmod_use_binary nn_Grammar_Synchronous nn_Grammar_ContextFree dobj_use_Grammar appos_Grammar_bSCFG prep_Grammar_based dep_based_on nn__Inversion nn__Transduction nn__Grammar abbrev__ITG pobj_on_ aux_define_to xcomp_use_define det_set_the dobj_define_set prep_set_of amod_segmentations_eligible pobj_of_segmentations prep_define_for det_pair_an amod_pair_aligned nn_pair_sentence pobj_for_pair ' +p,1207,'Albeit simple , the algorithm has proven to be very efficient and accurate for the task of parse selection ',Collins,'amod_simple_Albeit nsubj_proven_simple det_algorithm_the appos_simple_algorithm aux_proven_has aux_efficient_to cop_efficient_be advmod_efficient_very xcomp_proven_efficient cc_efficient_and conj_efficient_accurate prep_efficient_for det_task_the pobj_for_task prep_task_of amod_selection_parse pobj_of_selection ' +o,1208,'1 Introduction Over the past decade , researchers at IBM have developed a series of increasingly sophisticated statistical models for machine translation ',Dunning,'dep_developed_1 dep_Over_Introduction prep_developed_Over det_decade_the amod_decade_past pobj_Over_decade nsubj_developed_researchers prep_researchers_at pobj_at_IBM aux_developed_have det_series_a dobj_developed_series prep_series_of advmod_sophisticated_increasingly amod_models_sophisticated amod_models_statistical pobj_of_models prep_developed_for nn__machine nn__translation pobj_for_ ' +o,1209,'Given phrase p1 and its paraphrase p2 , we compute Score3 -LRB- p1 , p2 -RRB- by relative frequency : Score3 -LRB- p1 , p2 -RRB- = p -LRB- p2 p1 -RRB- = count -LRB- p2 , p1 -RRB- P pprime count -LRB- pprime , p1 -RRB- -LRB- 7 -RRB- People may wonder why we do not use the same method on the monolingual parallel and comparable corpora ',Koehn,'prep_compute_Given nn_p1_phrase dep_Given_p1 cc_p1_and poss_p2_its amod_p2_paraphrase conj_p1_p2 nsubj_compute_we dobj_compute_Score3 appos_Score3_p1 dep_p1_p2 prep_compute_by amod__relative nn__frequency pobj_by_ dobj_compute_Score3 appos_Score3_p1 dep_p1_p2 dep_Score3_= iobj_=_p amod_p1_p2 appos_count_p1 amod_count_= dobj_=_count appos_count_p2 dep_p2_p1 nn_count_P nn_count_pprime nn_People_count appos_count_pprime dep_pprime_p1 appos_count_7 nsubj_wonder_People aux_wonder_may dep_compute_wonder advmod_use_why nsubj_use_we aux_use_do neg_use_not advcl_wonder_use det_method_the amod_method_same dobj_use_method prep_use_on det_corpora_the amod_corpora_monolingual amod_corpora_parallel cc_parallel_and conj_parallel_comparable pobj_on_corpora ' +o,1210,'These include cube pruning , cube growing , early pruning , closing spans , coarse-to-fine methods , pervasive laziness , and many more ',Huang,'nsubj_include_These nn__cube nn__pruning dobj_include_ nn__cube nn__growing conj__ amod__early nn__pruning conj__ nn__closing nn__spans conj__ amod__coarse-to-fine nn__methods conj__ amod__pervasive nn__laziness conj__ cc__and amod_more_many conj__more ' +n,1211,'It also differs from previous proposals on lexical acquisition using statistical measures such as which either deny the prior existence of linguistic knowledge or use linguistic knowledge in ad hoc ways ',Brown,'nsubj_differs_It advmod_differs_also prep_differs_from amod_proposals_previous pobj_from_proposals prep_proposals_on amod_acquisition_lexical pobj_on_acquisition xcomp_differs_using amod_measures_statistical dobj_using_measures dep_as_such prep_measures_as pobj_as_ nsubj_deny_which preconj_deny_either rcmod__deny det_existence_the amod_existence_prior dobj_deny_existence prep_existence_of amod_knowledge_linguistic pobj_of_knowledge cc_deny_or conj_deny_use amod_knowledge_linguistic dobj_use_knowledge prep_use_in amod_ways_ad dep_ad_hoc pobj_in_ways ' +o,1212,'In recent years , sentiment classification has drawn much attention in the NLP field and it has many useful applications , such as opinion mining and summarization ',Titov,'prep_drawn_In amod_years_recent pobj_In_years nn_classification_sentiment nsubj_drawn_classification aux_drawn_has amod_attention_much dobj_drawn_attention prep_drawn_in det_field_the nn_field_NLP pobj_in_field cc_drawn_and nsubj_has_it conj_drawn_has amod_applications_many amod_applications_useful dobj_has_applications dep_as_such prep_applications_as nn_mining_opinion pobj_as_mining cc_mining_and conj_mining_summarization ' +o,1213,'To counteract this , we introduce two brevity penalty measures -LRB- BP -RRB- inspired by BLEU which we incorporate into the loss function , using a product , loss = 1PrecBP : BP1 = exp -LRB- 1max -LRB- 1 , rc -RRB- -RRB- -LRB- 6 -RRB- BP2 = exp -LRB- 1max -LRB- cr , rc -RRB- -RRB- where r is the reference length and c is the candidate length ',Papineni,'aux_counteract_To dobj_counteract_this nsubj_introduce_we dep_counteract_introduce num_measures_two amod_measures_brevity nn_measures_penalty dobj_introduce_measures abbrev_measures_BP partmod_measures_inspired prep_inspired_by nn__BLEU pobj_by_ rel_incorporate_which nsubj_incorporate_we rcmod__incorporate prep_incorporate_into det_function_the nn_function_loss pobj_into_function dep_counteract_using det_product_a dobj_using_product appos_product_loss dep_1PrecBP_= rcmod_loss_1PrecBP nsubj_=_BP1 parataxis_counteract_= dobj_=_exp dep_exp_1max appos_exp_1 dep_1_rc appos_exp_6 nsubj_=_BP2 rcmod_exp_= dobj_=_exp dep_exp_1max appos_exp_cr dep_cr_rc advmod_length_where nsubj_length_r cop_length_is det_length_the nn_length_reference rcmod_exp_length cc_counteract_and nsubj_length_c cop_length_is det_length_the nn_length_candidate conj_counteract_length ' +o,1214,'129 5 Active learning Whereas a passive supervised learning algorithm is provided with a collection of training examples that are typically drawn at random , an active learner has control over the labeled data that it obtains ',Brown,'num_5_129 dep_5_Active xcomp_Active_learning mark_provided_Whereas det_algorithm_a amod_algorithm_passive amod_algorithm_supervised nn_algorithm_learning nsubjpass_provided_algorithm auxpass_provided_is advcl_has_provided prep_provided_with det_collection_a pobj_with_collection prep_collection_of nn_examples_training pobj_of_examples nsubjpass_drawn_that auxpass_drawn_are advmod_drawn_typically rcmod_collection_drawn prep_drawn_at pobj_at_random det_learner_an amod_learner_active nsubj_has_learner ccomp_learning_has dobj_has_control prep_control_over det_data_the amod_data_labeled pobj_over_data complm_obtains_that nsubj_obtains_it ccomp_Active_obtains acomp_obtains_ ' +o,1215,'The Penn Treebank documentation defines a commonly used set of tags ',Marcus,'det__The nn__Penn nn__Treebank nn__documentation nsubj_defines_ det_set_a advmod_used_commonly amod_set_used dobj_defines_set prep_set_of pobj_of_tags ' +o,1216,'One of the main directions is sentiment classification , which classifies the whole opinion document -LRB- eg , a product review -RRB- as positive or negative ',McDonald,'nsubj_classification_One prep_One_of det_directions_the amod_directions_main pobj_of_directions cop_classification_is nn_classification_sentiment nsubj_classifies_which rcmod_classification_classifies det_document_the amod_document_whole nn_document_opinion nsubj_positive_document appos_document_eg det_review_a nn_review_product dep_eg_review advmod_positive_as xcomp_classifies_positive cc_positive_or conj_positive_negative ' +o,1217,'Finally , we are investigating several avenues for using this system output for Machine Translation -LRB- MT -RRB- including : -LRB- 1 -RRB- aiding word alignment for other MT system ; and -LRB- 2 -RRB- aiding the creation various MT models involving analyzed text , eg , ',Shen,'advmod_investigating_Finally nsubj_investigating_we aux_investigating_are amod_avenues_several dobj_investigating_avenues prep_avenues_for pcomp_for_using det_output_this nn_output_system dobj_using_output prep_output_for nn_Translation_Machine pobj_for_Translation dep_using_MT xcomp_using_including dep_alignment_1 amod_alignment_aiding nn_alignment_word dep_avenues_alignment prep_alignment_for amod__other nn__MT nn__system pobj_for_ cc_avenues_and dep_models_2 advmod_models_aiding det_models_the nn_models_creation amod_models_various amod_models_MT conj_avenues_models prep_models_involving amod_text_analyzed pobj_involving_text appos_text_eg ' +o,1218,'Hence we use a beam-search decoder during training and testing ; our idea is similar to that of who used a beam-search decoder as part of a perceptron parsing model ',Collins,'advmod_use_Hence nsubj_use_we det_decoder_a amod_decoder_beam-search dobj_use_decoder prep_use_during pobj_during_training cc_training_and conj_training_testing poss_idea_our nsubj_similar_idea cop_similar_is parataxis_use_similar prep_similar_to pobj_to_that prep_that_of pobj_of_ nsubj_used_who rcmod_that_used det_decoder_a amod_decoder_beam-search dobj_used_decoder prep_used_as pobj_as_part prep_part_of det_model_a nn_model_perceptron amod_model_parsing pobj_of_model ' +o,1219,'2.2 Corpus occurrence In order to get a feel for the relative frequency of VPCs in the corpus targeted for extraction, namely 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 VPC types (\%) Corpus frequency Figure 1: Frequency distribution of VPCs in the WSJ Tagger correctextracted Prec Rec Ffl=1 Brill 135135 1.000 0.177 0.301 Penn 667800 0.834 0.565 0.673 Table 1: POS-based extraction results the WSJ section of the Penn Treebank, we took a random sample of 200 VPCs from the Alvey Natural Language Tools grammar (Grover et al. , 1993) and did a manual corpus search for each.',Marcus,'' +o,1220,'However , the pb features yields no noticeable improvement unlike in prefect lexical choice scenario ; this is similar to the findings in ',Koehn,'advmod_yields_However det_features_the nn_features_pb nsubj_yields_features det_improvement_no amod_improvement_noticeable dobj_yields_improvement prep_yields_unlike dep_unlike_in amod_scenario_prefect amod_scenario_lexical nn_scenario_choice pobj_in_scenario nsubj_similar_this cop_similar_is parataxis_yields_similar prep_similar_to det_findings_the pobj_to_findings prep_similar_in ' +o,1221,' gave a systematic examination of the efficacy of unigram , bigram and trigram features drawn from different representations surface text , constituency parse tree and dependency parse tree ',Jiang,'advmod_gave_ det_examination_a amod_examination_systematic dobj_gave_examination prep_examination_of det_efficacy_the pobj_of_efficacy prep_efficacy_of nn_features_unigram conj_unigram_bigram cc_unigram_and conj_unigram_trigram pobj_of_features partmod_features_drawn prep_drawn_from amod_representations_different pobj_from_representations nn_text_surface nsubj_gave_text nn_tree_constituency nn_tree_parse appos_text_tree cc_tree_and amod_tree_dependency nn_tree_parse conj_tree_tree ' +o,1222,'Consider the lexical model pw -LRB- ry rx -RRB- , defined following , with a denoting the most frequent word alignment observed for the rule in the training set ',Koehn,'det_pw_the amod_pw_lexical nn_pw_model dobj_Consider_pw nn_rx_ry appos_pw_rx partmod_pw_defined prep_defined_following pobj_following_ prep_pw_with det_denoting_a pobj_with_denoting det_alignment_the advmod_frequent_most amod_alignment_frequent nn_alignment_word dep_denoting_alignment partmod_denoting_observed prep_observed_for det_rule_the pobj_for_rule prep_rule_in det_set_the nn_set_training pobj_in_set ' +o,1223,'Statistical Model In SIFTs statistical model , augmented parse trees are generated according to a process similar to that described in ',Collins,'nn_Model_Statistical prep_generated_In nn_model_SIFTs amod_model_statistical pobj_In_model amod_trees_augmented nn_trees_parse nsubjpass_generated_trees auxpass_generated_are dep_Model_generated prep_generated_according dep_according_to det_process_a pobj_to_process amod_process_similar prep_similar_to pobj_to_that partmod_that_described prep_described_in ' +o,1224,'3.5 Regularization We apply lscript1 regularization (Ng, 2004; Gao et al., 2007) to make learning more robust to noise and control the effective dimensionality of the feature spacebysubtractingaweightedsumofabsolutevalues of parameter weights from the log-likelihood of the training data w = argmaxw LL(w) summationdisplay i Ci|wi| (6) We optimize the objective using a variant of the orthant-wise limited-memory quasi-Newton algorithm proposed by Andrew & Gao (2007).3 All values Ci are set to 1 in most of the experiments below, although we apply stronger regularization (Ci = 3) to reordering features.',Gao,'' +o,1225,'Our method does not suppose a uniform distribution over all possible phrase segmentationsas since each phrase tree has a probability ',Koehn,'poss_method_Our nsubj_suppose_method aux_suppose_does neg_suppose_not det_distribution_a amod_distribution_uniform dobj_suppose_distribution prep_suppose_over det__all amod__possible nn__phrase nn__segmentationsas pobj_over_ mark_has_since det_tree_each nn_tree_phrase nsubj_has_tree advcl_suppose_has det_probability_a dobj_has_probability ' +o,1226,'After this conversion , we had 1000 positive and 1000 negative examples for each domain , the same balanced composition as the polarity dataset ',Pang,'prep_had_After det_conversion_this pobj_After_conversion nsubj_had_we num_examples_1000 amod_examples_positive cc_positive_and conj_positive_1000 amod_examples_negative dobj_had_examples prep_examples_for det_domain_each pobj_for_domain det_composition_the amod_composition_same amod_composition_balanced appos_domain_composition prep_composition_as det_dataset_the amod_dataset_polarity pobj_as_dataset ' +o,1227,'The MT community has developed not only an extensive literature on alignment , but also standard , proven alignment tools such as GIZA + + ',Brown,'det_community_The nn_community_MT nsubj_developed_community aux_developed_has cc_literature_not dep_not_only det_literature_an amod_literature_extensive dobj_developed_literature prep_literature_on amod__alignment pobj_on_ dep_also_but cc_literature_also conj_literature_standard partmod_standard_proven amod_tools_alignment dobj_proven_tools dep_as_such prep_tools_as pobj_as_GIZA amod_GIZA_+ cc_+_+ conj_+_ ' +o,1228,'51 Evaluation of Translation Translations are evaluated on two automatic metrics : Bleu and PER , position independent error-rate ',Papineni,'num_Evaluation_51 nsubjpass_evaluated_Evaluation prep_Evaluation_of nn_Translations_Translation pobj_of_Translations auxpass_evaluated_are prep_evaluated_on num_metrics_two amod_metrics_automatic pobj_on_metrics nn__Bleu dep_metrics_ cc__and nn__PER nn__position amod__independent amod__error-rate conj__ ' +o,1229,'These methods have been used in machine translation , terminology research and translation aids , bilingual lexicography , collocation studies , word-sense disambiguation and information retrieval in a multilingual environment ',Brown,'det_methods_These nsubjpass_used_methods aux_used_have auxpass_used_been prep_used_in nn__machine nn__translation pobj_in_ amod_research_terminology conj__research cc_research_and nn__translation nn__aids conj_research_ amod__bilingual nn__lexicography conj__ nn__collocation nn__studies conj__ amod__word-sense nn__disambiguation conj__ cc__and nn_retrieval_information conj__retrieval prep_used_in det__a amod__multilingual nn__environment pobj_in_ ' +o,1230,'For instance , for Maximum Entropy , I picked for the basic theory , for an application -LRB- POS tagging in this case -RRB- , and for more advanced topics such as optimization and smoothing ',Berger,'prep_picked_For pobj_For_instance prep_picked_for nn_Entropy_Maximum pobj_for_Entropy nsubj_picked_I ccomp_picked_ prep__for det_theory_the amod_theory_basic pobj_for_theory conj__ prep__for det_application_an pobj_for_application dep_application_POS amod_POS_tagging prep_POS_in det_case_this pobj_in_case cc__and conj__ prep__for advmod_advanced_more amod_topics_advanced pobj_for_topics dep_as_such prep_topics_as pobj_as_optimization cc_optimization_and conj_optimization_smoothing ' +o,1231,'In comparison with shallow semantic analysis tasks, such as wordsense disambiguation (Ide and Jeaneronis, 1998) and semantic role labeling (Gildea and Jurafsky, 2002; Carreras and M`arquez, 2005), which only partially tackle this problem by identifying the meanings of target words or finding semantic roles of predicates, semantic parsing (Kate et al. , 2005; Ge and Mooney, 2005; Zettlemoyer and Collins, 2005) pursues a more ambitious goal mapping natural language sentences to complete formal meaning representations (MRs), where the meaning of each part of a sentence is analyzed, including noun phrases, verb phrases, negation, quantifiers and so on.',Collins,'' +o,1232,'The kappa value was used to evaluate the agreement among the judges and to estimate how difficult the evaluation task was ',Carletta,'det__The nn__kappa nn__value nsubjpass_used_ auxpass_used_was aux_evaluate_to ccomp_used_evaluate det_agreement_the dobj_evaluate_agreement prep_evaluate_among det_judges_the pobj_among_judges cc_evaluate_and aux_estimate_to conj_evaluate_estimate advmod_difficult_how dep_was_difficult det_task_the nn_task_evaluation nsubj_was_task ccomp_estimate_was ' +o,1233,'Table look-up using an explicit translation lexicon is sufficient and preferable for many multilingual NLP applications , including ` crummy \' MT on the World Wide Web , certain machine-assisted translation tools -LRB- eg ',Dunning,'nn_look-up_Table nsubj_sufficient_look-up partmod_look-up_using det_lexicon_an amod_lexicon_explicit nn_lexicon_translation dobj_using_lexicon cop_sufficient_is cc_sufficient_and conj_sufficient_preferable prep_sufficient_for amod_applications_many amod_applications_multilingual nn_applications_NLP pobj_for_applications prep_sufficient_including amod_MT_crummy pobj_including_MT prep_MT_on det__the nn__World nn__Wide nn__Web pobj_on_ amod_translation_certain amod_translation_machine-assisted appos_MT_translation nn_eg_tools dep_translation_eg ' +o,1234,'The local dependencies between sentiment labels on sentences is similar to the work of where soft local consistency constraints were created between every sentence in adocument and inference wassolved using a min-cut algorithm ',Pang,'det_dependencies_The amod_dependencies_local nsubj_similar_dependencies prep_dependencies_between nn_labels_sentiment pobj_between_labels prep_labels_on pobj_on_sentences cop_similar_is prep_similar_to det_work_the pobj_to_work prep_work_of pobj_of_ advmod_created_where amod_constraints_soft amod_constraints_local nn_constraints_consistency nsubjpass_created_constraints auxpass_created_were advcl_similar_created prep_created_between det_sentence_every pobj_between_sentence prep_sentence_in pobj_in_adocument cc_adocument_and conj_adocument_inference partmod_sentence_wassolved xcomp_wassolved_using det_algorithm_a amod_algorithm_min-cut dobj_using_algorithm ' +o,1235,'Training Set -LRB- Labeled English Reviews -RRB- : There are many labeled English corpora available on the Web and we used the corpus constructed for multi-domain sentiment classification 9 , because the corpus was large-scale and it was within similar domains as the test set ',Blitzer,'dobj_Training_Set nn_Reviews_Labeled nn_Reviews_English appos_Set_Reviews expl_are_There rcmod_Set_are acomp_are_many dep_Training_labeled amod_corpora_English dep_available_corpora dep_labeled_available prep_available_on det_Web_the pobj_on_Web cc_available_and nsubj_used_we conj_available_used det_corpus_the dobj_used_corpus partmod_corpus_constructed prep_constructed_for amod__multi-domain nn__sentiment nn__classification pobj_for_ nsubj_Training_9 mark_large-scale_because det_corpus_the nsubj_large-scale_corpus cop_large-scale_was dep_9_large-scale cc_large-scale_and nsubj_was_it conj_large-scale_was prep_was_within amod_domains_similar pobj_within_domains mark_set_as det_test_the nsubj_set_test advcl_was_set ' +o,1236,'Atthefinestlevel , thisinvolvesthealignment of words and phrases within two sentences that are known to be translations ',Brown,'dep_Atthefinestlevel_thisinvolvesthealignment prep_Atthefinestlevel_of pobj_of_words cc_words_and conj_words_phrases prep_Atthefinestlevel_within num_sentences_two pobj_within_sentences nsubjpass_known_that auxpass_known_are rcmod_sentences_known aux_translations_to cop_translations_be xcomp_known_translations ' +o,1237,'The parser is coupled with an on-line averaged perceptron as the learning method ',Collins,'det_parser_The nsubjpass_coupled_parser auxpass_coupled_is prep_coupled_with det__an amod__on-line amod__averaged nn__perceptron pobj_with_ prep__as det_method_the nn_method_learning pobj_as_method ' +o,1238,'Statistical parsers have been developed for TAG , LFG , and HPSG , among others ',Cahill,'nn_parsers_Statistical nsubjpass_developed_parsers aux_developed_have auxpass_developed_been prep_developed_for nn__TAG pobj_for_ nn__LFG conj__ cc__and nn__HPSG conj__ prep_developed_among pobj_among_others ' +o,1239,'For the give source text , S , it finds the most probable alignment set , A , and target text , T = Aa SaTpSTp -RRB- , -LRB- -RRB- -LRB- -LRB- 1 -RRB- Brown proposed five alignment models , called IBM Model , for an English-French alignment task based on equa68 tion -LRB- 1 -RRB- ',Brown,'prep_called_For pcomp_For_the ccomp_finds_give nn_text_source dobj_give_text dobj_give_S nsubj_finds_it dep_the_finds det_set_the advmod_probable_most amod_set_probable nn_set_alignment dobj_finds_set conj_set_A cc_set_and nn_text_target conj_set_text nn_SaTpSTp_T nn_SaTpSTp_= nn_SaTpSTp_Aa dobj_the_SaTpSTp dep_Brown_1 dep_-RRB-_Brown prep_Brown_ amod_five_proposed pobj__five amod_models_alignment dep_SaTpSTp_models nn_Model_IBM nsubj_called_Model appos_Model_for det_task_an amod_task_English-French nn_task_alignment pobj_for_task prep_for_based dep_based_on amod_tion_equa68 pobj_on_tion appos_tion_1 ' +o,1240,'Large volumes of training data of this kind are indispensable for constructing statistical translation models , acquiring bilingual lexicon , and building example-based machine translation -LRB- EBMT -RRB- systems ',Brown,'amod_volumes_Large nsubj_indispensable_volumes prep_volumes_of nn_data_training pobj_of_data prep_data_of det_kind_this pobj_of_kind cop_indispensable_are prep_indispensable_for pcomp_for_constructing amod_models_statistical nn_models_translation dobj_constructing_models advmod_constructing_ conj_constructing_acquiring amod__bilingual nn__lexicon dobj_acquiring_ cc_constructing_and conj_constructing_building amod_translation_example-based nn_translation_machine nn_systems_translation abbrev_systems_EBMT nsubj__systems xcomp_building_ ' +o,1241,'We use the default configuration of the measure in WordNet : : Similarity-012 package , and , with a single exception , the measure performed below Gic ; see BP in table 1 ',Pedersen,'nsubj_use_We det_configuration_the nn_configuration_default dobj_use_configuration prep_configuration_of det_measure_the pobj_of_measure prep_measure_in pobj_in_WordNet amod_package_Similarity-012 nsubj__package parataxis_use_ cc_use_and prep_see_with det_exception_a amod_exception_single pobj_with_exception det_measure_the ccomp_see_measure partmod_measure_performed prep_performed_below pobj_below_Gic conj_use_see dobj_see_BP prep_BP_in pobj_in_table num_table_1 ' +o,1242,'The previous studies , with the exception of , used smaller gazetteers than ours ',Kazama,'det_studies_The amod_studies_previous nsubj_used_studies prep_studies_with det_exception_the pobj_with_exception prep_exception_of pobj_of_ amod_gazetteers_smaller dobj_used_gazetteers prep_used_than pobj_than_ours ' +o,1243,'This approach is similar to that of seed words -LRB- eg , -RRB- or hook words -LRB- eg , -RRB- in previous work ',Davidov,'det_approach_This nsubj_similar_approach cop_similar_is prep_similar_to pobj_to_that prep_that_of nn_words_seed pobj_of_words appos_words_eg dep_eg_ cc_words_or nn_words_hook conj_words_words appos_words_eg dep_eg_ prep_similar_in amod_work_previous pobj_in_work ' +o,1244,'The features we use are shown in Table 2 , which are based on the features used by and Uchimoto et al ',Ratnaparkhi,'det_features_The nsubjpass_shown_features nsubj_use_we rcmod_features_use auxpass_shown_are prep_shown_in pobj_in_Table num_Table_2 nsubjpass_based_which auxpass_based_are rcmod_Table_based prep_based_on det_features_the pobj_on_features partmod_features_used prep_used_by pobj_by_ cc__and conj__Uchimoto cc_Uchimoto_et conj_Uchimoto_al ' +o,1245,'Consequently , here we employ multiple references to evaluate MT systems like BLEU and NIST ',Papineni,'advmod_employ_Consequently advmod_we_here nsubj_employ_we amod_references_multiple nsubj_evaluate_references aux_evaluate_to xcomp_employ_evaluate nn_systems_MT dobj_evaluate_systems prep_evaluate_like nn__BLEU pobj_like_ cc__and nn__NIST conj__ ' +o,1246,'31 A simple solution suggests that in order to have an ITG take advantage of a known partial structure , one can simply stop the parser from using any spans that would violate the structure ',Wu,'det__A amod__simple nn__solution nsubj_suggests_ dep_31_suggests complm_stop_that mark_have_in dep_have_order aux_have_to dep_stop_have det_ITG_an nsubj_take_ITG ccomp_have_take dobj_take_advantage prep_advantage_of det_structure_a amod_structure_known amod_structure_partial pobj_of_structure nsubj_stop_one aux_stop_can advmod_stop_simply ccomp_suggests_stop det_parser_the dobj_stop_parser prep_stop_from pcomp_from_using det_spans_any dobj_using_spans nsubj_violate_that aux_violate_would rcmod_spans_violate det_structure_the dobj_violate_structure ' +p,1247,'It is an online training algorithm and has been successfully used in many NLP tasks , such as POS tagging , parsing , Chinese word segmentation , and so on ',Collins,'nsubjpass_algorithm_It cop_algorithm_is det_algorithm_an amod_algorithm_online nn_algorithm_training cc_algorithm_and aux_used_has auxpass_used_been advmod_used_successfully conj_algorithm_used prep_used_in amod_tasks_many nn_tasks_NLP pobj_in_tasks dep_as_such dep_tasks_as dep__POS amod__tagging pobj_as_ partmod__parsing amod__ amod__Chinese nn__word nn__segmentation dobj_parsing_ cc_as_and mark_on_so conj_as_on ' +o,1248,'We then rank-order the P X|Y MI XY M Z Pr Z|Y MI ZY G092log [P X P Y P X P Y ] f Y [P XY P XY ] f XY [P XY P XY ] f XY M iG13X,X} jG13Y,Y} (f ij G09 ij ) 2 ij f XY G09 XY XY (1G09( XY /N)) f XY G09 XY f XY (1G09(f XY /N)) Table 1: Probabilistic Approaches METHOD FORMULA Frequency (Guiliano, 1964) f XY Pointwise Mutual Information (MI) (Fano, 1961; Church and Hanks, 1990) log (P / PP) 2XY XY Selectional Association (Resnik, 1996) Symmetric Conditional Probability (Ferreira and Pereira, 1999) P / PP XY X Y 2 Dice Formula (Dice, 1945) 2 f / (f +f ) XY X Y Log-likelihood (Dunning, 1993; (Daille, 1996).',Church,'' +o,1249,'Similarly , propose a relative distortion model to be used with a phrase decoder ',Koehn,'advmod_propose_Similarly nsubj_propose_ det_model_a amod_model_relative nn_model_distortion nsubjpass_used_model aux_used_to auxpass_used_be xcomp_propose_used prep_used_with det_decoder_a nn_decoder_phrase pobj_with_decoder ' +p,1250,'We wish to minimize this error function , so we select accordingly : argmin summationdisplay a E -LRB- a -RRB- -LRB- a , -LRB- argmax a p -LRB- a , f e -RRB- -RRB- -RRB- -LRB- 4 -RRB- Maximizing performance for all of the weights at once is not computationally tractable , but has described an efficient one-dimensional search for a similar problem ',Och,'nsubj_wish_We aux_minimize_to xcomp_wish_minimize det_function_this nn_function_error dobj_minimize_function dep_wish_so nsubj_select_we dep_wish_select dobj_select_accordingly nn_summationdisplay_argmin dep_accordingly_summationdisplay det_E_a dep_summationdisplay_E dep_-LRB-_a dep_summationdisplay_a dep_summationdisplay_argmax det_p_a dep_argmax_p dep_p_a nn_e_f appos_a_e dep_tractable_4 nn_performance_Maximizing nsubj_tractable_performance prep_performance_for pobj_for_all prep_all_of det_weights_the pobj_of_weights prep_weights_at pobj_at_once cop_tractable_is neg_tractable_not advmod_tractable_computationally dep_wish_tractable cc_tractable_but nsubj_described_ aux_described_has conj_tractable_described det_search_an amod_search_efficient amod_search_one-dimensional dobj_described_search prep_search_for det_problem_a amod_problem_similar pobj_for_problem ' +o,1251,'An extension to WordNet was presented by ',Snow,'det_extension_An nsubjpass_presented_extension prep_extension_to pobj_to_WordNet auxpass_presented_was prep_presented_by ' +o,1252,'Discovering orientations of context dependent opinion comparative words is related to identifying domain opinion words ',Kanayama,'amod_orientations_Discovering prep_orientations_of nn_opinion_context amod_opinion_dependent pobj_of_opinion amod_words_comparative nsubjpass_related_words auxpass_related_is dep_orientations_related prep_related_to pcomp_to_identifying nn_words_domain nn_words_opinion dobj_identifying_words ' +o,1253,'Given a weight vector w , the score wf -LRB- x , y -RRB- ranks possible labelings of x , and we denote by Yk , w -LRB- x -RRB- the set of k top scoring labelings for x We use the standard B , I , O encoding for named entities ',Ramshaw,'prep_ranks_Given det_w_a nn_w_weight nn_w_vector dep_Given_w det_wf_the nn_wf_score nsubj_ranks_wf dep_wf_x appos_x_y amod_labelings_possible dobj_ranks_labelings prep_labelings_of pobj_of_x cc_ranks_and nsubj_denote_we conj_ranks_denote prep_denote_by pobj_by_Yk appos_Yk_w dep_labelings_x det_set_the nsubj_labelings_set prep_set_of amod_scoring_k amod_scoring_top pobj_of_scoring dep_ranks_labelings prep_labelings_for pobj_for_x nsubj_use_We rcmod_x_use det_B_the amod_B_standard nsubj_O_B appos_B_I ccomp_use_O xcomp_O_encoding prep_encoding_for amod_entities_named pobj_for_entities ' +o,1254,'To use the data from NANC , we use self-training ',McClosky,'aux_use_To dep_use_use det_data_the dobj_use_data prep_use_from pobj_from_NANC nsubj_use_we acomp_use_self-training ' +o,1255,'2 Related work recently advocated the need for a uniform approach to corpus-based semantic tasks ',Turney,'num__2 amod__Related nn__work dep_advocated_ advmod_advocated_recently det_need_the dobj_advocated_need prep_advocated_for det_approach_a amod_approach_uniform pobj_for_approach prep_advocated_to amod_tasks_corpus-based amod_tasks_semantic pobj_to_tasks ' +o,1256,'Similar to work in image retrieval , we cast the problem in terms of Machine Translation : given a paired corpus of words and a set of video event representations to which they refer , we make the IBM Model 1 assumption and use the expectation-maximization method to estimate the parameters : = + = m j ajm jvideowordpl Cvideowordp 1 -RRB- -LRB- -RRB- 1 -LRB- -RRB- -LRB- -LRB- 1 -RRB- This paired corpus is created from a corpus of raw video by first abstracting each video into the feature streams described above ',Brown,'ccomp_cast_Similar aux_work_to xcomp_Similar_work prep_work_in nn__image nn__retrieval pobj_in_ nsubj_cast_we det_problem_the dobj_cast_problem prep_cast_in pobj_in_terms prep_terms_of nn_Translation_Machine pobj_of_Translation prep_make_given det_corpus_a amod_corpus_paired dep_given_corpus prep_corpus_of pobj_of_words cc_corpus_and det_set_a conj_corpus_set prep_set_of amod_representations_video nn_representations_event pobj_of_representations rel_refer_to pobj_to_which nsubj_refer_they rcmod_set_refer nsubj_make_we parataxis_cast_make det_assumption_the nn_assumption_IBM nn_assumption_Model num_assumption_1 dobj_make_assumption cc_make_and conj_make_use det_method_the amod_method_expectation-maximization dobj_use_method aux__to dep__estimate det__the nn__parameters xcomp_use_ nn_Cvideowordp_= nn_Cvideowordp_+ nn_Cvideowordp_= nn_Cvideowordp_m nn_Cvideowordp_j nn_Cvideowordp_ajm amod_Cvideowordp_jvideowordpl dep__Cvideowordp dep_Cvideowordp_1 number_-RRB-_1 nsubjpass_created_1 measure_paired_This amod_1_paired advmod_created_corpus auxpass_created_is dep__created prep_created_from det_corpus_a pobj_from_corpus prep_corpus_of amod_video_raw pobj_of_video prep_created_by advmod_abstracting_first pcomp_by_abstracting det_video_each dobj_abstracting_video prep_abstracting_into det_streams_the nn_streams_feature pobj_into_streams partmod_streams_described prep_described_above ' +o,1257,'In this paper , sentence pairs are extracted by a simple model that is based on the so-called IBM Model1 ',Brown,'prep_extracted_In det_paper_this pobj_In_paper nn_pairs_sentence nsubjpass_extracted_pairs auxpass_extracted_are prep_extracted_by det_model_a amod_model_simple pobj_by_model nsubjpass_based_that auxpass_based_is rcmod_model_based prep_based_on det_Model1_the amod_Model1_so-called nn_Model1_IBM pobj_on_Model1 ' +o,1258,'The piecewise linearity observation made in is no longer applicable since we can not move the log operation into the expected value ',Papineni,'det_observation_The amod_observation_piecewise nn_observation_linearity nsubj_applicable_observation partmod_observation_made prep_made_in pobj_in_ cop_applicable_is advmod_applicable_no advmod_no_longer mark_move_since nsubj_move_we aux_move_can neg_move_not advcl_applicable_move det_operation_the nn_operation_log dobj_move_operation prep_move_into det_value_the amod_value_expected pobj_into_value ' +o,1259,'More specifically, by using translation probabilities, we can rewrite equation (11) and (12) as follow: nullnullnullnullnull null nullnull null nullnullnull null null nullnullnullnull null nullnull null null nullnull null nullnull null null | null null null null nullnull null nullnull null nullnull null null null null nullnull null nullnull null null null 1nullnull null nullnull null null null nullnull|nullnull (13) nullnullnullnullnull null nullnull null nullnullnull null null nullnullnullnull null nullnull null null nullnull null nullnull null null | null null null null nullnull null nullnull null nullnull null null null null nullnull null nullnull null null null 1nullnull null nullnull null null null nullnull|nullnull (14) where nullnullnullnull|null null null denotes the probability that topic term null is the translation of null null . In our experiments, to estimate the probability nullnullnullnull|null null null , we used the collections of question titles and question descriptions as the parallel corpus and the IBM model 1 (Brown et al., 1993) as the alignment model.',Brown,'' +o,1260,'Table 4 shows the linguistic features of the resulting model compared to the models of Carroll and Rooth , , and Charniak -LRB- 2000 -RRB- ',Collins,'nsubj_shows_Table num_Table_4 det_features_the amod_features_linguistic dobj_shows_features prep_features_of det_model_the amod_model_resulting pobj_of_model prep_model_compared dep_compared_to det_models_the pobj_to_models prep_models_of pobj_of_Carroll cc_Carroll_and conj_Carroll_Rooth num_Rooth_ num_Rooth_ cc_models_and conj_models_Charniak appos_Charniak_2000 ' +o,1261,'2 Related Work There has been a large and diverse body of research in opinion mining , with most research at the text , sentence or word level ',Pang,'num_Work_2 amod_Work_Related expl_been_There aux_been_has dep_Work_been det_body_a amod_body_large cc_large_and conj_large_diverse dobj_been_body prep_body_of pobj_of_research prep_research_in nn_mining_opinion pobj_in_mining prep_been_with amod_research_most pobj_with_research prep_been_at det__the nn__text pobj_at_ nn__sentence conj__ cc__or nn_level_word nn_level_ conj__level ' +o,1262,'A hierarchical alignment algorithm is a type of synchronous parser where , instead of constraining inferences by the production rules of a grammar , the constraints come from word alignments and possibly other sources ',Wu,'det_algorithm_A amod_algorithm_hierarchical nn_algorithm_alignment nsubj_type_algorithm cop_type_is det_type_a prep_type_of amod_parser_synchronous pobj_of_parser advmod_come_where dep_of_instead dep_come_of pcomp_of_constraining dobj_constraining_inferences prep_constraining_by det_rules_the nn_rules_production pobj_by_rules prep_rules_of det_grammar_a pobj_of_grammar det_constraints_the nsubj_come_constraints rcmod_type_come prep_come_from nn_alignments_word pobj_from_alignments cc_alignments_and advmod_sources_possibly amod_sources_other conj_alignments_sources ' +o,1263,'1 Introduction on measures for inter-rater reliability , on frameworks for evaluating spoken dialogue agents and on the use of different corpora in the development of a particular system -LRB- The Carnegie-Mellon Communicator , Eskenazi et al ',Carletta,'num_Introduction_1 dep_Introduction_on pobj_on_measures prep_measures_for amod__inter-rater nn__reliability pobj_for_ prep_measures_on pobj_on_frameworks prep_frameworks_for pcomp_for_evaluating ccomp_evaluating_spoken nn_agents_dialogue dobj_spoken_agents advmod_spoken_ cc_spoken_and conj_spoken_on det_use_the pobj_on_use prep_use_of amod_corpora_different pobj_of_corpora prep_on_in det_development_the pobj_in_development prep_development_of det_system_a amod_system_particular pobj_of_system det_Communicator_The nn_Communicator_Carnegie-Mellon dep_system_Communicator dep_Communicator_Eskenazi cc_Eskenazi_et conj_Eskenazi_al ' +o,1264,'In the future , we will experiment with semantic -LRB- rather than positional -RRB- clustering of premoditiers , using techniques such as those proposed in ',Marcus,'prep_experiment_In det_future_the pobj_In_future nsubj_experiment_we aux_experiment_will prep_experiment_with amod_clustering_semantic dep_than_rather dep_clustering_than pobj_than_positional pobj_with_clustering prep_clustering_of pobj_of_premoditiers partmod_clustering_using dobj_using_techniques dep_as_such prep_techniques_as pobj_as_those partmod_those_proposed prep_proposed_in pobj_in_ ' +o,1265,'Our learning method is an extension of Collinss perceptron-based method for sequence labeling ',Collins,'poss_method_Our nn_method_learning nsubj_extension_method cop_extension_is det_extension_an prep_extension_of nn_method_Collinss amod_method_perceptron-based pobj_of_method prep_extension_for nn_labeling_sequence pobj_for_labeling ' +o,1266,'These feature vectors and the associated parser actions are used to train maximum entropy models ',Berger,'det_vectors_These nn_vectors_feature nsubjpass_used_vectors cc_vectors_and det_actions_the amod_actions_associated nn_actions_parser conj_vectors_actions auxpass_used_are aux_train_to xcomp_used_train amod_models_maximum nn_models_entropy dobj_train_models ' +o,1267,'prime 1 1 1 05 05 1 05 05 01 01 01 00001 00001 01 00001 00001 Further , we ran each setting of each estimator at least 10 times -LRB- from randomly jittered initial starting points -RRB- for at least 1,000 iterations , as showed that some estimators require many iterations to converge ',Johnson,'nn_00001_prime num_00001_1 num_00001_1 num_00001_1 num_00001_05 num_00001_05 num_00001_1 num_00001_05 num_00001_05 num_00001_01 num_00001_01 num_00001_01 num_00001_00001 num_00001_00001 num_00001_01 nn_00001_00001 advmod_ran_Further nsubj_ran_we dep_00001_ran det_setting_each dobj_ran_setting prep_setting_of det_estimator_each pobj_of_estimator quantmod_10_at dep_at_least num_times_10 dep_estimator_times dep_estimator_from advmod_points_randomly amod_points_jittered amod_points_initial amod_points_starting pobj_from_points prep_estimator_for quantmod_1,000_at dep_at_least num_iterations_1,000 pobj_for_iterations mark_showed_as nsubj_showed_ dep_iterations_showed complm_require_that det_estimators_some nsubj_require_estimators ccomp_showed_require amod_iterations_many nsubj_converge_iterations aux_converge_to xcomp_require_converge ' +o,1268,'We use the discriminative perceptron learning algorithm to train the values of vectorw ',Collins,'nsubj_use_We det_perceptron_the amod_perceptron_discriminative dobj_use_perceptron partmod_perceptron_learning nn__algorithm dobj_learning_ aux_train_to xcomp_learning_train det_values_the dobj_train_values prep_values_of pobj_of_vectorw ' +o,1269,'7 Experiments To show the effectiveness of cross-language mention propagation information in improving mention detection system performance in Arabic , Chinese and Spanish , we use three SMT systems with very competitive performance in terms of BLEU11 ',Papineni,'num_Experiments_7 aux_show_To dep_use_show det_effectiveness_the dobj_show_effectiveness prep_effectiveness_of amod_information_cross-language nn_information_mention nn_information_propagation pobj_of_information prep_show_in pcomp_in_improving nn_performance_mention nn_performance_detection nn_performance_system dobj_improving_performance prep_improving_in pobj_in_Arabic conj_Arabic_Chinese cc_Arabic_and conj_Arabic_Spanish nsubj_use_we dep_Experiments_use num_systems_three nn_systems_SMT dobj_use_systems prep_use_with advmod_competitive_very amod_performance_competitive pobj_with_performance prep_performance_in pobj_in_terms prep_terms_of pobj_of_BLEU11 ' +o,1270,'The tagger described in this paper is based on the standard Hidden Markov Model architecture ',Marcus,'det_tagger_The nsubjpass_based_tagger partmod_tagger_described prep_described_in det_paper_this pobj_in_paper auxpass_based_is prep_based_on det__the amod__standard nn__Hidden nn__Markov nn__Model nn__architecture pobj_on_ ' +o,1271,'We measured associations using the log-likelihood measure for each combination of target category and semantic class by converting each cell of the contingency into a 22 contingency table ',Dunning,'nsubj_measured_We dobj_measured_associations partmod_associations_using det__the amod__log-likelihood nn__measure dobj_using_ prep__for det_combination_each pobj_for_combination prep_combination_of nn_category_target pobj_of_category cc_category_and amod_class_semantic conj_category_class prep_using_by pcomp_by_converting det_cell_each dobj_converting_cell prep_cell_of det_contingency_the pobj_of_contingency prep_converting_into det_table_a num_table_22 nn_table_contingency pobj_into_table ' +o,1272,'32 Rare Word Accuracy For these experiments , we use the Wall Street Journal portion of the Penn Treebank ',Marcus,'num_Accuracy_32 nn_Accuracy_Rare nn_Accuracy_Word prep_use_For det_experiments_these pobj_For_experiments nsubj_use_we dep_Accuracy_use det_portion_the nn_portion_Wall nn_portion_Street nn_portion_Journal dobj_use_portion prep_portion_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,1273,' extracts rules from non-anaphoric noun phrases and noun phrases patterns , which are then applied to test data to identify existential noun phrases ',Bean,'nn_rules_extracts dobj__rules prep__from amod_phrases_non-anaphoric nn_phrases_noun pobj_from_phrases cc_phrases_and nn_phrases_noun conj_phrases_phrases nsubj__patterns nsubjpass_applied_which auxpass_applied_are advmod_applied_then rcmod_patterns_applied aux_test_to xcomp_applied_test nsubj_identify_data aux_identify_to xcomp_test_identify amod_phrases_existential nn_phrases_noun dobj_identify_phrases ' +o,1274,'2 Previous Approaches method of estimating phrasetranslation probabilities is very simple ',Koehn,'num_Approaches_2 amod_Approaches_Previous amod_method_ nsubj_simple_method prep_method_of pcomp_of_estimating nn_probabilities_phrasetranslation dobj_estimating_probabilities cop_simple_is advmod_simple_very dep_Approaches_simple ' +o,1275,'The form of the maximum entropy probability model is identical to the one used in : k f $ -LRB- wi,wi-1 , wi-2 , at ~ ri -RRB- YIj = I Otj p -LRB- wilwi-l , wi-2 , attri -RRB- = Z -LRB- Wi-l , wi-2 , attri -RRB- k to t j = l where wi ranges over V t3 stop ',Berger,'det_form_The nsubj_identical_form prep_form_of det_model_the amod_model_maximum amod_model_entropy nn_model_probability pobj_of_model cop_identical_is prep_identical_to det_one_the pobj_to_one partmod_one_used prep_used_in pobj_in_ dep_k_k dep_k_f nsubj_k_$ nn_wi-2_wi,wi-1 dep_=_wi-2 dep_wi-2_at amod_p_~ nn_p_ri nn_p_YIj nn_p_= nn_p_I nn_p_Otj pobj_at_p nn_attri_wilwi-l dep_attri_wi-2 appos_wi-2_attri dep_Z_= num_$_Z appos_$_Wi-l dep_Wi-l_wi-2 dep_Wi-l_attri parataxis_identical_k prep_k_to nn_j_t pobj_to_j amod_l_= tmod_k_l advmod_stop_where nn_ranges_wi nsubj_stop_ranges prep_ranges_over nn_t3_V pobj_over_t3 advcl_k_stop ' +o,1276,'Instead , researchers routinely use automatic metrics like Bleu as the sole evidence of improvement to translation quality ',Papineni,'advmod_use_Instead nsubj_use_researchers advmod_use_routinely amod_metrics_automatic dobj_use_metrics prep_metrics_like nn__Bleu pobj_like_ prep__as det_evidence_the amod_evidence_sole pobj_as_evidence prep_evidence_of pobj_of_improvement prep_use_to nn_quality_translation pobj_to_quality ' +o,1277,'Rapp , -RRB- but using cosine rather than cityblock distance to measure profile similarity ',Dunning,'nn__Rapp amod__ prep__but dep_but_using dobj_using_cosine cc_cosine_rather dep_rather_than nn_distance_cityblock conj_cosine_distance aux_measure_to xcomp_using_measure nn_similarity_profile dobj_measure_similarity ' +o,1278,'32 Maximum Entropy ME models implement the intuition that the best model will be the one that is consistent with the set of constrains imposed by the evidence , but otherwise is as uniform as possible ',Berger,'num_models_32 amod_models_Maximum nn_models_Entropy nn_models_ME nsubj_implement_models det_intuition_the dobj_implement_intuition complm_one_that det_model_the amod_model_best nsubj_one_model aux_one_will cop_one_be det_one_the ccomp_implement_one nsubj_consistent_that cop_consistent_is rcmod_one_consistent prep_consistent_with det_set_the pobj_with_set prep_set_of pobj_of_constrains partmod_constrains_imposed prep_imposed_by det_evidence_the pobj_by_evidence cc_implement_but advmod_uniform_otherwise cop_uniform_is advmod_uniform_as conj_implement_uniform prep_uniform_as pobj_as_possible ' +o,1279,'The first model , referred to as Maxent1 below , is a loglinear combination of a trigram language model with a maximum entropy translation component that is an analog of the IBM translation model 2 ',Brown,'det_model_The amod_model_first nsubj_combination_model partmod_model_referred prep_referred_to quantmod_Maxent1_as pobj_to_Maxent1 advmod_referred_below cop_combination_is det_combination_a nn_combination_loglinear prep_combination_of det_model_a nn_model_trigram nn_model_language pobj_of_model prep_model_with det_component_a amod_component_maximum amod_component_entropy nn_component_translation pobj_with_component nsubj_analog_that cop_analog_is det_analog_an rcmod_component_analog prep_analog_of det_model_the nn_model_IBM nn_model_translation pobj_of_model num_model_2 ' +o,1280,'In Yarowsky \'s experiment , an average of 3936 examples were used to disambiguate between two senses ',Yarowsky,'prep_used_In poss__Yarowsky possessive_Yarowsky_\'s nn__experiment pobj_In_ det_average_an nsubjpass_used_average prep_average_of num_examples_3936 pobj_of_examples auxpass_used_were aux_disambiguate_to xcomp_used_disambiguate prep_disambiguate_between num_senses_two pobj_between_senses ' +o,1281,'= = = = = m aj j m j aj l i i l i ii m j j mlajdeft en pp m ap 0:1 11 1 2 0 0 0 -RRB- , -LRB- -RRB- , , -LRB- -RRB- -LRB- ! -RRB- -LRB- -RRB- , Pr -LRB- -RRB- , -LRB- 00 eef -LRB- 3 -RRB- 1 A cept is defined as the set of target words connected to a source word ',Brown,'nsubj_=_= dobj_=_= punct_=_= nn_=_= nn_=_m nn_=_aj nn_=_j nn_=_m nn_=_j nn_=_aj nn_=_l nn_=_i nn_=_i nn_=_l nn_=_i nn_=_ii nn_=_m nn_=_j nn_=_j nn_=_mlajdeft nn_=_en nn_=_pp nn_=_m dep_=_ap number_11_0:1 amod_=_11 number_2_1 dep_11_2 prep_2_0 num_0_0 pobj_0_0 nn_-LRB-_Pr advmod_eef_00 parataxis_-LRB-_eef dobj_eef_3 dep_defined_1 det_cept_A nsubjpass_defined_cept auxpass_defined_is dep_=_defined prep_defined_as det_set_the pobj_as_set prep_set_of nn_words_target pobj_of_words partmod_words_connected prep_connected_to det_word_a nn_word_source pobj_to_word ' +o,1282,'For Hw6 , students compared their POS tagging results with the ones reported in ',Ratnaparkhi,'prep_compared_For pobj_For_Hw6 nsubj_compared_students poss_results_their dep_results_POS nn_results_tagging dobj_compared_results prep_compared_with det_ones_the pobj_with_ones partmod_ones_reported prep_reported_in ' +o,1283,'A statistical language model a lexicalized PCFG is derived from the analysis grammar by processing a corpus using the same grammar with no statistical model and recording frequencies of substructures built by each rule ',Collins,'det_model_A amod_model_statistical nn_model_language det__a amod__lexicalized nn__PCFG nsubjpass_derived_ auxpass_derived_is rcmod_model_derived prep_derived_from det_grammar_the nn_grammar_analysis pobj_from_grammar prep_derived_by pcomp_by_processing det_corpus_a dobj_processing_corpus partmod_corpus_using det_grammar_the amod_grammar_same dobj_using_grammar prep_using_with det_model_no amod_model_statistical pobj_with_model cc_model_and amod_frequencies_recording conj_model_frequencies prep_model_of pobj_of_substructures partmod_substructures_built prep_built_by det_rule_each pobj_by_rule ' +o,1284,'We use GIZA + + to do m-to-n word-alignment and adopt heuristic grow-diag-final-and to do refinement ',Och,'nsubj_use_We nsubj_do_GIZA dep_do_+ cc_+_+ conj_+_ aux_do_to xcomp_use_do amod_word-alignment_m-to-n dobj_do_word-alignment cc_do_and conj_do_adopt amod_grow-diag-final-and_heuristic dobj_adopt_grow-diag-final-and aux_do_to xcomp_adopt_do dobj_do_refinement ' +o,1285,'The surface heuristic can define consistency according to any word alignment ; but most often , the alignment is provided by GIZA + + ',Och,'det_heuristic_The nn_heuristic_surface nsubj_define_heuristic aux_define_can dobj_define_consistency prep_define_according dep_according_to det_alignment_any nn_alignment_word pobj_to_alignment cc_provided_but advmod_often_most advmod_provided_often det_alignment_the nsubjpass_provided_alignment auxpass_provided_is parataxis_define_provided prep_provided_by nn_+_GIZA nn_+_+ pobj_by_+ ' +o,1286,'154 2 Translation Models 21 Standard Phrase-based Model Most phrase-based translation models rely on a pre-existing set of word-based alignments from which they induce their parameters ',Koehn,'number_2_154 num_models_2 nn_models_Translation nn_models_Models num_models_21 nn_models_Standard nn_models_Phrase-based num_models_Model advmod_phrase-based_Most amod_models_phrase-based nn_models_translation nsubj__models dep__rely prep_rely_on det_set_a amod_set_pre-existing pobj_on_set prep_set_of amod_alignments_word-based pobj_of_alignments prep_rely_from dobj_induce_which nsubj_induce_they pcomp_from_induce poss_parameters_their dobj_induce_parameters ' +o,1287,'Overall \% agreement among judges for 250 propositions 601 A commonly used metric for evaluating interrater reliability in categorization of data is the kappa statistic ',Carletta,'dep_\%_Overall amod_agreement_\% nsubj_statistic_agreement prep_agreement_among pobj_among_judges prep_judges_for num_propositions_250 pobj_for_propositions partmod_propositions_601 det_used_A advmod_used_commonly dobj_601_used amod_used_metric prep_used_for pcomp_for_evaluating nn_reliability_interrater dobj_evaluating_reliability prep_evaluating_in pobj_in_categorization prep_categorization_of pobj_of_data cop_statistic_is det_statistic_the nn_statistic_kappa ' +p,1288,'For French\\/English translation we use a state of the art phrase-based MT system similar to ',Och,'prep_use_For nn_translation_French\\/English pobj_For_translation nsubj_use_we det_state_a nsubj_similar_state prep_state_of det_system_the nn_system_art amod_system_phrase-based nn_system_MT pobj_of_system xcomp_use_similar prep_similar_to ' +o,1289,'A more optimistic view can be found in ; they argue that a near-100 \% interjudge agreement is possible , provided the part-of-speech annotation is done carefully by experts ',Marcus,'det_view_A advmod_optimistic_more amod_view_optimistic nsubjpass_found_view aux_found_can auxpass_found_be prep_found_in pobj_in_ nsubj_argue_they parataxis_found_argue complm_possible_that det_agreement_a num_\%_near-100 measure_interjudge_\% amod_agreement_interjudge nsubj_possible_agreement cop_possible_is ccomp_argue_possible prep_possible_provided det_annotation_the amod_annotation_part-of-speech nsubjpass_done_annotation auxpass_done_is dep_provided_done advmod_done_carefully prep_done_by pobj_by_experts ' +o,1290,'From the extracted n-grams , those with a flequc ` ncy of 3 or more were kept -LRB- other approaches get rid of n-grams of such low frequencies -RRB- ',Smadja,'prep_kept_From det_n-grams_the amod_n-grams_extracted pobj_From_n-grams nsubjpass_kept_those prep_those_with det_ncy_a amod_ncy_flequc pobj_with_ncy prep_ncy_of pobj_of_3 cc_3_or conj_3_more auxpass_kept_were amod_approaches_other nsubj_get_approaches parataxis_kept_get amod__rid prep_rid_of pobj_of_n-grams prep_n-grams_of amod_frequencies_such amod_frequencies_low pobj_of_frequencies dobj_get_ ' +o,1291,'This was expected , as it has been observed before that very simple smoothing techniques can perform well on large data sets , such as web data ',Brants,'nsubjpass_expected_This auxpass_expected_was mark_observed_as nsubjpass_observed_it aux_observed_has auxpass_observed_been advcl_expected_observed advmod_perform_before dep_perform_that advmod_simple_very amod_techniques_simple nn_techniques_smoothing nsubj_perform_techniques aux_perform_can ccomp_observed_perform advmod_perform_well prep_perform_on amod_sets_large nn_sets_data pobj_on_sets dep_as_such prep_sets_as nn_data_web pobj_as_data ' +o,1292,'On the other hand , proposed an algorithm , borrowed to the field of dynamic programming and based on the output of their previous work , to find the best alignment , subject to certain constraints , between words in parallel sentences ',Brown,'prep_proposed_On det_hand_the amod_hand_other pobj_On_hand nsubj_proposed_ det_algorithm_an dobj_proposed_algorithm dep_algorithm_borrowed prep_borrowed_to det_field_the pobj_to_field prep_field_of amod_programming_dynamic pobj_of_programming cc_borrowed_and conj_borrowed_based prep_based_on det_output_the pobj_on_output prep_output_of poss_work_their amod_work_previous pobj_of_work aux_find_to infmod_algorithm_find det_alignment_the amod_alignment_best dobj_find_alignment amod_alignment_subject prep_subject_to amod_constraints_certain pobj_to_constraints prep_alignment_between pobj_between_words prep_words_in amod_sentences_parallel pobj_in_sentences ' +p,1293,'In agreement with recent results on parsing with lexicalised probabilistic grammars , our main result is that statistics over lexical features best correspond to independently established truman intuitive preferences and experimental findings ',Collins,'prep_is_In pobj_In_agreement prep_agreement_with amod_results_recent pobj_with_results prep_results_on pcomp_on_parsing prep_parsing_with amod__lexicalised amod__probabilistic nn__grammars pobj_with_ poss_result_our amod_result_main nsubj_is_result complm_features_that nsubj_features_statistics prep_statistics_over pobj_over_lexical ccomp_is_features advmod_features_best dep_features_correspond aux_independently_to xcomp_correspond_independently amod_preferences_established amod_preferences_truman dep_truman_intuitive dobj_independently_preferences cc_preferences_and amod_findings_experimental conj_preferences_findings ' +o,1294,'21 Minimum Error Rate Training The predominant approach to reconciling the mismatch between the MAP decision rule and the evaluation metric has been to train the parameters of the exponential model to correlate the MAP choice with the maximum score as indicated by the evaluation metric on a development set with known references ',Och,'num_Rate_21 nn_Rate_Minimum nn_Rate_Error nsubj_been_Rate partmod_Rate_Training det_approach_The amod_approach_predominant dobj_Training_approach prep_Training_to pcomp_to_reconciling det_mismatch_the dobj_reconciling_mismatch prep_mismatch_between det_rule_the nn_rule_MAP nn_rule_decision pobj_between_rule cc_rule_and det_metric_the nn_metric_evaluation conj_rule_metric aux_been_has aux_train_to xcomp_been_train det_parameters_the dobj_train_parameters prep_parameters_of det_model_the amod_model_exponential pobj_of_model aux_correlate_to xcomp_train_correlate det_choice_the nn_choice_MAP dobj_correlate_choice prep_correlate_with det_score_the amod_score_maximum pobj_with_score mark_indicated_as advcl_correlate_indicated prep_indicated_by det_metric_the nn_metric_evaluation pobj_by_metric prep_indicated_on det_development_a pobj_on_development partmod_development_set prep_set_with amod_references_known pobj_with_references ' +o,1295,'In earlier work only singletons were used as seed words ; varying their number allows us to test whether multiple seed words have a positive effect in detection performance ',Turney,'prep_used_In amod__earlier nn__work pobj_In_ amod_singletons_only nsubjpass_used_singletons auxpass_used_were prep_used_as nn_words_seed pobj_as_words advmod_allows_varying poss_number_their nsubj_allows_number parataxis_used_allows nsubj_test_us aux_test_to xcomp_allows_test complm_have_whether amod_words_multiple nn_words_seed nsubj_have_words ccomp_test_have det_effect_a amod_effect_positive dobj_have_effect prep_effect_in nn_performance_detection pobj_in_performance ' +o,1296,'stituent alignments ',Galley,'amod_alignments_stituent ' +o,1297,'Many strategies have been proposed to integrate morphology information in SMT , including factored translation models , adding a translation dictionary containing inflected forms to the training data , entirely replacing surface forms by representations built on lemmas and POS tags , morphemes learned in an unsupervised manner , and using Porter stems and even 4-letter prefixes for word alignment ',Koehn,'amod_strategies_Many nsubjpass_proposed_strategies aux_proposed_have auxpass_proposed_been aux_integrate_to xcomp_proposed_integrate amod_information_morphology dobj_integrate_information prep_integrate_in pobj_in_SMT dep_integrate_including amod_models_factored nn_models_translation dobj_including_models advmod_including_ dep_including_adding det_dictionary_a nn_dictionary_translation dobj_adding_dictionary xcomp_adding_containing amod_forms_inflected dobj_containing_forms prep_containing_to det__the nn__training nn__data pobj_to_ advmod_including_entirely dep_including_replacing nn_forms_surface dobj_replacing_forms prep_replacing_by pobj_by_representations partmod_representations_built prep_built_on pobj_on_lemmas cc_lemmas_and dep__POS nn__tags conj_lemmas_ nsubj_learned_morphemes parataxis_replacing_learned prep_learned_in det__an amod__unsupervised nn__manner pobj_in_ cc_replacing_and conj_replacing_using nn_stems_Porter dobj_using_stems cc_stems_and advmod_stems_even amod_prefixes_4-letter dep_stems_prefixes prep_prefixes_for nn__word nn__alignment pobj_for_ ' +o,1298,'Manually defined heuristics are used to automatically annotate each tree in the treebank with partially specified HPSG derivation trees: Head/argument/modifier distinctions are made for each node in the tree based on Magerman (1994) and Collins (1997); 336 ODonovan et al. Large-Scale Induction and Evaluation of Lexical Resources the whole tree is then converted to a binary tree; heuristics are applied to deal with phenomena such as LDDs and coordination and to correct some errors in the treebank, and finally an HPSG category is assigned to each node in the tree in accordance with its CFG category.',Collins,'' +o,1299,'A synchronous 363 binarization method is proposed in whose basic idea is to build a left-heavy binary synchronous tree with a left-to-right shift-reduce algorithm ',Zhang,'det_method_A amod_method_synchronous tmod_synchronous_363 nn_method_binarization nsubjpass_proposed_method auxpass_proposed_is prep_proposed_in pobj_in_ poss_idea_whose amod_idea_basic nsubj_is_idea rcmod__is aux_build_to xcomp_is_build det__a amod__left-heavy advmod_synchronous_binary amod__synchronous nn__tree dobj_build_ prep__with det_algorithm_a amod_algorithm_left-to-right amod_algorithm_shift-reduce pobj_with_algorithm ' +o,1300,'This hypothesized relationship between distributional similarity and semantic similarity has given rise to a large body of work on automatic thesaurus generation ',Hindle,'det_relationship_This amod_relationship_hypothesized nsubj_given_relationship prep_relationship_between amod_similarity_distributional pobj_between_similarity cc_similarity_and amod_similarity_semantic conj_similarity_similarity aux_given_has dobj_given_rise prep_given_to det_body_a amod_body_large pobj_to_body prep_body_of pobj_of_work prep_given_on amod_generation_automatic nn_generation_thesaurus pobj_on_generation ' +o,1301,'The details of the algorithm can be found in the literature for statistical translation models , such as ',Brown,'det_details_The nsubjpass_found_details prep_details_of det_algorithm_the pobj_of_algorithm aux_found_can auxpass_found_be prep_found_in det_literature_the pobj_in_literature prep_literature_for amod_models_statistical nn_models_translation pobj_for_models advmod_as_such advmod_found_as ' +o,1302,'For comparing the sentence generator sample to the English sample , we compute log-likelihood statistics on neighboring words that at least co-occur twice ',Dunning,'prep_compute_For pcomp_For_comparing det_sample_the nn_sample_sentence nn_sample_generator dobj_comparing_sample prep_comparing_to det_sample_the amod_sample_English pobj_to_sample nsubj_compute_we amod_statistics_log-likelihood dobj_compute_statistics amod_statistics_ prep__on amod_words_neighboring pobj_on_words dep_at_that dep_statistics_at dep_at_least dep_at_co-occur advmod_co-occur_twice ' +o,1303,'To support distributed computation , we further split the N-gram data into shards by hash values of the first bigram ',Brants,'aux_support_To dep_split_support amod__distributed nn__computation dobj_support_ nsubj_split_we advmod_split_further det_data_the amod_data_N-gram dobj_split_data prep_split_into pobj_into_shards prep_split_by amod_values_hash pobj_by_values prep_values_of det_bigram_the amod_bigram_first pobj_of_bigram ' +o,1304,'See for an application of the boosting approach to named entity recognition , and Walker , Rambow , and Rogati for the application of boosting techniques for ranking in the context of natural language generation ',Collins,'dobj_See_ prep__for det_application_an pobj_for_application prep_application_of det_approach_the amod_approach_boosting pobj_of_approach prep_approach_to amod_recognition_named nn_recognition_entity pobj_to_recognition cc__and nn_Rambow_Walker conj__Rambow cc_Rambow_and nn__Rogati conj_Rambow_ prep_Rambow_for det_application_the pobj_for_application prep_application_of pcomp_of_boosting dobj_boosting_techniques prep_boosting_for pcomp_for_ranking prep_ranking_in det_context_the pobj_in_context prep_context_of amod_generation_natural nn_generation_language pobj_of_generation ' +o,1305,'The first stage parser is a best-first PCFG parser trained on sections 2 through 22 , and 24 of the Penn WSJ treebank ',Marcus,'det_parser_The amod_parser_first nn_parser_stage nsubj_parser_parser cop_parser_is det_parser_a amod_parser_best-first nn_parser_PCFG partmod_parser_trained prep_trained_on pobj_on_sections num_sections_2 prep_trained_through pobj_through_22 cc_parser_and conj_parser_24 prep_24_of det_treebank_the nn_treebank_Penn nn_treebank_WSJ pobj_of_treebank ' +o,1306,'35 The Experiments We have ran LexTract on the one-millionword English Penn Treebank and got two Treebank grammars ',Marcus,'det_Experiments_The nsubj__Experiments nsubj_ran_We aux_ran_have rcmod_Experiments_ran dobj_ran_LexTract prep_LexTract_on det_Treebank_the amod_Treebank_one-millionword nn_Treebank_English nn_Treebank_Penn pobj_on_Treebank dep_35_ cc__and conj__got num_grammars_two nn_grammars_Treebank dobj_got_grammars ' +o,1307,'The results are comparable to other results reported using the Inside\\/Outside method -LRB- see Table 7 ',Ramshaw,'det_results_The nsubj_comparable_results cop_comparable_are prep_comparable_to amod_results_other pobj_to_results partmod_results_reported xcomp_reported_using det__the nn__Inside\\/Outside nn__method dobj_using_ dep_using_see advmod_see_Table dobj_see_7 ' +o,1308,'Meanwhile , it is common for NP chunking tasks to represent a chunk -LRB- eg , NP -RRB- with two labels , the begin -LRB- eg , B-NP -RRB- and inside -LRB- eg , I-NP -RRB- of a chunk ',Ramshaw,'advmod_common_Meanwhile nsubj_common_it cop_common_is prep_common_for pobj_for_NP xcomp_common_chunking dobj_chunking_tasks aux_represent_to infmod_tasks_represent det_chunk_a dobj_represent_chunk appos_chunk_eg dep_eg_NP prep_chunk_with num_labels_two pobj_with_labels nsubj_begin_the dep_represent_begin appos_tasks_eg dep_eg_B-NP cc_tasks_and conj_tasks_inside appos_inside_eg dep_eg_I-NP prep_inside_of det_chunk_a pobj_of_chunk ' +o,1309,'Portage is a statistical phrase-based SMT system similar to Pharaoh ',Koehn,'nsubj_system_Portage cop_system_is det_system_a amod_system_statistical amod_system_phrase-based nn_system_SMT amod_system_similar prep_similar_to pobj_to_Pharaoh ' +o,1310,'Like , we used mutual information to measure the cohesion between two words ',Church,'prep_used_Like pobj_Like_ nsubj_used_we amod_information_mutual dobj_used_information aux_measure_to xcomp_used_measure det_cohesion_the dobj_measure_cohesion prep_cohesion_between num_words_two pobj_between_words ' +o,1311,'Workshop Towards Genre-Enabled Search Engines 13 -- 20 In G Rehm and M Santini , editors ork on an intra-document , or page segment level because a single document can contain instances of multiple genres , eg , contact information , list of publications , CV , see ',Pang,'nn_Engines_Workshop nn_Engines_Towards nn_Engines_Genre-Enabled nn_Engines_Search nsubj__Engines number_13_ dobj__13 number__20 num__ dep_13_ prep_see_In nn_Rehm_G pobj_In_Rehm cc_Rehm_and nn_Santini_M conj_Rehm_Santini nn_ork_editors nn_ork_ amod_ork_ nn_ork_ conj_Rehm_ork prep_ork_on det_intra-document_an pobj_on_intra-document cc_Rehm_or nn_level_page nn_level_segment conj_Rehm_level mark_contain_because det_document_a amod_document_single nsubj_contain_document aux_contain_can dep_level_contain dobj_contain_instances prep_instances_of amod_genres_multiple pobj_of_genres nsubj_see_eg nn_information_contact appos_eg_information appos_eg_list prep_list_of pobj_of_publications appos_eg_CV dep__see acomp_see_ ' +o,1312,'Again , we find the clearest patterns in the graphs for precision , where Malt has very low precision near the root but improves with increasing depth , while MST shows the opposite trend ',McDonald,'advmod_find_Again nsubj_find_we det_patterns_the amod_patterns_clearest dobj_find_patterns prep_patterns_in det_graphs_the pobj_in_graphs prep_graphs_for pobj_for_precision advmod_has_where nsubj_has_Malt rcmod_precision_has advmod_low_very amod_precision_low dobj_has_precision prep_precision_near det_root_the pobj_near_root cc_has_but conj_has_improves prep_find_with amod_depth_increasing pobj_with_depth mark_shows_while nsubj_shows_MST dep_depth_shows det_trend_the amod_trend_opposite dobj_shows_trend ' +o,1313,'Fortunately , using distributional characteristics of term contexts , it is feasible to induce part-of-speech categories directly from a corpus of suf cient size , as several papers have made clear ',Brown,'dep_feasible_Fortunately dep_Fortunately_using amod_characteristics_distributional dobj_Fortunately_characteristics prep_characteristics_of nn_contexts_term pobj_of_contexts nsubj_feasible_it cop_feasible_is aux_induce_to xcomp_feasible_induce amod_categories_part-of-speech dobj_induce_categories advmod_induce_directly prep_induce_from det_corpus_a pobj_from_corpus prep_corpus_of nn_size_suf nn_size_cient pobj_of_size mark_made_as amod_papers_several nsubj_made_papers aux_made_have advcl_feasible_made dep_made_clear ' +o,1314,'Abney notes important problems with the soundness of the approach when a unification-based grammar is actually determining the derivations , motivating the use of log-linear models for parse ranking that Johnson and colleagues further developed ',Collins,'advmod_notes_Abney advmod_notes_ amod_problems_important dobj_notes_problems prep_notes_with det_soundness_the pobj_with_soundness prep_soundness_of det_approach_the pobj_of_approach advmod_determining_when det_grammar_a amod_grammar_unification-based nsubj_determining_grammar aux_determining_is advmod_determining_actually dep_approach_determining det_derivations_the dobj_determining_derivations dep_determining_motivating det_use_the dobj_motivating_use prep_use_of amod_models_log-linear pobj_of_models dobj_determining_ prep_determining_for amod_ranking_parse pobj_for_ranking dep_developed_that nsubj_developed_Johnson cc_Johnson_and conj_Johnson_colleagues advmod_developed_further dep_notes_developed acomp_developed_ ' +o,1315,'6 Discourse Context pointed out that the sense of a target word is highly consistent within any given document -LRB- one sense per discourse -RRB- ',Yarowsky,'num__6 nn__Discourse nn__Context nsubj_pointed_ prt_pointed_out complm_consistent_that det_sense_the nsubj_consistent_sense prep_sense_of det_word_a nn_word_target pobj_of_word cop_consistent_is advmod_consistent_highly ccomp_pointed_consistent prep_consistent_within det_document_any amod_document_given pobj_within_document num_sense_one dep_document_sense prep_sense_per pobj_per_discourse ' +o,1316,'3 Model As an extension to commonly used lexical word pair probabilities p -LRB- f e -RRB- as introduced in , we define our model to operate on word triplets ',Brown,'num_Model_3 mark_introduced_As det_extension_an nsubj_introduced_extension prep_extension_to advmod_used_commonly amod_probabilities_used amod_probabilities_lexical nn_probabilities_word nn_probabilities_pair pobj_to_probabilities dep_as_p dep_-LRB-_f dep_f_e advmod_introduced_as advcl_define_introduced prep_introduced_in pobj_in_ nsubj_define_we dep_Model_define poss_model_our dobj_define_model aux_operate_to infmod_model_operate prep_operate_on nn_triplets_word pobj_on_triplets ' +o,1317,'To derive the joint counts c -LRB- ? s , ? t -RRB- from which p -LRB- ? s ? t -RRB- and p -LRB- ? t ? s -RRB- are estimated , we use the phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Koehn,'aux_derive_To dep_use_derive det_counts_the amod_counts_joint dobj_derive_counts nsubjpass_estimated_c dep_s_s dep_s_t prep_s_from dep_p_which pobj_from_p nn_t_s appos_s_t cc_s_and conj_s_p pobj_?_t dep_c_s auxpass_estimated_are rcmod_counts_estimated nsubj_use_we det_algorithm_the nn_algorithm_phrase nn_algorithm_induction nsubj_described_algorithm ccomp_use_described prep_described_in pobj_in_ prep_described_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments partmod_alignments_generated xcomp_generated_using nn_model_IBM nsubj__model dep__2 xcomp_using_ ' +p,1318,'Since human evaluation is costly and difficult to do reliably , a major focus of research has been on automatic measures of MT quality , pioneered by BLEU and NIST ',Papineni,'mark_costly_Since amod_evaluation_human nsubj_costly_evaluation cop_costly_is advcl_been_costly cc_costly_and conj_costly_difficult aux_do_to xcomp_costly_do advmod_do_reliably det_focus_a amod_focus_major nsubj_been_focus prep_focus_of pobj_of_research aux_been_has prep_been_on amod_measures_automatic pobj_on_measures prep_measures_of nn_quality_MT pobj_of_quality dep_measures_pioneered prep_pioneered_by nn__BLEU pobj_by_ cc__and nn__NIST conj__ ' +o,1319,'We utilise the automatic annotation algorithm of to derive a version of Penn-II where each node in each tree is annotated with an LFG functional annotation -LRB- ie an attribute value structure equation -RRB- ',Cahill,'nsubj_utilise_We det_algorithm_the amod_algorithm_automatic nn_algorithm_annotation dobj_utilise_algorithm prep_algorithm_of pcomp_of_ aux_version_to dep_version_derive det_version_a xcomp__version prep_version_of pobj_of_Penn-II advmod_annotated_where det_node_each nsubjpass_annotated_node prep_node_in det_tree_each pobj_in_tree auxpass_annotated_is advcl_version_annotated prep_annotated_with det_annotation_an nn_annotation_LFG amod_annotation_functional pobj_with_annotation parataxis_utilise_ie det_equation_an nn_equation_attribute nn_equation_value nn_equation_structure dobj_ie_equation ' +p,1320,'It is interesting to note that, while the study of how the granularity of context-free grammars (CFG) affects the performance of a parser (e.g. in the form 86 n1:IP [=] n2:NP [SUBJ=] n4:NR [=] GSC4ES JiangZemin n3:VP [=] n5:VV [=] ESDO interview n6:NP [OBJ=] n7:NR [ADJUNCT] AIC1 Thai n8:NN [=] D3D2 president f1 PRED ESDO SUBJ f2 PRED GSC4ESNTYPE proper NUM sg OBJ f3 PRED D3D2 NTYPE common NUM sg ADJUNCT f4 PRED AIC1NTYPE proper NUM sg : N F (n1)=(n3)=(n5)=f1 (n2)=(n4)=f2 (n6)=(n8)=f3 (n7)=f4 Figure 1: Cand f-structures with links for the sentence GSC4ESESDOAIC1D3D2 of grammar transforms (Johnson, 1998) and lexicalisation (Collins, 1997)) has attracted substantial attention, to our knowledge, there has been a lot less research on this subject for surface realisation, a process that is generally regarded as the reverse process of parsing.',Collins,'' +o,1321,'Most previous work with CRFs containing nonlocal dependencies used approximate probabilistic inference techniques , including TRP and Gibbs sampling ',Finkel,'amod_work_Most amod_work_previous nsubj_used_work prep_work_with pobj_with_CRFs partmod_CRFs_containing amod_dependencies_nonlocal dobj_containing_dependencies amod_techniques_approximate amod_techniques_probabilistic nn_techniques_inference dobj_used_techniques prep_used_including nn__TRP pobj_including_ cc__and conj__Gibbs partmod__sampling ' +o,1322,'We tokenized sentences using the standard treebank tokenization script , and then we performed part-of-speech tagging using MXPOST tagger ',Ratnaparkhi,'nsubj_tokenized_We dobj_tokenized_sentences xcomp_tokenized_using det_script_the amod_script_standard nn_script_treebank nn_script_tokenization dobj_using_script cc_tokenized_and advmod_performed_then nsubj_performed_we conj_tokenized_performed amod_tagging_part-of-speech dobj_performed_tagging xcomp_performed_using amod_tagger_MXPOST dobj_using_tagger ' +o,1323,'Training of the phrase translation model builds on top of a standard statistical word alignment over the training corpus of parallel text for identifying corresponding word blocks , assuming no further linguistic analysis of the source or target language ',Brown,'nsubj_builds_Training prep_Training_of det_model_the nn_model_phrase nn_model_translation pobj_of_model prep_builds_on pobj_on_top prep_top_of det_alignment_a amod_alignment_standard amod_alignment_statistical nn_alignment_word pobj_of_alignment prep_alignment_over det_corpus_the nn_corpus_training pobj_over_corpus prep_corpus_of amod__parallel nn__text pobj_of_ prep_builds_for pcomp_for_identifying amod_blocks_corresponding nn_blocks_word dobj_identifying_blocks xcomp_builds_assuming det_analysis_no amod_analysis_further amod_analysis_linguistic dobj_assuming_analysis prep_analysis_of det_language_the nn_language_source cc_source_or conj_source_target pobj_of_language ' +o,1324,'Section 7 considers recent efforts to induce effective procedures for automated sense labeling of discourse relations that are not lexically marked ',Marcu,'nsubj_considers_Section num_Section_7 amod_efforts_recent dobj_considers_efforts aux_induce_to xcomp_considers_induce amod_procedures_effective dobj_induce_procedures prep_induce_for amod_labeling_automated nn_labeling_sense pobj_for_labeling prep_labeling_of nn_relations_discourse pobj_of_relations nsubj_marked_that cop_marked_are neg_marked_not advmod_marked_lexically rcmod_labeling_marked ' +o,1325,'However , most of them fail to utilize non-syntactic phrases well that are proven useful in the phrase-based methods ',Koehn,'advmod_fail_However nsubj_fail_most prep_most_of pobj_of_them aux_phrases_to dep_phrases_utilize amod_phrases_non-syntactic xcomp_fail_phrases advmod_phrases_well complm_proven_that auxpass_proven_are ccomp_phrases_proven dep_proven_useful prep_proven_in det_methods_the amod_methods_phrase-based pobj_in_methods ' +o,1326,'Words are encoded through an automatic clustering algorithm while tags , labels and extensions are normally encoded using diagonal bits ',Brown,'nsubjpass_encoded_Words auxpass_encoded_are prep_encoded_through det__an amod__automatic nn__clustering nn__algorithm pobj_through_ mark_encoded_while nsubjpass_encoded_tags conj_tags_labels cc_tags_and conj_tags_extensions auxpass_encoded_are advmod_encoded_normally advcl_encoded_encoded xcomp_encoded_using amod_bits_diagonal dobj_using_bits ' +o,1327,'4 Features For our experiments we use the features proposed , motivated and described in detail by ',Nenkova,'num_Features_4 dobj_use_Features prep_Features_For poss_experiments_our pobj_For_experiments nsubj_use_we det_features_the dobj_use_features partmod_features_proposed conj_proposed_motivated cc_proposed_and conj_proposed_described prep_use_in pobj_in_detail prep_use_by ' +o,1328,' presented a thorough discussion on the Yarowsky algorithm ',Abney,'nsubj_presented_ det_discussion_a amod_discussion_thorough dobj_presented_discussion prep_presented_on det_algorithm_the nn_algorithm_Yarowsky pobj_on_algorithm ' +o,1329,'In the field of statistical analysis of natural language data , it is common to use measures of lexical association , such as the informationtheoretic measure of mutual information , to extract useful relationships between words -LRB- eg -RRB- ',Church,'prep_common_In det_field_the pobj_In_field prep_field_of amod_analysis_statistical pobj_of_analysis prep_analysis_of amod_data_natural nn_data_language pobj_of_data nsubj_common_it cop_common_is aux_use_to xcomp_common_use dobj_use_measures prep_measures_of amod_association_lexical pobj_of_association dep_as_such prep_measures_as det_measure_the amod_measure_informationtheoretic pobj_as_measure prep_measure_of amod_information_mutual pobj_of_information aux_extract_to xcomp_use_extract amod_relationships_useful dobj_extract_relationships prep_relationships_between pobj_between_words number__eg dep_words_ ' +p,1330,'The former term P -LRB- E -RRB- is called a language model , representing the likelihood of E The latter term P -LRB- J E -RRB- is called a translation model , representing the generation probability from E into J As an implementation of P -LRB- J E -RRB- , the word alignment based statistical translation has been successfully applied to similar language pairs , such as FrenchEnglish and German English , but not to drastically dierent ones , such as JapaneseEnglish ',Brown,'det_P_The amod_P_former nn_P_term nsubjpass_called_P appos_P_E auxpass_called_is det_model_a nn_model_language dobj_called_model xcomp_called_representing det_likelihood_the dobj_representing_likelihood prep_likelihood_of pobj_of_E det_P_The amod_P_latter nn_P_term nsubjpass_called_P nn_E_J appos_P_E auxpass_called_is dep_called_called det_model_a nn_model_translation dobj_called_model xcomp_called_representing det_probability_the nn_probability_generation dobj_representing_probability prep_representing_from pobj_from_E prep_representing_into pobj_into_J dep_J_As det_implementation_an pobj_As_implementation prep_implementation_of pobj_of_P nn_E_J appos_P_E det_alignment_the nn_alignment_word nsubj_based_alignment dep_J_based amod__statistical nn__translation nsubjpass_applied_ aux_applied_has auxpass_applied_been advmod_applied_successfully ccomp_based_applied prep_applied_to amod_pairs_similar nn_pairs_language pobj_to_pairs dep_as_such prep_J_as pobj_as_FrenchEnglish cc_FrenchEnglish_and nn_English_German conj_FrenchEnglish_English dep_not_but cc_J_not aux_dierent_to advmod_dierent_drastically conj_J_dierent dobj_dierent_ones dep_as_such prep_ones_as pobj_as_JapaneseEnglish ' +o,1331,'We hence chose transformation-based learning to create this -LRB- shallow -RRB- segmentation grammar , converting the segmentation task into a tagging task -LRB- as is done in 85 , inter alia -RRB- ',Ramshaw,'nsubj_chose_We advmod_chose_hence amod_learning_transformation-based dobj_chose_learning aux_create_to xcomp_chose_create det_grammar_this appos_grammar_shallow nn_grammar_segmentation dobj_create_grammar partmod_grammar_converting det_task_the nn_task_segmentation dobj_converting_task prep_converting_into det_task_a amod_task_tagging pobj_into_task mark_done_as auxpass_done_is dep_task_done prep_done_in num__85 pobj_in_ nn_alia_inter appos__alia ' +o,1332,'In particular , previous work has investigated the use of Markov random fields -LRB- MRFs -RRB- or log-linear models as probabilistic models with global features for parsing and other NLP tasks ',Collins,'prep_investigated_In pobj_In_particular amod__previous nn__work nsubj_investigated_ aux_investigated_has det_use_the dobj_investigated_use prep_use_of nn_fields_Markov amod_fields_random pobj_of_fields abbrev_fields_MRFs cc_use_or amod_models_log-linear conj_use_models prep_models_as amod_models_probabilistic pobj_as_models prep_investigated_with amod_features_global pobj_with_features prep_features_for nn_tasks_parsing cc_parsing_and conj_parsing_other nn_tasks_NLP pobj_for_tasks ' +o,1333,' reports results for different numbers of hidden states but it is unclear how to make this choice a priori , while Goldwater & Griffiths leave this question as future work ',Johnson,'amod_reports_ nsubj_results_reports prep_results_for amod_numbers_different pobj_for_numbers prep_numbers_of amod_states_hidden pobj_of_states cc_results_but nsubj_unclear_it cop_unclear_is conj_results_unclear advmod_make_how aux_make_to ccomp_unclear_make det_choice_this dobj_make_choice dep_priori_a advmod_make_priori mark_leave_while nn__Goldwater cc_Goldwater_& conj_Goldwater_Griffiths nsubj_leave_ advcl_make_leave det_question_this dobj_leave_question prep_question_as amod_work_future pobj_as_work ' +o,1334,'1 Introduction Word Sense Disambiguation -LRB- WSD -RRB- competitions have focused on general domain texts , as attested in the last Senseval and Semeval competitions ','Daume III','num_Disambiguation_1 nn_Disambiguation_Introduction nn_Disambiguation_Word nn_Disambiguation_Sense nn_competitions_Disambiguation abbrev_competitions_WSD nsubj_focused_competitions aux_focused_have prep_focused_on amod_texts_general nn_texts_domain pobj_on_texts prep_focused_as amod__attested prep_attested_in det_Senseval_the amod_Senseval_last pobj_in_Senseval cc_Senseval_and nn_competitions_Semeval conj_Senseval_competitions pobj_as_ ' +n,1335,'In pursuit of better translation , phrase-based models havesignificantlyimprovedthe quality over classical word-based models ',Brown,'prep__In pobj_In_pursuit prep_pursuit_of amod_translation_better pobj_of_translation amod_models_phrase-based nsubj__models amod_quality_havesignificantlyimprovedthe dobj__quality prep__over amod_models_classical amod_models_word-based pobj_over_models ' +o,1336,'History-based models for predicting the next parser action 3 ',Collins,'amod_models_History-based nsubj__models prep_models_for pcomp_for_predicting det_action_the amod_action_next nn_action_parser dobj_predicting_action dobj__3 ' +o,1337,'33 Features Similar to the default features in Pharaoh , we used following features to estimate the weight of our grammar rules ',Och,'num_Features_33 measure_Similar_Features amod_features_Similar prep_Similar_to det_default_the pobj_to_default prep_used_in nn__Pharaoh pobj_in_ nsubj_used_we dep_features_used amod_features_following dobj_used_features aux_estimate_to xcomp_used_estimate det_weight_the dobj_estimate_weight prep_weight_of poss_rules_our nn_rules_grammar pobj_of_rules ' +o,1338,'Following the framework of global linear models in , we cast this task as learning a mapping F from input verses x X to a text-reuse hypothesis y Y -LCB- epsilon1 -RCB- ',Collins,'prep_cast_Following det_framework_the pobj_Following_framework prep_framework_of amod_models_global amod_models_linear pobj_of_models prep_models_in pobj_in_ nsubj_cast_we det_task_this dobj_cast_task prep_cast_as pcomp_as_learning det_mapping_a dobj_learning_mapping dep_X_F prep_F_from nn_verses_input pobj_from_verses dep_X_x dep_learning_X prep_X_to det_epsilon1_a amod_epsilon1_text-reuse nn_epsilon1_hypothesis nn_epsilon1_y nn_epsilon1_Y pobj_to_epsilon1 ' +o,1339,'We obtained 47,025 50-dimensional reduced vectors from the SVD and clustered them into 200 classes using the fast clustering algorithm Buckshot -LRB- group average agglomeration applied to a sample -RRB- ',Cutting,'nsubj_obtained_We num_vectors_47,025 amod_vectors_50-dimensional amod_vectors_reduced dep_obtained_vectors prep_vectors_from det_SVD_the pobj_from_SVD cc_vectors_and conj_vectors_clustered dobj_clustered_them prep_clustered_into num_classes_200 pobj_into_classes partmod_classes_using det_algorithm_the amod_algorithm_fast nn_algorithm_clustering dobj_using_algorithm nn__Buckshot dobj_vectors_ nn_agglomeration_group amod_agglomeration_average dep__agglomeration partmod_agglomeration_applied prep_applied_to det_sample_a pobj_to_sample ' +o,1340,'On the other hand , according to the data-driven approach , a frequency-based language model is acquired from corpora and has the forms of ngrams , rules , decision trees or neural networks ',Cutting,'prep_acquired_On det_hand_the amod_hand_other pobj_On_hand prep_acquired_according dep_according_to det_approach_the amod_approach_data-driven pobj_to_approach det_model_a amod_model_frequency-based nn_model_language nsubjpass_acquired_model auxpass_acquired_is prep_acquired_from pobj_from_corpora cc_acquired_and conj_acquired_has det_forms_the dobj_has_forms prep_forms_of amod__ngrams pobj_of_ nn__rules conj__ nn__decision nn__trees conj__ cc__or amod__neural nn__networks conj__ ' +o,1341,'Several frameworks for finding translation equivalents or translation units in machine translation , such as and other example-based MT approaches , might be used to select the preferred mapping ',Marcus,'amod_frameworks_Several nsubjpass_used_frameworks prep_frameworks_for pcomp_for_finding nn_equivalents_translation dobj_finding_equivalents cc_equivalents_or nn_units_translation conj_equivalents_units prep_finding_in nn_translation_machine pobj_in_translation dep_as_such dep_finding_as pobj_as_ cc__and amod_approaches_other amod_approaches_example-based nn_approaches_MT conj__approaches aux_used_might auxpass_used_be aux_select_to purpcl_used_select det_mapping_the amod_mapping_preferred dobj_select_mapping ' +o,1342,'-LRB- b -RRB- MEDLINE DT JJ VBN NNS IN DT NN NNS VBP The oncogenic mutated forms of the ras proteins are RB JJ CC VBP IN JJ NN NN constitutively active and interfere with normal signal transduction Figure 1 : Part of speech-tagged sentences from both corpora we investigate its use in part of speech -LRB- PoS -RRB- tagging ',Ratnaparkhi,'dep_-RRB-_b nn_NNS_MEDLINE nn_NNS_DT nn_NNS_JJ nn_NNS_VBN prep_NNS_IN nn_VBP_DT nn_VBP_NN nn_VBP_NNS pobj_IN_VBP det_forms_The amod_forms_oncogenic amod_forms_mutated nsubj_VBP_forms prep_forms_of det_proteins_the nn_proteins_ras pobj_of_proteins cop_VBP_are nn_VBP_RB nn_VBP_JJ nn_VBP_CC rcmod_VBP_VBP prep_VBP_IN nn_NN_JJ nn_NN_NN pobj_IN_NN advmod_active_constitutively amod_NN_active cc_VBP_and conj_VBP_interfere prep_interfere_with amod_Figure_normal nn_Figure_signal nn_Figure_transduction pobj_with_Figure num_Figure_1 dep_NNS_Part prep_Part_of amod_sentences_speech-tagged pobj_of_sentences prep_sentences_from det_corpora_both pobj_from_corpora nsubj_investigate_we dep_NNS_investigate poss_use_its dobj_investigate_use prep_investigate_in pobj_in_part prep_part_of pobj_of_speech appos_speech_PoS partmod_speech_tagging ' +o,1343,'Introduction There has been considerable recent interest in the use of statistical methods for grouping words in large on-line corpora into categories which capture some of our intuitions about the reference of the words we use and the relationships between them ',Brown,'expl_been_There aux_been_has dep_Introduction_been amod_interest_considerable amod_interest_recent dobj_been_interest prep_been_in det_use_the pobj_in_use prep_use_of amod_methods_statistical pobj_of_methods prep_been_for pcomp_for_grouping dobj_grouping_words prep_words_in amod_corpora_large amod_corpora_on-line pobj_in_corpora prep_grouping_into pobj_into_categories nsubj_capture_which rcmod_categories_capture dobj_capture_some prep_some_of poss_intuitions_our pobj_of_intuitions prep_capture_about det_reference_the pobj_about_reference prep_reference_of det_words_the pobj_of_words nsubj_use_we rcmod_words_use cc_categories_and det_relationships_the conj_categories_relationships prep_relationships_between pobj_between_them ' +o,1344,'Our intuition is that we can not apply our binarization to ',Collins,'poss_intuition_Our nsubj_is_intuition complm_apply_that nsubj_apply_we aux_apply_can neg_apply_not ccomp_is_apply poss_binarization_our dobj_apply_binarization prep_apply_to ' +p,1345,'The best previous result is an accuracy of 561 \% ',Turney,'det_result_The advmod_result_best amod_result_previous nsubj_accuracy_result cop_accuracy_is det_accuracy_an prep_accuracy_of num_\%_561 pobj_of_\% ' +o,1346,'Table 3 : Example compressions Compression AvgLen Rating Baseline 970 193 BT-2-Step 2206 321 Spade 1909 310 Humans 2007 383 Table 4 : Mean ratings for automatic compressions nally , we added a simple baseline compression algorithm proposed by which removed all prepositional phrases , clauses , toinfinitives , and gerunds ',Jing,'num_Table_3 nsubj_compressions_Example ccomp_added_compressions nn_Baseline_Compression nn_Baseline_AvgLen nn_Baseline_Rating dobj_compressions_Baseline number_193_970 num_310_193 amod_310_BT-2-Step number_321_2206 num_310_321 amod_310_Spade num_310_1909 nsubj_Humans_310 rcmod_Baseline_Humans num_383_2007 iobj_Humans_383 dobj_Humans_Table num_Table_4 amod_ratings_Mean dobj_Humans_ratings prep_ratings_for amod_compressions_automatic pobj_for_compressions advmod_compressions_nally nsubj_added_we dep_Table_added det_algorithm_a amod_algorithm_simple nn_algorithm_baseline nn_algorithm_compression dobj_added_algorithm partmod_algorithm_proposed prep_proposed_by pobj_by_ nsubj_removed_which rcmod_algorithm_removed det_phrases_all amod_phrases_prepositional dobj_removed_phrases conj_phrases_clauses conj_phrases_toinfinitives cc_phrases_and conj_phrases_gerunds ' +o,1347,'Church and Hanks employed mutual information to extract both adjacent and distant bi-grams that tend to co-occur within a fixed-size window ',Church,'nn__Church cc_Church_and conj_Church_Hanks nsubj_employed_ amod_information_mutual dobj_employed_information aux_extract_to infmod_information_extract det_bi-grams_both amod_bi-grams_adjacent cc_adjacent_and conj_adjacent_distant dobj_extract_bi-grams nsubj_tend_that rcmod_bi-grams_tend aux_co-occur_to xcomp_tend_co-occur prep_co-occur_within det_window_a amod_window_fixed-size pobj_within_window ' +o,1348,'Previous research has addressed revision in single-document summaries and has suggested that revising summaries can make them more informative and correct errors ',Jing,'amod_research_Previous nsubj_addressed_research aux_addressed_has dobj_addressed_revision prep_revision_in amod_summaries_single-document pobj_in_summaries nsubj__ dep_addressed_ cc__and aux_suggested_has conj__suggested complm_make_that amod_summaries_revising nsubj_make_summaries aux_make_can ccomp_suggested_make nsubj_informative_them advmod_informative_more xcomp_make_informative cc_make_and conj_make_correct dobj_correct_errors ' +o,1349,'2 Three New Features for MT Evaluation Since our source-sentence constrained n-gram precision and discriminative unigram precision are both derived from the normal n-gram precision , it is worth describing the original n-gram precision metric , BLEU ',Papineni,'num_Features_2 num_Features_Three nn_Features_New nsubj_Evaluation_Features prep_Features_for pobj_for_MT mark_derived_Since poss_precision_our amod_precision_source-sentence amod_precision_constrained nn_precision_n-gram nsubjpass_derived_precision cc_precision_and nn_precision_discriminative nn_precision_unigram conj_precision_precision auxpass_derived_are advmod_derived_both advcl_worth_derived prep_derived_from det_precision_the amod_precision_normal amod_precision_n-gram pobj_from_precision nsubj_worth_it cop_worth_is ccomp_Evaluation_worth xcomp_worth_describing det_precision_the amod_precision_original amod_precision_n-gram dobj_describing_precision amod_precision_metric dep_metric_BLEU ' +o,1350,'This concept of alignment has been also used for tasks like authomatic vocabulary derivation and corpus alignment ',Brown,'det_concept_This nsubjpass_used_concept prep_concept_of pobj_of_alignment aux_used_has auxpass_used_been advmod_used_also prep_used_for pobj_for_tasks prep_tasks_like amod_derivation_authomatic nn_derivation_vocabulary pobj_like_derivation cc_derivation_and nn__corpus nn__alignment conj_derivation_ ' +o,1351,'We use the Stanford parser with its default Chinese grammar , the GIZA + + alignment package with its default settings , and the ME tool developed by ',Och,'nsubj_use_We det__the nn__Stanford nn__parser dobj_use_ prep__with poss_grammar_its nn_grammar_default amod_grammar_Chinese pobj_with_grammar det_package_the amod_package_GIZA cc_GIZA_+ conj_GIZA_+ amod_package_ nn_package_alignment conj__package prep_package_with poss_settings_its nn_settings_default pobj_with_settings cc__and det_tool_the nn_tool_ME conj__tool partmod_tool_developed prep_developed_by pobj_by_ ' +o,1352,'The idea is that the translation of a sentence x into a sentence y can be performed in the following steps1 : -LRB- a -RRB- If x is small enough , IBMs model 1 is employed for the translation ',Brown,'det_idea_The nsubj_is_idea complm_performed_that det_translation_the nsubjpass_performed_translation prep_translation_of det_x_a dep_x_sentence pobj_of_x dep_x_into det_y_a nn_y_sentence pobj_into_y aux_performed_can auxpass_performed_be ccomp_is_performed prep_performed_in det_steps1_the amod_steps1_following pobj_in_steps1 dep_-RRB-_a mark_small_If nsubj_small_x cop_small_is dep_employed_small advmod_small_enough nn_model_IBMs nsubjpass_employed_model num_model_1 advmod_employed_ auxpass_employed_is parataxis_is_employed prep_employed_for det_translation_the pobj_for_translation ' +o,1353,'1 A cept is defined as the set of target words connected to a source word ',Brown,'det_cept_A nsubjpass_defined_cept auxpass_defined_is dep_1_defined prep_defined_as det_set_the pobj_as_set prep_set_of nn_words_target pobj_of_words partmod_words_connected prep_connected_to det_word_a nn_word_source pobj_to_word ' +o,1354,'Measures of attributional similarity have been studied extensively , due to their applications in problems such as recognizing synonyms , information retrieval , determining semantic orientation , grading student essays , measuring textual cohesion , and word sense disambiguation ',Turney,'nsubjpass_studied_Measures prep_Measures_of amod_similarity_attributional pobj_of_similarity aux_studied_have auxpass_studied_been advmod_studied_extensively advmod_studied_due dep_due_to poss_applications_their pobj_to_applications prep_applications_in pobj_in_problems dep_as_such prep_problems_as pcomp_as_recognizing amod__synonyms dobj_recognizing_ nn__information nn__retrieval appos__ conj_recognizing_determining amod__semantic nn__orientation dobj_determining_ conj_recognizing_grading nn_essays_student dobj_grading_essays advmod_grading_ conj_recognizing_measuring amod__textual nn__cohesion dobj_measuring_ cc_recognizing_and conj_recognizing_word nn__sense nn__disambiguation dobj_word_ ' +o,1355,'Note that the algorithm from was designed for discriminatively training an HMM-style tagger ',Collins,'complm_designed_that det_algorithm_the nsubjpass_designed_algorithm prep_algorithm_from pobj_from_ auxpass_designed_was ccomp_Note_designed prep_designed_for advmod_training_discriminatively pcomp_for_training det_tagger_an amod_tagger_HMM-style dobj_training_tagger ' +o,1356,'In this years shared task we evaluated a number of different automatic metrics : Bleu Bleu remains the de facto standard in machine translation evaluation ',Papineni,'prep_evaluated_In det_years_this pobj_In_years partmod_years_shared dobj_shared_task nsubj_evaluated_we det_number_a dobj_evaluated_number prep_number_of amod_metrics_different amod_metrics_automatic pobj_of_metrics nn_Bleu_Bleu nn_Bleu_ nsubj_standard_Bleu cop_standard_remains det_standard_the amod_standard_de dep_de_facto parataxis_evaluated_standard prep_standard_in nn_evaluation_machine nn_evaluation_translation pobj_in_evaluation ' +p,1357,'1 Introduction Hierarchical approaches to machine translation have proven increasingly successful in recent years , and often outperform phrase-based systems on target-language fluency and adequacy ',Shen,'num_approaches_1 nn_approaches_Introduction nn_approaches_Hierarchical nsubj_proven_approaches prep_approaches_to nn_translation_machine pobj_to_translation aux_proven_have advmod_successful_increasingly acomp_proven_successful prep_proven_in amod__recent nn__years pobj_in_ cc_proven_and advmod_outperform_often conj_proven_outperform amod__phrase-based nn__systems dobj_outperform_ prep__on amod_fluency_target-language pobj_on_fluency cc_fluency_and conj_fluency_adequacy ' +o,1358,'In computational linguistics , our pattern discovery procedure extends over previous approaches that use surface patterns as indicators of semantic relations between nouns or verbs -LRB- inter alia -RRB- ',Turney,'prep_extends_In amod_linguistics_computational pobj_In_linguistics poss_procedure_our nn_procedure_pattern nn_procedure_discovery nsubj_extends_procedure prep_extends_over amod_approaches_previous pobj_over_approaches nsubj_use_that rcmod_approaches_use nn_patterns_surface dobj_use_patterns prep_use_as pobj_as_indicators prep_indicators_of amod_relations_semantic pobj_of_relations prep_relations_between pobj_between_nouns cc_nouns_or conj_nouns_verbs num_alia_ nn_alia_inter appos_approaches_alia ' +o,1359,'SMT has evolved from the original word-based approach into phrase-based approaches and syntax-based approaches ',Koehn,'nsubj_evolved_SMT aux_evolved_has prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep_evolved_into amod__phrase-based nn__approaches pobj_into_ cc__and amod__syntax-based nn__approaches conj__ ' +o,1360,'Various clustering techniques have been proposed which perform automatic word clustering optimizing a maximum-likelihood criterion with iterative clustering algorithms ',Brown,'amod_techniques_Various nn_techniques_clustering nsubj_been_techniques aux_been_have amod__proposed dobj_been_ nsubj_perform_which rcmod__perform amod_word_automatic dobj_perform_word partmod_word_clustering xcomp_clustering_optimizing det_criterion_a amod_criterion_maximum-likelihood dobj_optimizing_criterion prep_optimizing_with amod_algorithms_iterative amod_algorithms_clustering pobj_with_algorithms ' +p,1361,'Throughout , the likelihood ratio is used as significance measure because of its stable performance in various evaluations , yet many more measures are possible ',Dunning,'prep_used_Throughout det__the nn__likelihood nn__ratio nsubjpass_used_ auxpass_used_is prep_used_as nn_measure_significance pobj_as_measure dep_of_because prep_measure_of poss_performance_its amod_performance_stable pobj_of_performance prep_performance_in amod_evaluations_various pobj_in_evaluations advmod_used_yet amod_measures_many amod_measures_more nsubj_possible_measures cop_possible_are dep_used_possible ' +o,1362,' , and Lee , Wilson et al ',Pang,'nn__ conj__ cc__and conj__Lee appos__Wilson cc_Wilson_et conj_Wilson_al ' +o,1363,'The approach is able to achieve 94 \% precision and recall for base NPs derived from the Penn Treebank Wall Street Journal ',Marcus,'det_approach_The nsubj_able_approach cop_able_is aux_achieve_to xcomp_able_achieve num_\%_94 dobj_achieve_\% nsubj_derived_precision cc_precision_and conj_precision_recall prep_precision_for nn_NPs_base pobj_for_NPs dep_able_derived prep_derived_from det_Journal_the nn_Journal_Penn nn_Journal_Treebank nn_Journal_Wall nn_Journal_Street pobj_from_Journal ' +o,1364,'org\\/pubs\\/citations \\/ j ournals\\/toms\\/1986 -12 -2 \\/ p154-meht a \\/ Mutual Information Given the definition of Mutual Information , I -LRB- x , y -RRB- = log 2 P -LRB- x , y -RRB- P -LRB- x -RRB- P -LRB- y -RRB- \' we consider the distribution of a window word according to the contingency table -LRB- a -RRB- in Table 4 ',Church,'amod_p154-meht_\\/ nn_p154-meht_j nn_p154-meht_ournals\\/toms\\/1986 num_p154-meht_-12 num_p154-meht_-2 nn_p154-meht_\\/ dobj_org\\/pubs\\/citations_p154-meht det_Information_a nn_Information_\\/ nn_Information_Mutual nsubj_org\\/pubs\\/citations_Information prep_Information_Given det_definition_the dep_Given_definition prep_definition_of nn__Mutual nn__Information pobj_of_ appos__I dep_I_x appos_x_y nsubj_log_= rcmod_I_log number_P_2 num_P_P dep_P_x appos_x_y nn_P_P appos_P_x dobj_log_P appos_P_y nsubj_consider_we rcmod_definition_consider det_distribution_the dobj_consider_distribution prep_distribution_of det_word_a nn_word_window pobj_of_word prep_consider_according dep_according_to det_table_the nn_table_contingency pobj_to_table dep_-LRB-_a prep_table_in pobj_in_Table num_Table_4 ' +o,1365,'There has been recent work on discovering allomorphic phenomena automatically ',Dasgupta,'expl_been_There aux_been_has amod_work_recent dobj_been_work prep_work_on pcomp_on_discovering amod_phenomena_allomorphic dobj_discovering_phenomena advmod_discovering_automatically ' +o,1366,'Thus the alignment set is denoted as -RCB- & -RRB- ,1 -LRB- -RRB- , -LCB- -LRB- ialiaiA ii = We adapt the bilingual word alignment model , IBM Model 3 , to monolingual word alignment ',Dunning,'det_alignment_the dep_Thus_alignment partmod_alignment_set auxpass_denoted_is ccomp_set_denoted prep_set_as dep_to_& rcmod_&_,1 nn_=_ialiaiA nn_=_ii nsubj_alignment_= dep_adapt_We prep_alignment_adapt det_word_the amod_word_bilingual pobj_adapt_word advcl_,1_alignment dobj_alignment_model nn_Model_IBM dobj_alignment_Model dep__3 amod_Model_ prep_set_to amod_alignment_monolingual nn_alignment_word pobj_to_alignment ' +o,1367,'4 Experimental Work A part of the Wall Street Journal -LRB- WSJ -RRB- which had been processed in the Penn Treebanck Project was used in the experiments ',Marcus,'num_part_4 nn_part_Experimental nn_part_Work nn_part_A nsubjpass_used_part prep_part_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal abbrev_Journal_WSJ nsubjpass_processed_which aux_processed_had auxpass_processed_been rcmod_Journal_processed prep_processed_in det__the nn__Penn nn__Treebanck nn__Project pobj_in_ auxpass_used_was prep_used_in det_experiments_the pobj_in_experiments ' +p,1368,'However , the study of provides interesting insights into what makes a good distributional similarity measure in the contexts of semantic similarity prediction and language modeling ',Weeds,'advmod_provides_However det_study_the nsubj_provides_study prep_study_of pobj_of_ amod_insights_interesting dobj_provides_insights prep_provides_into nsubj_makes_what pcomp_into_makes det_measure_a amod_measure_good amod_measure_distributional nn_measure_similarity dobj_makes_measure prep_measure_in det_contexts_the pobj_in_contexts prep_contexts_of amod_prediction_semantic nn_prediction_similarity pobj_of_prediction cc_prediction_and nn_modeling_language conj_prediction_modeling ' +o,1369,'In , features are selected according to part-of-speech labels ',Turney,'prep_selected_In pobj_In_ nsubjpass_selected_features auxpass_selected_are prep_selected_according dep_according_to amod_labels_part-of-speech pobj_to_labels ' +o,1370,'In particular , since we treat each individual speech within a debate as a single document , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ',Pang,'prep_considering_In pobj_In_particular mark_treat_since nsubj_treat_we dep_considering_treat det_speech_each amod_speech_individual dobj_treat_speech prep_treat_within det_debate_a pobj_within_debate prep_debate_as det_document_a amod_document_single pobj_as_document nsubj_considering_we aux_considering_are det_version_a dobj_considering_version prep_version_of amod_classification_document-level amod_classification_sentiment-polarity pobj_of_classification advmod_considering_namely advmod_considering_automatically dep_considering_distinguishing prep_distinguishing_between amod_documents_positive cc_positive_and conj_positive_negative pobj_between_documents ' +o,1371,'In this paper we show how the extraction process can be scaled to the complete Wall Street Journal -LRB- WSJ -RRB- section of the Penn-II treebank , with about 1 million words in 50,000 sentences , based on the automatic LFG f-structure annotation algorithm described in ',Cahill,'prep_show_In det_paper_this pobj_In_paper nsubj_show_we advmod_scaled_how det_process_the nn_process_extraction nsubjpass_scaled_process aux_scaled_can auxpass_scaled_be ccomp_show_scaled prep_scaled_to det_section_the amod_section_complete nn_section_Wall nn_section_Street nn_section_Journal abbrev_section_WSJ pobj_to_section prep_section_of det_treebank_the amod_treebank_Penn-II pobj_of_treebank prep_scaled_with quantmod_million_about number_million_1 num_words_million pobj_with_words prep_words_in num_sentences_50,000 pobj_in_sentences prep_scaled_based dep_based_on det_algorithm_the amod_algorithm_automatic nn_algorithm_LFG nn_algorithm_f-structure nn_algorithm_annotation pobj_on_algorithm partmod_algorithm_described prep_described_in ' +o,1372,'1 A bilingual language model ITG has proposed a bilingual language model called Inversion Transduction Grammar -LRB- ITG -RRB- , which can be used to parse bilingual sentence pairs simultaneously ',Wu,'dep_model_1 det_model_A amod_model_bilingual nn_model_language nn__ITG nsubj_proposed_ aux_proposed_has dep_model_proposed det_model_a amod_model_bilingual nn_model_language nsubj_called_model ccomp_proposed_called nn_Grammar_Inversion nn_Grammar_Transduction dobj_called_Grammar abbrev_Grammar_ITG nsubjpass_used_which aux_used_can auxpass_used_be rcmod_Grammar_used aux_parse_to purpcl_used_parse amod_pairs_bilingual nn_pairs_sentence dobj_parse_pairs advmod_parse_simultaneously ' +o,1373,'5http : \\/ \\/ opennlpsourceforgenet \\/ We use the standard four-reference NIST MTEval data sets for the years 2003 , 2004 and 2005 -LRB- henceforth MT03 , MT04 and MT05 , respectively -RRB- for testing and the 2002 data set for tuning6 BLEU4 , METEOR and multiple-reference Word Error Rate scores are reported ',Banerjee,'amod_\\/_\\/ amod_\\/_\\/ nn_\\/_opennlpsourceforgenet nsubjpass_reported_\\/ nsubj_use_We rcmod_\\/_use det_data_the amod_data_standard amod_data_four-reference nn_data_NIST nn_data_MTEval nsubj_sets_data ccomp_use_sets prep_sets_for dep_years_the pobj_for_years tmod_years_2003 num_2003_2004 dep_2003_and dep_2003_2005 amod_MT03_henceforth dep_years_MT03 conj_MT03_MT04 cc_MT03_and conj_MT03_MT05 advmod_MT03_respectively prep_sets_for pobj_for_testing cc_\\/_and det_scores_the num_data_2002 dep_set_data amod_scores_set prep_set_for amod__tuning6 nn__BLEU4 pobj_for_ nn__METEOR conj__ cc__and amod_Rate_multiple-reference nn_Rate_Word nn_Rate_Error conj__Rate conj_\\/_scores auxpass_reported_are dep_5http_reported ' +o,1374,'4 Experiments 41 Experiment Settings A series of experiments were run to compare the performance of the three SWD models against the baseline , which is the standard phrase-based approach to SMT as elaborated in ',Koehn,'num_Experiments_4 nsubjpass_run_Experiments num_Settings_41 nn_Settings_Experiment dep_Experiments_Settings det_series_A dep_Experiments_series prep_series_of pobj_of_experiments auxpass_run_were aux_compare_to xcomp_run_compare det_performance_the dobj_compare_performance prep_performance_of det_models_the num_models_three nn_models_SWD pobj_of_models prep_compare_against det_baseline_the pobj_against_baseline nsubj_approach_which cop_approach_is det_approach_the amod_approach_standard amod_approach_phrase-based rcmod_baseline_approach aux_SMT_to infmod_approach_SMT advmod_elaborated_as acomp_SMT_elaborated prep_elaborated_in ' +o,1375,'We used the heuristic combination described in and extracted phrasal translation pairs from this combined alignment as described in ',Koehn,'nsubj_used_We det_combination_the amod_combination_heuristic dobj_used_combination dep_combination_described prep_described_in pobj_in_ cc_described_and conj_described_extracted amod_pairs_phrasal nn_pairs_translation dobj_extracted_pairs prep_extracted_from det_alignment_this amod_alignment_combined pobj_from_alignment advmod_described_as amod_alignment_described prep_described_in ' +o,1376,'Parse selection constitutes an important part of many parsing systems ',McClosky,'amod_selection_Parse nsubj_constitutes_selection det_part_an amod_part_important dobj_constitutes_part prep_part_of amod_systems_many amod_systems_parsing pobj_of_systems ' +o,1377,'To generate phrase pairs from a parallel corpus , we use the ` diag-and \' phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Brown,'aux_generate_To dep_use_generate nn_pairs_phrase dobj_generate_pairs prep_generate_from det_corpus_a amod_corpus_parallel pobj_from_corpus nsubj_use_we dep_described_use det_algorithm_the amod_algorithm_diag-and nn_algorithm_phrase nn_algorithm_induction dobj_use_algorithm prep_described_in pobj_in_ prep_described_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments partmod_alignments_generated xcomp_generated_using nn_model_IBM dobj_using_model nsubj_described_2 ' +o,1378,'In NLP community , it has been shown that having more data results in better performance ',Turney,'prep_shown_In nn_community_NLP pobj_In_community nsubjpass_shown_it aux_shown_has auxpass_shown_been complm_having_that ccomp_shown_having amod_results_more nn_results_data dobj_having_results prep_results_in amod_performance_better pobj_in_performance ' +o,1379,'Such a coding procedure covers , for example , how segmentation of a corpus is performed , if multiple tagging is allowed and if so , is it unlimited or are there just certain combinations of tags not allowed , is look ahead permitted , etc For further information on coding procedures we want to refer to and for good examples of coding books see , for example , , , or ',Carletta,'predet_covers_Such det_covers_a amod_covers_coding nn_covers_procedure nsubj_look_covers mark_unlimited_for nsubjpass_unlimited_example advmod_performed_how nsubjpass_performed_segmentation prep_segmentation_of det_corpus_a pobj_of_corpus auxpass_performed_is dep_example_performed mark_allowed_if amod_tagging_multiple nsubjpass_allowed_tagging auxpass_allowed_is dep_performed_allowed cc_performed_and dep_so_if conj_performed_so cop_unlimited_is measure_unlimited_it dep_covers_unlimited cc_unlimited_or auxpass_allowed_are advmod_just_there advmod_allowed_just amod_combinations_certain dep_just_combinations prep_combinations_of pobj_of_tags neg_allowed_not conj_unlimited_allowed cop_look_is advmod_permitted_ahead partmod_look_permitted dep_look_etc dep_etc_For amod_information_further pobj_For_information prep_information_on amod_procedures_coding pobj_on_procedures nsubj_want_we dep_look_want aux_refer_to xcomp_want_refer prep_refer_to pobj_to_ cc_to_and conj_to_for amod_examples_good pobj_for_examples prep_examples_of pcomp_of_coding nsubj_see_books ccomp_coding_see prep_see_for pobj_for_example dobj_see_ conj__ cc__or conj__ ' +o,1380,'2 Statistical Word Alignment According to the IBM models , the statistical word alignment model can be generally represented as in Equation -LRB- 1 -RRB- ',Brown,'num_Alignment_2 nn_Alignment_Statistical nn_Alignment_Word prep_represented_According dep_According_to det__the nn__IBM nn__models pobj_to_ det_model_the amod_model_statistical nn_model_word nn_model_alignment nsubjpass_represented_model aux_represented_can auxpass_represented_be advmod_represented_generally dep_Alignment_represented prep_represented_as dep_as_in pobj_in_Equation appos_Equation_1 ' +o,1381,'2 Data 21 The US Congressional Speech Corpus The text used in the experiments is from the United States Congressional Speech corpus , which is an XML formatted version of the electronic United States Congressional Record from the Library of Congress1 ',Thomas,'num_Data_2 nsubj_is_Data num_Data_21 det_Speech_The nn_Speech_US nn_Speech_Congressional nsubj_Corpus_Speech rcmod_Data_Corpus det_text_The dobj_Corpus_text partmod_text_used prep_used_in det_experiments_the pobj_in_experiments prep_is_from det__the nn__United nn__States nn__Congressional nn__Speech nn__corpus pobj_from_ nsubj_version_which cop_version_is det_version_an nn_version_XML amod_version_formatted rcmod__version prep_version_of det_Record_the amod_Record_electronic nn_Record_United nn_Record_States nn_Record_Congressional pobj_of_Record prep_Record_from det_Library_the pobj_from_Library prep_Library_of pobj_of_Congress1 ' +o,1382,'Among the four steps , the hypothesis alignment presents the biggest challenge to the method due to the varying word orders between outputs from different MT systems ',Rosti,'prep_presents_Among det_steps_the num_steps_four pobj_Among_steps det_alignment_the nn_alignment_hypothesis nsubj_presents_alignment det_challenge_the amod_challenge_biggest dobj_presents_challenge prep_presents_to det_method_the pobj_to_method amod_method_due prep_due_to det_orders_the amod_orders_varying nn_orders_word pobj_to_orders prep_orders_between pobj_between_outputs prep_outputs_from amod_systems_different nn_systems_MT pobj_from_systems ' +o,1383,' has proposed a bootstrapping method for word sense disambiguation ',Yarowsky,'nsubj_proposed_ aux_proposed_has det_method_a amod_method_bootstrapping dobj_proposed_method prep_method_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +o,1384,'Language modeling , noun-clustering , constructing syntactic rules for SMT , and finding analogies are examples of some of the problems where we need to compute relative frequencies ',Turney,'amod__Language nn__modeling amod__noun-clustering appos__ dep__constructing amod_rules_syntactic dobj_constructing_rules prep_constructing_for nn__SMT pobj_for_ cc_constructing_and conj_constructing_finding dobj_finding_analogies nsubj_examples_ cop_examples_are rcmod_analogies_examples prep_examples_of pobj_of_some prep_some_of det_problems_the pobj_of_problems advmod_need_where nsubj_need_we rcmod_examples_need aux_compute_to xcomp_need_compute amod_frequencies_relative dobj_compute_frequencies ' +o,1385,'This has been now an active research area for a couple of decades ',Hindle,'nsubj_been_This aux_been_has advmod_area_now det_area_an amod_area_active nn_area_research dobj_been_area prep_area_for det_couple_a pobj_for_couple prep_couple_of pobj_of_decades ' +o,1386,'Giza + + is a freely available implementation of IBM Models 1-5 and the HMM alignment , along with various improvements and modifications motivated by experimentation by Och & Ney ',Brown,'nn_+_Giza nn_+_+ nsubj_implementation_+ cop_implementation_is det_implementation_a advmod_available_freely amod_implementation_available prep_implementation_of nn__IBM nn__Models nn__1-5 pobj_of_ cc__and det__the nn__HMM nn__alignment conj__ advmod_motivated_along dep_along_with amod_improvements_various pobj_with_improvements cc_improvements_and conj_improvements_modifications dep_implementation_motivated prep_motivated_by pobj_by_experimentation prep_motivated_by nn__Och cc_Och_& conj_Och_Ney pobj_by_ ' +o,1387,'Although the BLEU score from Finnish to English is 218 , the score in the reverse direction is reported as 130 which is one of the lowest scores in 11 European languages scores ',Papineni,'mark_218_Although det_score_the amod_score_BLEU amod_score_ nsubj_218_score prep_score_from pobj_from_Finnish prep_Finnish_to pobj_to_English cop_218_is advcl_reported_218 det_score_the nsubjpass_reported_score prep_score_in det_direction_the amod_direction_reverse pobj_in_direction auxpass_reported_is prep_reported_as pobj_as_130 nsubj_one_which cop_one_is rcmod_130_one prep_one_of det_scores_the amod_scores_lowest pobj_of_scores prep_scores_in num_scores_11 amod_scores_European nn_scores_languages pobj_in_scores advmod_one_ ' +o,1388,' for English , but not identical to strictly anaphoric ones5 , since a non-anaphoric NP can corefer with a previous mention ',Bean,'prep__for pobj_for_English cc__but conj__not advmod__identical dep_identical_to amod__strictly amod__anaphoric amod__ones5 pobj_to_ mark_corefer_since det_NP_a amod_NP_non-anaphoric nsubj_corefer_NP aux_corefer_can dep__corefer prep_corefer_with det_mention_a amod_mention_previous pobj_with_mention ' +o,1389,'We report case sensitive Bleu scoreBleuCforallexperiments ',Papineni,'nsubj_report_We nsubj_sensitive_case xcomp_report_sensitive nn__Bleu nsubj_scoreBleuCforallexperiments_ dep_report_scoreBleuCforallexperiments ' +o,1390,'Using GIZA + + model 4 alignments and Pharaoh , we achieved a BLEU score of 03035 ',Koehn,'dep_achieved_Using amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ amod_alignments_model num_alignments_4 dobj_Using_alignments cc_alignments_and nn__Pharaoh conj_alignments_ nsubj_achieved_we det_score_a amod_score_BLEU dobj_achieved_score prep_score_of pobj_of_03035 ' +o,1391,'Making such an assumption is reasonable since POS taggers that can achieve accuracy of 96 \% are readily available to assign POS to unrestricted English sentences ',Cutting,'predet_assumption_such det_assumption_an nsubj_reasonable_assumption cop_reasonable_is ccomp_Making_reasonable mark_available_since dep_taggers_POS nsubj_available_taggers nsubj_achieve_that aux_achieve_can rcmod_taggers_achieve dobj_achieve_accuracy prep_accuracy_of num_\%_96 pobj_of_\% cop_available_are advmod_available_readily advcl_reasonable_available aux_assign_to xcomp_available_assign nsubj_unrestricted_POS aux_unrestricted_to xcomp_assign_unrestricted amod_sentences_English dobj_unrestricted_sentences ' +o,1392,'3 OverviewofExtractionWork 31 English As one mightexpect , the bulk of the collocation extractionwork concernsthe English language : , amongmany others1 ',Smadja,'num_language_3 measure_English_OverviewofExtractionWork num_OverviewofExtractionWork_31 amod_language_English prep_English_As num_mightexpect_one pobj_As_mightexpect det_bulk_the appos_mightexpect_bulk prep_bulk_of det_concernsthe_the nn_concernsthe_collocation nn_concernsthe_extractionwork pobj_of_concernsthe amod_language_English amod_others1_ amod_others1_amongmany dep_language_others1 ' +o,1393,'6 Experiments We evaluated the translation quality of the system using the BLEU metric ',Papineni,'num_Experiments_6 nsubj_evaluated_We rcmod_Experiments_evaluated det_quality_the nn_quality_translation dobj_evaluated_quality prep_quality_of det_system_the pobj_of_system partmod_system_using det_metric_the amod_metric_BLEU dobj_using_metric ' +o,1394,'The loglinear model feature weights were learned using minimum error rate training -LRB- MERT -RRB- with BLEU score as the objective function ',Papineni,'det_model_The nn_model_loglinear nsubj_feature_model nsubjpass_learned_weights auxpass_learned_were ccomp_feature_learned xcomp_learned_using amod__minimum nn__error nn__rate nn__training abbrev__MERT dobj_using_ prep__with amod__BLEU nn__score pobj_with_ prep_using_as det_function_the amod_function_objective pobj_as_function ' +o,1395,'Sentiment classification at the sentence-level has also been studied ',Stoyanov,'nn_classification_Sentiment nsubjpass_studied_classification prep_classification_at det_sentence-level_the pobj_at_sentence-level aux_studied_has advmod_studied_also auxpass_studied_been ' +o,1396,'This difference was highlighted in the 3http : \\/ \\/ w3msivxuse\\/jha\\/maltparser \\/ studyof , whichshowed that the difference is reflected directly in the error distributions of the parsers ',McDonald,'det_difference_This nsubjpass_highlighted_difference auxpass_highlighted_was prep_highlighted_in det_3http_the pobj_in_3http advmod_\\/_\\/ parataxis_highlighted_\\/ amod__w3msivxuse\\/jha\\/maltparser amod__\\/ nn__studyof dobj_\\/_ partmod__whichshowed complm_reflected_that det_difference_the nsubjpass_reflected_difference auxpass_reflected_is ccomp_whichshowed_reflected advmod_reflected_directly prep_reflected_in det_distributions_the nn_distributions_error pobj_in_distributions prep_distributions_of det_parsers_the pobj_of_parsers ' +o,1397,'43 Relaxing Length Restrictions Increasing the maximum phrase length in standard phrase-based translation does not improve BLEU ',Koehn,'num_Restrictions_43 nn_Restrictions_Relaxing nn_Restrictions_Length nsubj_improve_Restrictions partmod_Restrictions_Increasing det_length_the amod_length_maximum nn_length_phrase dobj_Increasing_length prep_Increasing_in amod_translation_standard amod_translation_phrase-based pobj_in_translation aux_improve_does neg_improve_not dobj_improve_BLEU ' +o,1398,'The importance of including single nonheadwords is now also uncontroversial , and the current paper has shown the importance of including two and more nonheadwords ',Collins,'det_importance_The nsubj__importance prep_importance_of pcomp_of_including amod_nonheadwords_single dobj_including_nonheadwords cop__is advmod__now advmod_uncontroversial_also amod__uncontroversial cc__and det_paper_the amod_paper_current nsubj_shown_paper aux_shown_has conj__shown det_importance_the dobj_shown_importance prep_importance_of pcomp_of_including num_nonheadwords_two cc_two_and conj_two_more dobj_including_nonheadwords ' +p,1399,'While does not discuss distinguishing more than 2 senses of a word , there is no immediate reason to doubt that the ` one sense per collocation \' rule would still hold for a larger number of senses ',Yarowsky,'mark_discuss_While nsubj_discuss_ aux_discuss_does neg_discuss_not advcl_is_discuss xcomp_discuss_distinguishing dep_than_more quantmod_2_than num_senses_2 dobj_distinguishing_senses prep_senses_of det_word_a pobj_of_word expl_is_there det_reason_no amod_reason_immediate nsubj_is_reason aux_doubt_to infmod_reason_doubt complm_hold_that det_sense_the nn_sense_one nsubj_hold_sense prep_sense_per poss__collocation nn__rule pobj_per_ aux_hold_would advmod_hold_still ccomp_doubt_hold prep_hold_for det_number_a amod_number_larger pobj_for_number prep_number_of pobj_of_senses ' +o,1400,'We annotated with the BIO tagging scheme used in syntactic chunkers ',Ramshaw,'nsubj_annotated_We prep_annotated_with det_scheme_the nn_scheme_BIO amod_scheme_tagging pobj_with_scheme partmod_scheme_used prep_used_in amod_chunkers_syntactic pobj_in_chunkers ' +o,1401,'We follow the approach of bootstrapping from a model with a narrower parameter space as is done in , eg Och and Ney and ',Fraser,'nsubj_follow_We det_approach_the dobj_follow_approach prep_approach_of pcomp_of_bootstrapping prep_bootstrapping_from det_model_a pobj_from_model prep_bootstrapping_with det_space_a amod_space_narrower nn_space_parameter pobj_with_space mark_done_as auxpass_done_is dep_space_done prep_done_in amod_Och_eg pobj_in_Och cc_Och_and nn_and_Ney nn_and_ conj_Och_and ' +o,1402,'6 Results We trained on the standard Penn Treebank WSJ corpus ',Marcus,'num_Results_6 nsubj_trained_We rcmod_Results_trained prep_trained_on det_corpus_the amod_corpus_standard nn_corpus_Penn nn_corpus_Treebank nn_corpus_WSJ pobj_on_corpus ' +o,1403,'For instance , BLEU and ROUGE are based on n-gram precisions , METEOR and STM use word-class or structural information , Kauchak -LRB- 2006 -RRB- leverages on paraphrases , and TER uses edit-distances ',Liu,'dep_leverages_For pobj_For_instance nn__BLEU cc_BLEU_and conj_BLEU_ROUGE nsubjpass_based_ auxpass_based_are dep_leverages_based prep_based_on amod_precisions_n-gram pobj_on_precisions nn__METEOR conj_precisions_ cc__and nn_word-class_STM nn_word-class_ nn_word-class_use conj__word-class cc_precisions_or amod_information_structural conj_precisions_information nsubj_leverages_Kauchak appos_Kauchak_2006 prep_leverages_on pobj_on_paraphrases cc_leverages_and nn__TER nsubj_uses_ conj_leverages_uses dobj_uses_edit-distances ' +o,1404,'Rulesize and lexicalization affect parsing complexity whether the grammar is binarized explicitly or implicitly binarized using Early-style intermediate symbols ',Zhang,'nsubj_complexity_Rulesize cc_Rulesize_and conj_Rulesize_lexicalization dep_complexity_affect amod_complexity_parsing complm_binarized_whether det_grammar_the nsubjpass_binarized_grammar auxpass_binarized_is ccomp_complexity_binarized advmod_binarized_explicitly acomp_binarized_ cc__or advmod_binarized_implicitly conj__binarized xcomp_binarized_using amod_symbols_Early-style amod_symbols_intermediate dobj_using_symbols advmod_using_ ' +o,1405,'Many methods for calculating the similarity have been proposed ',Banerjee,'amod_methods_Many nsubjpass_proposed_methods prep_methods_for pcomp_for_calculating det_similarity_the dobj_calculating_similarity aux_proposed_have auxpass_proposed_been ' +o,1406,'This idea of employing n-gram co-occurrence statistics to score the output of a computer system against one or more desired reference outputs has its roots in the BLEU metric for machine translation and the ROUGE metric for summarization ',Papineni,'det_idea_This prep_idea_of pcomp_of_employing amod_statistics_n-gram amod_statistics_co-occurrence dobj_employing_statistics prep_employing_to pobj_to_score det_output_the nsubj_desired_output prep_output_of det_system_a nn_system_computer pobj_of_system prep_system_against pobj_against_one cc_one_or conj_one_more dep_idea_desired nn_outputs_reference nsubj_has_outputs ccomp_desired_has poss_roots_its dobj_has_roots prep_has_in det_metric_the amod_metric_BLEU pobj_in_metric prep_metric_for nn__machine nn__translation pobj_for_ cc_metric_and det_metric_the nn_metric_ROUGE nn_metric_ conj_metric_metric prep_metric_for pobj_for_summarization ' +o,1407,'Nakagawa and also showed the effectiveness of global features in improving the accuracy of graph-based parsing , using the approximate Gibbs sampling method and a reranking approach , respectively ',Hall,'nn__Nakagawa nsubj_showed_ cc__and nn__ conj__ advmod_showed_also det_effectiveness_the dobj_showed_effectiveness prep_effectiveness_of amod_features_global pobj_of_features prep_showed_in pcomp_in_improving det_accuracy_the dobj_improving_accuracy prep_accuracy_of amod_parsing_graph-based pobj_of_parsing dep_improving_using det_method_the amod_method_approximate nn_method_Gibbs amod_method_sampling dobj_using_method cc_method_and det_approach_a amod_approach_reranking conj_method_approach advmod_using_respectively ' +o,1408,'One interesting approach to extending the current system is to introduce a statistical translation model to filter out irrelevant translation candidates and to extract the most appropriate subpart from a long English sequence as the translation by locally aligning the Japanese and English sequences ',Brown,'num_approach_One amod_approach_interesting nsubj_is_approach prep_approach_to pcomp_to_extending det_system_the amod_system_current dobj_extending_system aux_introduce_to xcomp_is_introduce det__a amod__statistical nn__translation nn__model dobj_introduce_ aux_filter_to dep_introduce_filter prt_filter_out amod_candidates_irrelevant nn_candidates_translation dobj_filter_candidates cc_filter_and aux_extract_to conj_filter_extract det_subpart_the advmod_appropriate_most amod_subpart_appropriate dobj_extract_subpart prep_extract_from det_sequence_a amod_sequence_long amod_sequence_English pobj_from_sequence prep_sequence_as det_translation_the pobj_as_translation prep_extract_by advmod_aligning_locally pcomp_by_aligning det_sequences_the amod_sequences_Japanese cc_Japanese_and conj_Japanese_English dobj_aligning_sequences ' +p,1409,'Similarity-based smoothing provides an intuitively appealing approach to language modeling ',Brown,'amod__Similarity-based nn__smoothing nsubj_provides_ det_approach_an advmod_appealing_intuitively amod_approach_appealing dobj_provides_approach prep_approach_to nn_modeling_language pobj_to_modeling ' +o,1410,'294 Fraser and Marcu Measuring Word Alignment Quality for Statistical Machine Translation 22 Measuring Translation Performance Changes Caused By Alignment In phrased-based SMT the knowledge sources which vary with the word alignment are the phrase translation lexicon -LRB- which maps source phrases to target phrases using counts from the word alignment -RRB- and some of the word level translation parameters -LRB- sometimes called lexical smoothing -RRB- ',Och,'nsubj_Measuring_Fraser cc_Fraser_and conj_Fraser_Marcu dep_294_Measuring nn_Quality_Word nn_Quality_Alignment nsubj_Caused_Quality prep_Quality_for nn_Changes_Statistical nn_Changes_Machine nn_Changes_Translation num_Changes_22 nn_Changes_Measuring nn_Changes_Translation nn_Changes_Performance pobj_for_Changes xcomp_Measuring_Caused dep_Caused_By pobj_By_Alignment dep_Caused_In amod__phrased-based nn__SMT pobj_In_ det_sources_the nn_sources_knowledge nsubj_lexicon_sources nsubj_vary_which rcmod_sources_vary prep_vary_with det_alignment_the nn_alignment_word pobj_with_alignment cop_lexicon_are det_lexicon_the nn_lexicon_phrase nn_lexicon_translation dep_294_lexicon dobj_target_which amod_phrases_maps nn_phrases_source nsubj_target_phrases aux_target_to dep_lexicon_target dobj_target_phrases partmod_phrases_using dobj_using_counts prep_using_from det_alignment_the nn_alignment_word pobj_from_alignment cc_lexicon_and conj_lexicon_some prep_some_of det_parameters_the nn_parameters_word nn_parameters_level nn_parameters_translation pobj_of_parameters advmod_called_sometimes dep_some_called amod_smoothing_lexical dobj_called_smoothing ' +o,1411,'Such techniques are currently being applied in many areas , including language identification , authorship attribution , text genre classification , topic identification , and subjective sentiment classification ',Turney,'amod_techniques_Such nsubjpass_applied_techniques aux_applied_are advmod_applied_currently auxpass_applied_being prep_applied_in amod_areas_many pobj_in_areas prep_applied_including nn_identification_language pobj_including_identification nn__authorship nn__attribution conj_identification_ nn__text nn__genre nn__classification conj_identification_ nn__topic nn__identification conj_identification_ cc_identification_and amod_classification_subjective nn_classification_sentiment conj_identification_classification ' +p,1412,' , which is the classic work on collocation extraction , uses a two-stage filtering model in which , in the first step , n-gram statistics determine possible collocations and , in the second step , these candidates are submitted to a syntactic valida7Of course , lexical material is always at least partially dependent on the domain in question ',Smadja,'nsubj_uses_ nsubj_work_which cop_work_is det_work_the amod_work_classic rcmod__work prep_work_on nn_extraction_collocation pobj_on_extraction det_model_a amod_model_two-stage amod_model_filtering dobj_uses_model rel_determine_in pobj_in_which dep_determine_in det_step_the amod_step_first pobj_in_step amod_statistics_n-gram nsubj_determine_statistics rcmod_model_determine amod_collocations_possible dobj_determine_collocations cc_collocations_and prep_,_in det_step_the amod_step_second pobj_in_step det_candidates_these nsubjpass_submitted_candidates auxpass_submitted_are ccomp_determine_submitted prep_submitted_to det_course_a amod_course_syntactic amod_course_valida7Of pobj_to_course amod_material_lexical nsubj_dependent_material cop_dependent_is advmod_dependent_always advmod_dependent_at dep_at_least advmod_dependent_partially dep_determine_dependent prep_dependent_on det_domain_the pobj_on_domain prep_domain_in pobj_in_question ' +o,1413,' present a probabilistic model for pronoun resolution trained on a small subset of the Penn Treebank Wall Street Journal corpus ',Marcus,'advmod_present_ det_model_a amod_model_probabilistic dobj_present_model prep_model_for nn_resolution_pronoun pobj_for_resolution partmod_resolution_trained prep_trained_on det_subset_a amod_subset_small pobj_on_subset prep_subset_of det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_of_corpus ' +o,1414,'This feature is implemented by using the IBM-1 lexical parameters ',Brown,'det_feature_This nsubjpass_implemented_feature auxpass_implemented_is prep_implemented_by pcomp_by_using det_parameters_the nn_parameters_IBM-1 nn_parameters_lexical dobj_using_parameters ' +o,1415,'In this paper we present results on using a recent phrase-based SMT system , PHARAOH , for NLG1 Although moderately effec1We also tried IBM Model 4\\/REWRITE , a word-based SMT system , but it gave much worse results ',Koehn,'prep_present_In det_paper_this pobj_In_paper nsubj_present_we dobj_present_results prep_present_on pcomp_on_using det_system_a amod_system_recent amod_system_phrase-based nn_system_SMT dobj_using_system nn__PHARAOH appos_system_ prep_system_for pobj_for_NLG1 mark_tried_Although amod_effec1We_moderately nsubj_tried_effec1We advmod_tried_also advcl_gave_tried nn__IBM nn__Model num__4\\/REWRITE dobj_tried_ det_system_a amod_system_word-based nn_system_SMT conj_gave_system cc_gave_but conj_gave_it dep_present_gave advmod_worse_much amod_results_worse dobj_gave_results ' +o,1416,'In phrase-based SMT systems , foreign sentences are firstly segmented into phrases which consists of adjacent words ',Koehn,'prep_segmented_In amod__phrase-based nn__SMT nn__systems pobj_In_ amod_sentences_foreign nsubj_segmented_sentences cop_segmented_are advmod_segmented_firstly prep_segmented_into pobj_into_phrases nsubj_consists_which rcmod_phrases_consists prep_consists_of amod_words_adjacent pobj_of_words ' +o,1417,'Automatic identification of subjective content often relies on word indicators , such as unigrams or predetermined sentiment lexica ',Pang,'nn_identification_Automatic nsubj_relies_identification prep_identification_of amod_content_subjective pobj_of_content advmod_relies_often prep_relies_on nn_indicators_word pobj_on_indicators dep_as_such prep_indicators_as amod__unigrams pobj_as_ cc__or amod__predetermined nn__sentiment nn__lexica conj__ ' +o,1418,'Some regarded Wikipedia as the corpora and applied hand-crafted or machine-learned rules to acquire semantic relations ',Kazama,'nsubj_regarded_Some dobj_regarded_Wikipedia prep_regarded_as det_corpora_the pobj_as_corpora cc_regarded_and conj_regarded_applied amod_rules_hand-crafted cc_hand-crafted_or conj_hand-crafted_machine-learned dobj_applied_rules aux_acquire_to xcomp_applied_acquire amod_relations_semantic dobj_acquire_relations ' +o,1419,'Since Soon started the trend of using the machine learning approach by using a binary classifier in a pairwise manner for solving co-reference resolution problem , many machine learning-based systems have been built , using both supervised and , unsupervised learning methods ',Haghighi,'mark_started_Since nn__Soon nsubj_started_ advcl_built_started det_trend_the dobj_started_trend prep_trend_of pcomp_of_using det_machine_the dobj_using_machine partmod_machine_learning dobj_learning_approach prep_learning_by pcomp_by_using det_classifier_a amod_classifier_binary dobj_using_classifier prep_classifier_in det_manner_a amod_manner_pairwise pobj_in_manner prep_using_for pcomp_for_solving amod_problem_co-reference nn_problem_resolution dobj_solving_problem amod_systems_many nn_systems_machine amod_systems_learning-based nsubjpass_built_systems aux_built_have auxpass_built_been xcomp_built_using nsubj_supervised_both ccomp_using_supervised cc_supervised_and conj_supervised_unsupervised xcomp_unsupervised_learning dobj_learning_methods ' +o,1420,'We used these weights in a beam search decoder to produce translations for the test sentences , which we compared to the WMT07 gold standard using Bleu ',Papineni,'nsubj_used_We det_weights_these dobj_used_weights prep_used_in det_decoder_a nn_decoder_beam nn_decoder_search pobj_in_decoder aux_produce_to xcomp_used_produce dobj_produce_translations prep_translations_for det_sentences_the nn_sentences_test pobj_for_sentences dep_compared_which nsubj_compared_we dep_used_compared prep_compared_to det_standard_the nn_standard_WMT07 nn_standard_gold pobj_to_standard xcomp_compared_using dobj_using_Bleu ' +p,1421,'One of the most effective taggers based on a pure HMM is that developed at Xerox ',Cutting,'nsubj_is_One prep_One_of det_taggers_the advmod_effective_most amod_taggers_effective pobj_of_taggers partmod_taggers_based prep_based_on det_HMM_a amod_HMM_pure pobj_on_HMM complm_developed_that ccomp_is_developed prep_developed_at pobj_at_Xerox ' +p,1422,'The success of recent high-quality parsers relies on the availability of such treebank corpora ',Collins,'det_success_The nsubj_relies_success prep_success_of amod__recent amod__high-quality nn__parsers pobj_of_ prep_relies_on det_availability_the pobj_on_availability prep_availability_of amod_corpora_such nn_corpora_treebank pobj_of_corpora ' +o,1423,'The training methods of LRM-F and SVM-F were useful to improve the F M - scores of LRM and SVM , as reported in ',Jansche,'det_methods_The nn_methods_training nsubj_useful_methods prep_methods_of pobj_of_LRM-F cc_LRM-F_and conj_LRM-F_SVM-F cop_useful_were aux_improve_to xcomp_useful_improve det_M_the nn_M_F dobj_improve_M dep_M_scores prep_scores_of pobj_of_LRM cc_LRM_and conj_LRM_SVM mark_reported_as advcl_improve_reported prt_reported_in ' +o,1424,'These categories were automatically generated using the labeled parses in Penn Treebank and the labeled semantic roles of PropBank ',Marcus,'det_categories_These nsubjpass_generated_categories auxpass_generated_were advmod_generated_automatically xcomp_generated_using det_parses_the amod_parses_labeled dobj_using_parses prep_using_in nn__Penn nn__Treebank pobj_in_ cc__and det_roles_the amod_roles_labeled amod_roles_semantic conj__roles prep_roles_of nn__PropBank pobj_of_ ' +o,1425,'or cooking , which agrees with the knowledge presented in previous work ',Ostler,'cc_cooking_or nsubj_presented_cooking nsubj_agrees_which dep_cooking_agrees prep_cooking_with det_knowledge_the pobj_with_knowledge prep_presented_in amod_work_previous pobj_in_work ' +o,1426,'This ITG constraint is characterized by the two forbidden structures shown in Figure 1 ',Wu,'det_constraint_This nn_constraint_ITG nsubjpass_characterized_constraint auxpass_characterized_is prep_characterized_by det_structures_the num_structures_two amod_structures_forbidden pobj_by_structures partmod_structures_shown prep_shown_in pobj_in_Figure num_Figure_1 ' +o,1427,'One is distortion model which penalizes translations according to their jump distance instead of their content ',Och,'nsubj__One cop__is nn__distortion nn__model nsubj_penalizes_which rcmod__penalizes dobj_penalizes_translations prep_penalizes_according dep_according_to poss_distance_their nn_distance_jump pobj_to_distance dep_of_instead prep_distance_of poss_content_their pobj_of_content ' +o,1428,'-LRB- ii -RRB- Apply some statistical tests such as the Binomial Hypothesis Test and loglikelihood ratio score to SCCs to filter out false SCCs on the basis of their reliability and likelihood ',Dunning,'dep_Apply_ii dobj_Apply_some amod_tests_statistical dep_Apply_tests dep_as_such prep_tests_as det__the nn__Binomial nn__Hypothesis nn__Test pobj_as_ cc__and amod_score_loglikelihood nn_score_ratio conj__score partmod__ prep__to pobj_to_SCCs aux_filter_to xcomp__filter prt_filter_out amod_SCCs_false dobj_filter_SCCs prep_filter_on det_basis_the pobj_on_basis prep_basis_of poss_reliability_their pobj_of_reliability cc_reliability_and conj_reliability_likelihood ' +o,1429,'Second , it can be applied to control the quality of parallel bilingual sentences mined from the Web , which are critical sources for a wide range of applications , such as statistical machine translation and cross-lingual information retrieval ',Brown,'advmod_applied_Second nsubjpass_applied_it aux_applied_can auxpass_applied_be aux_control_to purpcl_applied_control det_quality_the dobj_control_quality prep_quality_of amod_sentences_parallel amod_sentences_bilingual pobj_of_sentences partmod_sentences_mined prep_mined_from det_Web_the pobj_from_Web nsubj_sources_which cop_sources_are amod_sources_critical rcmod_Web_sources prep_sources_for det_range_a amod_range_wide pobj_for_range prep_range_of pobj_of_applications dep_as_such prep_range_as amod__statistical nn__machine nn__translation pobj_as_ cc__and amod__cross-lingual nn__information nn__retrieval conj__ ' +o,1430,'For example , the word alignment computed by GIZA + + and used as a basis to extract the TTS templates in most SSMT systems has been observed to be a problem for SSMT , due to the fact that the word-based alignment models are not aware of the syntactic structure of the sentences and could produce many syntax-violating word alignments ',DeNero,'prep_observed_For pobj_For_example det_alignment_the nn_alignment_word nsubjpass_observed_alignment dep_alignment_computed prep_computed_by nn_+_GIZA nn_+_+ pobj_by_+ cc_computed_and conj_computed_used prep_used_as det_basis_a pobj_as_basis aux_extract_to infmod_basis_extract det_templates_the nn_templates_TTS dobj_extract_templates prep_extract_in amod_systems_most amod_systems_SSMT pobj_in_systems aux_observed_has auxpass_observed_been aux_problem_to cop_problem_be det_problem_a xcomp_observed_problem prep_problem_for nn__SSMT pobj_for_ dep_observed_due prep_due_to det_fact_the pobj_to_fact dep_aware_that det_models_the amod_models_word-based nn_models_alignment nsubj_aware_models cop_aware_are neg_aware_not ccomp_observed_aware prep_aware_of det_structure_the amod_structure_syntactic pobj_of_structure prep_structure_of det_sentences_the pobj_of_sentences cc_observed_and aux_produce_could conj_observed_produce amod_alignments_many amod_alignments_syntax-violating nn_alignments_word dobj_produce_alignments ' +o,1431,'Baseline We use the Moses MT system as a baseline and closely follow the example training procedure given for the WMT-07 and WMT-08 shared tasks4 In particular , we perform word alignment in each direction using GIZA + + , apply the grow-diag-finaland heuristic for symmetrization and use a maximum phrase length of 7 ',Och,'advmod_use_Baseline nsubj_use_We det__the nn__Moses nn__MT nn__system dobj_use_ prep__as det_baseline_a pobj_as_baseline cc_use_and advmod_use_closely conj_use_follow det_procedure_the nn_procedure_example nn_procedure_training dobj_follow_procedure prep_follow_given dep_given_for det_tasks4_the amod_tasks4_WMT-07 cc_WMT-07_and conj_WMT-07_WMT-08 amod_tasks4_shared pobj_for_tasks4 prep_perform_In pobj_In_particular nsubj_perform_we dep_use_perform nn_alignment_word dobj_perform_alignment prep_perform_in det_direction_each pobj_in_direction partmod_direction_using dobj_using_GIZA amod_GIZA_+ cc_+_+ conj_+_ conj_perform_apply det_heuristic_the amod_heuristic_grow-diag-finaland dobj_apply_heuristic prep_heuristic_for pobj_for_symmetrization cc_perform_and conj_perform_use det_length_a amod_length_maximum nn_length_phrase dobj_use_length prep_length_of pobj_of_7 ' +o,1432,'2 Related Work This method is similar to block-orientation modeling and maximum entropy based phrase reordering model , in which local orientations -LRB- left\\/right -RRB- of phrase pairs -LRB- blocks -RRB- are learned via MaxEnt classifiers ',Tillmann,'num_Work_2 amod_Work_Related det_method_This nsubj_similar_method cop_similar_is dep_Work_similar prep_similar_to amod__block-orientation nn__modeling pobj_to_ cc__and nn_entropy_maximum conj__entropy partmod__based nn__phrase nn__reordering nn__model dobj_based_ rel_learned_in pobj_in_which amod_orientations_local nsubjpass_learned_orientations appos_orientations_left\\/right prep_orientations_of nn_pairs_phrase pobj_of_pairs appos_pairs_blocks auxpass_learned_are rcmod__learned prep_learned_via nn_classifiers_MaxEnt pobj_via_classifiers ' +n,1433,'Several studies have shown that large-margin methods can be adapted to the special complexities of the task However , the capacity of these algorithms to improve over state-of-the-art baselines is currently limited by their lack of robust dimensionality reduction ',Liang,'amod_studies_Several nsubj_shown_studies aux_shown_have complm_adapted_that amod_methods_large-margin nsubjpass_adapted_methods aux_adapted_can auxpass_adapted_be ccomp_shown_adapted prep_adapted_to det_complexities_the amod_complexities_special pobj_to_complexities prep_complexities_of det__the nn__task pobj_of_ advmod_limited_However det_capacity_the nsubjpass_limited_capacity prep_capacity_of det_algorithms_these pobj_of_algorithms aux_improve_to infmod_algorithms_improve prep_improve_over amod_baselines_state-of-the-art pobj_over_baselines auxpass_limited_is advmod_limited_currently dep_shown_limited prep_limited_by poss_lack_their pobj_by_lack prep_lack_of amod_reduction_robust amod_reduction_dimensionality pobj_of_reduction ' +o,1434,'The straight-forward way is to first generate the best BTG tree for each sentence pair using the way of , then annotate each BTG node with linguistic elements by projecting source-side syntax tree to BTG tree , and finally extract rules from these annotated BTG trees ',Wu,'det_way_The amod_way_straight-forward nsubj_is_way aux_generate_to advmod_generate_first xcomp_is_generate det_tree_the amod_tree_best nn_tree_BTG dobj_generate_tree prep_tree_for det_pair_each nn_pair_sentence pobj_for_pair partmod_pair_using det_way_the dobj_using_way prep_way_of pobj_of_ advmod_annotate_then conj_generate_annotate det_node_each nn_node_BTG dobj_annotate_node prep_annotate_with amod_elements_linguistic pobj_with_elements prep_annotate_by pcomp_by_projecting amod_tree_source-side nn_tree_syntax dobj_projecting_tree prep_projecting_to nn_tree_BTG pobj_to_tree cc_generate_and advmod_extract_finally conj_generate_extract dobj_extract_rules prep_extract_from det_trees_these amod_trees_annotated nn_trees_BTG pobj_from_trees ' +o,1435,'Measurement of Beliability The Kappa Statistic Following Jean , we use the kappa statistic to measure degree of agreement among subjects ',Carletta,'dep_use_Measurement prep_Measurement_of pobj_of_Beliability det_Statistic_The nn_Statistic_Kappa dobj_Measurement_Statistic prep_Statistic_Following nn__Jean pobj_Following_ nsubj_use_we det__the nn__kappa nn__statistic nsubj_measure_ aux_measure_to xcomp_use_measure dobj_measure_degree prep_degree_of pobj_of_agreement prep_measure_among pobj_among_subjects ' +o,1436,'As point out , WordNet does not encode antonymy across part-of-speech -LRB- for example , legallyembargo -RRB- ',Marcu,'mark_point_As nsubj_point_ advcl_encode_point prt_point_out nsubj_encode_WordNet aux_encode_does neg_encode_not advmod_encode_antonymy prep_encode_across pobj_across_part-of-speech dep_part-of-speech_for pobj_for_example appos_example_legallyembargo ' +o,1437,'In comparison , we deployed the GIZA + + MT modeling tool kit , which is an implementation of the IBM Models 1 to 4 ',Brown,'prep_deployed_In pobj_In_comparison nsubj_deployed_we det_kit_the nn_kit_GIZA nn_kit_+ nn_kit_+ nn_kit_MT nn_kit_modeling nn_kit_tool dobj_deployed_kit nsubj_implementation_which cop_implementation_is det_implementation_an rcmod_kit_implementation prep_implementation_of det_Models_the nn_Models_IBM pobj_of_Models number_4_1 dep_4_to num_Models_4 ' +o,1438,'This is the shared task baseline system for the 2006 NAACL\\/HLT workshop on statistical machine translation and consists of the Pharaoh decoder , SRILM , GIZA + + , mkcls , Carmel ,1 and a phrase model training code ',Och,'nsubj_system_This cop_system_is det_system_the amod_system_shared nn_system_task nn_system_baseline prep_system_for det_workshop_the num_workshop_2006 nn_workshop_NAACL\\/HLT pobj_for_workshop prep_system_on amod__statistical nn__machine nn__translation pobj_on_ cc_system_and conj_system_consists prep_consists_of det__the nn__Pharaoh nn__decoder pobj_of_ nn__SRILM conj__ nn__GIZA nn__+ nn__+ conj__ amod__mkcls conj__ conj__Carmel num_Carmel_,1 cc__and det_code_a nn_code_phrase nn_code_model nn_code_training conj__code ' +o,1439,' claimed that this approximation achieved essentially equivalent performance to that obtained when directly using the loss as the objective , O = lscript ',Och,'nsubj_claimed_ complm_lscript_that det_approximation_this nsubj_lscript_approximation partmod_approximation_achieved advmod_equivalent_essentially amod_performance_equivalent dobj_achieved_performance prep_achieved_to nsubj_obtained_that pcomp_to_obtained advmod_using_when advmod_using_directly advcl_obtained_using det_loss_the dobj_using_loss prep_using_as det_objective_the pobj_as_objective appos_objective_O dep_lscript_= ccomp_claimed_lscript ' +o,1440,'There also have been prior work on maintaining approximate counts for higher-order language models -LRB- LMs -RRB- -LRB- -RRB- operates under the model that the goal is to store a compressed representation of a disk-resident table of counts and use this compressed representation to answer count queries approximately ',Talbot,'expl_been_There advmod_been_also aux_been_have dep_operates_been amod_work_prior dobj_been_work prep_work_on pcomp_on_maintaining amod_counts_approximate dobj_maintaining_counts prep_maintaining_for amod_models_higher-order nn_models_language pobj_for_models abbrev_models_LMs appos_models_ prep_operates_under det_model_the pobj_under_model complm_is_that det_goal_the nsubj_is_goal ccomp_operates_is aux_store_to xcomp_is_store det_representation_a amod_representation_compressed dobj_store_representation prep_representation_of det_table_a amod_table_disk-resident pobj_of_table prep_table_of pobj_of_counts cc_store_and conj_store_use det_representation_this amod_representation_compressed dobj_use_representation aux_answer_to xcomp_use_answer nn_queries_count dobj_answer_queries nsubj_operates_approximately ' +o,1441,'Therefore the probability of alignment aj for position j should have a dependence on the previous alignment position O j_l : P -LRB- -LRB- \\/ j -LRB- -LRB- \\/ j-1 -RRB- A similar approach has been chosen by and ',Brown,'det_probability_the dep_Therefore_probability prep_probability_of amod_aj_alignment pobj_of_aj mark_have_for nn_j_position nsubj_have_j aux_have_should dep_probability_have det_dependence_a nsubj_P_dependence prep_dependence_on det_j_l_the amod_j_l_previous nn_j_l_alignment nn_j_l_position nn_j_l_O pobj_on_j_l ccomp_have_P nsubj_j_\\/ dep_P_j amod_j-1_\\/ nsubjpass_chosen_j-1 dep_similar_A amod_j-1_similar advmod_chosen_approach aux_chosen_has auxpass_chosen_been ccomp_j_chosen prep_chosen_by pobj_by_ cc_j_and conj_j_ ' +o,1442,'54 Domain Adaptation 541 Feature-Based Approaches Onewayofadaptingalearnertoanewdomainwithout using any unlabeled data is to only include features that are expected to transfer well ',Dredze,'nn_Onewayofadaptingalearnertoanewdomainwithout_Domain nn_Onewayofadaptingalearnertoanewdomainwithout_Adaptation nn_Onewayofadaptingalearnertoanewdomainwithout_541 nn_Onewayofadaptingalearnertoanewdomainwithout_Feature-Based nn_Onewayofadaptingalearnertoanewdomainwithout_Approaches nsubj_using_Onewayofadaptingalearnertoanewdomainwithout dep_54_using det_data_any amod_data_unlabeled nsubj_is_data ccomp_using_is aux_include_to advmod_include_only xcomp_is_include dobj_include_features nsubjpass_expected_that auxpass_expected_are rcmod_features_expected aux_transfer_to xcomp_expected_transfer advmod_transfer_well ' +o,1443,'Our test set is 3718 sentences from the English Penn treebank which were translated into German ',Brown,'poss_set_Our nn_set_test nsubj_sentences_set cop_sentences_is num_sentences_3718 prep_sentences_from det__the nn__English nn__Penn nn__treebank pobj_from_ nsubjpass_translated_which auxpass_translated_were rcmod__translated prep_translated_into pobj_into_German ' +o,1444,'Given a set of evidences E over all the relevant word pairs , in , the probabilistic taxonomy learning task is defined as the problem of finding the taxonomy hatwideT that maximizes the 67 probability of having the evidences E , ie : hatwideT = arg max T P -LRB- E T -RRB- In , this maximization problem is solved with a local search ',Snow,'prep_defined_Given det_set_a dep_Given_set prep_set_of amod_E_evidences pobj_of_E prep_E_over predet_pairs_all det_pairs_the amod_pairs_relevant nn_pairs_word pobj_over_pairs prep_pairs_in pobj_in_ det_task_the amod_task_probabilistic amod_task_taxonomy nn_task_learning nsubjpass_defined_task auxpass_defined_is prep_defined_as det_problem_the pobj_as_problem prep_problem_of pcomp_of_finding det_hatwideT_the amod_hatwideT_taxonomy dobj_finding_hatwideT nsubj_maximizes_that rcmod_hatwideT_maximizes det_probability_the num_probability_67 dobj_maximizes_probability prep_probability_of pcomp_of_having det_P_the amod_P_evidences nn_P_E dep_P_ie nn_P_hatwideT amod_P_= nn_P_arg nn_P_max nn_P_T dobj_having_P nn_T_E appos_P_T prep_solved_In pobj_In_ det_problem_this nn_problem_maximization nsubjpass_solved_problem auxpass_solved_is rcmod_P_solved prep_solved_with det_search_a amod_search_local pobj_with_search ' +o,1445,'Inter-annotator agreement was measured using the kappa -LRB- K -RRB- statistics on 1,502 instances -LRB- three Switchboard dialogues -RRB- marked by two annotators who followed specific written guidelines ',Carletta,'nn_agreement_Inter-annotator nsubjpass_measured_agreement auxpass_measured_was xcomp_measured_using det_kappa_the dobj_using_kappa appos_kappa_K nsubj__statistics dep_measured_ prep__on num_instances_1,502 pobj_on_instances num_dialogues_three nn_dialogues_Switchboard appos_instances_dialogues partmod_instances_marked prep_marked_by num_annotators_two pobj_by_annotators nsubj_followed_who rcmod_annotators_followed amod_guidelines_specific amod_guidelines_written dobj_followed_guidelines ' +o,1446,' noted that the unigram unpredictable might have a positive sentiment in a movie review -LRB- eg unpredictable plot -RRB- , but could be negative in the review of an automobile -LRB- eg unpredictable steering -RRB- ',Turney,'nsubj_noted_ complm_have_that det_unpredictable_the nn_unpredictable_unigram nsubj_have_unpredictable aux_have_might ccomp_noted_have det_sentiment_a amod_sentiment_positive dobj_have_sentiment prep_sentiment_in det_review_a nn_review_movie pobj_in_review amod_plot_eg amod_plot_unpredictable appos_review_plot cc_noted_but aux_negative_could cop_negative_be conj_noted_negative prep_negative_in det_review_the pobj_in_review prep_review_of det_automobile_an pobj_of_automobile amod_steering_eg amod_steering_unpredictable appos_automobile_steering ' +p,1447,'Synchronous binarization solves this problem by simultaneously binarizing both source and target-sides of a synchronous rule , making sure of contiguous spans on both sides whenever possible ',Zhang,'amod__Synchronous nn__binarization nsubj_solves_ det_problem_this dobj_solves_problem prep_solves_by advmod_binarizing_simultaneously pcomp_by_binarizing preconj_source_both dobj_binarizing_source cc_source_and conj_source_target-sides prep_source_of det_rule_a amod_rule_synchronous pobj_of_rule xcomp_solves_making amod_spans_sure prep_sure_of pobj_of_contiguous dobj_making_spans prep_spans_on det_sides_both pobj_on_sides advmod_possible_whenever dep_making_possible ' +o,1448,'To prune away those pairs , we used the log-likelihood-ratio algorithm to compute the degree of association between the verb and the noun in each pair ',Dunning,'aux_prune_To dep_used_prune prt_prune_away det_pairs_those dobj_prune_pairs nsubj_used_we det__the amod__log-likelihood-ratio nn__algorithm dobj_used_ aux_compute_to xcomp_used_compute det_degree_the dobj_compute_degree prep_degree_of pobj_of_association prep_association_between det_verb_the pobj_between_verb cc_verb_and det_noun_the conj_verb_noun prep_compute_in det_pair_each pobj_in_pair ' +o,1449,'An alternative approach to extracting the informal phrases is to use a bootstrapping algorithm -LRB- eg , -RRB- ',Yarowsky,'det_approach_An amod_approach_alternative nsubj_is_approach prep_approach_to pcomp_to_extracting det_phrases_the amod_phrases_informal dobj_extracting_phrases aux_use_to xcomp_is_use det_algorithm_a amod_algorithm_bootstrapping dobj_use_algorithm appos_algorithm_eg dep_eg_ ' +o,1450,'Accuracy on sentiment classification in other domains exceeds 80 \% ',Turney,'nsubj_exceeds_Accuracy prep_Accuracy_on nn_classification_sentiment pobj_on_classification prep_classification_in amod_domains_other pobj_in_domains num_\%_80 dobj_exceeds_\% ' +o,1451,'Unlike , the ke ~ vord rnay be part of a Chinese word ',Smadja,'prep_part_Unlike pobj_Unlike_ det_~_the nn_~_ke nsubj_part_~ aux_part_vord advmod_part_rnay cop_part_be prep_part_of det_word_a amod_word_Chinese pobj_of_word ' +o,1452,'Re-ordering effects across languages have been modeled in several ways , including word-based , template-based and syntax-based ',Brown,'amod_effects_Re-ordering nsubjpass_modeled_effects prep_effects_across pobj_across_languages aux_modeled_have auxpass_modeled_been prep_modeled_in amod_ways_several pobj_in_ways prep_modeled_including amod__word-based pobj_including_ amod__template-based conj__ cc__and amod__syntax-based conj__ ' +o,1453,' generate ill-formed sentences by sampling a probabilistic language model and end up with pseudo-negative examples which resemble machine translation output more than they do learner texts ',Okanohara,'advmod_generate_ amod_sentences_ill-formed dobj_generate_sentences prep_generate_by pcomp_by_sampling det_model_a amod_model_probabilistic nn_model_language dobj_sampling_model cc_generate_and conj_generate_end prt_end_up prep_end_with amod_examples_pseudo-negative pobj_with_examples nsubj_output_which cop_output_resemble nn_output_machine nn_output_translation rcmod_examples_output advmod_output_more mark_do_than nsubj_do_they dep_more_do nn_texts_learner dobj_do_texts ' +o,1454,'For the combined set -LRB- ALL -RRB- , we also show the 95 \% BLEU confidence interval computed using bootstrap resampling ',Och,'prep_show_For det_set_the amod_set_combined pobj_For_set abbrev_set_ALL nsubj_show_we advmod_show_also det_confidence_the dep_\%_95 amod_confidence_\% nn_confidence_BLEU dobj_show_confidence amod_confidence_interval dep_interval_computed xcomp_show_using nn_resampling_bootstrap dobj_using_resampling ' +o,1455,'33 Model Construction The head transducer model was trained and evaluated on English-to-Mandarin Chinese translation of transcribed utterances from the ATIS corpus ',Brown,'num_Construction_33 nn_Construction_Model det_model_The nn_model_head nn_model_transducer nsubjpass_trained_model auxpass_trained_was dep_Construction_trained cc_trained_and conj_trained_evaluated prep_trained_on amod_translation_English-to-Mandarin amod_translation_Chinese pobj_on_translation prep_translation_of amod_utterances_transcribed pobj_of_utterances prep_utterances_from det__the nn__ATIS nn__corpus pobj_from_ ' +o,1456,'We have applied it to the two data sets mentioned in ',Ramshaw,'nsubj_applied_We aux_applied_have dobj_applied_it prep_applied_to det_sets_the num_sets_two nn_sets_data pobj_to_sets partmod_sets_mentioned prep_mentioned_in ' +o,1457,'Bilingual bracketing methods were used to produce a word alignment in ',Wu,'amod_methods_Bilingual nn_methods_bracketing nsubjpass_used_methods auxpass_used_were aux_produce_to xcomp_used_produce det_alignment_a nn_alignment_word dobj_produce_alignment prt_produce_in ' +o,1458,'Probabilistic generative models like IBM 1-5 (Brown et al., 1993), HMM (Vogel et al., 1996), ITG (Wu, 1997), and LEAF (Fraser and Marcu, 2007) define formulas for P(f | e) or P(e, f), with ok-voon ororok sprok at-voon bichat dat erok sprok izok hihok ghirok totat dat arrat vat hilat ok-drubel ok-voon anok plok sprok at-drubel at-voon pippat rrat dat ok-voon anok drok brok jok at-voon krat pippat sat lat wiwok farok izok stok totat jjat quat cat lalok sprok izok jok stok wat dat krat quat cat lalok farok ororok lalok sprok izok enemok wat jjat bichat wat dat vat eneat lalok brok anok plok nok iat lat pippat rrat nnat wiwok nok izok kantok ok-yurp totat nnat quat oloat at-yurp lalok mok nok yorok ghirok clok wat nnat gat mat bat hilat lalok nok crrrok hihok yorok zanzanok wat nnat arrat mat zanzanat lalok rarok nok izok hihok mok wat nnat forat arrat vat gat Figure 1: Word alignment exercise (Knight, 1997).',Fraser,'' +o,1459,'Still , however , such techniques often require seeds , or prototypes -LRB- cf , -RRB- which are used to prune search spaces or direct learners ',Haghighi,'advmod_require_Still advmod_require_however amod_techniques_such nsubj_require_techniques advmod_require_often dobj_require_seeds cc_seeds_or conj_seeds_prototypes appos_seeds_cf dep_cf_ nsubjpass_used_which auxpass_used_are rcmod_seeds_used aux_prune_to xcomp_used_prune nn_spaces_search dobj_prune_spaces cc_spaces_or amod_learners_direct conj_spaces_learners ' +o,1460,'As resolving direct anaphoric descriptions -LRB- the ones where anaphor and antecedent have the same head noun -RRB- is a much simpler problem with high performance rates as shown in previous results , these heuristics should be applied first in a system that resolves definite descriptions ',Bean,'mark_problem_As csubj_problem_resolving amod_descriptions_direct amod_descriptions_anaphoric dobj_resolving_descriptions det_ones_the dep_descriptions_ones advmod_have_where nsubj_have_anaphor cc_anaphor_and conj_anaphor_antecedent rcmod_ones_have det_noun_the amod_noun_same nn_noun_head dobj_have_noun cop_problem_is det_problem_a advmod_simpler_much amod_problem_simpler advcl_applied_problem prep_problem_with amod_rates_high nn_rates_performance pobj_with_rates mark__as csubj__shown prep_shown_in amod_results_previous pobj_in_results advcl_problem_ det_heuristics_these nsubjpass_applied_heuristics aux_applied_should auxpass_applied_be advmod_applied_first prep_applied_in det_system_a pobj_in_system nsubj_resolves_that rcmod_system_resolves amod_descriptions_definite dobj_resolves_descriptions ' +o,1461,'et al , 2004 ; Collins-Thompson and Callan , 2005 ; and Ramage , 2007 -RRB- ',Hughes,'nn_al_et appos_al_2004 dep_al_Collins-Thompson cc_Collins-Thompson_and conj_Collins-Thompson_Callan appos_Collins-Thompson_2005 dep_2005_ cc__and conj__Ramage appos__2007 ' +o,1462,'We benchmark our results against a model -LRB- Hiero -RRB- which was directly trained to optimise BLEUNIST using the standard MERT algorithm and the full set of translation and lexical weight features described for the Hiero model ',Och,'nsubj_benchmark_We poss_results_our dobj_benchmark_results prep_benchmark_against det_model_a pobj_against_model appos_model_Hiero nsubjpass_trained_which auxpass_trained_was advmod_trained_directly rcmod_model_trained aux_optimise_to xcomp_trained_optimise dobj_optimise_BLEUNIST partmod_BLEUNIST_using det__the amod__standard nn__MERT nn__algorithm dobj_using_ cc_BLEUNIST_and det_set_the amod_set_full conj_BLEUNIST_set prep_set_of nn_features_translation cc_translation_and conj_translation_lexical nn_features_weight pobj_of_features partmod_set_described prep_described_for det__the nn__Hiero nn__model pobj_for_ ' +o,1463,'Although to a lesser extent , measures of word relatedness have also been applied on other languages , including German , Chinese , Dutch and others ',Mohammad,'mark_applied_Although prep_applied_to det_extent_a amod_extent_lesser pobj_to_extent nsubjpass_applied_measures prep_measures_of nn_relatedness_word pobj_of_relatedness aux_applied_have advmod_applied_also auxpass_applied_been prep_applied_on amod_languages_other pobj_on_languages prep_applied_including amod__German pobj_including_ amod__Chinese conj__ amod__Dutch conj__ cc__and conj__others ' +p,1464,' reported very high results -LRB- 96 \% on the Brown corpus -RRB- for unsupervised POS tagging using Hidden Markov Models -LRB- HMMs -RRB- by exploiting hand-built tag dictionaries and equivalence classes ',Cutting,'nsubj_reported_ advmod_high_very amod_results_high dobj_reported_results num_\%_96 dep_results_\% prep_\%_on det_corpus_the nn_corpus_Brown pobj_on_corpus prep_reported_for amod_POS_unsupervised pobj_for_POS partmod_POS_tagging xcomp_tagging_using nn_Models_Hidden nn_Models_Markov dobj_using_Models abbrev_Models_HMMs prep_using_by pcomp_by_exploiting amod_dictionaries_hand-built nn_dictionaries_tag dobj_exploiting_dictionaries cc_dictionaries_and nn_classes_equivalence conj_dictionaries_classes ' +o,1465,'This method is very similar to some ideas in domain adaptation , but we argue that the underlying problems are quite different ','Daume III','det_method_This nsubj_similar_method cop_similar_is advmod_similar_very prep_similar_to det_ideas_some pobj_to_ideas prep_ideas_in nn__domain nn__adaptation pobj_in_ cc_similar_but nsubj_argue_we conj_similar_argue complm_different_that det_problems_the amod_problems_underlying nsubj_different_problems cop_different_are advmod_different_quite ccomp_argue_different ' +o,1466,'This wrong translation of content words is similar to the incorrect omission reported in , which both hurt translation adequacy ',Och,'det_translation_This amod_translation_wrong nsubj_similar_translation prep_translation_of nn_words_content pobj_of_words cop_similar_is prep_similar_to det_omission_the amod_omission_incorrect pobj_to_omission partmod_omission_reported prep_reported_in pobj_in_ nsubj_hurt_which dep_hurt_both rcmod__hurt nn_adequacy_translation dobj_hurt_adequacy ' +o,1467,'1 Introduction In global linear models -LRB- GLMs -RRB- for structured prediction , -LRB- eg , -RRB- , the optimal label y for an input x is y = arg max yY -LRB- x -RRB- w f -LRB- x , y -RRB- -LRB- 1 -RRB- where Y -LRB- x -RRB- is the set of possible labels for the input x ; f -LRB- x , y -RRB- Rd is a feature vector that represents the pair -LRB- x , y -RRB- ; and w is a parameter vector ',Collins,'num_Introduction_1 prep_yY_In amod_models_global amod_models_linear pobj_In_models abbrev_models_GLMs prep_models_for amod_prediction_structured pobj_for_prediction appos_prediction_eg dep_eg_ det_y_the amod_y_optimal nn_y_label nsubj_yY_y prep_y_for det_input_an pobj_for_input nsubjpass_y_x auxpass_y_is rcmod_input_y dep_max_= nn_max_arg ccomp_y_max dep_Introduction_yY nsubj_w_x ccomp_yY_w dobj_w_f dep_f_x appos_x_y appos_f_1 advmod_set_where nsubj_set_Y dep_Y_x cop_set_is det_set_the rcmod_f_set prep_set_of amod_labels_possible pobj_of_labels prep_labels_for det_input_the pobj_for_input dep_input_x dobj_vector_f dep_f_x appos_x_y nsubj_vector_Rd cop_vector_is det_vector_a nn_vector_feature dep_x_vector nsubj_represents_that rcmod_vector_represents det_pair_the dobj_represents_pair dep_pair_x appos_x_y cc_yY_and nsubj_vector_w cop_vector_is det_vector_a nn_vector_parameter conj_yY_vector ' +o,1468,'We use the same preprocessing steps as Turian and Melamed : during both training and testing , the parser is given text POS-tagged by the tagger of , with capitalization stripped and outermost punctuation removed ',Ratnaparkhi,'nsubj_use_We det_steps_the amod_steps_same amod_steps_preprocessing dobj_use_steps prep_steps_as nn__Turian cc_Turian_and conj_Turian_Melamed pobj_as_ prep_given_during preconj_training_both pobj_during_training cc_training_and conj_training_testing det_parser_the nsubjpass_given_parser auxpass_given_is parataxis_use_given dobj_given_text amod_text_POS-tagged prep_given_by det_tagger_the pobj_by_tagger prep_tagger_of pobj_of_ prep_given_with pobj_with_capitalization partmod_capitalization_stripped cc_use_and nn_punctuation_outermost nsubj_removed_punctuation conj_use_removed ' +o,1469,'2 Phrase-based SMT We use a phrase-based SMT system , Pharaoh , , which is based on a log-linear formulation ',Och,'num_SMT_2 amod_SMT_Phrase-based nsubj_use_We dep_SMT_use det_system_a amod_system_phrase-based nn_system_SMT dobj_use_system appos_system_Pharaoh appos_system_ nsubjpass_based_which auxpass_based_is rcmod_system_based prep_based_on det_formulation_a amod_formulation_log-linear pobj_on_formulation ' +o,1470,'To quickly -LRB- and approximately -RRB- evaluate this phenomenon , we trained the statistical IBM wordalignment model 4 ,1 using the GIZA + + software for the following language pairs : ChineseEnglish , Italian English , and DutchEnglish , using the IWSLT-2006 corpus for the first two language pairs , and the Europarl corpus for the last one ',Brown,'aux_evaluate_To advmod_evaluate_quickly dep_-LRB-_and dep_-LRB-_approximately dep_trained_evaluate det_phenomenon_this dobj_evaluate_phenomenon nsubj_trained_we det_model_the amod_model_statistical nn_model_IBM nn_model_wordalignment iobj_trained_model number__4 dobj_trained_ dobj_trained_,1 xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__software dobj_using_ prep__for det_pairs_the amod_pairs_following nn_pairs_language pobj_for_pairs nn_English_ChineseEnglish amod_English_Italian dep__English cc_English_and conj_English_DutchEnglish partmod_English_using det__the nn__IWSLT-2006 nn__corpus dobj_using_ prep__for det_pairs_the amod_pairs_first num_pairs_two nn_pairs_language pobj_for_pairs cc__and det__the nn__Europarl nn__corpus conj__ prep__for det_one_the amod_one_last pobj_for_one ' +o,1471,'Alignment spaces can emerge from generative stories , from syntactic notions , or they can be imposed to create competition between links ',Brown,'amod_spaces_Alignment nsubj_emerge_spaces aux_emerge_can prep_emerge_from amod__generative nn__stories pobj_from_ prep_emerge_from amod__syntactic nn__notions pobj_from_ cc_emerge_or nsubjpass_imposed_they aux_imposed_can auxpass_imposed_be conj_emerge_imposed aux_create_to purpcl_imposed_create dobj_create_competition prep_competition_between nn__links pobj_between_ ' +o,1472,'1 Introduction Many different statistical tests have been proposed to measure the strength of word similarity or word association in natural language texts ',Dunning,'num_Introduction_1 amod_tests_Many amod_tests_different amod_tests_statistical nsubjpass_proposed_tests aux_proposed_have auxpass_proposed_been dep_Introduction_proposed aux_measure_to xcomp_proposed_measure det_strength_the dobj_measure_strength prep_strength_of nn_similarity_word pobj_of_similarity cc_similarity_or nn_association_word conj_similarity_association prep_similarity_in amod_texts_natural nn_texts_language pobj_in_texts ' +o,1473,'For example , it has been observed that texts often contain multiple opinions on different topics , which makes assignment of the overall sentiment to the whole document problematic ',Turney,'prep_observed_For pobj_For_example nsubjpass_observed_it aux_observed_has auxpass_observed_been complm_contain_that nsubj_contain_texts advmod_contain_often ccomp_observed_contain amod_opinions_multiple dobj_contain_opinions prep_contain_on amod__different nn__topics pobj_on_ nsubj_makes_which rcmod__makes nsubj_problematic_assignment prep_assignment_of det_sentiment_the amod_sentiment_overall pobj_of_sentiment prep_sentiment_to det_document_the amod_document_whole pobj_to_document xcomp_makes_problematic ' +o,1474,'In each experiment , performance IMutu \' , d Information provides an estimate of the magnitude of the ratio t -RRB- ctw -LRB- -LRB- - n the joint prol -RRB- ability P -LRB- verb\\/noun ,1 -RRB- reposition -RRB- , and the joint probability a ~ suming indcpendcnce P -LRB- verb\\/noun -RRB- P -LRB- prcl -RRB- osition -RRB- s -LRB- : -LRB- , ',Church,'prep_provides_In det_IMutu_each nn_IMutu_experiment dep_IMutu_performance pobj_In_IMutu nn_Information_d nsubj_provides_Information det_estimate_an dobj_provides_estimate prep_estimate_of det_magnitude_the pobj_of_magnitude prep_magnitude_of det_ctw_the nn_ctw_ratio nn_ctw_t pobj_of_ctw dep_ability_n det_prol_the amod_prol_joint dep_n_prol prep_ctw_ability dep_ability_P appos_P_verb\\/noun num_verb\\/noun_,1 dep_P_reposition cc_magnitude_and det_probability_the amod_probability_joint conj_magnitude_probability det_P_a punct_P_~ nn_P_suming nn_P_indcpendcnce nn_P_P nn_P_verb\\/noun nsubj_s_P appos_P_prcl dep_P_osition rcmod_probability_s ' +o,1475,'We use the IBM Model 1 -LRB- uniform distribution -RRB- and the Hidden Markov Model -LRB- HMM , first-order dependency , -RRB- to estimate the alignment model ',Brown,'nsubj_use_We det__the nn__IBM nn__Model num__1 nsubj_estimate_ amod_distribution_uniform appos__distribution cc__and det_Model_the nn_Model_Hidden nn_Model_Markov conj__Model appos_Model_HMM amod_dependency_first-order dep_HMM_dependency dep_HMM_ aux_estimate_to xcomp_use_estimate det_model_the amod_model_alignment dobj_estimate_model ' +o,1476,'The parameters , j , were trained using minimum error rate training to maximize the BLEU score on a 150 sentence development set ',Och,'det_parameters_The nsubjpass_trained_parameters appos_parameters_j auxpass_trained_were xcomp_trained_using amod__minimum nn__error nn__rate nn__training dobj_using_ aux_maximize_to xcomp_using_maximize det__the amod__BLEU nn__score dobj_maximize_ prep_maximize_on det_set_a num_set_150 nn_set_sentence nn_set_development pobj_on_set ' +o,1477,'For these first SMT systems , translation-model probabilities at the sentence level were approximated from word-based translation models that were trained by using bilingual corpora ',Brown,'prep_approximated_For det_systems_these amod_systems_first nn_systems_SMT pobj_For_systems amod_probabilities_translation-model nsubjpass_approximated_probabilities prep_probabilities_at det_level_the nn_level_sentence pobj_at_level auxpass_approximated_were prep_approximated_from amod_models_word-based nn_models_translation pobj_from_models nsubjpass_trained_that auxpass_trained_were rcmod_models_trained prep_trained_by pcomp_by_using amod_corpora_bilingual dobj_using_corpora ' +o,1478,' binarize grammars into CNF normal form , while allow only Griebach-Normal form grammars ',Zhang,'amod_grammars_binarize dobj__grammars prep__into pobj_into_CNF amod_form_normal nsubj__form mark_allow_while nsubj_allow_ dep_form_allow advmod_grammars_only amod_grammars_Griebach-Normal nn_grammars_form dobj_allow_grammars ' +o,1479,'Equation -LRB- 10 -RRB- is of interest because the ratio p -LRB- C v , r -RRB- \\/ p -LRB- C r -RRB- can be interpreted as a measure of association between the verb v and class C This ratio is similar to pointwise mutual information and also forms part of Resniks association score , which will be introduced in Section 6 ',Church,'nsubj_is_Equation appos_Equation_10 prep_is_of pobj_of_interest mark_interpreted_because det_p_the nn_p_ratio nn_p_p nn_v_C dep_p_v appos_v_r amod_p_\\/ nsubjpass_interpreted_p nn_r_C appos_p_r aux_interpreted_can auxpass_interpreted_be advcl_is_interpreted prep_interpreted_as det_measure_a pobj_as_measure prep_measure_of pobj_of_association prep_association_between det_v_the nn_v_verb pobj_between_v cc_v_and nn_C_class conj_v_C det_ratio_This nsubj_similar_ratio cop_similar_is dep_is_similar aux_pointwise_to xcomp_similar_pointwise amod__mutual nn__information dobj_pointwise_ cc_similar_and advmod_forms_also conj_similar_forms dobj_forms_part prep_part_of nn_score_Resniks nn_score_association pobj_of_score nsubjpass_introduced_which aux_introduced_will auxpass_introduced_be rcmod_score_introduced prep_introduced_in pobj_in_Section num_Section_6 ' +o,1480,'In , lexical 72 features were limited on each single side due to the feature space problem ',He,'prep_limited_In pobj_In_ amod_features_lexical num_features_72 nsubjpass_limited_features auxpass_limited_were prep_limited_on det_side_each amod_side_single pobj_on_side amod_side_due prep_due_to det_problem_the nn_problem_feature nn_problem_space pobj_to_problem ' +o,1481,'The SPECIALIST minimal commitment parser relies on the SPECIALIST Lexicon as well as the Xerox stochastic tagger ',Cutting,'det_parser_The amod_parser_SPECIALIST amod_parser_minimal nn_parser_commitment nsubj_relies_parser prep_relies_on det_Lexicon_the nn_Lexicon_SPECIALIST pobj_on_Lexicon dep_well_as cc_Lexicon_well dep_well_as det_tagger_the nn_tagger_Xerox amod_tagger_stochastic conj_Lexicon_tagger ' +o,1482,'The idea caught on very quickly : Suhm and Waibel , Mast et aL , Warnke et al ',Berger,'det_idea_The partmod_idea_caught prep_caught_on advmod_quickly_very dep_on_quickly dep_idea_Suhm cc_Suhm_and nn__Waibel nn__ conj__Mast cc__et conj__aL conj_Suhm_ conj__Warnke cc__et conj__al ' +o,1483,' integrated a WSD system into a phrase-based SMT system , Pharaoh ',Carpuat,'nsubj_integrated_ det_system_a nn_system_WSD dobj_integrated_system prep_integrated_into det_system_a amod_system_phrase-based nn_system_SMT pobj_into_system nn__Pharaoh appos_system_ ' +o,1484,'Recently , it has gained renewed attention as empirical methods in parsing have emphasized the importance of relations between words -LRB- see , eg , -RRB- , which is what dependency grammars model explicitly , but context-free phrase-structure grammars do not ',Collins,'advmod_gained_Recently nsubj_gained_it aux_gained_has amod_attention_renewed dobj_gained_attention mark_emphasized_as amod_methods_empirical nsubj_emphasized_methods prep_methods_in pobj_in_parsing aux_emphasized_have advcl_gained_emphasized det_importance_the dobj_emphasized_importance prep_importance_of pobj_of_relations prep_relations_between pobj_between_words dep_importance_see dep_see_eg dep_see_ nsubj_is_which rcmod_importance_is dobj_grammars_what nsubj_grammars_dependency ccomp_is_grammars dobj_grammars_model advmod_grammars_explicitly cc_emphasized_but amod_grammars_context-free amod_grammars_phrase-structure nsubj_do_grammars conj_emphasized_do neg_do_not ' +o,1485,'It is clear that Appendix B contains far fewer true non-compositional phrases than Appendix A 7 Related Work There have been numerous previous research on extracting collocations from corpus , eg , and ',Dunning,'nsubj_clear_It cop_clear_is complm_contains_that nn_B_Appendix nsubj_contains_B ccomp_clear_contains advmod_fewer_far amod_phrases_fewer amod_phrases_true amod_phrases_non-compositional dobj_contains_phrases prep_phrases_than nn_Work_Appendix nn_Work_A num_Work_7 nn_Work_Related pobj_than_Work expl_been_There aux_been_have dep_clear_been amod_research_numerous amod_research_previous dobj_been_research prep_research_on pcomp_on_extracting dobj_extracting_collocations prep_extracting_from pobj_from_corpus conj_corpus_eg conj_corpus_ cc_corpus_and conj_corpus_ ' +o,1486,'model reranking has also been established , both for synchronous binarization and for target-only binarization ',Zhang,'nn_reranking_model nsubjpass_established_reranking aux_established_has advmod_established_also auxpass_established_been preconj_for_both prep_established_for amod__synchronous nn__binarization pobj_for_ cc_for_and conj_for_for amod__target-only nn__binarization pobj_for_ ' +o,1487,'From wordlevel alignments , such systems extract the grammar rules consistent either with the alignments and parse trees for one of languages , or with the the word-level alignments alone without reference to external syntactic analysis , which is the scenario we address here ',Galley,'prep_extract_From amod_alignments_wordlevel pobj_From_alignments amod_systems_such nsubj_extract_systems det_rules_the nn_rules_grammar nsubj_consistent_rules xcomp_extract_consistent preconj_with_either prep_consistent_with det_alignments_the pobj_with_alignments cc_alignments_and nn_trees_parse conj_alignments_trees prep_alignments_for pobj_for_one prep_one_of pobj_of_languages partmod_languages_ cc_with_or conj_with_with det__the det_alignments_the amod_alignments_word-level measure_alone_alignments advmod__alone dep_alone_without pobj_without_reference dep_without_to amod_analysis_external amod_analysis_syntactic pobj_to_analysis pobj_with_ nsubj_scenario_which cop_scenario_is det_scenario_the rcmod__scenario nsubj_address_we dep_extract_address advmod_address_here ' +o,1488,'A remedy is to aggressively limit the feature space , eg to syntactic labels or a small fraction of the bi-lingual features available , as in , but that reduces the benefit of lexical features ',He,'det_remedy_A nsubj_is_remedy aux_limit_to advmod_limit_aggressively xcomp_is_limit det_space_the nn_space_feature dobj_limit_space conj_space_eg prep_eg_to amod_labels_syntactic pobj_to_labels cc_space_or det_fraction_a amod_fraction_small conj_space_fraction prep_fraction_of det_features_the amod_features_bi-lingual pobj_of_features amod_features_available dep_features_as dep_as_in pobj_in_ cc_as_but nsubj_reduces_that conj_as_reduces det_benefit_the dobj_reduces_benefit prep_benefit_of amod_features_lexical pobj_of_features ' +o,1489,' Classification allows a word to align with a target word using the collective translation tendency of words in the same class ',Brown,'nn_Classification_ nsubj_allows_Classification det_word_a dobj_allows_word aux_align_to infmod_word_align prep_align_with det_word_a nn_word_target pobj_with_word partmod_word_using det_tendency_the amod_tendency_collective nn_tendency_translation dobj_using_tendency prep_tendency_of pobj_of_words prep_words_in det_class_the amod_class_same pobj_in_class ' +o,1490,'51 The Prague Dependency Tree Bank -LRB- PDT in the sequel -RRB- , which has been inspired by the build-up of the Penn Treebank , is aimed at a complex annotation of -LRB- a part of -RRB- the Czech National Corpus -LRB- CNC in the sequel -RRB- , the creation of which is under progress at the Department of Czech National Corpus at the Faculty of Philosophy , Charles University -LRB- the corpus currently comprises about 100 million tokens of word forms -RRB- ',Marcus,'det_Bank_The nn_Bank_Prague nn_Bank_Dependency nn_Bank_Tree nsubjpass_aimed_Bank abbrev_Bank_PDT prep_PDT_in det_sequel_the pobj_in_sequel nsubjpass_inspired_which aux_inspired_has auxpass_inspired_been rcmod_Bank_inspired prep_inspired_by det_build-up_the pobj_by_build-up prep_build-up_of det__the nn__Penn nn__Treebank pobj_of_ auxpass_aimed_is dep_51_aimed prep_aimed_at det_annotation_a amod_annotation_complex pobj_at_annotation prep_annotation_of det_part_a dep_of_part prep_part_of det_Corpus_the nn_Corpus_Czech nn_Corpus_National pobj_of_Corpus abbrev_annotation_CNC prep_CNC_in det_sequel_the pobj_in_sequel det_creation_the nsubj_is_creation prep_creation_of pobj_of_which rcmod_annotation_is prep_is_under pobj_under_progress prep_progress_at det_Department_the pobj_at_Department prep_progress_of nn_Corpus_Czech nn_Corpus_National pobj_of_Corpus prep_Corpus_at det_Faculty_the pobj_at_Faculty prep_progress_of nn_University_Philosophy appos_University_Charles pobj_of_University det_corpus_the nsubj_comprises_corpus advmod_comprises_currently dep_progress_comprises quantmod_million_about number_million_100 num_tokens_million dobj_comprises_tokens prep_tokens_of nn_forms_word pobj_of_forms ' +o,1491,'The overall POS tag distribution learned by EM is relatively uniform , as noted by , and it tends to assign equal number of tokens to each tag label whereas the real tag distribution is highly skewed ',Johnson,'det_distribution_The amod_distribution_overall dep_distribution_POS nn_distribution_tag nsubj_uniform_distribution partmod_distribution_learned prep_learned_by pobj_by_EM cop_uniform_is advmod_uniform_relatively mark_noted_as advcl_uniform_noted prep_noted_by pobj_by_ cc_uniform_and nsubj_tends_it conj_uniform_tends aux_number_to dep_number_assign amod_number_equal xcomp_tends_number prep_number_of pobj_of_tokens prep_number_to det_label_each nn_label_tag pobj_to_label mark_skewed_whereas det_distribution_the amod_distribution_real nn_distribution_tag nsubj_skewed_distribution cop_skewed_is advmod_skewed_highly advcl_number_skewed ' +o,1492,'1 Introduction Since 1995 , a few statistical parsing algorithms demonstrated a breakthrough in parsing accuracy , as measured against the University of Pennsylvania TREEBANK as a gold standard ',Collins,'num_Introduction_1 prep_demonstrated_Since pobj_Since_1995 det__a amod__few amod__statistical nn__parsing nn__algorithms nsubj_demonstrated_ dep_Introduction_demonstrated det_breakthrough_a dobj_demonstrated_breakthrough prep_demonstrated_in amod_accuracy_parsing pobj_in_accuracy mark_measured_as advcl_demonstrated_measured prep_measured_against det_University_the pobj_against_University prep_University_of nn_TREEBANK_Pennsylvania pobj_of_TREEBANK prep_TREEBANK_as det_standard_a amod_standard_gold pobj_as_standard ' +o,1493,'The results are quite promising : our extraction method discovered 89 \% of the WordNet cousins , and the sense partitions in our lexicon yielded better values than arbitrary sense groupings on the agreement data ',Carletta,'det_results_The nsubj_promising_results aux_promising_are advmod_promising_quite poss_method_our nn_method_extraction nsubj_discovered_method parataxis_promising_discovered num_\%_89 dobj_discovered_\% prep_\%_of det_cousins_the nn_cousins_WordNet pobj_of_cousins cc_promising_and det_partitions_the nn_partitions_sense nsubj_yielded_partitions prep_partitions_in poss_lexicon_our pobj_in_lexicon conj_promising_yielded amod_values_better dobj_yielded_values dep_than_ prep_yielded_than amod_groupings_arbitrary nn_groupings_sense pobj_than_groupings prep_yielded_on det_data_the nn_data_agreement pobj_on_data ' +o,1494,'We use the following features for our rules: sourceand target-conditioned neg-log lexical weights as described in (Koehn et al. , 2003b) neg-log relative frequencies: left-handside-conditioned, target-phrase-conditioned, source-phrase-conditioned Counters: n.o. rule applications, n.o. target words Flags: IsPurelyLexical (i.e. , contains only terminals), IsPurelyAbstract (i.e. , contains only nonterminals), IsXRule (i.e. , non-syntactical span), IsGlueRule 139 Penalties: rareness penalty exp(1 RuleFrequency); unbalancedness penalty |MeanTargetSourceRatio n.o. source words n.o. target words| 4 Parsing Our SynCFG rules are equivalent to a probabilistic context-free grammar and decoding is therefore an application of chart parsing.',Koehn,'' +o,1495,'We examine Structural Correspondence Learning -LRB- SCL -RRB- for this task , and compare it to several variants of Self-training ',McClosky,'nsubj_examine_We nn__Structural nn__Correspondence nn__Learning abbrev__SCL dobj_examine_ prep__for det_task_this pobj_for_task cc_examine_and conj_examine_compare dobj_compare_it prep_compare_to amod_variants_several pobj_to_variants prep_variants_of pobj_of_Self-training ' +o,1496,'We observe that AER is loosely correlated to BLEU -LRB- = 081 -RRB- though the relation is weak , as observed earlier by ',Fraser,'nsubj_observe_We complm_correlated_that nsubj_correlated_AER cop_correlated_is advmod_correlated_loosely ccomp_observe_correlated prep_correlated_to pobj_to_BLEU number_081_= dep_correlated_081 mark_weak_though det_relation_the nsubj_weak_relation cop_weak_is ccomp_correlated_weak advmod_observed_as dep_weak_observed advmod_weak_earlier dep_earlier_by ' +o,1497,'1 Introduction Robust statistical syntactic parsers , made possible by new statistical techniques and by the availability of large , hand-annotated training corpora such as WSJ and Switchboard , have had a major impact on the field of natural language processing ',Dunning,'num_parsers_1 nn_parsers_Introduction nn_parsers_Robust amod_parsers_statistical amod_parsers_syntactic nsubj_had_parsers partmod_parsers_made acomp_made_possible prep_made_by amod__new amod__statistical nn__techniques pobj_by_ cc_by_and conj_by_by det_availability_the pobj_by_availability prep_availability_of amod_corpora_large amod_corpora_hand-annotated nn_corpora_training pobj_of_corpora dep_as_such prep_availability_as nn__WSJ pobj_as_ cc__and nn__Switchboard conj__ aux_had_have det_impact_a amod_impact_major dobj_had_impact prep_had_on det_field_the pobj_on_field prep_field_of amod_processing_natural nn_processing_language pobj_of_processing ' +o,1498,'The data set consisting of 249,994 TFSs was generated by parsing the Figure 3 : The size of Dpi ; for the size of the data set 800 bracketed sentences in the Wall Street Journal corpus -LRB- the first 800 sentences in Wall Street Journal 00 -RRB- in the Penn Treebank with the XHPSG grammar ',Marcus,'det_data_The nsubjpass_generated_data partmod_data_set xcomp_set_consisting prep_consisting_of num_TFSs_249,994 pobj_of_TFSs auxpass_generated_was prep_generated_by pcomp_by_parsing det_Figure_the dobj_parsing_Figure num_Figure_3 det_size_The dep_Figure_size prep_size_of pobj_of_Dpi prep_parsing_for det_size_the pobj_for_size prep_size_of det_data_the pobj_of_data partmod_data_set num_sentences_800 amod_sentences_bracketed dobj_set_sentences prep_sentences_in det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_in_corpus det_sentences_the amod_sentences_first num_sentences_800 dep_corpus_sentences prep_sentences_in nn_Journal_Wall nn_Journal_Street pobj_in_Journal num_Journal_00 prep_set_in det__the nn__Penn nn__Treebank pobj_in_ prep_set_with det__the nn__XHPSG nn__grammar pobj_with_ ' +o,1499,'In this method , the decision list -LRB- DL -RRB- learning algorithm is used ',Yarowsky,'prep_used_In det_method_this pobj_In_method det_list_the nn_list_decision nsubjpass_used_list abbrev_list_DL partmod_list_learning nn__algorithm dobj_learning_ auxpass_used_is ' +o,1500,' dealt with this problem largely by producing an unsupervised learning algorithm that generates probabilistic decision list models of word senses from seed collocates ',Yarowsky,'nsubj_dealt_ prep_dealt_with det_problem_this pobj_with_problem dep_by_largely prep_dealt_by pcomp_by_producing det_algorithm_an amod_algorithm_unsupervised nn_algorithm_learning dobj_producing_algorithm nsubj_generates_that rcmod_algorithm_generates amod_models_probabilistic nn_models_decision nn_models_list dobj_generates_models prep_models_of nn_senses_word pobj_of_senses prep_generates_from nn_collocates_seed pobj_from_collocates ' +o,1501,'and CAUS ate slgmficantly different for unaccusattve and object-dtop verbs, indicating that we need additional featules that have different values across these two classes In Section 2 1, we noted the differing semantic role asmgnments for the verb classes, and hypothesized that these differences would affect the expression of syntactic features that ate countable in a corpus For example, the c ~bs feature approximates sen]antic role reformation b.~ encoding the oxerlap beh~een nouns that can occur m the ~ubject and object positions of a cau~ative xetb Here x~e suggest another feature, that of ammacy of subject, that is intended to distinguish nouns that receive an Agent role flora those that receive a Theme role Recall that objectdrop verbs assign Agent to their subject in both the transitive and intransitive alternations, while unaccusattves assign Agent to their subject only in the transitive, and Theme m the intransitive We expect then that object-drop verbs will occur more often with an animate subject Note again that ~e are 20 II Features [Acc\% SE\% II I VBD ACT INTR CAUS I 63 7 0 6 ] VBD ACT INTR CAUS PRO 70 7 0 4 Table 6 Percentage Accuracy (Acc\%) and Standard Error (SE\%) of C5 0, W~th and W~thout New PRO Feature, All Verb Classes (33 8\% basehne) making use of frequency dmtnbutmns--the clatm ~s not that only Agents can be ammate, but rather that nouns that receive the Agent role will more often be ammate than nouns that receive the Theme role A problem w~th a feature hke ammacy ~s that ~t requires etther manual determmatmn of the antmacy of extracted subjects, or reference to an on-hne resource such as WordNet for determining ammacy To approximate ammacy w~th a feature that can be extracted automatically, and w~thout reference to a resource external to the corpus, we instead count pronouns (other than ~t) m subject positron The assumptmn ~s that the words I, we, you, she, he, and they most often refer to ammate ent~tms The values for the new feature, P~.O, were determined by automatmally extracting all subject/verb tuples including our 59 examples verbs (from the WSJ88 parsed corpus), and computing the ratm of occurrences of pronouns to all subjects We again apply t-tests to our new data to determine whether the sets of PRo values d~ffer across the verb classes Interestingly, we find that the Prto values for unaccusat~ve verbs (the only class to ass~gn Theme role to the sub tect m one of tts alternatmns) are s~gmficantly dtffe~ent from those for both unergatlve and object-drop verbs (p< 05) Moreover, the PRo values for unergat~ve and object-drop verbs (whose subjects are Agents m bo~h alternatmns) are not s~gmficantly d~fferent Th~s pattern confirms the abd~ty of the feature to capture the thematm d~stmctmn between unaccusat~ve verbs and the other two classes Table 6 shows the result of applying C5 0 (10-fold eross-vahdatmn repeated 50 t~mes) to the three-x~ay classfficatmn task using the PRo feature m conjunctmn w~th the four previous features ~.ccuracy ranproves to over 70\%, a teductmn m the error rate of almost 20\% due to th~s single nex~ feature Moteover, classifying the unaccusat~ve an2 object-drop verbs using the new feature m conjunctmn w~th the prevmus four leads to accuracy of over 68\% (compared to 58\% w~thout PRo) We conclude that this feature ~s ~mportant in d~stmgmshlng unaccusat~ve and object-drop verbs, and hkely contributes to the tmprovement m the three-way classtficatton because of th~s Future work wdl examine the performance w~thm the verb classes of th~s new set of features to see whether accuracy has also tmproved for unergatire verbs 5 Conclusions In thin paper, we have presented an m-depth case study, m whmh we investigate varmus machine learnmg techmques to automatically classify a set of verbs, based on dlstnbutmnal features extracted from a very large corpus Results show that a small number of hngmstlcally motivated grammatical features are sufficmnt to reduce the error rate by mote than 50\% over chance, acluevmg a 70\% acctuacy rate m a three-way classfficatmn task Tins leads us to conclude that corpus data is a usable repository of verb class mformatmn On one hand ~e observe that semantlc propemes of verb classes (such as causatlvlty, or ammacy of subject) may be usefully approximated through countable syntactic features Even with some noise, lexmal propertms are reflected m the corpus robustly enough to positively contribute m classlficatmn On the other hand, however, we remark that deep hngumtm analysis cannot be ehmmated--m our approach, it is embedded m the selection of the features to count We also think that using hngumtlcally motivated features makes the approach very effective and easdy scalable we report a 56\% reductmn m error rate, w~th only five features that are relatwely straightforward to count Acknowledgements This research was partly sponsored by the S~ lss Natmnal Scmnce Foundatmn, under fello~slup 821046569 to Paola Merlo, by the US Natmnal Scmnce Foundatmn, under grants #9702331 and #9818322 to $uzanne Stevenson, and by the Infotmatton Sciences Councd of Rutgers Umverslty ~,~,e thank Martha Palmer for getting us started on tlus ~ork and Mmhael Colhns for gwmg us access to the output of his parser We gratefully acknowledge the help of Ixlva Dickinson, ~ho calculated no~mahzatmns of the corpus data Appendix A The une~gatx~es are manner of morton ~erbs jumptd rushed, malched, leaped floated, laced, huslwd uandered, vaulted, paraded, galloped, gl,ded, hzked hopped jogged, scooted, ncurlzed, ~kzpped, hptoed, trotted The unaccusau~es are verbs of change of state opened, exploded, flooded, dzs~olved, cracked, hardened bozled, melted,.fractured,,ol,dzfied, collapsed cooled folded, w~dened, changed, clealed, dzwded, ~,mmered stabdzzed The object-dlop verbs are unspecffied object altelnatron verbs played, painted, k,cked, carved, reaped, washed, danced, yelled, typed, kmtted bolrowed mhet21 tted, organtzed, rented, sketched, cleaned, packed, studted, swallowed, called References Thomas G Bever 1970 The cogmtwe basis for hngmstlc structure In J R Hayes, e&tor, Cognttson and the Development of Language John Wdey, New York Michael Brent 1993 From grammar to le~con Unsupervmed learmng of [ex~cal syntax Computational Linguistics, 19(2) 243-262 Edward Bnscoe and Ann Copestake 1995 Lex~cal rules m the TDFS framework Techmcal report, AcquflexI I Working\' Papers Anne-Marm Brousseau and Ehzabeth R~tter 1991 A non-umfied analysis of agent~ve verbs In West Coast Conference on Formal Lmgutstzcs, number 20, pages 53-64 M~chael John Colhns 1997 Three generaUve, lexacahsed models for statistical parsmg In Proc of the ~5th Annual Meeting of the ACL, pages 16-23 Hoa Trang Dang, Kann K~pper, Martha Palmer, and Joseph Rosenzwe~g 1998 Investtgatmg regular sense extenmons based on mteresecttve Levm classes In Proc of the 361h Annual Meeting of the ACL and the 171h [nternatwnal Conference on Computatwnal L,ngu,st,cs (COLING-A CL \'98), pages 293-299, Montreal, Canada Umvers~t6 de Montreal Bonme Dorr and Doug Jones 1996 Role of word sense d~samb~guatmn m lexacal acqms~tmn Predmtmg semantics from syntactic cues In Proc of the 161h Internattonal Conference on Computat*onal Lmgutsttcs, pages 322-327, Copenhagen, Denmark COLING Bonnie Dorr 1997 Large-scale chctmnary constructmn for foreign language tutonng and mterhngual machine translatmn Machine Translatton, 12 1-55 Hana Fd~p M~chael Tanenhaus, Greg Carlson, Paul AIlopenna, and Joshua Blatt 1999 Reduced relatives judged hard require constraint-based analyses In P Merlo and S Stevenson, echtors, Sentence Processmg and the Lextcon Formal, Computational, and Ezpertmental Perspectives, John Benjamms, Holland Ken Hale and Jay Keyser 1993 On argument structure and the lexacal representatmn of s:~ ntact~c relatmns In K Hale and J Keyser, editors, The t\',ew from Budding ~0, pages 53-110 MIT Press Juchth L Ixlavans and Martin Chodorow 1992 Degrees of stat~vlty The lexacal representatmn of verb aspect In Proceedmg~ of the Fourteenth International Conference on Computahonal Lmgmst,cs Juchth Ixlavans and Mm-Yen Kan 1998 Role of ~erbs m document analysis In Proc of the 361h Annual Meeting of the ACL and the 171h [nternatzonal Conference on Computational Lmgutsttcs ( C O L L\'v G4 C L \'98), pages 680-686, Montreal, Canada Umvers~te de Montreal Beth Levm and/Vlalka Rappapti(t\'Hovav 1995 (Jnaccusatwlty MIT Press, Cambridge, MA Beth Le~m 1993 Enghsh Verb Clas~e~ and 4lternatwns Chacago Umvers~ty Press, Chicago, IL Maryellen C MacDonald 1994 Probablhstlc constramts and syntactic amblgtuty resolution Language and Cognltzve Processes, 9(2) 157-201 Paola Merlo and Suzanne Stevenson 1998 What grammars tell us about corpora the case of reduced relative clauses In P1oceedmgs of the Slzth Workshop on Very Large Corpora, pages 134-142, Montreal, CA George Miller, R Beckw~th, C Fellbaum, D Gross, and Ix I~hller 1990 Fwe papers on Wordnet Techmcal report, Cogmtzve Scmnce Lab, Princeton Ual~erstt~ Martha Palmer 1999 Coasmtent criteria for sense distmctmns Computmg ]or the Hamamttes Fernando Perelra, Naftah Tlshby, and Ldhan Lee 1993 Dlstrabutmnal clustering of enghsh words [n Proc of the 31th 4nnual Meeting of the 4CL, pages 183-190 Fernando Perexra, Ido Dagan, and Lalhan Lee 1997 Slmdanty-based methods for word sense dlsamblguatmn In Proc of the 35th Annual Meeting of the 4 CL and the 8th Conf of the E 4 CL (A CL/EA CL \'97) pages 56 -63 Geoffrey K Pullum 1996 Learnabthty, hyperlearnrag, and the poverty of the sttmulus In Jan Johnson, Matthew L Jute, and Jen L Moxley, editors, ~nd Annual Meeting of the Berkeley Lmgutstzcs Soctety General Sesston and Parasesswn on the Role of Learnabdzty m Grammatzcal Theory, pages 498-513, Berkeley, Cahforma Berkeley Linguistics Socmty James Pustejovsky 1995 The Generatwe Lexicon MIT Press J Ross Qumlan 1992 C$ 5 Programs fo~ Machine Learning Series m Machme Learning Morgan Ixaufmann, San Mateo, C 4.',Collins,'' +o,1502,'Algorithm 1 SCL 1 : Select m pivot features ',Blitzer,'nn_SCL_Algorithm num_SCL_1 amod_SCL_ tmod__1 nn_features_Select nn_features_m nn_features_pivot dep_SCL_features ' +o,1503,' applies this approach to the so-called IBM Candide system to build context dependent models , compute automatic sentence splitting and to improve word reordering in translation ',Berger,'advmod_applies_ det_approach_this dobj_applies_approach prep_applies_to det_system_the amod_system_so-called nn_system_IBM nn_system_Candide pobj_to_system aux_build_to xcomp_applies_build dobj_build_context amod_models_dependent nsubj_applies_models amod_splitting_compute amod_splitting_automatic nn_splitting_sentence appos_models_splitting cc_splitting_and aux_improve_to conj_splitting_improve nn_reordering_word dobj_improve_reordering prep_improve_in pobj_in_translation ' +o,1504,'The techniques examined are Structural Correspondence Learning -LRB- SCL -RRB- and Self-training ',Blitzer,'det_techniques_The nsubj_Structural_techniques partmod_techniques_examined aux_Structural_are nn__Correspondence nn__Learning abbrev__SCL dobj_Structural_ cc__and nn__Self-training conj__ ' +o,1505,'The perceptron has been used in many NLP tasks , such as POS tagging , Chinese word segmentation and so on ',Collins,'det_perceptron_The nsubjpass_used_perceptron aux_used_has auxpass_used_been prep_used_in amod_tasks_many nn_tasks_NLP pobj_in_tasks dep_as_such prep_tasks_as pobj_as_POS partmod_POS_tagging amod__ amod__Chinese nn__word nn__segmentation dobj_tagging_ cc_so_and advmod_tagging_so dep_so_on ' +o,1506,'However, in order to cope with the prediction errors of the classi er, we approximate a74a51a18a77a76 a28 with an a80 -gram language model on sequences of the re ned tag labels: a38a58a39 a41 a81 a43a82a44a47a46a83a48a47a50a75a44a15a52 a53a9a54a49a84 a53a9a54a83a84a49a85a9a86a13a87a89a88a91a90 a55a57a56 a38a40a39 a81 a59a60a42a61 (2) a92 a44a47a46a83a48a47a50a75a44a15a52 a53a9a54 a84 a53a9a54a83a84a49a85a9a86a13a87a89a88a91a90 a93 a94a96a95 a55a57a56a98a97a66a99 a95 a59a100a27a61 (3) In order to estimate the conditional distribution a101 a18a20a19a15a21 a1 a68 a72 a28 we use the general technique of choosing the maximum entropy (maxent) distribution that properly estimates the average of each feature over the training data (Berger et al. , 1996).',Berger,'' +o,1507,'Presently , many systems , , , focus on online recognition of proper nouns , and have achieved inspiring results in newscorpus but will be deteriorated in special text , such as spoken corpus , novels ',Dunning,'advmod__Presently amod_systems_many nsubjpass__systems amod_focus_ amod_focus_ amod_focus_ conj__focus prep_focus_on amod_recognition_online pobj_on_recognition prep_recognition_of amod_nouns_proper pobj_of_nouns cc__and aux_achieved_have conj__achieved amod_results_inspiring dobj_achieved_results prep_achieved_in pobj_in_newscorpus cc_achieved_but aux_deteriorated_will auxpass_deteriorated_be conj_achieved_deteriorated prep_deteriorated_in amod_text_special pobj_in_text dep_as_such prep_text_as amod_corpus_spoken pobj_as_corpus appos_corpus_novels ' +o,1508,'For MCE learning , we selected the reference compression that maximize the BLEU score -LRB- = argmax rR BLEU -LRB- r , R r -RRB- -RRB- from the set of reference compressions and used it as correct data for training ',Papineni,'prep_selected_For nsubj_learning_MCE pcomp_For_learning nsubj_selected_we det_compression_the nn_compression_reference dobj_selected_compression complm_maximize_that ccomp_selected_maximize det__the amod__BLEU nn__score dobj_maximize_ amod_BLEU_= nn_BLEU_argmax nn_BLEU_rR dep__BLEU dep__r nn_r_R appos_r_r prep__from det_set_the pobj_from_set prep_set_of nn_compressions_reference pobj_of_compressions cc_maximize_and conj_maximize_used dobj_used_it prep_used_as amod_data_correct pobj_as_data prep_used_for pobj_for_training ' +o,1509,'In addition to adapting the idea of Head Word Chains , we also compared the input sentences argument structures against the treebank for certain syntactic categories ',Liu,'prep_compared_In pobj_In_addition prep_addition_to pcomp_to_adapting det_idea_the dobj_adapting_idea prep_idea_of nn__Head nn__Word nn__Chains pobj_of_ nsubj_compared_we advmod_compared_also det_sentences_the nn_sentences_input nsubj_structures_sentences nn_structures_argument xcomp_compared_structures prep_structures_against det_treebank_the pobj_against_treebank prep_treebank_for amod_categories_certain amod_categories_syntactic pobj_for_categories ' +o,1510,'312 Kappa Kappa is an evaluation measure which is increasingly used in NLP annotation work ',Carletta,'nn__312 nn__Kappa nn__Kappa nsubj_measure_ cop_measure_is det_measure_an nn_measure_evaluation nsubjpass_used_which auxpass_used_is advmod_used_increasingly rcmod_measure_used prep_used_in nn_work_NLP nn_work_annotation pobj_in_work ' +o,1511,'1 Introduction Text-to-text generation is an emerging area of research in NLP ',Jing,'num_generation_1 nn_generation_Introduction nn_generation_Text-to-text nsubj_area_generation cop_area_is det_area_an amod_area_emerging prep_area_of pobj_of_research prep_research_in pobj_in_NLP ' +o,1512,'Having a single , canonical tree structure for each possible alignment can help when flattening binary trees , as it indicates arbitrary binarization decisions ',Wu,'det_structure_a amod_structure_single amod_structure_canonical nn_structure_tree dobj_Having_structure prep_structure_for det_alignment_each amod_alignment_possible pobj_for_alignment aux_help_can dep_Having_help advmod_flattening_when advcl_help_flattening advmod_flattening_binary nsubj_Having_trees mark_indicates_as nsubj_indicates_it dep_trees_indicates amod_decisions_arbitrary dep_arbitrary_binarization dobj_indicates_decisions ' +o,1513,'We use GIZA + + to train generative directed alignment models : HMM and IBM Model4 from training record-text pairs ',Brown,'nsubj_use_We nsubj_train_GIZA dep_train_+ cc_+_+ conj_+_ aux_train_to xcomp_use_train dobj_train_generative partmod_generative_directed amod_models_alignment dobj_directed_models nn_Model4_HMM cc_HMM_and conj_HMM_IBM nsubj__Model4 parataxis_use_ prep__from pcomp_from_training amod_pairs_record-text dobj_training_pairs ' +o,1514,' introduced a transformationbased learning method which considered chunking as a kind of tagging problem ',Ramshaw,'nsubj_introduced_ det_method_a amod_method_transformationbased nn_method_learning dobj_introduced_method nsubj_considered_which rcmod_method_considered xcomp_considered_chunking prep_chunking_as det_kind_a pobj_as_kind prep_kind_of amod_problem_tagging pobj_of_problem ' +o,1515,'The Penn Treebank annotation was chosen to be the first among equals : it is the starting point for the merger and data from other annotations are attached at tree nodes ',Marcus,'det__The nn__Penn nn__Treebank nn__annotation nsubjpass_chosen_ auxpass_chosen_was aux_first_to cop_first_be det_first_the xcomp_chosen_first prep_first_among pobj_among_equals nsubj_point_it cop_point_is det_point_the amod_point_starting parataxis_chosen_point prep_point_for det_merger_the pobj_for_merger cc_chosen_and nsubjpass_attached_data prep_data_from amod_annotations_other pobj_from_annotations auxpass_attached_are conj_chosen_attached prep_attached_at nn_nodes_tree pobj_at_nodes ' +o,1516,'We can then use this newly identified set to : -LRB- 1 -RRB- use s method to find the orientation for the terms and employ the terms and their scores in a classifier , and -LRB- 2 -RRB- use s method to find the orientation for the terms and add the new terms as additional seed terms for a second iteration As opposed to , we do not use the web as a resource to find associations , rather we apply the method directly to in-domain data ',Turney,'nsubj_use_We aux_use_can advmod_use_then det_set_this advmod_identified_newly amod_set_identified dobj_use_set prep_use_to dep_use_1 dep_use_use amod_method_ nn_method_s dobj_use_method aux_find_to xcomp_use_find det_orientation_the dobj_find_orientation prep_orientation_for det_terms_the pobj_for_terms cc_find_and conj_find_employ det_terms_the dobj_employ_terms cc_employ_and poss_scores_their conj_employ_scores prep_scores_in det_classifier_a pobj_in_classifier cc_employ_and dep_use_2 conj_employ_use amod_method_ nn_method_s dobj_use_method aux_find_to xcomp_use_find det_orientation_the dobj_find_orientation prep_orientation_for det_terms_the pobj_for_terms cc_find_and conj_find_add det_terms_the amod_terms_new dobj_add_terms prep_add_as amod_terms_additional nn_terms_seed pobj_as_terms prep_terms_for det_iteration_a amod_iteration_second pobj_for_iteration mark_opposed_As advcl_use_opposed prep_opposed_to number__ pobj_to_ nsubj_use_we aux_use_do neg_use_not dep_use_use det_web_the dobj_use_web prep_use_as det_resource_a pobj_as_resource aux_find_to infmod_resource_find dobj_find_associations advmod_apply_rather nsubj_apply_we dep_use_apply det_method_the dobj_apply_method advmod_apply_directly prep_apply_to amod_data_in-domain pobj_to_data ' +o,1517,'The second uses Lin dependency similarity , a syntacticdependency based distributional word similarity resource described in 9 ',Lin,'det_second_The nsubj_uses_second nn_similarity_Lin nn_similarity_dependency dobj_uses_similarity det_resource_a advmod_based_syntacticdependency amod_resource_based amod_resource_distributional nn_resource_word nn_resource_similarity appos_similarity_resource partmod_resource_described prep_described_in pobj_in_ num__9 ' +o,1518,'A monotonous segmentation copes with monotonous alignments , that is , j -LRB- k aj -LRB- ak following the notation of ',Brown,'det_copes_A amod_copes_monotonous nn_copes_segmentation nsubj_alignments_copes prep_copes_with pobj_with_monotonous dep_alignments_that dep_that_is ccomp_alignments_j dobj_j_k dep_j_aj dobj_aj_ak xcomp_aj_following det_notation_the dobj_following_notation prep_notation_of ' +o,1519,'We also can not use prior graph construction methods for the document level -LRB- such as physical proximity of sentences , used in -RRB- at the word sense level ',Pang,'nsubj_use_We advmod_use_also aux_use_can neg_use_not amod_methods_prior nn_methods_graph nn_methods_construction dobj_use_methods prep_methods_for det_level_the nn_level_document pobj_for_level dep_as_such dep_methods_as amod_proximity_physical pobj_as_proximity prep_proximity_of pobj_of_sentences partmod_sentences_used prep_used_in pobj_in_ prep_methods_at det_level_the nn_level_word nn_level_sense pobj_at_level ' +o,1520,'The model scaling factors are optimized on the development corpus with respect to mWER similar to ',Och,'det_factors_The amod_factors_model amod_factors_scaling nsubjpass_optimized_factors auxpass_optimized_are prep_optimized_on det_corpus_the nn_corpus_development pobj_on_corpus prep_corpus_with pobj_with_respect aux_mWER_to xcomp_optimized_mWER acomp_mWER_similar prep_similar_to ' +o,1521,'For these classications , we calculated a kappa statistic of 0528 ',Carletta,'prep_calculated_For det_classications_these pobj_For_classications nsubj_calculated_we det_statistic_a nn_statistic_kappa dobj_calculated_statistic prep_statistic_of pobj_of_0528 ' +o,1522,'Feature weights of both systems are tuned on the same data set3 For Pharaoh , we use the standard minimum error-rate training ; and for our system , since there are only two independent features -LRB- as we always fix = 1 -RRB- , we use a simple grid-based line-optimization along the language-model weight axis ',Och,'amod_weights_Feature nsubjpass_tuned_weights prep_weights_of det_systems_both pobj_of_systems auxpass_tuned_are prep_tuned_on det_set3_the amod_set3_same nn_set3_data pobj_on_set3 prep_use_For pobj_For_Pharaoh nsubj_use_we dep_tuned_use det_training_the amod_training_standard amod_training_minimum amod_training_error-rate dobj_use_training advmod_use_ cc_use_and prep_use_for poss_system_our pobj_for_system mark_are_since expl_are_there dep_use_are advmod_are_only num_features_two amod_features_independent nsubj_are_features mark_fix_as nsubj_fix_we advmod_fix_always dep_features_fix dep_1_= ccomp_fix_1 nsubj_use_we conj_use_use det_line-optimization_a amod_line-optimization_simple amod_line-optimization_grid-based dobj_use_line-optimization prep_line-optimization_along det_axis_the amod_axis_language-model nn_axis_weight pobj_along_axis ' +o,1523,'In most statistical machine translation -LRB- SMT -RRB- models , some of measure words can be generated without modification or additional processing ',Koehn,'prep_generated_In advmod_statistical_most amod__statistical nn__machine nn__translation abbrev__SMT nn__models pobj_In_ nsubjpass_generated_some prep_some_of nn_words_measure pobj_of_words aux_generated_can auxpass_generated_be prep_generated_without pobj_without_modification cc_modification_or amod_processing_additional conj_modification_processing ' +o,1524,'The resulting corpus contains 385 documents of American English selected from the Penn Treebank , annotated in the framework of Rhetorical Structure Theory ',Marcus,'det_corpus_The amod_corpus_resulting nsubj_contains_corpus amod_documents_385 dobj_contains_documents prep_documents_of nn_English_American pobj_of_English partmod_English_selected prep_selected_from det__the nn__Penn nn__Treebank pobj_from_ partmod__annotated prep_annotated_in det_framework_the pobj_in_framework prep_framework_of nn_Theory_Rhetorical nn_Theory_Structure pobj_of_Theory ' +o,1525,'We solve SAT analogies with a simplified version of the method of ',Turney,'nsubj_solve_We nn_analogies_SAT dobj_solve_analogies prep_solve_with det_version_a amod_version_simplified pobj_with_version prep_version_of det_method_the pobj_of_method prep_method_of ' +o,1526,'Also , the aspect of generalizing features across different products is closely related to fully supervised domain adaptation , and we plan to combine our approach with the idea from Daume III -LRB- 2007 -RRB- to gain insights into whether the composite back-off features exhibit different behavior in domain-general versus domain-specific feature sub-spaces ','Daume III','advmod_related_Also det_aspect_the nsubjpass_related_aspect prep_aspect_of amod_features_generalizing pobj_of_features prep_features_across amod_products_different pobj_across_products auxpass_related_is advmod_related_closely prep_related_to advmod__fully amod__supervised nn__domain nn__adaptation pobj_to_ cc_related_and nsubj_plan_we conj_related_plan aux_combine_to xcomp_plan_combine poss_approach_our dobj_combine_approach prep_combine_with det_idea_the pobj_with_idea prep_combine_from nn_III_Daume pobj_from_III appos_III_2007 aux_gain_to xcomp_combine_gain dobj_gain_insights prep_gain_into complm_exhibit_whether det_features_the amod_features_composite amod_features_back-off nsubj_exhibit_features pcomp_into_exhibit amod_behavior_different dobj_exhibit_behavior prep_exhibit_in pobj_in_domain-general prep_domain-general_versus amod_sub-spaces_domain-specific nn_sub-spaces_feature pobj_versus_sub-spaces ' +o,1527,'1 Introduction The task of sentence compression -LRB- or sentence reduction -RRB- can be defined as summarizing a single sentence by removing information from it ',Jing,'num_Introduction_1 det_task_The nsubjpass_defined_task prep_task_of nn_compression_sentence pobj_of_compression dep_sentence_or dep_compression_sentence dep_sentence_reduction aux_defined_can auxpass_defined_be dep_Introduction_defined prep_defined_as pcomp_as_summarizing det_sentence_a amod_sentence_single dobj_summarizing_sentence prep_summarizing_by pcomp_by_removing dobj_removing_information prep_removing_from pobj_from_it ' +p,1528,'All the enumerated segment pairs are listed in the following table : Feature x , y Feature x , y AM1 +1 c1 , c0 AM2 +1 c2c1 , c0 AM1 +2 c1 , c0c1 AM2 +2 c2c1 , c0c1 AM1 +3 c1 , c0c1c2 AM3 +1 c3c2c1 , c0 We use Dunnings method because it does not depend on the assumption of normality and it allows comparisons to be made between the signiflcance of the occurrences of both rare and common phenomenon ',Dunning,'predet_pairs_All det_pairs_the amod_pairs_enumerated nn_pairs_segment nsubjpass_listed_pairs auxpass_listed_are prep_listed_in det_table_the amod_table_following pobj_in_table nn_x_Feature nsubj_c0_x nn_x_y nn_x_Feature appos_x_x nn_AM1_y appos_x_AM1 num_c1_+1 dep_AM1_c1 amod_AM2_c0 appos_x_AM2 num_c2c1_+1 dep_AM2_c2c1 amod_AM1_c0 appos_x_AM1 num_c2c1_+2 amod_c2c1_c1 amod_c2c1_c0c1 nn_c2c1_AM2 num_c2c1_+2 dep_AM1_c2c1 amod_AM1_c0c1 appos_x_AM1 num_c3c2c1_+3 amod_c3c2c1_c1 amod_c3c2c1_c0c1c2 nn_c3c2c1_AM3 num_c3c2c1_+1 dep_AM1_c3c2c1 parataxis_listed_c0 nsubj_use_We ccomp_c0_use nn__Dunnings nn__method dobj_use_ mark_depend_because nsubj_depend_it aux_depend_does neg_depend_not advcl_use_depend prep_depend_on det_assumption_the pobj_on_assumption prep_assumption_of pobj_of_normality cc_listed_and nsubj_allows_it conj_listed_allows nsubjpass_made_comparisons aux_made_to auxpass_made_be xcomp_allows_made prep_made_between det_signiflcance_the pobj_between_signiflcance prep_signiflcance_of det_occurrences_the pobj_of_occurrences prep_occurrences_of det_phenomenon_both amod_phenomenon_rare cc_rare_and conj_rare_common pobj_of_phenomenon ' +o,1529,'Because of these kinds of results , the vast majority of statistical parsing work has focused on parsing as a supervised learning problem ',Collins,'prep_focused_Because dep_Because_of det_kinds_these pobj_of_kinds prep_kinds_of pobj_of_results det_majority_the amod_majority_vast nsubj_focused_majority prep_majority_of amod_work_statistical nn_work_parsing pobj_of_work aux_focused_has prep_focused_on pcomp_on_parsing prep_parsing_as det_problem_a amod_problem_supervised nn_problem_learning pobj_as_problem ' +o,1530,'5 The SemCor collection is a subset of the Brown Corpus and consists of 352 news articles distributed into three sets in which the nouns , verbs , adverbs , and adjectives have been manually tagged with their corresponding WordNet senses and part-of-speech tags using Brills tagger ',Marcus,'det__The nn__SemCor nn__collection nsubj_subset_ cop_subset_is det_subset_a rcmod_5_subset prep_subset_of det_Corpus_the nn_Corpus_Brown pobj_of_Corpus cc_subset_and conj_subset_consists prep_5_of amod_articles_352 nn_articles_news pobj_of_articles partmod_articles_distributed prep_distributed_into num_sets_three pobj_into_sets rel_tagged_in pobj_in_which det_nouns_the nsubjpass_tagged_nouns conj_nouns_verbs conj_nouns_adverbs cc_nouns_and conj_nouns_adjectives aux_tagged_have auxpass_tagged_been advmod_tagged_manually rcmod_sets_tagged prep_tagged_with poss_senses_their amod_senses_corresponding nn_senses_WordNet pobj_with_senses cc_senses_and amod_tags_part-of-speech conj_senses_tags xcomp_tagged_using nn__Brills nn__tagger dobj_using_ ' +o,1531,'There are also approaches to anaphora resolution using unsupervised methods to extract useful information , such as gender and number , or contextual role-knowledge ',Bean,'expl_are_There advmod_are_also nsubj_are_approaches prep_approaches_to nn_resolution_anaphora pobj_to_resolution partmod_resolution_using amod_methods_unsupervised dobj_using_methods aux_extract_to xcomp_using_extract amod_information_useful dobj_extract_information dep_as_such prep_approaches_as nn__gender cc_gender_and conj_gender_number pobj_as_ cc_approaches_or amod_role-knowledge_contextual conj_approaches_role-knowledge ' +o,1532,' discuss the influence of bias towards highor low-frequency items for different tasks -LRB- correlation with WordNet-derived neighbor sets and pseudoword disambiguation -RRB- , and it would not be surprising if the different high-frequency bias were leading to different results ',Weeds,'nsubj_discuss_ det_influence_the dobj_discuss_influence prep_influence_of pobj_of_bias prep_bias_towards amod_items_highor amod_items_low-frequency pobj_towards_items prep_items_for amod_tasks_different pobj_for_tasks dep_bias_correlation prep_correlation_with amod_sets_WordNet-derived nn_sets_neighbor pobj_with_sets cc_sets_and amod_disambiguation_pseudoword conj_sets_disambiguation cc_discuss_and nsubj_surprising_it aux_surprising_would neg_surprising_not cop_surprising_be conj_discuss_surprising mark_leading_if det_bias_the amod_bias_different amod_bias_high-frequency nsubj_leading_bias aux_leading_were advcl_surprising_leading prep_leading_to amod_results_different pobj_to_results ' +o,1533,'Some of these have been previously employed for various tasks by Gabrilovich and Markovitch , ; Overell and Ruger , , and Suchanek et al ',Cucerzan,'nsubjpass_employed_Some prep_Some_of pobj_of_these aux_employed_have auxpass_employed_been advmod_employed_previously prep_employed_for amod_tasks_various pobj_for_tasks prep_employed_by pobj_by_Gabrilovich cc_Gabrilovich_and conj_Gabrilovich_Markovitch nn__ conj__Overell cc__and conj__Ruger conj_Gabrilovich_ conj_Gabrilovich_ cc_Gabrilovich_and conj_Gabrilovich_Suchanek cc_Suchanek_et conj_Suchanek_al ' +o,1534,'Model 4 of is also a first-order alignment model -LRB- along the source positions -RRB- like the HMM , trot includes also fertilities ',Brown,'nsubj_model_Model num_Model_4 prep_Model_of pobj_of_ cop_model_is advmod_model_also det_model_a amod_model_first-order nn_model_alignment prep_model_along det_positions_the nn_positions_source pobj_along_positions prep_includes_like det_HMM_the pobj_like_HMM nsubj_includes_trot dep_model_includes advmod_includes_also dobj_includes_fertilities ' +o,1535,', ie : -LRB- ll -RRB- Lj = ~ maz -LRB- zi -LRB- j , u -RRB- -RRB- i = I where xi -LRB- j , u -RRB- E Qi and max -LRB- xi -LRB- j , u -RRB- -RRB- is the highest score in the line of the matrix Qi which corresponds to the head word sense j n is the number of modifiers of the head word h at the current tree level , and k i Lj = j ~ l Lj where k is the number of senses of the head word h The reason why gj -LRB- I0 -RRB- is calculated as a sum of the best scores -LRB- ll -RRB- , rather than by using the traditional maximum likelihood estimate -LRB- Gah eta -LRB- ',Berger,'dep_,_ie dep_~_ll nn_=_Lj nsubj_~_= dobj_~_maz dep_maz_zi dep_maz_j dep_j_u partmod_maz_i dep_I_= ccomp_i_I advmod_number_where nsubj_number_xi dep_xi_j dep_j_u nn_Qi_E nsubj_score_Qi cc_Qi_and conj_Qi_max dep_max_xi dep_max_j dep_j_u cop_score_is det_score_the amod_score_highest rcmod_xi_score prep_score_in det_line_the pobj_in_line prep_line_of det_Qi_the nn_Qi_matrix pobj_of_Qi nsubj_corresponds_which rcmod_score_corresponds prep_corresponds_to det_n_the nn_n_head nn_n_word nn_n_sense nn_n_j pobj_to_n cop_number_is det_number_the advcl_i_number prep_number_of pobj_of_modifiers prep_modifiers_of det_h_the nn_h_head nn_h_word pobj_of_h prep_h_at det_level_the amod_level_current nn_level_tree pobj_at_level cc_maz_and nn_Lj_k nn_Lj_i nn_Lj_Lj nn_Lj_= nn_Lj_j nn_Lj_~ nn_Lj_l conj_maz_Lj advmod_number_where nsubj_number_k cop_number_is det_number_the rcmod_Lj_number prep_number_of pobj_of_senses prep_senses_of det_h_the nn_h_head nn_h_word pobj_of_h det_reason_The nsubj_estimate_reason advmod_calculated_why nsubjpass_calculated_gj appos_gj_I0 auxpass_calculated_is rcmod_reason_calculated prep_calculated_as det_sum_a pobj_as_sum prep_sum_of det_scores_the amod_scores_best pobj_of_scores appos_scores_ll cc_as_rather dep_rather_than conj_as_by pcomp_by_using det_likelihood_the amod_likelihood_traditional nn_likelihood_maximum dobj_using_likelihood rcmod_number_estimate nn_Gah_ dobj_estimate_Gah partmod_Gah_eta ' +o,1536,'SIGHAN , the Special Interest Group for Chinese Language Processing of the Association for Computational Linguistics , conducted three prior word segmentation bakeoffs , in 2003 , 2005 and 2006 , which established benchmarks for word segmentation and named entity recognition ',Levow,'nsubj_conducted_SIGHAN det_Group_the amod_Group_Special nn_Group_Interest appos_SIGHAN_Group prep_Group_for nn_Processing_Chinese nn_Processing_Language pobj_for_Processing prep_Processing_of det_Association_the pobj_of_Association prep_Association_for nn_Linguistics_Computational pobj_for_Linguistics num_bakeoffs_three amod_bakeoffs_prior nn_bakeoffs_word nn_bakeoffs_segmentation dobj_conducted_bakeoffs dep__in pobj_in_2003 num_2003_2005 cc_2003_and conj_2003_2006 rcmod_bakeoffs_ nsubj_established_which rcmod_bakeoffs_established dobj_established_benchmarks prep_established_for nn_segmentation_word pobj_for_segmentation cc_established_and conj_established_named nn_recognition_entity dobj_named_recognition ' +o,1537,'We show translation results in terms of the automatic BLEU evaluation metric on the MT03 Arabic-English DARPA evaluation test set consisting of a212a89a212a89a87 sentences with a98a89a212a161a213a89a214a89a215 Arabic words with a95 reference translations ',Papineni,'nsubj_show_We nn_results_translation iobj_show_results prep_results_in pobj_in_terms prep_terms_of det_evaluation_the amod_evaluation_automatic amod_evaluation_BLEU pobj_of_evaluation amod__metric dobj_show_ prep_show_on det_test_the nn_test_MT03 nn_test_Arabic-English nn_test_DARPA nn_test_evaluation pobj_on_test partmod_test_set xcomp_set_consisting prep_consisting_of amod_sentences_a212a89a212a89a87 pobj_of_sentences prep_sentences_with amod_words_a98a89a212a161a213a89a214a89a215 nn_words_Arabic pobj_with_words prep_consisting_with amod_translations_a95 nn_translations_reference pobj_with_translations ' +p,1538,'3 Extending Bleu and Ter with Flexible Matching Many widely used metrics like Bleu and Ter are based on measuring string level similarity between the reference translation and translation hypothesis , just like Meteor Most of them , however , depend on finding exact matches between the words in two strings ',Papineni,'dep_based_3 nn_Bleu_Extending nsubjpass_based_Bleu cc_Bleu_and amod_metrics_Ter prep_Ter_with nn_Matching_Flexible pobj_with_Matching amod_metrics_Many advmod_used_widely amod_metrics_used conj_Bleu_metrics prep_metrics_like nn__Bleu pobj_like_ cc__and nn__Ter conj__ auxpass_based_are prep_based_on pcomp_on_measuring nn_similarity_string nn_similarity_level dobj_measuring_similarity prep_measuring_between det_translation_the nn_translation_reference pobj_between_translation cc_translation_and nn_hypothesis_translation conj_translation_hypothesis advmod_measuring_just prep_depend_like nn_Most_Meteor pobj_like_Most prep_Most_of pobj_of_them advmod_depend_however dep_based_depend prep_depend_on pcomp_on_finding amod_matches_exact dobj_finding_matches prep_matches_between det_words_the pobj_between_words prep_words_in num_strings_two pobj_in_strings ' +o,1539,'Wu and Weld and calculate the overlap between contexts of named entities and candidate articles from Wikipedia , using overlap ratios or similarity scores in a vector space model , respectively ',Cucerzan,'nn__Wu cc_Wu_and conj_Wu_Weld nsubj_calculate_ cc__and nn__ conj__ det_overlap_the dobj_calculate_overlap prep_overlap_between pobj_between_contexts prep_contexts_of amod_entities_named pobj_of_entities cc_entities_and nn_articles_candidate conj_entities_articles prep_calculate_from pobj_from_Wikipedia xcomp_calculate_using amod_ratios_overlap dobj_using_ratios cc_ratios_or nn_scores_similarity conj_ratios_scores prep_using_in det_model_a nn_model_vector nn_model_space pobj_in_model advmod_using_respectively ' +o,1540,'In our approach , equation -LRB- 1 -RRB- is further normalized so that the probability for different lengths of F is comparable at the word level : m m j n i ijm eft l EFP \\/ 1 10 -RRB- -LRB- -RRB- 1 -LRB- 1 -RRB- -LRB- + = = = -LRB- 2 -RRB- The alignment models described in are all based on the notion that an alignment aligns each source word to exactly one target word ',Brown,'prep_normalized_In poss_approach_our pobj_In_approach nsubjpass_normalized_equation appos_equation_1 auxpass_normalized_is advmod_normalized_further dep_based_normalized advmod_comparable_so dep_comparable_that det_probability_the nsubj_comparable_probability prep_probability_for amod_lengths_different pobj_for_lengths prep_lengths_of pobj_of_F cop_comparable_is ccomp_normalized_comparable prep_comparable_at det_level_the nn_level_word pobj_at_level nn_EFP_m nn_EFP_m nn_EFP_j nn_EFP_n nn_EFP_i nn_EFP_ijm nn_EFP_eft nn_EFP_l dep_level_EFP partmod_EFP_\\/ number_10_1 dobj_\\/_10 nsubj_+_1 appos_1_1 conj_normalized_+ ccomp_+_= prep_+_= pcomp_=_= dep_described_2 det_described_The amod_described_alignment nn_described_models dobj_+_described prep_described_in pobj_in_ auxpass_based_are advmod_based_all prep_based_on det_notion_the pobj_on_notion complm_aligns_that det_alignment_an nsubj_aligns_alignment ccomp_based_aligns det_word_each nn_word_source dobj_aligns_word prep_aligns_to dep_to_exactly num_word_one nn_word_target nsubjpass_based_word ' +o,1541,'3http : \\/ \\/ wwwopenofficeorg Another corpora based method due to Turney and Littman tries to measure the semantic orientation O -LRB- t -RRB- for a term t by O -LRB- t -RRB- = summationdisplay tiS + PMI -LRB- t , ti -RRB- summationdisplay tjS PMI -LRB- t , tj -RRB- where S + and S are minimal sets of polar terms that contain prototypical positive and negative terms respectively , and PMI -LRB- t , ti -RRB- is the pointwise mutual information between the terms t and ti ',Lin,'dep_tries_3http nn_\\/_\\/ nsubj_tries_\\/ partmod_\\/_wwwopenofficeorg det_corpora_Another dobj_wwwopenofficeorg_corpora partmod_corpora_based nsubj_due_method xcomp_based_due prep_due_to nn__Turney cc_Turney_and conj_Turney_Littman pobj_to_ aux_measure_to xcomp_tries_measure det_O_the amod_O_semantic nn_O_orientation nsubj_=_O appos_O_t prep_O_for det_t_a nn_t_term pobj_for_t prep_t_by pobj_by_O appos_O_t ccomp_measure_= advmod_tiS_summationdisplay amod_PMI_tiS cc_tiS_+ conj_tiS_PMI dep_PMI_t dep_t_ti amod_PMI_summationdisplay nn_PMI_tjS dobj_=_PMI appos_PMI_t dep_t_tj advmod_sets_where nn_+_S nsubj_sets_+ cc_+_and conj_+_S cop_sets_are amod_sets_minimal rcmod_PMI_sets prep_sets_of amod_terms_polar pobj_of_terms nsubj_contain_that rcmod_sets_contain dobj_contain_prototypical dep_prototypical_positive cc_prototypical_and amod_terms_negative conj_prototypical_terms advmod_contain_respectively cc_tries_and nsubj__PMI appos_PMI_t dep_t_ti cop__is det__the amod__pointwise amod__mutual nn__information conj_tries_ prep__between det_t_the nn_t_terms pobj_between_t cc_t_and conj_t_ti ' +o,1542,'To evaluate sentence automatically generated with taking consideration word concatenation into by using references varied among humans , various metrics using n-gram precision and word accuracy have been proposed : word string precision for summarization through word extraction , ROUGE for abstracts , and BLEU for machine translation ',Papineni,'aux_evaluate_To dobj_evaluate_sentence advmod_generated_automatically partmod_sentence_generated prep_generated_with pcomp_with_taking nn_concatenation_consideration nn_concatenation_word dobj_taking_concatenation prep_taking_into dep_into_by pcomp_by_using dobj_using_references partmod_references_varied prep_varied_among pobj_among_humans amod_metrics_various conj_references_metrics partmod_metrics_using amod_precision_n-gram dobj_using_precision cc_precision_and nn_accuracy_word conj_precision_accuracy aux_proposed_have auxpass_proposed_been rcmod_precision_proposed nn_precision_word nn_precision_string dep_references_precision partmod_precision_ prep__for pobj_for_summarization prep__through nn_extraction_word pobj_through_extraction nn__ROUGE conj_references_ prep__for pobj_for_abstracts cc_references_and nn__BLEU conj_references_ prep__for nn_translation_machine pobj_for_translation ' +o,1543,'3 Evaluation We trained our model parameters on a subset of the provided dev2006 development set , optimizing for case-insensitive IBM-style BLEU with several iterations of minimum error rate training on n-best lists ',Papineni,'num_Evaluation_3 nsubj_trained_We dep_Evaluation_trained poss_parameters_our nn_parameters_model dobj_trained_parameters prep_trained_on det_subset_a pobj_on_subset prep_subset_of det_set_the amod_set_provided amod_set_dev2006 nn_set_development pobj_of_set xcomp_trained_optimizing prep_optimizing_for amod__case-insensitive amod__IBM-style amod__BLEU pobj_for_ prep__with amod_iterations_several pobj_with_iterations prep_iterations_of amod_training_minimum nn_training_error nn_training_rate pobj_of_training prep_optimizing_on amod_lists_n-best pobj_on_lists ' +o,1544,'Levin assumes that the syntactic realization of a verb \'s arguments is directly correlated with its meaning -LRB- cf ',Dunning,'nn__Levin nsubj_assumes_ complm_correlated_that det_realization_the amod_realization_syntactic nsubjpass_correlated_realization prep_realization_of det_verb_a poss_arguments_verb possessive_verb_\'s pobj_of_arguments auxpass_correlated_is advmod_correlated_directly ccomp_assumes_correlated prep_correlated_with pobj_with_its nn_cf_meaning dep_its_cf ' +o,1545,'The f-structures are created automatically by annotating nodes in the gold standard WSJ trees with LFG functional equations and then passing these equations through a constraint solver ',Cahill,'det_f-structures_The nsubjpass_created_f-structures auxpass_created_are advmod_created_automatically prep_created_by pcomp_by_annotating dobj_annotating_nodes prep_annotating_in det_trees_the amod_trees_gold amod_trees_standard nn_trees_WSJ pobj_in_trees prep_trees_with nn_equations_LFG amod_equations_functional pobj_with_equations cc_annotating_and advmod_annotating_then conj_annotating_passing det_equations_these dobj_passing_equations prep_passing_through det_solver_a nn_solver_constraint pobj_through_solver ' +o,1546,'First , we trained a finitestate shallow parser on base phrases extracted from the Penn Wall St Journal -LRB- WSJ -RRB- Treebank ',Marcus,'advmod_Treebank_First nsubj_Treebank_we partmod_we_trained det_parser_a amod_parser_finitestate nn_parser_shallow dobj_trained_parser prep_trained_on nn_phrases_base pobj_on_phrases partmod_phrases_extracted prep_extracted_from det_Journal_the nn_Journal_Penn nn_Journal_Wall nn_Journal_St pobj_from_Journal abbrev_we_WSJ ' +o,1547,'-LRB- Termbased versions of this premise have motivated much sentiment-analysis work for over a decade -RRB- ',Turney,'nn_versions_Termbased nsubj_motivated_versions prep_versions_of det_premise_this pobj_of_premise aux_motivated_have advmod_sentiment-analysis_much amod_work_sentiment-analysis dobj_motivated_work prep_motivated_for dep_for_over det__a nn__decade pobj_over_ ' +o,1548,'In particular , knowing a little about the structure of a language can help in developing annotated corpora and tools , since a little knowledge can go a long way in inducing accurate structure and annotations ',Haghighi,'prep_help_In pobj_In_particular csubj_help_knowing det_little_a dobj_knowing_little prep_knowing_about det_structure_the pobj_about_structure prep_structure_of det_language_a pobj_of_language aux_help_can ccomp_go_help prep_help_in pcomp_in_developing amod_corpora_annotated dobj_developing_corpora cc_corpora_and conj_corpora_tools prep_go_since det_knowledge_a amod_knowledge_little pobj_since_knowledge aux_go_can det_way_a amod_way_long dobj_go_way prep_go_in pcomp_in_inducing acomp_inducing_accurate nsubj_go_structure cc_structure_and conj_structure_annotations ' +o,1549,'As shown in , using this representation , a linear classifier can not distinguish sentences sampled from a trigram and real sentences ',Okanohara,'mark_shown_As advcl_distinguish_shown prep_shown_in pobj_in_ dep_shown_using det_representation_this dobj_using_representation det_classifier_a amod_classifier_linear nsubj_distinguish_classifier aux_distinguish_can neg_distinguish_not dobj_distinguish_sentences partmod_sentences_sampled prep_sampled_from det_trigram_a pobj_from_trigram cc_trigram_and amod_sentences_real conj_trigram_sentences ' +o,1550,'Researchers have focused on learning adjectives or adjectival phrases and verbs , but no previous work has focused on learning nouns ',Turney,'nsubj_focused_Researchers aux_focused_have prep_focused_on pcomp_on_learning dobj_learning_adjectives cc_adjectives_or nn_phrases_adjectival conj_adjectives_phrases nsubj__ cc__and conj__verbs dep_focused_ cc__but det_work_no amod_work_previous nsubj_focused_work aux_focused_has conj__focused prep_focused_on pcomp_on_learning dobj_learning_nouns ' +o,1551,'Probabilistic translation models generally seek to find the translation string e that maximizes the probability Pra5 ea6fa7 , given the source string f ',Brown,'amod_translation_Probabilistic nsubj_seek_models advmod_seek_generally rcmod_translation_seek aux_find_to xcomp_seek_find det_string_the nn_string_translation dobj_find_string dep_maximizes_e dep_maximizes_that dep_ea6fa7_maximizes det_ea6fa7_the nn_ea6fa7_probability nn_ea6fa7_Pra5 dep_translation_ea6fa7 partmod_ea6fa7_given det_source_the iobj_given_source nn_f_string dobj_given_f ' +o,1552,'joint likelihood (JL) productdisplay i p parenleftBig xi,yi | vector parenrightBig conditional likelihood (CL) productdisplay i p parenleftBig yi | xi,vector parenrightBig classification accuracy (Juang and Katagiri, 1992) summationdisplay i (yi, y(xi)) expected classification accuracy (Klein and Manning, 2002) summationdisplay i p parenleftBig yi | xi,vector parenrightBig negated boosting loss (Collins, 2000) summationdisplay i p parenleftBig yi | xi,vector parenrightBig1 margin (Crammer and Singer, 2001) s.t. bardbl vectorbardbl 1;i,y negationslash= yi, vector (vectorf(xi,yi ) vectorf(xi,y)) expected local accuracy (Altun et al. , 2003) productdisplay i productdisplay j p parenleftBig lscriptj(Y ) = lscriptj(yi ) | xi,vector parenrightBig Table 1: Various supervised training criteria.',Klein,'' +o,1553,'Most work on discriminative training for SMT has focussed on linear models , often with margin based algorithms , or rescaling a product of sub-models ',Och,'amod_work_Most nsubj_focussed_work prep_work_on amod_training_discriminative pobj_on_training prep_training_for pobj_for_SMT aux_focussed_has prep_focussed_on amod_models_linear pobj_on_models advmod_based_often dep_often_with pobj_with_margin partmod_models_based acomp_based_algorithms dep_algorithms_ cc_algorithms_or conj_algorithms_rescaling det_product_a dobj_rescaling_product prep_product_of pobj_of_sub-models ' +o,1554,'Bikel and Chiang in fact contains two parsers : one is a lexicalized probabilistic contextfree grammar -LRB- PCFG -RRB- similar to ; the other is based on statistical TAG ',Collins,'nn__Bikel cc_Bikel_and conj_Bikel_Chiang nsubj_contains_ prep__in pobj_in_fact num_parsers_two dobj_contains_parsers nsubj_grammar_one cop_grammar_is det_grammar_a amod_grammar_lexicalized amod_grammar_probabilistic nn_grammar_contextfree dep_parsers_grammar abbrev_grammar_PCFG amod_grammar_similar dep_similar_to dep_similar_ det_other_the nsubjpass_based_other auxpass_based_is parataxis_contains_based prep_based_on amod__statistical nn__TAG pobj_on_ ' +o,1555,'Finally , it should be noted that in the current implementation , we have not applied any of the possible optimizations that appear in the literature to speed up normalization of the probability distribution q These improvements take advantage of a models structure to simplify the evaluation of the denominator in -LRB- 1 -RRB- ',Berger,'advmod_noted_Finally nsubjpass_noted_it aux_noted_should auxpass_noted_be complm_applied_that prep_applied_in det_implementation_the amod_implementation_current pobj_in_implementation nsubj_applied_we aux_applied_have neg_applied_not ccomp_noted_applied dobj_applied_any prep_any_of det_optimizations_the amod_optimizations_possible pobj_of_optimizations nsubj_appear_that rcmod_optimizations_appear prep_appear_in det__the nn__literature pobj_in_ aux_speed_to xcomp_appear_speed prt_speed_up dobj_speed_normalization prep_normalization_of det_q_the nn_q_probability nn_q_distribution pobj_of_q det_improvements_These nsubj_take_improvements dep_noted_take nsubj_simplify_advantage prep_advantage_of det_structure_a nn_structure_models pobj_of_structure aux_simplify_to xcomp_take_simplify det_evaluation_the dobj_simplify_evaluation prep_evaluation_of det_denominator_the pobj_of_denominator prep_denominator_in pobj_in_1 ' +o,1556,'1510 5 Related Work In recent years , many research has been done on extracting relations from free text -LRB- eg , -RRB- ; however , almost all of them require some language-dependent parsers or taggers for English , which restrict the language of their extractions to English only -LRB- or languages that have these parsers -RRB- ',Snow,'number_5_1510 num_Work_5 amod_Work_Related prep_done_In amod_years_recent pobj_In_years amod_research_many nsubjpass_done_research aux_done_has auxpass_done_been dep_Work_done prep_done_on pcomp_on_extracting dobj_extracting_relations prep_extracting_from amod_text_free pobj_from_text appos_text_eg dep_eg_ advmod_require_however advmod_all_almost nsubj_require_all prep_all_of pobj_of_them parataxis_done_require det_parsers_some amod_parsers_language-dependent dobj_require_parsers cc_parsers_or conj_parsers_taggers prep_parsers_for pobj_for_English nsubj_restrict_which rcmod_English_restrict det_language_the dobj_restrict_language prep_language_of poss_extractions_their pobj_of_extractions prep_restrict_to pobj_to_English advmod_English_only dep_languages_or dep_English_languages nsubj_have_that rcmod_languages_have det_parsers_these dobj_have_parsers ' +o,1557,'To compare the performance of system , we recorded the total training time and the BLEU score , which is a standard automatic measurement of the translation qualit ',Papineni,'aux_compare_To dep_recorded_compare det_performance_the dobj_compare_performance prep_performance_of pobj_of_system nsubj_recorded_we det_time_the amod_time_total nn_time_training dobj_recorded_time cc_time_and det_score_the amod_score_BLEU conj_time_score nsubj_measurement_which cop_measurement_is det_measurement_a amod_measurement_standard amod_measurement_automatic rcmod_time_measurement prep_measurement_of det_qualit_the nn_qualit_translation pobj_of_qualit ' +o,1558,'Typically , a small set of seed polar phrases are prepared , and new polar phrases are detected based on the strength of co-occurrence with the seeds ',Kanayama,'advmod_prepared_Typically det_set_a amod_set_small nsubj_prepared_set prep_set_of nn_phrases_seed amod_phrases_polar pobj_of_phrases cop_prepared_are cc_prepared_and amod_phrases_new amod_phrases_polar nsubjpass_detected_phrases auxpass_detected_are conj_prepared_detected prep_detected_based dep_based_on det_strength_the pobj_on_strength prep_strength_of pobj_of_co-occurrence prep_co-occurrence_with det_seeds_the pobj_with_seeds ' +o,1559,'The basic LCS has a problem that it does not differentiate LCSes of different spatial relations within their embedding sequences ',Lin,'det_LCS_The amod_LCS_basic nsubj_has_LCS det_problem_a dobj_has_problem complm_differentiate_that nsubj_differentiate_it aux_differentiate_does neg_differentiate_not dep_problem_differentiate dobj_differentiate_LCSes prep_LCSes_of amod_relations_different amod_relations_spatial pobj_of_relations prep_differentiate_within poss_sequences_their amod_sequences_embedding pobj_within_sequences ' +o,1560,' , -RRB- ',Ponzetto,'appos__ ' +o,1561,'3 Parse Tree Features We tagged each candidate transcription with -LRB- 1 -RRB- part-of-speech tags , using the tagger documented in ; and -LRB- 2 -RRB- a full parse tree , using the parser documented in Collins -LRB- 1999 -RRB- ',Collins,'num_Features_3 nn_Features_Parse nn_Features_Tree nsubj_tagged_We dep_Features_tagged det_transcription_each nn_transcription_candidate dobj_tagged_transcription prep_tagged_with appos_tags_1 amod_tags_part-of-speech pobj_with_tags dep_tagged_using det_tagger_the dobj_using_tagger partmod_tagger_documented prep_documented_in pobj_in_ cc_tagger_and dep_tree_2 det_tree_a amod_tree_full nn_tree_parse conj_tagger_tree dep_tagged_using det_parser_the dobj_using_parser partmod_parser_documented prep_tagged_in pobj_in_Collins appos_Collins_1999 ' +o,1562,'We use the by now standard a0 statistic to quantify the degree of above-chance agreement between multiple annotators , and the a1 statistic for analysis of sources of unreliability ',Carletta,'nsubj_use_We dep_by_the prep_use_by advmod__now amod__standard nn__a0 nn__statistic pobj_by_ aux_quantify_to xcomp_use_quantify det_degree_the dobj_quantify_degree prep_degree_of amod_agreement_above-chance pobj_of_agreement prep_agreement_between amod_annotators_multiple pobj_between_annotators cc_degree_and det_statistic_the amod_statistic_a1 conj_degree_statistic prep_statistic_for pobj_for_analysis prep_analysis_of pobj_of_sources prep_sources_of amod__unreliability pobj_of_ ' +o,1563,'Other methods that have been proposed are one based on using the gain and an approximate method for selecting informative features , and several criteria for feature selection were proposed and compared with other criteria ',Berger,'amod_methods_Other nsubj_one_methods nsubjpass_proposed_that aux_proposed_have auxpass_proposed_been rcmod_methods_proposed cop_one_are prep_one_based dep_based_on pcomp_on_using det__the nn__gain dobj_using_ cc_one_and det_method_an amod_method_approximate nsubj__method prep_method_for pcomp_for_selecting amod_features_informative dobj_selecting_features conj_one_ cc_one_and amod_criteria_several nsubjpass_proposed_criteria prep_criteria_for nn_selection_feature pobj_for_selection auxpass_proposed_were conj_one_proposed cc_proposed_and conj_proposed_compared prep_proposed_with amod_criteria_other pobj_with_criteria ' +o,1564,'As our basic data source , we use 500 000 sentences from the Wikipedia XML corpus ; this is the corpus used by Akhmatova and Dras , and related to one used in one set of experiments by et al ',Snow,'prep_use_As poss_source_our amod_source_basic nn_source_data pobj_As_source nsubj_use_we number_000_500 num_sentences_000 dobj_use_sentences prep_sentences_from det__the nn__Wikipedia nn__XML nn__corpus pobj_from_ nsubj_corpus_this cop_corpus_is det_corpus_the parataxis_use_corpus dep_corpus_used prep_used_by nn__Akhmatova cc_Akhmatova_and conj_Akhmatova_Dras pobj_by_ cc_used_and conj_used_related prep_related_to pobj_to_one partmod_one_used prep_used_in num_set_one pobj_in_set prep_set_of pobj_of_experiments prep_used_by pobj_by_ cc__et conj__al ' +p,1565,'Promising features might include those over source side reordering rules or source context features ',Carpuat,'amod_features_Promising nsubj_include_features aux_include_might dobj_include_those prep_those_over nn__source nn__side nn__reordering nn__rules pobj_over_ cc__or nn_features_source nn_features_context conj__features ' +o,1566,'Among all possible target strings , we will choose the one with the highest probability which is given by Bayes \' decision rule : , ~ = argmaxP , \' -LRB- e -RRB- ~ lfg ~ -RRB- -RCB- = argmax -LCB- P , \' -LRB- ef -RRB- ',Brown,'prep_choose_Among det_strings_all amod_strings_possible nn_strings_target pobj_Among_strings nsubj_choose_we aux_choose_will det_one_the dobj_choose_one dep_one_with det_probability_the amod_probability_highest pobj_with_probability nsubjpass_given_which auxpass_given_is rcmod_probability_given prep_given_by poss__Bayes nn__decision nn__rule pobj_by_ amod_argmaxP_~ amod_argmaxP_= nsubj_=_argmaxP dep_-LRB-_e nn_~_~ nn_~_lfg dep_argmaxP_~ rcmod_one_= dobj_=_argmax dep_argmax_P dep_argmax_ef ' +o,1567,'In the English all-words task of the previous SENSEVAL evaluations , the best performing English all-words task systems with the highest WSD accuracy were trained on SEMCOR ','Daume III','prep_trained_In det_task_the amod_task_English amod_task_all-words pobj_In_task prep_task_of det__the amod__previous amod__SENSEVAL nn__evaluations pobj_of_ det_best_the nsubjpass_trained_best partmod_best_performing amod_systems_English amod_systems_all-words nn_systems_task dobj_performing_systems prep_performing_with det_accuracy_the amod_accuracy_highest nn_accuracy_WSD pobj_with_accuracy auxpass_trained_were prep_trained_on nn__SEMCOR pobj_on_ ' +o,1568,'One is a phrase-based translation in which a phrasal unit is employed for translation ',Koehn,'nsubj_translation_One cop_translation_is det_translation_a amod_translation_phrase-based rel_employed_in pobj_in_which det_unit_a amod_unit_phrasal nsubjpass_employed_unit auxpass_employed_is rcmod_translation_employed prep_employed_for pobj_for_translation ' +o,1569,'Other languagesfor which this is the case include English -LRB- with the Penn treebank , the Susanne Corpus , and the British section of the ICE Corpus -RRB- and Italian -LRB- with ISST and TUT -RRB- ',Marcus,'amod_languagesfor_Other nsubj_include_languagesfor dobj_case_which nsubj_case_this cop_case_is det_case_the rcmod_languagesfor_case acomp_include_English dep_English_with det__the nn__Penn nn__treebank pobj_with_ det__the nn__Susanne nn__Corpus conj__ cc__and det_section_the amod_section_British conj__section prep_section_of det__the nn__ICE nn__Corpus pobj_of_ cc_English_and conj_English_Italian dep_Italian_with nn__ISST pobj_with_ cc__and nn__TUT conj__ ' +p,1570,'The BLEU metric and the closely related NIST metric along with WER and PER 48 have been widely used by many machine translation researchers ',Papineni,'det__The amod__BLEU amod__metric cc__and det_NIST_the advmod_related_closely amod_NIST_related conj__NIST amod__metric nsubjpass_used_ advmod_used_along dep_along_with pobj_with_WER cc_WER_and conj_WER_PER num_PER_48 aux_used_have auxpass_used_been advmod_used_widely dep__used prep_used_by amod_researchers_many nn_researchers_machine nn_researchers_translation pobj_by_researchers ' +o,1571,'The algorithm is based on the Machine Learning method for word categorisation , inspired by the well known study on basic-level categories , presented in ',Dunning,'det_algorithm_The nsubjpass_based_algorithm auxpass_based_is prep_based_on det_method_the nn_method_Machine nn_method_Learning pobj_on_method prep_method_for nn_categorisation_word pobj_for_categorisation partmod_method_inspired prep_inspired_by det_study_the advmod_known_well amod_study_known pobj_by_study prep_study_on amod__basic-level nn__categories pobj_on_ partmod_method_presented prep_presented_in pobj_in_ ' +o,1572,'Automated evaluation will utilize the standard DUC evaluation metric ROUGE which representsrecallovervariousn-gramsstatisticsfrom asystem-generatedsummaryagainstasetofhumangenerated peer summaries5 We compute ROUGE scores with and without stop words removed from peer and proposed summaries ',Lin,'amod_evaluation_Automated nsubj_utilize_evaluation aux_utilize_will det_evaluation_the amod_evaluation_standard nn_evaluation_DUC dobj_utilize_evaluation amod__metric amod__ROUGE tmod_utilize_ rel_removed_which amod_summaries5_representsrecallovervariousn-gramsstatisticsfrom amod_summaries5_asystem-generatedsummaryagainstasetofhumangenerated nn_summaries5_peer nsubj_removed_summaries5 nsubj_compute_We rcmod_summaries5_compute amod_scores_ROUGE dobj_compute_scores prep_compute_with cc_with_and conj_with_without nn_words_stop pobj_with_words dep__removed prep_removed_from pobj_from_peer cc_peer_and amod_summaries_proposed conj_peer_summaries ' +o,1573,'Furthermore , our model is not necessarily nativist ; these biases may be innate , but they may also be the product of some other earlier learning algorithm , as the results of Ellison and et al ',Brown,'advmod_nativist_Furthermore poss_model_our nsubj_nativist_model cop_nativist_is neg_nativist_not advmod_nativist_necessarily det_biases_these nsubj_innate_biases aux_innate_may cop_innate_be parataxis_nativist_innate cc_innate_but nsubj_product_they aux_product_may advmod_product_also cop_product_be det_product_the conj_innate_product prep_product_of det_algorithm_some amod_algorithm_other amod_algorithm_earlier nn_algorithm_learning pobj_of_algorithm prep_product_as det_results_the pobj_as_results prep_results_of nn__Ellison pobj_of_ cc__and conj__ cc__et conj__al ' +o,1574,'For example , Animal would be mapped to Aa , GM would again be mapped to AA The tagger was applied and trained in the same way as described in ',Ratnaparkhi,'prep_mapped_For pobj_For_example nsubjpass_mapped_Animal aux_mapped_would auxpass_mapped_be parataxis_mapped_mapped prep_mapped_to pobj_to_Aa nsubjpass_mapped_GM aux_mapped_would advmod_mapped_again auxpass_mapped_be prep_mapped_to pobj_to_AA det_tagger_The nsubjpass_applied_tagger auxpass_applied_was dep_mapped_applied cc_applied_and conj_applied_trained prep_trained_in det_way_the amod_way_same pobj_in_way advmod_described_as amod_way_described prep_trained_in ' +o,1575,'As a result , the empirical approach has been adopted by almost all contemporary part-of-speech programs : Bahl and Mercer , Leech , Garside , and Atwell , Jelinek , Deroualt and Merialdo , Garside , Leech , and Sampson , Church , DeRose , Hindle , Kupiec , Ayuso et al ',Brown,'prep_adopted_As det_result_a pobj_As_result det_approach_the amod_approach_empirical nsubjpass_adopted_approach aux_adopted_has auxpass_adopted_been prep_adopted_by advmod_programs_almost det_programs_all amod_programs_contemporary amod_programs_part-of-speech pobj_by_programs nn__Bahl cc_Bahl_and conj_Bahl_Mercer dep_programs_ nn_Garside_Leech appos__Garside cc_Garside_and nn__Atwell conj_Garside_ nn__Jelinek appos__ nn__Deroualt cc_Deroualt_and conj_Deroualt_Merialdo appos__ nn_Leech_Garside appos__Leech cc_Leech_and nn__Sampson conj_Leech_ nn__Church appos__ nn__DeRose appos__ nn__Hindle appos__ nn__Kupiec appos__ conj__Ayuso cc__et conj__al ' +o,1576,'Typicality was measured using the log-likelihood ratio test ',Dunning,'nsubjpass_measured_Typicality auxpass_measured_was xcomp_measured_using det_test_the amod_test_log-likelihood nn_test_ratio dobj_using_test ' +o,1577,'The translation system is a factored phrasebased translation system that uses the Moses toolkit for decoding and training , GIZA + + for word alignment , and SRILM for language models ',Och,'det_system_The nn_system_translation nsubj_system_system cop_system_is det_system_a amod_system_factored amod_system_phrasebased nn_system_translation nsubj_uses_that rcmod_system_uses det__the nn__Moses nn__toolkit dobj_uses_ prep__for pobj_for_decoding cc_decoding_and conj_decoding_training nn_+_GIZA nn_+_+ conj_system_+ prep_+_for nn__word nn__alignment pobj_for_ cc_system_and nn__SRILM conj_system_ prep__for nn_models_language pobj_for_models ' +o,1578,'In constrast with many previous approaches , our model does not try to capture how Source sentences can be mapped into Target sentences , but rather how Source and Target sentences can be generated simultaneously ',Brown,'prep_try_In amod__constrast prep_constrast_with amod_approaches_many amod_approaches_previous pobj_with_approaches pobj_In_ poss_model_our nsubj_try_model aux_try_does neg_try_not aux_capture_to xcomp_try_capture advmod_mapped_how nn_sentences_Source nsubjpass_mapped_sentences aux_mapped_can auxpass_mapped_be dep_capture_mapped prep_mapped_into nn_sentences_Target pobj_into_sentences dep_rather_but cc_mapped_rather advmod_generated_how nn_sentences_Source cc_Source_and conj_Source_Target nsubjpass_generated_sentences aux_generated_can auxpass_generated_be conj_mapped_generated advmod_generated_simultaneously ' +o,1579,'Table 2 shows results in lowercase BLEU for both the baseline -LRB- B -RRB- and the improved baseline systems -LRB- B5 -RRB- on development and held151 out evaluation sets ',Papineni,'dep_shows_Table tmod_Table_2 dobj_shows_results prep_shows_in amod__lowercase amod__BLEU pobj_in_ prep__for predet_baseline_both det_baseline_the pobj_for_baseline appos_baseline_B cc_baseline_and det_baseline_the amod_baseline_improved conj_baseline_baseline nsubj_shows_systems appos_systems_B5 prep_shows_on pobj_on_development cc_development_and conj_development_held151 prep_shows_out nn_sets_evaluation pobj_out_sets ' +o,1580,'This view is supported by , who concludes that correlations to human judgments were increased by using multiple references but using single reference summary with enough number of samples was a valid alternative ',Lin,'det_view_This nsubjpass_supported_view auxpass_supported_is prep_supported_by pobj_by_ nsubj_concludes_who rcmod__concludes complm_increased_that nsubjpass_increased_correlations prep_correlations_to amod_judgments_human pobj_to_judgments auxpass_increased_were ccomp_concludes_increased prep_increased_by pcomp_by_using amod_references_multiple dobj_using_references cc_supported_but csubj_alternative_using amod_summary_single nn_summary_reference dobj_using_summary prep_using_with amod_number_enough pobj_with_number prep_number_of pobj_of_samples cop_alternative_was det_alternative_a amod_alternative_valid conj_supported_alternative ' +o,1581,'To date , researchers have harvested , with varying success , several resources , including concept lists , topic signatures , facts , and word similarity lists ',Hindle,'prep_harvested_To pobj_To_date nsubj_harvested_researchers aux_harvested_have prep_harvested_with amod_success_varying pobj_with_success amod_resources_several appos_success_resources prep_success_including nn__concept nn__lists pobj_including_ nn__topic nn__signatures conj__ nn__facts conj__ cc__and nn_lists_word nn_lists_similarity conj__lists ' +o,1582,'Most existing methods treat word tokens as basic alignment units , however , many languages have no explicit word boundary markers , such as Chinese and Japanese ',Brown,'amod_methods_Most amod_methods_existing nsubj_treat_methods nn_tokens_word dobj_treat_tokens prep_treat_as amod_units_basic nn_units_alignment pobj_as_units nsubj_have_ advmod_have_however amod_languages_many nsubj_have_languages dep_treat_have det_markers_no amod_markers_explicit nn_markers_word nn_markers_boundary dobj_have_markers dep_as_such prep_markers_as pobj_as_Chinese cc_Chinese_and conj_Chinese_Japanese ' +o,1583,'To model aspects of co-occurrence association that might be obscured by raw frequency , the log-likelihood ratio G2 was also used to transform the feature space ',Dunning,'aux_model_To dep_used_model dobj_model_aspects prep_aspects_of amod_association_co-occurrence pobj_of_association nsubjpass_obscured_that aux_obscured_might auxpass_obscured_be rcmod_aspects_obscured prep_obscured_by amod_frequency_raw pobj_by_frequency det__the amod__log-likelihood nn__ratio nn__G2 nsubjpass_used_ auxpass_used_was advmod_used_also aux_transform_to xcomp_used_transform det_space_the nn_space_feature dobj_transform_space ' +o,1584,'In Table 1 , the MALINE row 3 shows that the English name has a palato-alveolar modification 2 As point out , these insights are not easy to come by : These rules are based on first author Dr Andrew Freemans experience with reading and translating Arabic language texts for more than 16 years ',Freeman,'prep_shows_In pobj_In_Table num_Table_1 det_row_the amod_row_MALINE nsubj_shows_row num_row_3 complm_has_that det_name_the nn_name_English nsubj_has_name ccomp_shows_has det_modification_a amod_modification_palato-alveolar dobj_has_modification dep_easy_2 mark_point_As nsubj_point_ dep_easy_point prt_point_out det_insights_these nsubj_easy_insights cop_easy_are neg_easy_not rcmod_modification_easy aux_come_to xcomp_easy_come prep_come_by det_rules_These nsubjpass_based_rules auxpass_based_are parataxis_easy_based prep_based_on amod_author_first nn_experience_author nn_experience_Dr nn_experience_Andrew nn_experience_Freemans pobj_on_experience prep_based_with pcomp_with_reading cc_reading_and conj_reading_translating nn_texts_Arabic nn_texts_language dobj_reading_texts prep_texts_for dep_than_more quantmod_16_than num_years_16 pobj_for_years ' +o,1585,'The ITG we apply in our experiments has more structural labels than the primitive bracketing grammar : it has a start symbol S , a single preterminal C , and two intermediate nonterminals A and B used to ensure that only one parse can generate any given word-level alignment , as discussed by and Zens and Ney -LRB- 2003 -RRB- ',Wu,'det_ITG_The nsubj_has_ITG nsubj_apply_we rcmod_ITG_apply prep_apply_in poss_experiments_our pobj_in_experiments amod_labels_more amod_labels_structural dobj_has_labels prep_labels_than det_grammar_the amod_grammar_primitive nn_grammar_bracketing pobj_than_grammar nsubj_has_it parataxis_has_has det_symbol_a nn_symbol_start dobj_has_symbol nsubj_used_S det_C_a amod_C_single amod_C_preterminal conj_S_C cc_S_and num_nonterminals_two amod_nonterminals_intermediate conj_S_nonterminals dep_nonterminals_A cc_A_and conj_A_B rcmod_symbol_used aux_ensure_to xcomp_used_ensure complm_generate_that advmod_parse_only num_parse_one nsubj_generate_parse aux_generate_can ccomp_ensure_generate det_alignment_any amod_alignment_given amod_alignment_word-level dobj_generate_alignment mark_discussed_as advcl_has_discussed prep_discussed_by pobj_by_ cc__and conj__Zens cc__and conj__Ney dep_discussed_2003 ' +o,1586,'They are a subset of the features used in ',Ratnaparkhi,'nsubj_subset_They cop_subset_are det_subset_a prep_subset_of det_features_the pobj_of_features partmod_features_used prep_used_in ' +o,1587,'Other corpus-based methods determine associations between words , which yields a basis for computing thesauri , or dictionaries of terminological expressions and multiword lexemes ',Dunning,'amod_methods_Other amod_methods_corpus-based nsubj_determine_methods dobj_determine_associations prep_associations_between nn__words pobj_between_ nsubj_yields_which rcmod__yields det_basis_a dobj_yields_basis prep_basis_for amod_thesauri_computing pobj_for_thesauri cc_basis_or conj_basis_dictionaries prep_dictionaries_of amod_expressions_terminological pobj_of_expressions cc_expressions_and amod__multiword nn__lexemes conj_expressions_ ' +o,1588,'The data set is same as in Section 51 , except that we also parsed the English-side using a variant of the parser , and then extracted 247M tree-to-string rules using the algorithm of ',Collins,'det_set_The nn_set_data nsubj_same_set cop_same_is prep_same_as dep_as_in pobj_in_Section num_Section_51 mark_parsed_except dep_parsed_that nsubj_parsed_we advmod_parsed_also advcl_same_parsed det_English-side_the dobj_parsed_English-side partmod_English-side_using det_variant_a dobj_using_variant prep_variant_of det_parser_the amod_parser_ pobj_of_parser cc_parsed_and advmod_extracted_then conj_parsed_extracted num_rules_247M amod_rules_tree-to-string dobj_extracted_rules partmod_rules_using det_algorithm_the dobj_using_algorithm prep_algorithm_of pobj_of_ ' +p,1589,'A later study found that performance increased to 872 \% when considering only those portions of the text deemed to be subjective ',Pang,'det__A amod__later nn__study nsubj_found_ complm_increased_that nsubj_increased_performance ccomp_found_increased prep_increased_to num_\%_872 pobj_to_\% advmod_deemed_when csubj_deemed_considering advmod_portions_only det_portions_those dobj_considering_portions prep_portions_of det_text_the pobj_of_text advcl_increased_deemed aux_subjective_to cop_subjective_be xcomp_deemed_subjective ' +o,1590,'The triplet lexicon model presented in this work can also be interpreted as an extension of the standard IBM model 1 with an additional trigger ',Brown,'det_model_The amod_model_triplet nn_model_lexicon nsubjpass_interpreted_model partmod_model_presented prep_presented_in det_work_this pobj_in_work aux_interpreted_can advmod_interpreted_also auxpass_interpreted_be prep_interpreted_as det_extension_an pobj_as_extension prep_extension_of det__the amod__standard nn__IBM nn__model num__1 pobj_of_ prep__with det_trigger_an amod_trigger_additional pobj_with_trigger ' +o,1591,'Both systems rely on the OpenNlp maximum-entropy part-of-speech tagger and chunker , but KNOWITALL applies them to pages downloaded from the Web based on the results of Google queries , whereas KNOWITNOW applies them once to crawled and indexed pages6 Overall , each of the above elements of KNOWITALL and KNOWITNOW are the same to allow for controlled experiments ',Ratnaparkhi,'det_systems_Both nsubj_rely_systems prep_rely_on det_tagger_the nn_tagger_OpenNlp amod_tagger_maximum-entropy amod_tagger_part-of-speech pobj_on_tagger cc_tagger_and nn__chunker conj_tagger_ cc_rely_but nsubj_applies_KNOWITALL conj_rely_applies dobj_applies_them prep_applies_to pobj_to_pages partmod_pages_downloaded prep_downloaded_from det_Web_the pobj_from_Web partmod_Web_based prep_based_on det_results_the pobj_on_results prep_results_of nn_queries_Google pobj_of_queries dep_rely_whereas nsubj_applies_KNOWITNOW dep_rely_applies dobj_applies_them advmod_applies_once prep_applies_to amod_pages6_crawled cc_crawled_and conj_crawled_indexed pobj_to_pages6 advmod_applies_Overall dobj_applies_each prep_each_of det_elements_the amod_elements_above pobj_of_elements prep_elements_of pobj_of_KNOWITALL cc_rely_and nsubj_same_KNOWITNOW cop_same_are det_same_the conj_rely_same aux_allow_to dep_same_allow prep_allow_for amod_experiments_controlled pobj_for_experiments ' +o,1592,'However , if we are willing to accept that occasionally our model will be unable to distinguish between distinct n-grams , then it is possible to store each parameter in constant space independent of both n and the vocabulary size , ',Talbot,'dep_willing_However mark_willing_if nsubj_willing_we cop_willing_are aux_accept_to xcomp_willing_accept complm_unable_that advmod_unable_occasionally poss_model_our nsubj_unable_model aux_unable_will cop_unable_be ccomp_accept_unable aux_distinguish_to xcomp_unable_distinguish prep_distinguish_between amod_n-grams_distinct pobj_between_n-grams advmod_unable_then nsubj_possible_it cop_possible_is dep_unable_possible aux_store_to xcomp_possible_store det_parameter_each dobj_store_parameter prep_store_in amod_independent_constant nn_independent_space pobj_in_independent prep_independent_of det_n_both pobj_of_n cc_n_and det__the nn__vocabulary nn__size conj_n_ ' +o,1593,'Models of that form include hidden Markov models as well as discriminative tagging models based on maximum entropy classification , conditional random fields , and large-margin techniques ',Ratnaparkhi,'nsubj_include_Models prep_Models_of det_form_that pobj_of_form amod_models_hidden nn_models_Markov nsubj__models ccomp_include_ dep_well_as cc__well dep_well_as conj__discriminative amod_models_tagging dobj_discriminative_models prep_discriminative_based dep_based_on amod__maximum amod__entropy nn__classification pobj_on_ amod_fields_conditional amod_fields_random nsubj__fields ccomp_include_ cc_include_and amod_techniques_large-margin nsubj__techniques conj_include_ ' +o,1594,'We enrich the semantic information available to the classifier by using semantic similarity measures based on the WordNet taxonomy ',Pedersen,'nsubj_enrich_We det_information_the amod_information_semantic dobj_enrich_information amod_information_available prep_available_to det_classifier_the pobj_to_classifier prep_enrich_by pcomp_by_using amod_measures_semantic nn_measures_similarity dobj_using_measures prep_using_based dep_based_on det_taxonomy_the nn_taxonomy_WordNet pobj_on_taxonomy ' +o,1595,'We can stipulate the time line to be linearly ordered -LRB- although it is not in approaches that build ignorance of relative times into the representation of time nor in approaches using branching futures -RRB- , and we can stipulate it to be dense -LRB- although it is not in the situation calculus -RRB- ',Hobbs,'nsubj_stipulate_We aux_stipulate_can det_line_the nn_line_time nsubjpass_ordered_line aux_ordered_to auxpass_ordered_be advmod_ordered_linearly xcomp_stipulate_ordered mark_is_although nsubj_is_it parataxis_stipulate_is neg_is_not prep_is_in pobj_in_approaches nsubj_build_that rcmod_approaches_build dobj_build_ignorance prep_ignorance_of amod_times_relative pobj_of_times prep_build_into det_representation_the nn_approaches_representation prep_representation_of nn__time pobj_of_ cc_representation_nor conj_representation_in pobj_into_approaches partmod_approaches_using xcomp_using_branching nn__futures dobj_branching_ cc_stipulate_and nsubj_stipulate_we aux_stipulate_can conj_stipulate_stipulate nsubj_dense_it aux_dense_to cop_dense_be xcomp_stipulate_dense mark_is_although nsubj_is_it parataxis_dense_is neg_is_not prep_is_in det_calculus_the nn_calculus_situation pobj_in_calculus ' +o,1596,'Typical approaches to conversion of constituent structures into dependencies are based on handconstructed head percolation rules , an idea that has its roots in lexicalized constituent parsing ',Collins,'amod_approaches_Typical nsubjpass_based_approaches prep_approaches_to pobj_to_conversion prep_conversion_of amod_structures_constituent pobj_of_structures prep_structures_into pobj_into_dependencies auxpass_based_are prep_based_on amod_rules_handconstructed nn_rules_head nn_rules_percolation pobj_on_rules det_idea_an appos_rules_idea nsubj_has_that rcmod_idea_has poss_roots_its dobj_has_roots prep_roots_in amod_constituent_lexicalized pobj_in_constituent amod_constituent_parsing ' +o,1597,'We extracted 181,250 case frames from the WSJ -LRB- Wall Street Journal -RRB- bracketed corpus of the Penn Tree Bank ',Marcus,'nsubj_extracted_We num_frames_181,250 nn_frames_case dep_from_frames prep_extracted_from det_corpus_the nn_corpus_WSJ nn_Journal_Wall nn_Journal_Street appos_corpus_Journal amod_corpus_bracketed pobj_from_corpus prep_corpus_of det_Bank_the nn_Bank_Penn nn_Bank_Tree pobj_of_Bank ' +o,1598,'c2008 Association for Computational Linguistics Refining Event Extraction through Cross-document Inference Heng Ji Ralph Grishman Computer Science Department New York University New York , NY 10003 , USA -LRB- hengji , grishman -RRB- @ csnyuedu Abstract We apply the hypothesis of One Sense Per Discourse to information extraction -LRB- IE -RRB- , and extend the scope of discourse from one single document to a cluster of topically-related documents ',Yarowsky,'nn_Association_c2008 prep_Association_for nn_Extraction_Computational nn_Extraction_Linguistics nn_Extraction_Refining nn_Extraction_Event pobj_for_Extraction prep_Extraction_through nn_USA_Cross-document nn_USA_Inference nn_USA_Heng nn_USA_Ji nn_USA_Ralph nn_USA_Grishman dep_USA_Computer dep_USA_Science dep_USA_Department nn_York_New nn_York_York nn_York_University nn_York_New dep_USA_York dep_York_NY num_NY_10003 pobj_through_USA appos_USA_hengji dep_hengji_grishman dep_Abstract_@ amod_Abstract_csnyuedu dep_apply_Abstract nsubj_apply_We dep_Association_apply det_hypothesis_the dobj_apply_hypothesis prep_hypothesis_of num_Sense_One pobj_of_Sense dep_hypothesis_Per nn__Discourse dep_hypothesis_ prep__to nn_extraction_information pobj_to_extraction abbrev_extraction_IE cc_apply_and conj_apply_extend det_scope_the dobj_extend_scope prep_scope_of pobj_of_discourse prep_extend_from num_document_one amod_document_single pobj_from_document prep_extend_to det_cluster_a pobj_to_cluster prep_cluster_of amod_documents_topically-related pobj_of_documents ' +o,1599,'Section 3 describes two standard lexicalized models , as well as an unlexicalized baseline model ',Collins,'nsubj_describes_Section num_Section_3 num__two amod__standard amod__lexicalized nn__models dobj_describes_ dep_well_as cc__well dep_well_as det_model_an amod_model_unlexicalized nn_model_baseline conj__model ' +o,1600,'Many researchers build alignment links with bilingual corpora ',Wu,'amod_researchers_Many nsubj_build_researchers amod_links_alignment dobj_build_links prep_build_with amod_corpora_bilingual pobj_with_corpora ' +o,1601,'The simplest version , called Dependency Model with Valence -LRB- DMV -RRB- , has been used in isolation and in combination with other models ',Smith,'det_version_The amod_version_simplest nsubjpass_used_version partmod_version_called nn_Model_Dependency dobj_called_Model prep_Model_with pobj_with_Valence abbrev_Valence_DMV aux_used_has auxpass_used_been prep_used_in pobj_in_isolation cc_in_and conj_in_in pobj_in_combination prep_used_with amod_models_other pobj_with_models ' +o,1602,'The translation table is obtained as described in (Koehn et al. , 2003), i.e. the alignment tool GIZA++ is run over the training data in both translation directions, and the two alignTest Setting BLEU B1 standard phrase-based SMT 29.22 B2 (B1) + clause splitting 29.13 Table 2: Experiment Baseline Test Setting BLEU BLEU 2-ary 2,3-ary 1 rule 29.77 30.31 2 ME (phrase label) 29.93 30.49 3 ME (left,right) 30.10 30.53 4 ME ((3)+head) 30.24 30.71 5 ME ((3)+phrase label) 30.12 30.30 6 ME ((4)+context) 30.24 30.76 Table 3: Tests on Various Reordering Models The 3rd column comprises the BLEU scores obtained by reordering binary nodes only, the 4th column the scores by reordering both binary and 3-ary nodes.',Koehn,'' +o,1603,'6 Related Work The most relevant previous works include word sense translation and translation disambiguation , frame semantic induction , and bilingual semantic mapping ',Green,'num_Work_6 amod_Work_Related dep_relevant_The advmod_relevant_most dep_include_relevant amod_works_previous nsubj_include_works dep_frame_include nn_translation_word nn_translation_sense dobj_include_translation cc_translation_and nn__translation nn__disambiguation conj_translation_ dep_Work_frame amod__semantic nn__induction dobj_frame_ cc__and amod__bilingual amod__semantic nn__mapping conj__ ' +o,1604,'We take the generator of as our baseline generator ',Cahill,'nsubj_take_We det_generator_the dobj_take_generator prep_generator_of pobj_of_ prep_take_as poss_generator_our nn_generator_baseline pobj_as_generator ' +o,1605,'translation systems and use Moses to search for the best target sentence ',Koehn,'nn_systems_translation nsubj__systems cc__and conj__use nn__Moses dobj__ prep__to pobj_to_search prep__for det_sentence_the amod_sentence_best nn_sentence_target pobj_for_sentence ' +o,1606,'Moreover , as P-DOP is formulated as an enrichment of the treebank Probabilistic Context-free Grammar -LRB- PCFG -RRB- , it allows for much easier comparison to alternative approaches to statistical parsing ',Collins,'advmod_allows_Moreover mark_formulated_as nsubjpass_formulated_P-DOP auxpass_formulated_is dep_allows_formulated prep_formulated_as det_enrichment_an pobj_as_enrichment prep_enrichment_of det_Grammar_the amod_Grammar_treebank nn_Grammar_Probabilistic nn_Grammar_Context-free pobj_of_Grammar abbrev_Grammar_PCFG nsubj_allows_it prep_allows_for advmod_easier_much amod_comparison_easier pobj_for_comparison prep_comparison_to amod_approaches_alternative pobj_to_approaches prep_allows_to amod_parsing_statistical pobj_to_parsing ' +o,1607,'This model shares some similarities with the stochastic inversion transduction grammars -LRB- SITG -RRB- presented by Wu in ',Wu,'det_model_This dep_similarities_model dep_model_shares dep_similarities_some dep_grammars_similarities prep_similarities_with det_inversion_the amod_inversion_stochastic pobj_with_inversion nn_grammars_transduction abbrev_grammars_SITG partmod_grammars_presented prep_presented_by pobj_by_Wu prep_presented_in ' +o,1608,'For Czech , we created a prototype of the first step of this process - the part-of-speech -LRB- POS -RRB- tagger - using Rank Xerox tools , ',Cutting,'prep_created_For pobj_For_Czech nsubj_created_we det_prototype_a dobj_created_prototype prep_prototype_of det_step_the amod_step_first pobj_of_step prep_step_of det_process_this pobj_of_process det_tagger_the amod_tagger_part-of-speech appos_tagger_POS dep_process_tagger parataxis_created_using nn_tools_Rank nn_tools_Xerox nsubj__tools xcomp_using_ ' +o,1609,'The rules are then treated as events in a relative frequency estimate4 We used Giza + + Model 4 to obtain word alignments , using the grow-diag-final-and heuristic to symmetrise the two directional predictions ',Och,'det_rules_The nsubjpass_treated_rules auxpass_treated_are advmod_treated_then prep_treated_as pobj_as_events prep_events_in det_estimate4_a amod_estimate4_relative nn_estimate4_frequency pobj_in_estimate4 nsubj_used_We dep_treated_used nn_Model_Giza nn_Model_+ nn_Model_+ nsubj_obtain_Model num_Model_4 aux_obtain_to xcomp_used_obtain nn_alignments_word dobj_obtain_alignments advmod_obtain_ xcomp_used_using det_heuristic_the amod_heuristic_grow-diag-final-and dobj_using_heuristic aux_symmetrise_to xcomp_using_symmetrise det__the num__two nn__directional nn__predictions dobj_symmetrise_ ' +o,1610,'Mutual Informatio n discussed the use of the mutual information statistics as a way to identify a variety of interesting linguistic phenomena , ranging from semanti c relations of the doctor\\/nurse type -LRB- content word\\/content word -RRB- to lexico-syntactic co-occurrence preferences between verbs and prepositions -LRB- content word\\/function word -RRB- ',Church,'amod__Mutual nn__Informatio nn__n partmod__discussed det_use_the dobj_discussed_use prep_use_of det_statistics_the amod_statistics_mutual nn_statistics_information pobj_of_statistics prep_discussed_as det_way_a pobj_as_way aux_identify_to infmod_way_identify det_variety_a dobj_identify_variety prep_variety_of amod_phenomena_interesting amod_phenomena_linguistic pobj_of_phenomena partmod_variety_ranging prep_ranging_from pobj_from_semanti dep_relations_c dep_ranging_relations prep_relations_of det_type_the amod_type_doctor\\/nurse pobj_of_type num_word_content nn_word_word\\/content appos_type_word prep_ranging_to amod_preferences_lexico-syntactic amod_preferences_co-occurrence pobj_to_preferences prep_preferences_between pobj_between_verbs cc_verbs_and conj_verbs_prepositions dep__content nn_word_word\\/function dobj_content_word ' +o,1611,'A constituent-based system using Collins parser ',Collins,'det_system_A amod_system_constituent-based dep_using_system nn_parser_Collins dobj_using_parser ' +o,1612,'This task evaluated parsing performance on 10 languages : Arabic , Basque , Catalan , Chinese , Czech , English , Greek , Hungarian , Italian , and Turkish using data originating from a wide variety of dependency treebanks , and transformations of constituency-based treebanks ',Marcus,'det_task_This nsubj_evaluated_task amod_performance_parsing dobj_evaluated_performance prep_performance_on num_languages_10 pobj_on_languages nn_Italian_Arabic appos_Italian_Basque appos_Italian_Catalan appos_Italian_Chinese appos_Italian_Czech appos_Italian_English amod_Italian_Greek amod_Italian_Hungarian dep_performance_Italian cc_Italian_and conj_Italian_Turkish partmod_Turkish_using dobj_using_data partmod_data_originating prep_originating_from det_variety_a amod_variety_wide pobj_from_variety prep_variety_of amod_treebanks_dependency pobj_of_treebanks cc_performance_and conj_performance_transformations prep_transformations_of amod_treebanks_constituency-based pobj_of_treebanks ' +o,1613,'Among the most widely studied is the Gibbs distribution ',Berger,'prep_is_Among dep_studied_the advmod_studied_most advmod_studied_widely pobj_Among_studied det_distribution_the nn_distribution_Gibbs nsubj_is_distribution partmod_distribution_ ' +o,1614,'Illustrative clusterings of this type can also be found in Pereira , Tishby , and Lee , , Kneser and Ney , and Brill et al ',Brown,'amod_clusterings_Illustrative nsubjpass_found_clusterings prep_clusterings_of det_type_this pobj_of_type aux_found_can advmod_found_also auxpass_found_be prep_found_in nn_Tishby_Pereira pobj_in_Tishby cc_Tishby_and nn__Lee conj_Tishby_ nn__ conj__Kneser cc__and conj__Ney conj_Tishby_ cc_Tishby_and conj_Tishby_Brill cc_Brill_et conj_Brill_al ' +o,1615,'Since then , supervised learning from sense-tagged corpora has since been used by several researchers : Zernik , Hearst , Leacock , Towell , and Voorhees , Gale , Church , and Yarowsky , Bruce and Wiebe , Miller et al ',Brown,'prep_used_Since pobj_Since_then amod_learning_supervised nsubjpass_used_learning prep_learning_from amod_corpora_sense-tagged pobj_from_corpora aux_used_has advmod_used_since auxpass_used_been prep_used_by amod_researchers_several pobj_by_researchers nn__Zernik dep_researchers_ nn_Towell_Hearst nn_Towell_ appos_Towell_Leacock conj_researchers_Towell cc_Towell_and nn__Voorhees conj_Towell_ nn_Church_Gale conj_researchers_Church cc_Church_and nn__Yarowsky conj_Church_ conj_researchers_Bruce cc_researchers_and nn__Wiebe conj_researchers_ conj__Miller cc__et conj__al ' +o,1616,'We tuned Pharaohs four parameters using minimum error rate training on DEV12 We obtained an increase of 08 9As in the POS features , we map each phrase pair to its majority constellation ',Och,'nsubj_Pharaohs_We advmod_Pharaohs_tuned num_parameters_four dobj_Pharaohs_parameters partmod_parameters_using amod__minimum nn__error nn__rate nn__training dobj_using_ prep_using_on pobj_on_DEV12 nsubj_obtained_We ccomp_map_obtained det_increase_an dobj_obtained_increase prep_increase_of num_9As_08 pobj_of_9As prep_obtained_in det_features_the dep_features_POS pobj_in_features nsubj_map_we dep_Pharaohs_map det_pair_each nn_pair_phrase dobj_map_pair prep_map_to poss_constellation_its nn_constellation_majority pobj_to_constellation ' +o,1617,'The performance figures given below are based on training each method on the 1-million-word Brown corpus and testing it on a 3\\/4-million-word corpus of Wall Street Journal text ',Marcus,'det_figures_The nn_figures_performance nsubjpass_based_figures prep_figures_given dep_given_below auxpass_based_are prep_based_on pcomp_on_training det_method_each dobj_training_method prep_method_on det__the amod__1-million-word nn__Brown nn__corpus pobj_on_ cc_training_and conj_training_testing dobj_testing_it prep_testing_on det_corpus_a amod_corpus_3\\/4-million-word pobj_on_corpus prep_corpus_of nn_text_Wall nn_text_Street nn_text_Journal pobj_of_text ' +o,1618,'Many researchers -LRB- ; -RRB- have suggested that the informationtheoretic notion of mutual information score -LRB- MIS -RRB- directly captures the idea of context ',Dunning,'amod_researchers_Many nsubj_suggested_researchers nn__ appos_researchers_ aux_suggested_have complm_captures_that det_notion_the amod_notion_informationtheoretic nsubj_captures_notion prep_notion_of amod_score_mutual nn_score_information pobj_of_score abbrev_score_MIS advmod_captures_directly ccomp_suggested_captures det_idea_the dobj_captures_idea prep_idea_of pobj_of_context ' +o,1619,'Word alignment and phrase extraction We used the GIZA + + word alignment software 3 to produce initial word alignments for our miniature bilingual corpus consisting of the source French file and the English reference file , and the refined word alignment strategy of to obtain improved word and phrase alignments ',Koehn,'amod_alignment_Word nsubj_extraction_alignment cc_alignment_and conj_alignment_phrase nsubj_used_We ccomp_extraction_used det_software_the amod_software_GIZA cc_GIZA_+ conj_GIZA_+ nn_software_word nn_software_alignment dobj_used_software dep_software_3 aux_produce_to xcomp_used_produce amod_alignments_initial nn_alignments_word dobj_produce_alignments prep_produce_for poss_corpus_our amod_corpus_miniature amod_corpus_bilingual pobj_for_corpus partmod_corpus_consisting prep_consisting_of det_file_the nn_file_source amod_file_French pobj_of_file cc_file_and det_file_the amod_file_English nn_file_reference conj_file_file cc_file_and det_strategy_the amod_strategy_refined nn_strategy_word nn_strategy_alignment conj_file_strategy prep_strategy_of pobj_of_ aux_obtain_to xcomp_consisting_obtain amod_word_improved dobj_obtain_word cc_word_and nn_alignments_phrase conj_word_alignments ' +o,1620,'For each training data size , we report the size of the resulting language model , the fraction of 5-grams from the test data that is present in the language model , and the BLEU score obtained by the machine translation system ',Papineni,'prep_report_For det_size_each nn_size_training nn_size_data pobj_For_size nsubj_report_we det_size_the dobj_report_size prep_size_of det_model_the amod_model_resulting nn_model_language pobj_of_model det_fraction_the conj_size_fraction prep_fraction_of amod_data_5-grams prep_5-grams_from det_test_the pobj_from_test pobj_of_data nsubj_present_that cop_present_is rcmod_fraction_present prep_present_in det_model_the nn_model_language pobj_in_model cc_size_and det__the amod__BLEU nn__score conj_size_ partmod__obtained prep_obtained_by det_system_the nn_system_machine nn_system_translation pobj_by_system ' +o,1621,'The approach is based on the hypothesis that positive words co-occur more than expected by chance , and so do negative words ; this hypothesis was validated , at least for strong positive\\/negative words , in ',Turney,'det_approach_The nsubjpass_based_approach auxpass_based_is prep_based_on det_hypothesis_the pobj_on_hypothesis complm_co-occur_that amod_words_positive nsubj_co-occur_words ccomp_based_co-occur advmod_expected_more dep_more_than dep_co-occur_expected prep_expected_by pobj_by_chance cc_expected_and advmod_do_so conj_expected_do amod_words_negative dobj_do_words det_hypothesis_this nsubjpass_validated_hypothesis auxpass_validated_was parataxis_based_validated advmod_validated_at dep_at_least dep_at_for amod_words_strong nn_words_positive\\/negative pobj_for_words prep_validated_in ' +o,1622,'Following initial work by and , an early , online distributional thesaurus presented in has been widely used and cited , and numerous authors since have explored thesaurus properties and parameters : see survey component of ',Weeds,'amod_work_initial pobj_Following_work prep_used_by pobj_by_ cc__and conj__ det_thesaurus_an amod_thesaurus_early amod_thesaurus_online amod_thesaurus_distributional nsubjpass_used_thesaurus partmod_thesaurus_presented prep_presented_in pobj_in_ aux_used_has auxpass_used_been advmod_used_widely rcmod_work_used cc_used_and conj_used_cited cc_used_and amod_authors_numerous nsubj_explored_authors advmod_explored_since aux_explored_have conj_used_explored nn_properties_thesaurus dobj_explored_properties cc_properties_and conj_properties_parameters dep_Following_see nn_component_survey dobj_see_component prep_see_of ' +p,1623,'1 Introduction We have seen rapid recent progress in machine translation through the use of rich features and the development of improved decoding algorithms , often based on grammatical formalisms1 If we view MT as a machine learning problem , features and formalisms imply structural independence assumptions , which are in turn exploited by efficient inference algorithms , including decoders ',Koehn,'num_Introduction_1 nsubj_seen_We aux_seen_have dep_Introduction_seen amod_progress_rapid amod_progress_recent dobj_seen_progress prep_progress_in nn_translation_machine pobj_in_translation prep_seen_through det_use_the pobj_through_use prep_use_of amod_features_rich pobj_of_features cc_use_and det_development_the conj_use_development prep_development_of amod_algorithms_improved nn_algorithms_decoding pobj_of_algorithms advmod_seen_often prep_seen_based dep_based_on amod_formalisms1_grammatical pobj_on_formalisms1 mark_view_If nsubj_view_we advcl_imply_view dobj_view_MT prep_MT_as det_machine_a pobj_as_machine partmod_machine_learning dobj_learning_problem nsubj_imply_features cc_features_and conj_features_formalisms dep_Introduction_imply amod_assumptions_structural nn_assumptions_independence dobj_imply_assumptions nsubj_are_which rcmod_assumptions_are prep_are_in pobj_in_turn partmod_turn_exploited prep_exploited_by amod_algorithms_efficient nn_algorithms_inference pobj_by_algorithms prep_turn_including pobj_including_decoders ' +o,1624,'The tree is produced by a state-of-the-art dependency parser trained on the Wall Street Journal Penn Treebank ',Marcus,'det_tree_The nsubjpass_produced_tree auxpass_produced_is prep_produced_by det__a amod__state-of-the-art amod__dependency nn__parser pobj_by_ partmod__trained prep_trained_on det_Treebank_the nn_Treebank_Wall nn_Treebank_Street nn_Treebank_Journal nn_Treebank_Penn pobj_on_Treebank ' +o,1625,'Part-of-speech tagging is an active area of research ; a great deal of work has been done in this area over the past few years ',Marcus,'nn_tagging_Part-of-speech nsubj_area_tagging cop_area_is det_area_an amod_area_active prep_area_of pobj_of_research det_deal_a amod_deal_great nsubjpass_done_deal prep_deal_of pobj_of_work aux_done_has auxpass_done_been parataxis_area_done prep_done_in det_area_this pobj_in_area prep_area_over det__the amod__past amod__few nn__years pobj_over_ ' +o,1626,'However , following the work of , Yarowsky -LRB- 1995 -RRB- , many supervised WSD systems use minimal information about syntactic structures , for the most part restricting the notion of context to topical and local features ',Yarowsky,'advmod_use_However prep_use_following det_work_the pobj_following_work prep_work_of nn_Yarowsky_ pobj_of_Yarowsky appos_Yarowsky_1995 amod_systems_many amod_systems_supervised nn_systems_WSD nsubj_use_systems amod_information_minimal dobj_use_information prep_information_about amod_structures_syntactic pobj_about_structures prep_use_for det_part_the amod_part_most pobj_for_part partmod_part_restricting det_notion_the dobj_restricting_notion prep_notion_of dep_of_context dep_of_to amod_features_topical cc_topical_and conj_topical_local pobj_of_features ' +o,1627,'Graphically speaking , parsing amounts to identifying rectangular crosslinguistic constituents by assembling smaller rectangles that will together cover the full string spans in both dimensions -LRB- compare -RRB- ',Wu,'amod_speaking_Graphically amod_amounts_parsing appos_speaking_amounts prep_amounts_to pcomp_to_identifying amod_constituents_rectangular amod_constituents_crosslinguistic dobj_identifying_constituents prep_identifying_by pcomp_by_assembling amod_rectangles_smaller dobj_assembling_rectangles nsubj_cover_that aux_cover_will advmod_cover_together rcmod_rectangles_cover det_spans_the amod_spans_full nn_spans_string dobj_cover_spans prep_cover_in det_dimensions_both pobj_in_dimensions dep_speaking_compare acomp_compare_ ' +n,1628,'The 746 \% final accuracy on apartments is higher than any result obtained by -LRB- the highest is 741 \% -RRB- , higher than the supervised HMM results reported by Grenager et al ',Haghighi,'det_accuracy_The num_\%_746 measure_final_\% amod_accuracy_final nsubj_higher_accuracy prep_accuracy_on pobj_on_apartments cop_higher_is mark_obtained_than det_result_any nsubj_obtained_result advcl_higher_obtained prep_obtained_by pobj_by_ det_highest_the nsubj_\%_highest cop_\%_is num_\%_741 dep__\% dep_than_higher prep_obtained_than det_results_the amod_results_supervised nn_results_HMM pobj_than_results partmod_results_reported prep_reported_by pobj_by_Grenager cc_Grenager_et conj_Grenager_al ' +o,1629,'Liang uses the discriminative perceptron algorithm to score whole character tag sequences , finding the best candidate by the global score ',Collins,'dobj_Liang_ dep_Liang_uses det__the amod__discriminative nn__perceptron nn__algorithm dobj_uses_ prep_uses_to pobj_to_score amod_sequences_whole nn_sequences_character nn_sequences_tag nsubj_Liang_sequences partmod_sequences_finding det_candidate_the amod_candidate_best dobj_finding_candidate prep_finding_by det_score_the amod_score_global pobj_by_score ' +p,1630,'Although state-of-the-art statistical parsers are more accurate , the simplicity and efficiency of deterministic parsers make them attractive in a number of situations requiring fast , light-weight parsing , or parsing of large amounts of data ',Collins,'mark_accurate_Although amod_parsers_state-of-the-art amod_parsers_statistical nsubj_accurate_parsers advmod_accurate_ cop_accurate_are advmod_accurate_more advcl_make_accurate det_simplicity_the nsubj_make_simplicity cc_simplicity_and conj_simplicity_efficiency prep_simplicity_of amod_parsers_deterministic pobj_of_parsers nsubj_requiring_them dep_requiring_attractive prep_attractive_in det_number_a pobj_in_number prep_number_of pobj_of_situations xcomp_make_requiring amod_parsing_fast dep_fast_light-weight dobj_requiring_parsing cc_requiring_or conj_requiring_parsing prep_parsing_of amod_amounts_large pobj_of_amounts prep_amounts_of pobj_of_data ' +o,1631,'The rationale for using Kappa is explained in ',Carletta,'det_rationale_The nsubjpass_explained_rationale prep_rationale_for pcomp_for_using dobj_using_Kappa auxpass_explained_is prep_explained_in ' +o,1632,'The measures vary from simple edge-counting to attempt to factor in peculiarities of the network structure by considering link direction , relative path , and density , such as vector , lesk , hso , lch , wup , path , res , lin and jcn ',Pedersen,'det_measures_The nsubj_vary_measures prep_vary_from amod_edge-counting_simple pobj_from_edge-counting aux_attempt_to xcomp_vary_attempt aux_factor_to xcomp_attempt_factor prep_factor_in pobj_in_peculiarities prep_peculiarities_of det_structure_the nn_structure_network pobj_of_structure prep_factor_by pcomp_by_considering nn_direction_link dobj_considering_direction amod_path_relative conj_direction_path cc_direction_and conj_direction_density dep_as_such prep_considering_as pobj_as_vector conj_vector_lesk conj_vector_hso conj_vector_lch conj_vector_wup conj_vector_path conj_vector_res conj_vector_lin cc_vector_and conj_vector_jcn ' +o,1633,'Variations of SCFGs go back to Aho and Ullman s Syntax-Directed Translation Schemata , but also include the Inversion Transduction Grammars in , which restrict grammar rules to be binary , the synchronous grammars in Chiang -LRB- 2005 -RRB- , which use only a single nonterminal symbol , and the Multitext Grammars in Melamed , which allow independent rewriting , as well as other tree-based models such as Yamada and Knight and Galley et al ',Wu,'nsubj_go_Variations prep_Variations_of pobj_of_SCFGs advmod_go_back prep_go_to pobj_to_Aho cc_Aho_and nn_Schemata_Ullman nn_Schemata_ nn_Schemata_s nn_Schemata_Syntax-Directed nn_Schemata_Translation conj_Aho_Schemata cc_go_but advmod_include_also conj_go_include det_Grammars_the nn_Grammars_Inversion nn_Grammars_Transduction dobj_include_Grammars prep_Grammars_in pobj_in_ nsubj_restrict_which rcmod__restrict nn_rules_grammar nsubj_binary_rules aux_binary_to cop_binary_be xcomp_restrict_binary det_grammars_the amod_grammars_synchronous conj_Grammars_grammars prep_grammars_in pobj_in_Chiang appos_Chiang_2005 nsubj_use_which rcmod_Chiang_use advmod_symbol_only det_symbol_a amod_symbol_single amod_symbol_nonterminal dobj_use_symbol cc_Grammars_and det_Grammars_the nn_Grammars_Multitext conj_Grammars_Grammars prep_Grammars_in nn__Melamed pobj_in_ nsubj_allow_which rcmod__allow amod_rewriting_independent dobj_allow_rewriting dep_well_as cc_Grammars_well dep_well_as amod_models_other amod_models_tree-based conj_Grammars_models dep_as_such prep_models_as nn__Yamada cc_Yamada_and conj_Yamada_Knight pobj_as_ cc__and conj__Galley cc_Galley_et conj_Galley_al ' +o,1634,'The interest reader is referred to , for a summary of ARIOSTO , an integrated tool for extensive acquisition of lexieal knowledge from corpora that we used to demonstrate and validate our approach ',Dunning,'det_reader_The nn_reader_interest nsubjpass_referred_reader auxpass_referred_is aux__to xcomp_referred_ prep__for det_summary_a pobj_for_summary prep_summary_of pobj_of_ARIOSTO det_tool_an amod_tool_integrated appos_ARIOSTO_tool prep_tool_for amod_acquisition_extensive pobj_for_acquisition prep_tool_of amod_knowledge_lexieal pobj_of_knowledge prep_tool_from pobj_from_corpora complm_used_that nsubj_used_we ccomp_referred_used aux_demonstrate_to xcomp_used_demonstrate cc_demonstrate_and conj_demonstrate_validate poss_approach_our dobj_demonstrate_approach ' +o,1635,'53 Baseline System We conducted experiments using different segmenters with a standard log-linear PB-SMT model : GIZA + + implementation of IBM word alignment model 4 , the refinement and phrase-extraction heuristics described in , minimum-errorrate training , a 5-gram language model with Kneser-Ney smoothing trained with SRILM on the English side of the training data , and Moses to translate both single best segmentation and word lattices ',Och,'num_System_53 nn_System_Baseline nsubj_conducted_We dep_System_conducted dobj_conducted_experiments partmod_experiments_using amod_segmenters_different dobj_using_segmenters prep_using_with det_model_a amod_model_standard amod_model_log-linear nn_model_PB-SMT pobj_with_model amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dep_model_implementation prep_implementation_of nn_model_IBM nn_model_word amod_model_alignment pobj_of_model dep__4 amod_implementation_ det_heuristics_the amod_heuristics_refinement cc_refinement_and conj_refinement_phrase-extraction appos_implementation_heuristics partmod_heuristics_described prep_described_in amod__ amod__minimum-errorrate nn__training pobj_in_ det_model_a amod_model_5-gram nn_model_language appos__model prep_model_with amod_smoothing_Kneser-Ney pobj_with_smoothing partmod_smoothing_trained prep_trained_with nn__SRILM pobj_with_ prep__on det_side_the amod_side_English pobj_on_side prep_side_of det_data_the nn_data_training pobj_of_data cc_conducted_and nsubj__Moses conj_conducted_ aux_translate_to xcomp__translate preconj_segmentation_both amod_segmentation_single dep_single_best dobj_translate_segmentation cc_segmentation_and nn_lattices_word conj_segmentation_lattices ' +o,1636,'The prior probability P0 is the prior distribution for the phrase probability which is estimated using the phrase normalized counts commonly used in conventional Phrasebased SMT systems , eg , ',Koehn,'det_P0_The amod_P0_prior nn_P0_probability nsubj_distribution_P0 cop_distribution_is det_distribution_the amod_distribution_prior prep_distribution_for det_probability_the nn_probability_phrase pobj_for_probability nsubjpass_estimated_which auxpass_estimated_is rcmod_distribution_estimated xcomp_estimated_using det_phrase_the nsubj_normalized_phrase ccomp_using_normalized dobj_normalized_counts advmod_used_commonly partmod_counts_used prep_used_in amod_systems_conventional nn_systems_Phrasebased nn_systems_SMT pobj_in_systems appos_systems_eg ' +o,1637,'Collocation map that is first suggested in is a sigmoid belief network with words as probabilistic variables ',Dunning,'nn_map_Collocation nsubj_network_map nsubjpass_suggested_that auxpass_suggested_is advmod_suggested_first rcmod_map_suggested prep_suggested_in pobj_in_ cop_network_is det_network_a amod_network_sigmoid nn_network_belief prep_network_with pobj_with_words prep_words_as amod_variables_probabilistic pobj_as_variables ' +o,1638,'The model of incorporated a latent variable for named entity class ',Haghighi,'det_model_The prep_model_of pobj_of_ partmod__incorporated det_variable_a amod_variable_latent dobj_incorporated_variable prep_variable_for amod_class_named nn_class_entity pobj_for_class ' +o,1639,'CIT -RRB- ',Wu,'nn_-RRB-_CIT ' +o,1640,'We thus introduce a multiplier to form the actual objective function that we minimize with respect to :4 summationdisplay iL logp , i -LRB- yi -RRB- + Nsummationdisplay inegationslashL H -LRB- p , i -RRB- -LRB- 4 -RRB- One may regard as a Lagrange multiplier that is used to constrain the classifiers uncertainty H to be low , as presented in the work on entropy regularization ',Jiao,'nsubj_introduce_We advmod_introduce_thus det_multiplier_a dobj_introduce_multiplier aux_form_to infmod_multiplier_form det_function_the amod_function_actual nn_function_objective dobj_form_function complm_minimize_that nsubj_minimize_we ccomp_form_minimize prep_minimize_with pobj_with_respect prep_minimize_to num_logp_:4 amod_logp_summationdisplay nn_logp_iL pobj_to_logp appos_logp_i appos_logp_yi nn_H_+ nn_H_Nsummationdisplay nn_H_inegationslashL dep_logp_H dep_multiplier_p appos_p_i dep_regard_4 nsubj_regard_One aux_regard_may dep_introduce_regard prep_regard_as det_multiplier_a nn_multiplier_Lagrange pobj_as_multiplier nsubjpass_used_that auxpass_used_is rcmod_multiplier_used aux_constrain_to xcomp_used_constrain det_H_the amod_H_classifiers nn_H_uncertainty nsubj_low_H aux_low_to cop_low_be xcomp_constrain_low mark_presented_as advcl_low_presented prep_presented_in det_work_the pobj_in_work prep_work_on amod_regularization_entropy pobj_on_regularization ' +o,1641,'Much work has gone into methods for measuring synset similarity ; early work in this direction includes , which attempted to discover sense similarities between dictionary senses ',Dolan,'amod_work_Much nsubj_gone_work aux_gone_has prep_gone_into pobj_into_methods prep_methods_for pcomp_for_measuring amod_similarity_synset dobj_measuring_similarity amod_work_early nsubj_includes_work prep_work_in det_direction_this pobj_in_direction parataxis_gone_includes dobj_includes_ nsubj_attempted_which rcmod__attempted aux_discover_to xcomp_attempted_discover nn_similarities_sense dobj_discover_similarities prep_similarities_between nn_senses_dictionary pobj_between_senses ' +o,1642,' considered some location constrains in meeting summarization evaluation , which utilizes speaker information to some extent ',Galley,'nsubj_considered_ det_location_some nsubj_constrains_location ccomp_considered_constrains prep_constrains_in nn_evaluation_meeting nn_evaluation_summarization pobj_in_evaluation nsubj_utilizes_which rcmod_evaluation_utilizes nn_information_speaker dobj_utilizes_information prep_utilizes_to det_extent_some pobj_to_extent ' +o,1643,'In contrast to the semi-supervised LEAF alignment algorithm of , which requires 1,5002,000 CPU days per iteration to align 84M ChineseEnglish sentences -LRB- anonymous , pc -RRB- , link deletion requires only 450 CPU hours to re-align such a corpus -LRB- after initial alignment by GIZA + + , which requires 20-24 CPU days -RRB- ',Fraser,'prep_requires_In dep_In_contrast dep_In_to det_algorithm_the amod_algorithm_semi-supervised nn_algorithm_LEAF nn_algorithm_alignment pobj_In_algorithm prep_algorithm_of pobj_of_ nsubj_requires_which dep_requires_requires num_days_1,5002,000 nn_days_CPU nsubj_align_days prep_days_per pobj_per_iteration aux_align_to xcomp_requires_align num_sentences_84M amod_sentences_ChineseEnglish dobj_align_sentences appos_sentences_anonymous dep_anonymous_pc nn_deletion_link nsubj_requires_deletion quantmod_450_only num_hours_450 nn_hours_CPU dobj_requires_hours aux_re-align_to xcomp_requires_re-align predet_corpus_such det_corpus_a dobj_re-align_corpus dep_requires_after amod_alignment_initial pobj_after_alignment prep_alignment_by nn_+_GIZA nn_+_+ pobj_by_+ nsubj_requires_which rcmod_+_requires num_days_20-24 nn_days_CPU tmod_requires_days ' +o,1644,'There are more sophisticated surface generation packages , such as FUF\\/SURGE , KPML , MUMBLE , and RealPro , which produce natural language text from an abstract semantic representation ',Berger,'expl_are_There advmod_sophisticated_more amod_packages_sophisticated nn_packages_surface nn_packages_generation nsubj_are_packages dep_as_such dep_packages_as nn__FUF\\/SURGE pobj_as_ nn__KPML conj__ nn__MUMBLE conj__ cc__and nn__RealPro conj__ nsubj_produce_which rcmod_packages_produce amod_text_natural nn_text_language dobj_produce_text prep_produce_from det_representation_an amod_representation_abstract amod_representation_semantic pobj_from_representation ' +o,1645,'Since so many concepts used in discourse are graindependent , a theory of granularity is also fundamental ',Hobbs,'mark_graindependent_Since advmod_concepts_so amod_concepts_many nsubj_graindependent_concepts partmod_concepts_used prep_used_in pobj_in_discourse cop_graindependent_are advcl_fundamental_graindependent det_theory_a nsubj_fundamental_theory prep_theory_of pobj_of_granularity cop_fundamental_is advmod_fundamental_also ' +o,1646,'63 Comparison with re-ranking approach Finally , we compared our algorithm with the reranking approach , where we rst generate the n-best candidates using a model with only local features -LRB- the rst model -RRB- and then re-rank the candidates using a model with non-local features -LRB- the second model -RRB- ',Collins,'num_Comparison_63 prep_compared_with pcomp_with_re-ranking dobj_re-ranking_approach advmod_re-ranking_Finally nsubj_compared_we dep_Comparison_compared poss_algorithm_our dobj_compared_algorithm prep_compared_with det__the amod__reranking nn__approach pobj_with_ advmod_rst_where nsubj_rst_we rcmod__rst xcomp_rst_generate det_candidates_the amod_candidates_n-best dobj_generate_candidates dep_Comparison_using det_model_a dobj_using_model prep_using_with advmod_features_only amod_features_local pobj_with_features det_model_the nn_model_rst appos_features_model cc_using_and advmod_using_then conj_using_re-rank det_candidates_the nsubj_using_candidates dep_re-rank_using det_model_a dobj_using_model prep_using_with amod_features_non-local pobj_with_features det_model_the amod_model_second appos_features_model ' +o,1647,'For this paper , we used POS tags that were provided either by the Treebank itself -LRB- gold standard tags -RRB- or by the perceptron POS tagger3 presented in ',Collins,'prep_used_For det_paper_this pobj_For_paper nsubj_used_we dep_tags_POS nsubj_presented_tags nsubjpass_provided_that auxpass_provided_were rcmod_tags_provided preconj_by_either prep_provided_by det_Treebank_the pobj_by_Treebank advmod_Treebank_itself amod_tags_gold amod_tags_standard appos_Treebank_tags cc_by_or conj_by_by det_tagger3_the nn_tagger3_perceptron dep_tagger3_POS pobj_by_tagger3 ccomp_used_presented prt_presented_in ' +o,1648,'The approach presented here has some resemblance to the bracketing transduction grammars -LRB- BTG -RRB- of , which have been applied to a phrase-based machine translation system in ',Wu,'det_approach_The nsubj_has_approach partmod_approach_presented advmod_presented_here det_resemblance_some dobj_has_resemblance prep_resemblance_to det_grammars_the amod_grammars_bracketing nn_grammars_transduction pobj_to_grammars abbrev_grammars_BTG prep_grammars_of pobj_of_ nsubjpass_applied_which aux_applied_have auxpass_applied_been rcmod__applied prep_applied_to det_system_a amod_system_phrase-based nn_system_machine nn_system_translation pobj_to_system prep_system_in pobj_in_ ' +o,1649,'In , an undirected graphical model for constituent parse reranking uses dependency relations to define the edges ',Koo,'prep_uses_In pobj_In_ det_model_an amod_model_undirected amod_model_graphical nsubj_uses_model prep_model_for amod_parse_constituent pobj_for_parse amod_parse_reranking amod_relations_dependency dobj_uses_relations aux_define_to xcomp_uses_define det_edges_the dobj_define_edges ' +o,1650,'For an HMM with a set of states T and a set of output symbols V : t T t Dir -LRB- 1 , T -RRB- -LRB- 1 -RRB- t T t Dir -LRB- 1 , V -RRB- -LRB- 2 -RRB- ti ti1 , ti1 Multi -LRB- ti1 -RRB- -LRB- 3 -RRB- wi ti , ti Multi -LRB- ti -RRB- -LRB- 4 -RRB- One advantage of the Bayesian approach is that the prior allows us to bias learning toward sparser structures , by setting the Dirichlet hyperparameters , to a value less than one ',Johnson,'prep_t_For det_T_an amod_T_HMM prep_HMM_with det_set_a pobj_with_set prep_set_of pobj_of_states pobj_For_T cc_T_and det_set_a conj_T_set prep_set_of nn_symbols_output pobj_of_symbols nsubj_t_V nn_Dir_t nn_Dir_T nn_Dir_t dep_V_Dir appos_Dir_1 dep_1_T appos_V_1 nn_Dir_T nn_Dir_t dobj_t_Dir appos_Dir_1 dep_1_V dep_ti_2 dep_Dir_ti num_ti_ti1 num_ti_ti1 prep_ti_Multi appos_Multi_ti1 dep_ti_3 amod_ti_wi dep_Multi_ti nn_Multi_ti appos_ti_Multi appos_Multi_ti dep_is_4 num_advantage_One nsubj_is_advantage prep_advantage_of det_approach_the nn_approach_Bayesian pobj_of_approach dep_t_is complm_allows_that det_prior_the nsubj_allows_prior ccomp_is_allows dobj_allows_us prep_allows_to pobj_to_bias partmod_bias_learning prep_learning_toward nn_structures_sparser pobj_toward_structures prep_learning_by pcomp_by_setting det_hyperparameters_the nn_hyperparameters_Dirichlet dobj_setting_hyperparameters prep_learning_to det_value_a pobj_to_value dep_than_less quantmod_one_than num_value_one ' +o,1651,'The above observations can be stated formally from the perspective of Model 2 ',Brown,'det_observations_The amod_observations_above nsubjpass_stated_observations aux_stated_can auxpass_stated_be advmod_stated_formally prep_stated_from det_perspective_the pobj_from_perspective prep_perspective_of amod_Model_ pobj_of_Model num_Model_2 ' +o,1652,'The progress in parsing technology are noteworthy , and in particular , various statistical dependency models have been proposed , , , ',Collins,'det_progress_The nsubj_noteworthy_progress prep_progress_in amod_technology_parsing pobj_in_technology cop_noteworthy_are cc_noteworthy_and prep_proposed_in pobj_in_particular amod_models_various amod_models_statistical nn_models_dependency nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been conj_noteworthy_proposed acomp_proposed_ dep__ advmod_proposed_ ' +o,1653,'Statistical and information theoretic approaches , , , Using lexical collocations to determine PPA with statistical techniques was first proposed by ',Marcus,'nn_approaches_Statistical cc_Statistical_and conj_Statistical_information nn_approaches_theoretic nn__ appos__ appos__ nsubjpass_proposed_ partmod__Using amod_collocations_lexical dobj_Using_collocations aux_determine_to xcomp_Using_determine dobj_determine_PPA prep_PPA_with amod_techniques_statistical pobj_with_techniques auxpass_proposed_was advmod_proposed_first dep_approaches_proposed prep_proposed_by pobj_by_ ' +n,1654,'One prominent constraint of the IBM word alignment models is functional alignment , that is each target word is mapped onto at most one source word ',Brown,'num_constraint_One amod_constraint_prominent nsubj_alignment_constraint prep_constraint_of det__the nn__IBM nn__word nn__alignment nn__models pobj_of_ cop_alignment_is amod_alignment_functional dep_mapped_alignment nsubj_word_that cop_word_is det_word_each nn_word_target rcmod_alignment_word auxpass_mapped_is prep_mapped_onto dep_onto_at dep_at_most num_word_one nn_word_source nsubjpass_mapped_word ' +o,1655,'Furthermore , recent studies revealed that word clustering is useful for semi-supervised learning in NLP ',Kazama,'advmod_revealed_Furthermore amod_studies_recent nsubj_revealed_studies complm_useful_that nn_clustering_word nsubj_useful_clustering cop_useful_is ccomp_revealed_useful prep_useful_for pobj_for_semi-supervised xcomp_useful_learning prep_learning_in pobj_in_NLP ' +o,1656,' , Johansson and Nugues , Prokopidis et al ',Marcus,'nn_Prokopidis_Johansson cc_Johansson_and conj_Johansson_Nugues nn_Prokopidis_ appos__Prokopidis cc_Prokopidis_et conj_Prokopidis_al ' +o,1657,'2.2 Corpus occurrence In order to get a feel for the relative frequency of VPCs in the corpus targeted for extraction, namely 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 VPC types (\%) Corpus frequency Figure 1: Frequency distribution of VPCs in the WSJ Tagger correctextracted Prec Rec Ffl=1 Brill 135135 1.000 0.177 0.301 Penn 667800 0.834 0.565 0.673 Table 1: POS-based extraction results the WSJ section of the Penn Treebank, we took a random sample of 200 VPCs from the Alvey Natural Language Tools grammar (Grover et al. , 1993) and did a manual corpus search for each.',Dunning,'' +o,1658,'Online discriminative training has already been studied by Tillmann and Zhang and et al ',Liang,'amod_training_Online nn_training_discriminative nsubjpass_studied_training aux_studied_has advmod_studied_already auxpass_studied_been prep_studied_by nn__Tillmann cc_Tillmann_and conj_Tillmann_Zhang pobj_by_ cc__and conj__ cc__et conj__al ' +o,1659,'Accordingly, in this section we describe a set of experiments which extends the work of (Way and Gough, 2005) by evaluating the Marker-based EBMT system of (Gough & Way, 2004b) against a phrase-based SMT system built using the following components: Giza++, to extract the word-level correspondences; The Giza++ word alignments are then refined and used to extract phrasal alignments ((Och & Ney, 2003); or (Koehn et al. , 2003) for a more recent implementation); Probabilities of the extracted phrases are calculated from relative frequencies; The resulting phrase translation table is passed to the Pharaoh phrase-based SMT decoder which along with SRI language modelling toolkit5 performs translation.',Och,'' +p,1660,' has been unable to find real examples of cases where hierarchical alignment would fail under these conditions , at least in fixed-word-order languages that are lightly inflected , such as English and Chinese -LRB- p 385 -RRB- ',Wu,'nsubj_unable_ aux_unable_has dep_unable_been aux_find_to xcomp_unable_find amod_examples_real dobj_find_examples prep_examples_of pobj_of_cases advmod_fail_where amod_alignment_hierarchical nsubj_fail_alignment aux_fail_would rcmod_cases_fail prep_fail_under det_conditions_these pobj_under_conditions advmod_fail_at dep_at_least dep_at_in amod_languages_fixed-word-order pobj_in_languages nsubjpass_inflected_that auxpass_inflected_are advmod_inflected_lightly rcmod_languages_inflected dep_as_such prep_languages_as pobj_as_English cc_English_and conj_English_Chinese number_385_p dep_cases_385 ' +n,1661,'Although the first three are particular cases where N = 1 and\\/or M = 1 , the distinction is relevant , because most word-based translation models -LRB- eg IBM models -RRB- can typically not accommodate general M-N alignments ',Brown,'mark_cases_Although det_three_the amod_three_first nsubj_cases_three cop_cases_are amod_cases_particular advcl_relevant_cases advmod_1_where nsubj_1_N dep_1_= rcmod_cases_1 cc_cases_and\\/or conj_cases_M dep_1_= ccomp_M_1 det_distinction_the nsubj_relevant_distinction cop_relevant_is mark_accommodate_because advmod_word-based_most amod_models_word-based nn_models_translation nsubj_accommodate_models amod__eg nn__IBM nn__models abbrev_models_ aux_accommodate_can advmod_accommodate_typically neg_accommodate_not advcl_relevant_accommodate amod_alignments_general amod_alignments_M-N dobj_accommodate_alignments ' +o,1662,'2 Related Work There has been a large and diverse body of research in opinion mining , with most research at the text , sentence or word level ',Turney,'num_Work_2 amod_Work_Related expl_been_There aux_been_has dep_Work_been det_body_a amod_body_large cc_large_and conj_large_diverse dobj_been_body prep_body_of pobj_of_research prep_research_in nn_mining_opinion pobj_in_mining prep_been_with amod_research_most pobj_with_research prep_been_at det__the nn__text pobj_at_ nn__sentence conj__ cc__or nn_level_word nn_level_ conj__level ' +o,1663,'Most previous work with CRFs containing nonlocal dependencies used approximate probabilistic inference techniques , including TRP and Gibbs sampling ',Lin,'amod_work_Most amod_work_previous nsubj_used_work prep_work_with pobj_with_CRFs partmod_CRFs_containing amod_dependencies_nonlocal dobj_containing_dependencies amod_techniques_approximate amod_techniques_probabilistic nn_techniques_inference dobj_used_techniques prep_used_including nn__TRP pobj_including_ cc__and conj__Gibbs partmod__sampling dobj_sampling_ ' +o,1664,'A more fine-grained distinction is made by and Vieira and Poesio -LRB- 2000 -RRB- to distinguish restrictive from non-restrictive postmodification by ommitting those modifiers that occur between commas , which should not be classified as chain starting ',Bean,'det_distinction_A advmod_fine-grained_more amod_distinction_fine-grained nsubjpass_made_distinction auxpass_made_is prep_made_by pobj_by_ cc__and conj__Vieira cc_Vieira_and conj_Vieira_Poesio appos_Vieira_2000 aux_distinguish_to xcomp_made_distinguish acomp_distinguish_restrictive prep_restrictive_from amod_postmodification_non-restrictive pobj_from_postmodification prep_distinguish_by pcomp_by_ommitting det_modifiers_those dobj_ommitting_modifiers nsubj_occur_that rcmod_modifiers_occur prep_occur_between pobj_between_commas nsubjpass_classified_which aux_classified_should neg_classified_not auxpass_classified_be rcmod_commas_classified prep_classified_as pobj_as_chain partmod_chain_starting ' +o,1665,'32 Training Algorithm We adopt the perceptron training algorithm of to learn a discriminative model mapping from inputs xX to outputs yY , where X is the set of sentences in the training corpus and Y is the set of corresponding labeled results ',Collins,'num_Algorithm_32 nn_Algorithm_Training nsubj_adopt_We dep_Algorithm_adopt det_algorithm_the nn_algorithm_perceptron nn_algorithm_training nsubj_learn_algorithm prep_algorithm_of pobj_of_ aux_learn_to xcomp_adopt_learn det_mapping_a amod_mapping_discriminative nn_mapping_model dobj_learn_mapping prep_learn_from amod_xX_inputs pobj_from_xX prep_learn_to amod_yY_outputs pobj_to_yY advmod_set_where nsubj_set_X cop_set_is det_set_the advcl_adopt_set prep_set_of pobj_of_sentences prep_sentences_in det_corpus_the nn_corpus_training pobj_in_corpus cc_adopt_and nsubj_set_Y cop_set_is det_set_the conj_adopt_set prep_set_of amod_results_corresponding amod_results_labeled pobj_of_results ' +p,1666,'1 Introduction Phrase-based method and syntaxbased method represent the state-of-the-art technologies in statistical machine translation -LRB- SMT -RRB- ',Shen,'num__1 nn__Introduction amod__Phrase-based nn__method nsubj_represent_ cc__and nn__syntaxbased nn__method conj__ det_technologies_the amod_technologies_state-of-the-art dobj_represent_technologies prep_technologies_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +o,1667,'Since so many concepts used in discourse are graindependent , a theory of granularity is also fundamental ',Hobbs,'mark_graindependent_Since advmod_concepts_so amod_concepts_many nsubj_graindependent_concepts partmod_concepts_used prep_used_in pobj_in_discourse cop_graindependent_are advcl_fundamental_graindependent det_theory_a nsubj_fundamental_theory prep_theory_of pobj_of_granularity cop_fundamental_is advmod_fundamental_also ' +o,1668,'22 Three Treebanks The Treebanks that we used in this paper are the English Penn Treebank II , the Chinese Penn Treebank , and the Korean Penn Treebank ',Marcus,'number_Three_22 num_Treebanks_Three det_Treebanks_The nsubj__Treebanks complm_used_that nsubj_used_we dep_Treebanks_used prep_used_in det_paper_this pobj_in_paper cop__are det__the nn__English nn__Penn nn__Treebank nn__II dep_Treebanks_ det__the amod__Chinese nn__Penn nn__Treebank conj__ cc__and det__the amod__Korean nn__Penn nn__Treebank conj__ ' +o,1669,'One of the theoretical problems with phrase based SMT models is that they can not effectively model the discontiguous translations and numerous attempts have been made on this issue ',Wellington,'nsubjpass_made_One prep_One_of det_problems_the amod_problems_theoretical pobj_of_problems prep_problems_with pobj_with_phrase partmod_phrase_based nn_models_SMT nsubj_is_models ccomp_based_is complm_model_that nsubj_model_they aux_model_can neg_model_not advmod_model_effectively ccomp_is_model det_translations_the amod_translations_discontiguous dobj_model_translations cc_translations_and amod_attempts_numerous conj_translations_attempts aux_made_have auxpass_made_been prep_made_on det_issue_this pobj_on_issue ' +o,1670,'Furthermore , we use averaged weights in Algorithm 1 ',Collins,'advmod_use_Furthermore nsubj_use_we ccomp_use_averaged nn__weights dobj_averaged_ prep_averaged_in pobj_in_Algorithm num_Algorithm_1 ' +o,1671,'Alternatively , one can train them with respect to the final translation quality measured by some error criterion ',Och,'advmod_train_Alternatively nsubj_train_one aux_train_can dobj_train_them prep_train_with pobj_with_respect prep_train_to det_quality_the amod_quality_final nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_by det_criterion_some nn_criterion_error pobj_by_criterion ' +o,1672,'(Blitzer et al., 2006; Jiang and Zhai, 2007).',Blitzer,'' +o,1673,'They propose two modifications to f-measure: varying the precision/recall tradeoff, and fully-connecting the alignment links before computing f-measure.11 Weighted Fully-Connected F-Measure Given a hypothesized set of alignment links H and a goldstandard set of alignment links G, we define H+ = fullyConnect(H) and G+ = fullyConnect(G), and then compute: f-measure(H+) = 1 precision(H+) + 1 recall(H+) For phrase-based Chinese-English and ArabicEnglish translation tasks, (Fraser and Marcu, 2007a) obtain the closest correlation between weighted fully-connected alignment f-measure and BLEU score using =0.5 and =0.1, respectively.',Fraser,'' +o,1674,'One of the steps in the analysis of English is named entity recognition using Stanford Named Entity Recognizer ',Finkel,'nsubjpass_named_One prep_One_of det_steps_the pobj_of_steps prep_steps_in det_analysis_the pobj_in_analysis prep_analysis_of pobj_of_English auxpass_named_is nn_recognition_entity dobj_named_recognition xcomp_named_using nn_Recognizer_Stanford nn_Recognizer_Named nn_Recognizer_Entity dobj_using_Recognizer ' +o,1675,' and -RRB- ',Cutting,'nn__ nn__and ' +o,1676,'52 Maximum Entropy Model We use the Maximum Entropy -LRB- ME -RRB- Model for our classification task ',Berger,'num_Model_52 nn_Model_Maximum nn_Model_Entropy nsubj_use_We dep_Model_use det__the nn_Model_Maximum nn_Model_Entropy abbrev_Model_ME nn__Model dobj_use_ prep__for poss_task_our nn_task_classification pobj_for_task ' +o,1677,'TheChinesesentencefromtheselected pair is used as the single reference to tune and evaluate the MT system with word-based BLEU-4 ',Papineni,'amod_pair_TheChinesesentencefromtheselected nsubjpass_used_pair auxpass_used_is prep_used_as det_reference_the amod_reference_single pobj_as_reference aux_tune_to xcomp_used_tune cc_tune_and conj_tune_evaluate det_system_the nn_system_MT dobj_tune_system prep_tune_with amod_BLEU-4_word-based pobj_with_BLEU-4 ' +o,1678,'The reliability of the annotations was checked using the kappa statistic ',Carletta,'det_reliability_The nsubjpass_checked_reliability prep_reliability_of det_annotations_the pobj_of_annotations auxpass_checked_was xcomp_checked_using det_statistic_the nn_statistic_kappa dobj_using_statistic ' +o,1679,'A major issue in MaxEnt training is how to select proper features and determine the feature targets ',Berger,'det_issue_A amod_issue_major nsubj_is_issue prep_issue_in nn_training_MaxEnt pobj_in_training advmod_select_how aux_select_to ccomp_is_select amod_features_proper dobj_select_features cc_select_and conj_select_determine det_targets_the nn_targets_feature dobj_determine_targets ' +o,1680,'For example , our system configuration for the shared task incorporates a wrapper around GIZA + + for word alignment and a wrapper around Moses for decoding ',Koehn,'prep_incorporates_For pobj_For_example poss_configuration_our nn_configuration_system nsubj_incorporates_configuration prep_configuration_for det_task_the amod_task_shared pobj_for_task det_wrapper_a dobj_incorporates_wrapper prep_incorporates_around nn__GIZA nn__+ nn__+ pobj_around_ prep__for nn_alignment_word pobj_for_alignment cc__and det_wrapper_a conj__wrapper prep_wrapper_around nn__Moses pobj_around_ prep__for pobj_for_decoding ' +o,1681,' and Chan and Ng -LRB- 2008 -RRB- use WordNet , and Zhou et al ',Banerjee,'nsubj_use_ cc__and conj__Chan cc_Chan_and conj_Chan_Ng appos_Chan_2008 dobj_use_WordNet cc_WordNet_and conj_WordNet_Zhou cc_Zhou_et conj_Zhou_al ' +o,1682,'The solution we employ here is the discriminative training procedure of ',Och,'det_solution_The nsubj_procedure_solution nsubj_employ_we rcmod_solution_employ advmod_employ_here cop_procedure_is det_procedure_the amod_procedure_discriminative nn_procedure_training prep_procedure_of ' +o,1683,' invented heuristic symmetriza57 FRENCH\\/ENGLISH ARABIC\\/ENGLISH SYSTEM F-MEASURE -LRB- = 04 -RRB- BLEU F-MEASURE -LRB- = 01 -RRB- BLEU GIZA + + 735 3063 758 5155 741 3140 791 5289 LEAF UNSUPERVISED 745 723 LEAF SEMI-SUPERVISED 763 3186 845 5434 Table 3 : Experimental Results tion of the output of a 1-to-N model and a M-to-1 model resulting in a M-to-N alignment , this was extended in ',Koehn,'advmod_invented_ amod_symmetriza57_heuristic dobj_invented_symmetriza57 nn_F-MEASURE_FRENCH\\/ENGLISH nn_F-MEASURE_ARABIC\\/ENGLISH nn_F-MEASURE_SYSTEM nsubj_invented_F-MEASURE number_04_= dep_F-MEASURE_04 amod_F-MEASURE_BLEU dep_F-MEASURE_F-MEASURE number_01_= dep_F-MEASURE_01 nn_GIZA_BLEU nsubjpass_extended_GIZA amod_GIZA_+ number_5434_+ cc_5434_735 conj_5434_3063 number_5434_758 number_5434_5155 number_5434_ number_5434_741 number_5434_3140 number_5434_791 number_5434_5289 number_5434_LEAF number_5434_UNSUPERVISED number_5434_745 cc_5434_723 conj_5434_LEAF number_5434_SEMI-SUPERVISED number_5434_763 number_5434_3186 number_5434_845 dep_+_5434 cc_5434_Table conj_5434_3 nn_tion_Experimental nn_tion_Results dep_5434_tion prep_tion_of det_output_the pobj_of_output prep_output_of det_model_a amod_model_1-to-N pobj_of_model cc_output_and det_model_a amod_model_M-to-1 conj_output_model partmod_model_resulting prep_resulting_in det_alignment_a amod_alignment_M-to-N pobj_in_alignment appos_alignment_this auxpass_extended_was rcmod_F-MEASURE_extended prep_extended_in ' +o,1684,'Independently , in AI an effort arose to encode large amounts of commonsense knowledge ',Hobbs,'dep_arose_Independently dep_Independently_in pobj_in_AI det_effort_an nsubj_arose_effort aux_encode_to xcomp_arose_encode amod_amounts_large dobj_encode_amounts prep_amounts_of amod_knowledge_commonsense pobj_of_knowledge ' +p,1685,'1 Introduction Phrase-based translation and hierarchical phrase-based translation are the state of the art in statistical machine translation -LRB- SMT -RRB- techniques ',Koehn,'num__1 amod__Introduction amod__Phrase-based nn__translation nsubj_state_ cc__and amod__hierarchical amod__phrase-based nn__translation conj__ cop_state_are det_state_the prep_state_of det_art_the nn_techniques_art prep_art_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT pobj_of_techniques ' +p,1686,'Although the Kappa coefficient has a number of advantages over percentage agreement -LRB- eg , it takes into account the expected chance interrater agreement ; see for details -RRB- , we also report percentage agreement as it allows us to compare straightforwardly the human performance and the automatic methods described below , whose performance will also be reported in terms of percentage agreement ',Carletta,'mark_has_Although det_coefficient_the nn_coefficient_Kappa nsubj_has_coefficient advcl_report_has det_number_a dobj_has_number prep_number_of pobj_of_advantages prep_advantages_over nn_agreement_percentage pobj_over_agreement nsubj_takes_eg nsubj_takes_it dep_advantages_takes prep_takes_into pobj_into_account det_agreement_the amod_agreement_expected nn_agreement_chance nn_agreement_interrater dobj_takes_agreement dep_agreement_see dobj_see_ prep__for pobj_for_details nsubj_report_we advmod_report_also nn_agreement_percentage dobj_report_agreement mark_allows_as nsubj_allows_it advcl_report_allows nsubj_compare_us aux_compare_to xcomp_allows_compare advmod_compare_straightforwardly det_performance_the amod_performance_human nsubj_described_performance cc_performance_and det_methods_the amod_methods_automatic conj_performance_methods dep_report_described advmod_described_below poss_performance_whose nsubjpass_reported_performance aux_reported_will advmod_reported_also auxpass_reported_be ccomp_described_reported prep_reported_in pobj_in_terms prep_terms_of nn_agreement_percentage pobj_of_agreement ' +o,1687,'Different methods have been proposed to reduce error propagation between pipelined tasks , both in general and for specific problems such as language modeling and utterance classification and labeling and chunking ',Finkel,'amod_methods_Different nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been aux_reduce_to xcomp_proposed_reduce nn_propagation_error dobj_reduce_propagation prep_propagation_between amod_tasks_pipelined pobj_between_tasks preconj_in_both prep_reduce_in amod__general pobj_in_ cc_in_and conj_in_for amod_problems_specific pobj_for_problems dep_as_such prep_problems_as nn_modeling_language pobj_as_modeling cc_modeling_and nn__utterance nn__classification conj_modeling_ cc_modeling_and conj_modeling_labeling cc_modeling_and nn__chunking conj_modeling_ ' +o,1688,'Both data were extracted from the Penn Treebank Wall Street Journal -LRB- WSJ -RRB- Corpus ',Marcus,'det_data_Both auxpass_extracted_were rcmod_data_extracted prep_extracted_from det_Wall_the nn_Wall_Penn nn_Wall_Treebank pobj_from_Wall nn_Corpus_Street nn_Corpus_Journal abbrev_Corpus_WSJ dep_data_Corpus ' +o,1689,'This preprocessing step can be accomplished by applying the GIZA + + toolkit that provides Viterbi alignments based on IBM Model-4 ',Och,'det_step_This amod_step_preprocessing nsubjpass_accomplished_step aux_accomplished_can auxpass_accomplished_be prep_accomplished_by pcomp_by_applying det__the nn__GIZA amod__+ cc_+_+ conj_+_toolkit dobj_applying_ nsubj_provides_that rcmod__provides nn_alignments_Viterbi dobj_provides_alignments prep_provides_based dep_based_on nn_Model-4_IBM pobj_on_Model-4 ' +o,1690,'Research on the automatic classification of movie or product reviews as positive or negative -LRB- eg , -RRB- is perhaps the most similar to our work ',Pang,'nsubj_similar_Research prep_Research_on det_classification_the amod_classification_automatic pobj_on_classification prep_classification_of nn_reviews_movie cc_movie_or conj_movie_product pobj_of_reviews advmod_positive_as amod_reviews_positive cc_positive_or conj_positive_negative dep_positive_eg dep_eg_ cop_similar_is advmod_similar_perhaps det_similar_the advmod_similar_most prep_similar_to poss_work_our pobj_to_work ' +o,1691,'The de-facto answer came during the 1990s from the research community on Statistical Machine Translation , who made use of statistical tools based on a noisy channel model originally developed for speech recognition ',Koehn,'det_answer_The amod_answer_de-facto nsubj_came_answer prep_came_during det_1990s_the pobj_during_1990s prep_came_from det_community_the nn_community_research pobj_from_community prep_came_on nn_Translation_Statistical nn_Translation_Machine pobj_on_Translation nsubj_made_who rcmod_Translation_made dobj_made_use prep_use_of amod_tools_statistical pobj_of_tools prep_made_based dep_based_on det_model_a amod_model_noisy nn_model_channel pobj_on_model advmod_developed_originally partmod_model_developed prep_developed_for nn_recognition_speech pobj_for_recognition ' +o,1692,'These problems formulations are similar to those studied in and , respectively ',Ramshaw,'det_formulations_These nn_formulations_problems nsubj_similar_formulations cop_similar_are prep_similar_to pobj_to_those partmod_those_studied prep_studied_in pobj_in_ cc__and conj__ advmod_studied_respectively ' +o,1693,' reported that it was appropriate in 722 \% of cases ',Kanayama,'nsubj_reported_ complm_appropriate_that nsubj_appropriate_it cop_appropriate_was ccomp_reported_appropriate prep_appropriate_in num_\%_722 pobj_in_\% prep_\%_of pobj_of_cases ' +o,1694,' 019-048 Leacock & Chodrow 036 036 Resnik -LRB- 1995 -RRB- 037 Proposed 0504 7 Conclusion We proposed a relational model to measure the semantic similarity between two words ',Lin,'number_019-048_ num_Leacock_019-048 cc_Leacock_& nn_Conclusion_Chodrow nn_Conclusion_ nn_Conclusion_036 nn_Conclusion_ nn_Conclusion_036 nn_Conclusion_Resnik appos_Conclusion_1995 nn_Conclusion_037 nn_Conclusion_Proposed num_Conclusion_0504 amod_Conclusion_7 conj_Leacock_Conclusion nsubj_proposed_We dep_Leacock_proposed det_model_a amod_model_relational dobj_proposed_model aux_measure_to xcomp_proposed_measure det_similarity_the amod_similarity_semantic dobj_measure_similarity prep_similarity_between num_words_two pobj_between_words ' +o,1695,'More recently , the integration of information sources , and the modeling of more complex language processing tasks in the statistical framework has increased the interest in smoothing methods ',Ratnaparkhi,'advmod_recently_More advmod_increased_recently det_integration_the nsubj_increased_integration prep_integration_of nn_sources_information pobj_of_sources cc_integration_and det_modeling_the conj_integration_modeling prep_modeling_of advmod_complex_more amod_tasks_complex nn_tasks_language amod_tasks_processing pobj_of_tasks prep_tasks_in det_framework_the amod_framework_statistical pobj_in_framework aux_increased_has det_interest_the dobj_increased_interest prep_interest_in amod_methods_smoothing pobj_in_methods ' +p,1696,'Recent innovations have greatly improved the efficiency of language model integration through multipass techniques , such as forest reranking , local search , and coarse-to-fine pruning ',Huang,'amod_innovations_Recent nsubj_improved_innovations aux_improved_have advmod_improved_greatly det_efficiency_the dobj_improved_efficiency prep_efficiency_of nn_integration_language nn_integration_model pobj_of_integration prep_improved_through amod_techniques_multipass pobj_through_techniques dep_as_such prep_improved_as nn__forest amod__reranking pobj_as_ amod__local nn__search conj__ cc__and amod__coarse-to-fine nn__pruning conj__ ' +o,1697,'As a sanity check , we duplicated baseline in which all unigrams that appear four or more times in the training documents are used as features ',Pang,'prep_duplicated_As det_check_a amod_check_sanity pobj_As_check nsubj_duplicated_we amod_baseline_ dobj_duplicated_baseline rel_used_in pobj_in_which det_unigrams_all nsubjpass_used_unigrams nsubj_appear_that rcmod_unigrams_appear num_times_four cc_four_or conj_four_more dep_in_times prep_appear_in det_documents_the amod_documents_training pobj_in_documents auxpass_used_are rcmod_baseline_used prep_used_as pobj_as_features ' +o,1698,'This was a difcult challenge as many participants in the task failed to obtain any meaningful gains from unlabeled data ',Dredze,'nsubj_challenge_This cop_challenge_was det_challenge_a amod_challenge_difcult mark_failed_as amod_participants_many nsubj_failed_participants prep_participants_in det_task_the pobj_in_task advcl_challenge_failed aux_obtain_to xcomp_failed_obtain det_gains_any amod_gains_meaningful dobj_obtain_gains prep_obtain_from amod_data_unlabeled pobj_from_data ' +o,1699,'The work most similar in spirit to ours that of ',Turney,'det_work_The advmod_similar_most amod_work_similar prep_work_in pobj_in_spirit dep_work_to pobj_to_ours dep_ours_that dep_work_of ' +o,1700,'Day 1 Day 2 No ASR adaptation 2939 2741 Unsupervised ASR adaptation 3155 2766 Supervised ASR adaptation 3219 2765 Table 2 : Impact of ASR adaptation to SMT Table 2 shows the impact of ASR adaptation on the performance of the translation system in BLEU ',Papineni,'num_Day_1 dep_adaptation_Day num_Day_2 det_adaptation_No nn_adaptation_ASR dep_Day_adaptation number_2741_2939 num_ASR_2741 nn_ASR_Unsupervised dep_adaptation_ASR dep_adaptation_adaptation num_adaptation_3155 num_adaptation_2766 nn_adaptation_Supervised nn_adaptation_ASR dep_Day_adaptation number_2765_3219 dep_adaptation_2765 dep_2765_Table num_Table_2 dep_Day_Impact prep_Impact_of pobj_of_ASR dep_impact_adaptation prep_adaptation_to nn_Table_SMT pobj_to_Table num_shows_2 dep_impact_shows det_impact_the dep_Impact_impact prep_impact_of nn_adaptation_ASR pobj_of_adaptation prep_Impact_on det_performance_the pobj_on_performance prep_performance_of det_system_the nn_system_translation pobj_of_system prep_system_in pobj_in_BLEU ' +p,1701,'However , since most of statistical translation models are symmetrical , it is relatively easy to train a translation system to translate from English to Chinese , except that weneed to train aChinese language model from the Chinese monolingual data ',Koehn,'advmod_easy_However mark_symmetrical_since nsubj_symmetrical_most prep_most_of amod_models_statistical nn_models_translation pobj_of_models advmod_symmetrical_ cop_symmetrical_are dep_easy_symmetrical nsubj_easy_it cop_easy_is advmod_easy_relatively aux_train_to xcomp_easy_train det_system_a nn_system_translation dobj_train_system aux_translate_to xcomp_train_translate prep_translate_from pobj_from_English prep_translate_to pobj_to_Chinese mark_weneed_except nsubj_weneed_that advcl_easy_weneed aux_train_to xcomp_weneed_train amod_model_aChinese nn_model_language dobj_train_model prep_train_from det_data_the amod_data_Chinese amod_data_monolingual pobj_from_data ' +n,1702,'Inside-out alignments , such as the one in Example 13 , can not be induced by any of these theories ; in fact , there seems to be no useful synchronous grammar formalisms available that handle inside-out alignments , with the possible exceptions of synchronous tree-adjoining grammars , Bertsch and Nederhof and generalized multitext grammars , which are all way more complex than ITG , STSG and -LRB- 2,2 -RRB- - BRCG ',Wu,'amod__Inside-out nn__alignments nsubjpass_induced_ dep_as_such prep__as det_one_the pobj_as_one prep_one_in pobj_in_Example num_Example_13 aux_induced_can neg_induced_not auxpass_induced_be prep_induced_by pobj_by_any prep_any_of det_theories_these pobj_of_theories prep_seems_in pobj_in_fact expl_seems_there parataxis_induced_seems aux_grammar_to cop_grammar_be det_grammar_no amod_grammar_useful dep_useful_synchronous xcomp_seems_grammar amod_grammar_formalisms dep_formalisms_available complm_handle_that ccomp_grammar_handle amod_alignments_inside-out dobj_handle_alignments prep_grammar_with det_exceptions_the amod_exceptions_possible pobj_with_exceptions prep_exceptions_of amod__synchronous amod__tree-adjoining amod__grammars dep_grammars_ nn__Bertsch cc_Bertsch_and conj_Bertsch_Nederhof conj_grammars_ cc_grammars_and conj_grammars_generalized amod__multitext nn__grammars pobj_of_ nsubj_complex_which cop_complex_are dep_way_all advmod_complex_way advmod_complex_more rcmod__complex prep_complex_than pobj_than_ITG conj_ITG_STSG cc_ITG_and appos_ITG_2,2 conj_ITG_BRCG ' +p,1703,'In terms of applying non-parametric Bayesian approaches to NLP , evaluated the clustering properties of DPMMs by performing anaphora resolution with good results ',Haghighi,'pobj_In_terms prep_terms_of pcomp_of_applying amod_approaches_non-parametric nn_approaches_Bayesian dobj_applying_approaches prep_applying_to pobj_to_NLP dep_evaluated_ dep_In_evaluated det_properties_the amod_properties_clustering dobj_evaluated_properties prep_properties_of pobj_of_DPMMs prep_evaluated_by pcomp_by_performing nn_resolution_anaphora dobj_performing_resolution prep_performing_with amod_results_good pobj_with_results ' +o,1704,'32 Learning Algorithm For learning coreference decisions , we used a Maximum Entropy model ',Berger,'num_Algorithm_32 nn_Algorithm_Learning prep_used_For pcomp_For_learning amod_decisions_coreference dobj_learning_decisions nsubj_used_we dep_Algorithm_used det_model_a amod_model_Maximum nn_model_Entropy nn_model_ dobj_used_model ' +o,1705,'For example it has been used to measure centrality in hyperlinked web pages networks , lexical networks , and semantic networks ',Mihalcea,'prep_used_For pobj_For_example nsubjpass_used_it aux_used_has auxpass_used_been aux_measure_to xcomp_used_measure advmod_measure_centrality prep_measure_in amod__hyperlinked nn__web nn__pages nn__networks pobj_in_ amod__lexical nn__networks conj__ cc__and amod_networks_semantic conj__networks ' +o,1706,'Two LUs close in the space are likely to be in a paradigmatic relation , ie to be close in a is-a hierarchy ',Lin,'num_close_Two nn_close_LUs nsubj_likely_close prep_close_in det_space_the pobj_in_space cop_likely_are aux_be_to xcomp_likely_be prep_be_in det_relation_a amod_relation_paradigmatic pobj_in_relation appos_relation_ie aux_close_to cop_close_be xcomp_be_close prep_close_in det_hierarchy_a amod_hierarchy_is-a pobj_in_hierarchy ' +o,1707,'The WSJNPVP set consists of part-of speech tagged Wall Street Journal material , supplemented with syntactic tags indicating noun phrase and verb phrase boundaries ',Marcus,'det_set_The nn_set_WSJNPVP nsubj_consists_set prep_consists_of amod_speech_part-of pobj_of_speech partmod_speech_tagged nn__Wall nn__Street nn__Journal nn__material dobj_tagged_ partmod__supplemented prep_supplemented_with amod_tags_syntactic pobj_with_tags partmod_tags_indicating nn_phrase_noun dobj_indicating_phrase cc_phrase_and nn_boundaries_verb nn_boundaries_phrase conj_phrase_boundaries advmod_indicating_ ' +o,1708,'For example , developed the Sub-Tree Metric -LRB- STM -RRB- over constituent parse trees and the Head-Word Chain Metric -LRB- HWCM -RRB- over dependency parse trees ',Liu,'prep_developed_For pobj_For_example nsubj_developed_ det_Metric_the nn_Metric_Sub-Tree dobj_developed_Metric abbrev_Metric_STM prep_Metric_over amod_trees_constituent nn_trees_parse pobj_over_trees cc_Metric_and det_Metric_the amod_Metric_Head-Word nn_Metric_Chain conj_Metric_Metric abbrev_Metric_HWCM prep_Metric_over amod_trees_dependency nn_trees_parse pobj_over_trees ' +o,1709,'Because our algorithm does not consider the context given by the preceding sentences , we have conducted the following experiment to see to what extent the discourse context could improve the performance of the wordsense disambiguation : Using the semantic concordance files , we have counted the occurrences of content words which previously appear in the same discourse file ',Marcus,'mark_consider_Because poss_algorithm_our nsubj_consider_algorithm aux_consider_does neg_consider_not advcl_conducted_consider det_context_the dobj_consider_context prep_consider_given dep_given_by det_sentences_the amod_sentences_preceding pobj_by_sentences nsubj_conducted_we aux_conducted_have det_experiment_the amod_experiment_following dobj_conducted_experiment aux_see_to xcomp_conducted_see prep_see_to dobj_improve_what dep_what_extent det_context_the nn_context_discourse nsubj_improve_context aux_improve_could pcomp_to_improve det_performance_the dobj_improve_performance prep_performance_of det_disambiguation_the amod_disambiguation_wordsense pobj_of_disambiguation dep_counted_Using det__the amod__semantic amod__concordance nn__files dobj_Using_ nsubj_counted_we aux_counted_have parataxis_conducted_counted det_occurrences_the dobj_counted_occurrences prep_occurrences_of nn_words_content pobj_of_words nsubj_appear_which advmod_appear_previously rcmod_occurrences_appear prep_appear_in det_file_the amod_file_same nn_file_discourse pobj_in_file ' +o,1710,'In some cases , class -LRB- or part of speech -RRB- n-grams are used instead of word n-gram ',Brown,'prep_used_In det_cases_some pobj_In_cases appos_cases_class dep_part_or dep_class_part prep_part_of pobj_of_speech nsubjpass_used_n-grams auxpass_used_are dep_of_instead prep_used_of nn_n-gram_word pobj_of_n-gram ' +o,1711,'It is mentioned that the limitation is largely caused by inconsistencies in the corpus ',Ratnaparkhi,'nsubjpass_mentioned_It auxpass_mentioned_is complm_caused_that det_limitation_the nsubjpass_caused_limitation auxpass_caused_is advmod_caused_largely ccomp_mentioned_caused prep_caused_by pobj_by_inconsistencies prep_inconsistencies_in det_corpus_the pobj_in_corpus ' +o,1712,'As our approach for incorporating unlabeled data , we basically follow the idea proposed in ',Suzuki,'prep_follow_As poss_approach_our pobj_As_approach prep_approach_for pcomp_for_incorporating amod_data_unlabeled dobj_incorporating_data nsubj_follow_we advmod_follow_basically det_idea_the nsubj_proposed_idea ccomp_follow_proposed prt_proposed_in ' +o,1713,'The training is performed by a single generalized perceptron ',Collins,'det_training_The nsubjpass_performed_training auxpass_performed_is prep_performed_by det_perceptron_a amod_perceptron_single amod_perceptron_generalized pobj_by_perceptron ' +o,1714,'Furthermore , as pointed out in , the sense division in an MRD is frequently too fine-grained for the purpose of WSD ',Dolan,'advmod_fine-grained_Furthermore mark_pointed_as dep_fine-grained_pointed prt_pointed_out prep_pointed_in pobj_in_ det_division_the nn_division_sense nsubj_fine-grained_division prep_division_in det_MRD_an pobj_in_MRD cop_fine-grained_is advmod_fine-grained_frequently advmod_fine-grained_too prep_fine-grained_for det_purpose_the pobj_for_purpose prep_purpose_of pobj_of_WSD ' +o,1715,'Generative word alignment models , initially developed at IBM , and then augmented by an HMM-based model , have provided powerful modeling capability for word alignment ',Dunning,'amod_models_Generative nn_models_word nn_models_alignment nsubj_provided_models advmod_developed_initially dep_models_developed prep_developed_at nn__IBM pobj_at_ cc_developed_and advmod_augmented_then conj_developed_augmented prep_augmented_by det__an amod__HMM-based nn__model pobj_by_ aux_provided_have amod_capability_powerful nn_capability_modeling dobj_provided_capability prep_capability_for nn_alignment_word pobj_for_alignment ' +o,1716,' and ',Och,'nn_and_ ' +o,1717,'We report results on the Boston University -LRB- BU -RRB- Radio Speech Corpus and Boston Directions Corpus -LRB- BDC -RRB- , two publicly available speech corpora with manual ToBI annotations intended for experiments in automatic prosody labeling ',Berger,'nsubj_report_We dobj_report_results prep_results_on dep__the nn_University_Boston nn__University abbrev__BU nn__Radio nn__Speech nn__Corpus pobj_on_ cc__and nn__Boston nn__Directions nn__Corpus abbrev__BDC conj__ num_corpora_two advmod_available_publicly amod_corpora_available nn_corpora_speech conj__corpora prep_report_with amod_annotations_manual nn_annotations_ToBI pobj_with_annotations partmod_annotations_intended prep_intended_for pobj_for_experiments prep_experiments_in amod_labeling_automatic amod_labeling_prosody pobj_in_labeling ' +o,1718,'The co-occurrence relation can also be based on distance in a bitext space , which is a more general representations of bitext correspondence , or it can be restricted to words pairs that satisfy some matching predicate , which can be extrinsic to the model ',Brown,'det_relation_The amod_relation_co-occurrence nsubjpass_based_relation aux_based_can advmod_based_also auxpass_based_be prep_based_on pobj_on_distance prep_based_in det_space_a amod_space_bitext pobj_in_space nsubj_representations_which cop_representations_is det_representations_a advmod_general_more amod_representations_general rcmod_space_representations prep_representations_of amod__bitext nn__correspondence pobj_of_ cc_based_or nsubjpass_restricted_it aux_restricted_can auxpass_restricted_be conj_based_restricted prep_restricted_to nn_pairs_words pobj_to_pairs nsubj_satisfy_that rcmod_pairs_satisfy det_predicate_some amod_predicate_matching dobj_satisfy_predicate nsubj_extrinsic_which aux_extrinsic_can cop_extrinsic_be rcmod_predicate_extrinsic prep_extrinsic_to det__the nn__model pobj_to_ ' +o,1719,'One option is what calls many-to-one -LRB- M-to-1 -RRB- accuracy , in which each induced tag is labeled with its most frequent gold tag ',Johnson,'num_option_One nsubj_is_option dobj_calls_what nsubj_calls_ ccomp_is_calls amod_accuracy_many-to-one appos_accuracy_M-to-1 dobj_calls_accuracy rel_labeled_in pobj_in_which det_tag_each amod_tag_induced nsubjpass_labeled_tag auxpass_labeled_is rcmod_accuracy_labeled prep_labeled_with poss_tag_its advmod_frequent_most amod_tag_frequent nn_tag_gold pobj_with_tag ' +o,1720,'We used the Maximum Entropy approach5 as a machine learner for this task ',Berger,'nsubj_used_We det__the amod__Maximum nn__Entropy nn__approach5 dobj_used_ prep_used_as det_learner_a nn_learner_machine pobj_as_learner prep_used_for det_task_this pobj_for_task ' +o,1721,'We report that our parsing framework achieved high accuracy -LRB- 886 \% -RRB- in dependency analysis of Japanese with a combination of an underspecified HPSG-based Japanese grammar , SLUNG and the maximum entropy method ',Berger,'nsubj_report_We complm_achieved_that poss_framework_our amod_framework_parsing nsubj_achieved_framework ccomp_report_achieved amod_accuracy_high dobj_achieved_accuracy num_\%_886 appos_accuracy_\% prep_achieved_in amod_analysis_dependency pobj_in_analysis prep_analysis_of pobj_of_Japanese prep_achieved_with det_combination_a pobj_with_combination prep_combination_of det_grammar_an amod_grammar_underspecified amod_grammar_HPSG-based amod_grammar_Japanese pobj_of_grammar nn__SLUNG conj_grammar_ cc_grammar_and det_method_the amod_method_maximum nn_method_entropy conj_grammar_method ' +o,1722,'The model weights of the transducer are tuned based on the development set using a grid-based line search , and the translation results are evaluated based on a single Chinese reference6 using BLEU-4 ',Papineni,'det_weights_The nn_weights_model nsubjpass_tuned_weights prep_weights_of det_transducer_the pobj_of_transducer auxpass_tuned_are prep_tuned_based dep_based_on det_development_the pobj_on_development partmod_development_set xcomp_set_using det_search_a amod_search_grid-based nn_search_line dobj_using_search cc_tuned_and det_results_the nn_results_translation nsubjpass_evaluated_results auxpass_evaluated_are conj_tuned_evaluated prep_evaluated_based dep_based_on det_reference6_a amod_reference6_single amod_reference6_Chinese pobj_on_reference6 partmod_reference6_using dobj_using_BLEU-4 ' +o,1723,' deserves the credit for bringing to the attention of computational linguists ',Carletta,'nsubj_deserves_ det_credit_the dobj_deserves_credit prep_deserves_for pcomp_for_bringing prep_bringing_to det_attention_the pobj_to_attention prep_attention_of amod_linguists_computational pobj_of_linguists ' +o,1724,'Thus , we can compute the source dependency LM score in the same way we compute the target side score , using a procedure described in ',Shen,'advmod_compute_Thus nsubj_compute_we aux_compute_can det_score_the nn_score_source nn_score_dependency nn_score_LM dobj_compute_score prep_compute_in det_way_the amod_way_same pobj_in_way nsubj_compute_we dep_compute_compute det_score_the nn_score_target nn_score_side dobj_compute_score xcomp_compute_using det_procedure_a dobj_using_procedure partmod_procedure_described prep_described_in ' +o,1725,'The typical problems like doctor-nurse could be avoided by using such information ',Church,'det_problems_The amod_problems_typical nsubjpass_avoided_problems prep_problems_like amod__doctor-nurse pobj_like_ aux_avoided_could auxpass_avoided_be prep_avoided_by pcomp_by_using amod_information_such dobj_using_information ' +o,1726,'Kanayama and Nasukawa used both intraand inter-sentential co-occurrence to learn polarity of words and phrases ',Kanayama,'nsubj_used_Kanayama cc_Kanayama_and conj_Kanayama_Nasukawa det_co-occurrence_both amod_co-occurrence_intraand amod_co-occurrence_inter-sentential dobj_used_co-occurrence aux_learn_to xcomp_used_learn dobj_learn_polarity prep_polarity_of pobj_of_words cc_words_and conj_words_phrases ' +o,1727,'For comparison purposes , we also computed the value of R 2 for adequacy using the BLEU score formula given in , for the 7 systems using the same one reference , and we obtain a similar value , 8391 \% ; computing the value of R 2 for adequacy using the BLEU scores computed with all 4 references available also yielded a lower value for R 2 , 6221 \% ',Papineni,'prep_computed_For nn_purposes_comparison pobj_For_purposes nsubj_computed_we advmod_computed_also det_value_the dobj_computed_value prep_value_of pobj_of_R num_R_2 prep_computed_for pobj_for_adequacy partmod_adequacy_using det_formula_the amod_formula_BLEU nn_formula_score dobj_using_formula prep_using_given dep_given_in pobj_in_ prep_computed_for det_systems_the num_systems_7 pobj_for_systems partmod_systems_using det_reference_the amod_reference_same num_reference_one dobj_using_reference cc_computed_and nsubj_obtain_we conj_computed_obtain det_value_a amod_value_similar dobj_obtain_value num_\%_8391 appos_value_\% csubj_yielded_computing det_value_the dobj_computing_value prep_value_of pobj_of_R num_R_2 prep_computing_for pobj_for_adequacy partmod_adequacy_using det_scores_the amod_scores_BLEU dobj_using_scores partmod_scores_computed prep_computed_with pobj_with_all num_references_4 measure_available_references amod_all_available advmod_yielded_also parataxis_computed_yielded det_value_a amod_value_lower dobj_yielded_value prep_yielded_for pobj_for_R num_R_2 num_\%_6221 appos_R_\% ' +o,1728,'They first extract English collocations using the Xtract systetn , and theu look for French coutlterparts ',Smadja,'nsubj_extract_They advmod_extract_first amod_collocations_English dobj_extract_collocations partmod_collocations_using det__the nn__Xtract nn__systetn dobj_using_ cc_extract_and nsubj_look_theu conj_extract_look prep_look_for amod_coutlterparts_French pobj_for_coutlterparts ' +o,1729,' , is to translate dependency parses into neo-Davidsonian-style quasilogical forms , and to perform weighted abductive theorem proving in the tradition of ',Hobbs,'nsubj_is_ aux_translate_to ccomp_is_translate amod_parses_dependency dobj_translate_parses prep_translate_into amod_forms_neo-Davidsonian-style amod_forms_quasilogical pobj_into_forms cc_translate_and aux_perform_to conj_translate_perform amod_theorem_weighted nn_theorem_abductive dobj_perform_theorem partmod_theorem_proving prep_proving_in det_tradition_the pobj_in_tradition prep_tradition_of ' +o,1730,'61 Evaluation of Translation Performance We use the BLEU score to evaluate our systems ',Papineni,'num_Evaluation_61 nsubj_Performance_Evaluation prep_Evaluation_of pobj_of_Translation nsubj_use_We ccomp_Performance_use det__the amod__BLEU nn__score nsubj_evaluate_ aux_evaluate_to xcomp_use_evaluate poss_systems_our dobj_evaluate_systems ' +o,1731,'It is often argued that the ability to translate discontiguous phrases is important to modeling translation , and it may be that this explains the results ',Quirk,'nsubjpass_argued_It auxpass_argued_is advmod_argued_often complm_important_that det_ability_the nsubj_important_ability aux_translate_to infmod_ability_translate amod_phrases_discontiguous dobj_translate_phrases cop_important_is ccomp_argued_important prep_important_to nn__modeling nn__translation pobj_to_ cc_argued_and nsubj_be_it aux_be_may conj_argued_be complm_explains_that nsubj_explains_this ccomp_be_explains det_results_the dobj_explains_results ' +p,1732,'The full model yields a stateof-the-art BLEU score of 08506 on Section 23 of the CCGbank , which is to our knowledge the best score reported to date 410 using a reversible , corpus-engineered grammar ',Papineni,'det_model_The amod_model_full nsubj_yields_model det_score_a amod_score_stateof-the-art amod_score_BLEU nn_score_ dobj_yields_score prep_score_of pobj_of_08506 prep_08506_on pobj_on_Section num_Section_23 prep_score_of det_CCGbank_the pobj_of_CCGbank nsubj_is_which rcmod_CCGbank_is prep_is_to poss_knowledge_our pobj_to_knowledge det_score_the amod_score_best nsubj_reported_score dep_yields_reported prep_reported_to pobj_to_date num_date_410 xcomp_reported_using det_grammar_a amod_grammar_reversible amod_grammar_corpus-engineered dobj_using_grammar ' +o,1733,'Clustering-based approaches usually represent word contexts as vectors and cluster words based on similarities of the vectors ',Brown,'amod_approaches_Clustering-based nsubj_represent_approaches advmod_represent_usually nn_contexts_word dobj_represent_contexts prep_represent_as pobj_as_vectors cc_vectors_and nn_words_cluster conj_vectors_words partmod_vectors_based prep_based_on pobj_on_similarities prep_similarities_of det_vectors_the pobj_of_vectors ' +n,1734,'Bilexical context-free grammars have been presented in as an abstraction of language models that have been adopted in several recent real-world parsers , improving state-of-the-art parsing accuracy ',Collins,'amod_grammars_Bilexical amod_grammars_context-free nsubjpass_presented_grammars aux_presented_have auxpass_presented_been prep_presented_in pobj_in_ prep_presented_as det_abstraction_an pobj_as_abstraction prep_abstraction_of nn_models_language pobj_of_models nsubjpass_adopted_that aux_adopted_have auxpass_adopted_been rcmod_abstraction_adopted prep_adopted_in amod_parsers_several amod_parsers_recent amod_parsers_real-world pobj_in_parsers xcomp_adopted_improving amod_accuracy_state-of-the-art amod_accuracy_parsing dobj_improving_accuracy ' +o,1735,'However , show that , in phrase-based translation , improvements in AER or f-measure do not necessarily correlate with improvements in BLEU score ',Fraser,'advmod_show_However nsubj_show_ complm_correlate_that dep_correlate_in amod_translation_phrase-based pobj_in_translation nsubj_correlate_improvements prep_improvements_in pobj_in_AER cc_AER_or conj_AER_f-measure aux_correlate_do neg_correlate_not advmod_correlate_necessarily ccomp_show_correlate prep_correlate_with pobj_with_improvements prep_improvements_in amod_score_BLEU pobj_in_score ' +o,1736,' 916 \% 916 \% F\\/3 = 1 9386 9326 928 9203 916 Table 3 : The overall pertbrmance of the majority voting combination of our best five systems -LRB- selected on tinting data perfbrnmnce -RRB- applied to the standard data set pnt tbrward by together with an overview of earlier work ',Ramshaw,'num_\%_916 dobj__\% num_\%_916 nsubj__\% number_916_F\\/3 cc_916_= conj_916_1 number_916_9386 number_916_9326 number_916_928 number_916_9203 num_\%_916 cc_916_Table conj_916_3 det_pertbrmance_The amod_pertbrmance_overall dep_\%_pertbrmance prep_pertbrmance_of det_combination_the nn_combination_majority amod_combination_voting pobj_of_combination prep_combination_of poss_systems_our amod_systems_best num_systems_five pobj_of_systems dep_combination_selected prep_selected_on pcomp_on_tinting nn_perfbrnmnce_data dobj_tinting_perfbrnmnce partmod_combination_applied prep_applied_to det_pnt_the amod_pnt_standard nn_pnt_data nn_pnt_set pobj_to_pnt partmod_pnt_tbrward prep_tbrward_by pcomp_by_ advmod__together prep__with det_overview_an pobj_with_overview prep_overview_of amod_work_earlier pobj_of_work ' +o,1737,'Given a sentence-pair -LRB- f , e -RRB- , the most likely -LRB- Viterbi -RRB- word alignment is found as : a = argmaxa P -LRB- f , a e -RRB- ',Brown,'det_sentence-pair_a dep_Given_sentence-pair dep_sentence-pair_f dep_word_e det_word_the advmod_likely_most amod_word_likely appos_word_Viterbi appos_f_word nsubjpass_found_alignment auxpass_found_is rcmod_word_found prep_found_as pobj_as_ det_P_a amod_P_= nn_P_argmaxa dep_word_P dep_P_f det_e_a appos_f_e ' +o,1738,'We directly model the conditional probability of the alignment a , given x and y , using the maximum entropy framework , P -LRB- a x , y -RRB- = exp -LCB- F -LRB- a , x , y -RRB- -RCB- summationdisplay aC -LRB- x , y -RRB- exp -LCB- F -LRB- a , x , y -RRB- -RCB- ',Berger,'advmod_model_directly dep_We_model det_probability_the amod_probability_conditional dobj_model_probability prep_probability_of det_a_the amod_a_alignment pobj_of_a partmod_a_given dobj_given_x cc_x_and conj_x_y xcomp_given_using det__the amod__maximum amod__entropy nn__framework nsubj_=_ appos__P det_x_a dep_P_x appos_x_y ccomp_using_= dobj_=_exp dep_exp_F dep_exp_a appos_a_x appos_a_y advmod_aC_summationdisplay partmod_exp_aC dep_We_x appos_x_y dep_We_exp dep_exp_F dep_exp_a appos_a_x appos_a_y ' +o,1739,'The first approaches are used for Penn Treebank and the KAIST language resource ',Marcus,'det_approaches_The amod_approaches_first nsubjpass_used_approaches auxpass_used_are prep_used_for nn__Penn nn__Treebank pobj_for_ cc__and det__the amod__KAIST nn__language nn__resource conj__ ' +o,1740,' introduced a statistical measurement called mutual information for extracting strongly associated or collocated words ',Church,'nsubj_introduced_ det_measurement_a amod_measurement_statistical nsubj_called_measurement ccomp_introduced_called amod_information_mutual dobj_called_information prep_called_for pcomp_for_extracting advmod_associated_strongly dep_extracting_associated cc_associated_or conj_associated_collocated dobj_associated_words ' +o,1741,'A variety of other measures of semantic relatedness have been proposed , including distributional similarity measures based on co-occurrence in a body of text see for a survey ',Weeds,'det_variety_A nsubjpass_proposed_variety prep_variety_of amod_measures_other pobj_of_measures prep_measures_of amod_relatedness_semantic pobj_of_relatedness aux_proposed_have auxpass_proposed_been prep_proposed_including amod_measures_distributional nn_measures_similarity pobj_including_measures partmod_measures_based prep_based_on pobj_on_co-occurrence prep_co-occurrence_in det_body_a pobj_in_body prep_body_of pobj_of_text ccomp_based_see dobj_see_ prep__for det_survey_a pobj_for_survey ' +o,1742,'The Collins parser does use dynamic programming in its search ',Collins,'det__The nn__Collins nn__parser nsubj_use_ aux_use_does amod_programming_dynamic dobj_use_programming prep_use_in poss_search_its pobj_in_search ' +o,1743,'On the other hand , both BLEU and NIST scores are higher for the baseline system -LRB- mteval-v11bpl -RRB- ',Papineni,'prep_higher_On det_hand_the amod_hand_other pobj_On_hand preconj__both num__BLEU nsubj_higher_ cc__and nn_scores_NIST nn_scores_ conj__scores cop_higher_are prep_higher_for det_system_the nn_system_baseline pobj_for_system dep_higher_mteval-v11bpl ' +o,1744,'The application of this algorithm to the basic problem using a parallel bilingual corpus aligned on the sentence level is described in ',Brown,'det_application_The nsubjpass_described_application prep_application_of det_algorithm_this pobj_of_algorithm prep_algorithm_to det_problem_the amod_problem_basic pobj_to_problem partmod_problem_using det_corpus_a amod_corpus_parallel amod_corpus_bilingual dobj_using_corpus partmod_corpus_aligned prep_aligned_on det_level_the nn_level_sentence pobj_on_level auxpass_described_is prep_described_in ' +o,1745,'We further note that our results are different from that of as they use extensive feature engineering and weight tuning during the graph generation process that we have not been able to reproduce ',Hughes,'nsubj_note_We advmod_note_further complm_different_that poss_results_our nsubj_different_results cop_different_are ccomp_note_different prep_different_from pobj_from_that prep_that_of pobj_of_ mark_use_as nsubj_use_they advcl_different_use amod_engineering_extensive nn_engineering_feature dobj_use_engineering cc_engineering_and nn_tuning_weight conj_engineering_tuning prep_use_during det_process_the nn_process_graph nn_process_generation pobj_during_process complm_able_that nsubj_able_we aux_able_have neg_able_not dep_able_been ccomp_use_able aux_reproduce_to xcomp_able_reproduce ' +o,1746,'Selectional preferences are estimated using grammatical collocation information from the British National Corpus -LRB- BNC -RRB- , obtained with the Word Sketch Engine -LRB- WSE -RRB- ',Pedersen,'amod_preferences_Selectional nsubjpass_estimated_preferences auxpass_estimated_are ccomp_obtained_estimated xcomp_estimated_using amod_information_grammatical nn_information_collocation dobj_using_information prep_using_from det_Corpus_the nn_Corpus_British nn_Corpus_National pobj_from_Corpus abbrev_Corpus_BNC prep_obtained_with det_Word_the pobj_with_Word nn__Sketch nn__Engine abbrev__WSE nsubj_obtained_ ' +o,1747,'1 Introduction The probabilistic relation between verbs and their arguments plays an important role in modern statistical parsers and supertaggers , and in psychological theories of language processing ',Marcus,'num_Introduction_1 det_relation_The amod_relation_probabilistic nsubj_plays_relation prep_relation_between pobj_between_verbs cc_verbs_and poss_arguments_their conj_verbs_arguments rcmod_Introduction_plays det_role_an amod_role_important dobj_plays_role prep_role_in amod_parsers_modern amod_parsers_statistical pobj_in_parsers cc_parsers_and amod__supertaggers conj_parsers_ cc_Introduction_and conj_Introduction_in amod_theories_psychological pobj_in_theories prep_theories_of nn__language nn__processing pobj_of_ ' +p,1748,'They have been successfully applied in several tasks , such as information retrieval and harvesting thesauri ',Lin,'nsubjpass_applied_They aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_in amod_tasks_several pobj_in_tasks dep_as_such prep_tasks_as nn__information nn__retrieval pobj_as_ cc__and amod_thesauri_harvesting conj__thesauri ' +o,1749,'Hence , either the best translation hypothesis is directly extracted from the word graph and output , or an N-best list of translations is computed ',Berger,'advmod_extracted_Hence dep_hypothesis_either det_hypothesis_the amod_hypothesis_best nn_hypothesis_translation nsubjpass_extracted_hypothesis auxpass_extracted_is advmod_extracted_directly prep_extracted_from det_graph_the nn_graph_word pobj_from_graph cc_graph_and conj_graph_output cc_extracted_or det_list_an amod_list_N-best nsubjpass_computed_list prep_list_of pobj_of_translations auxpass_computed_is conj_extracted_computed dep_computed_ ' +o,1750,'1 Introduction Parsing sentences using statistical information gathered from a treebank was first examined a decade ago in and is by now a fairly well-studied problem -LRB- , , -RRB- ',Collins,'num_sentences_1 nn_sentences_Introduction nn_sentences_Parsing nsubjpass_examined_sentences partmod_sentences_using amod_information_statistical dobj_using_information partmod_information_gathered prep_gathered_from det_treebank_a pobj_from_treebank auxpass_examined_was advmod_examined_first det_decade_a dep_ago_decade dep_in_ago prep_examined_in pobj_in_ cc_examined_and conj_examined_is prep_is_by advmod_problem_now det_problem_a advmod_well-studied_fairly amod_problem_well-studied pobj_by_problem appos_problem_ dep__ dep__ ' +o,1751,'Feature function weights in the loglinear model are set using Ochs minium error rate algorithm ',Och,'amod_weights_Feature nn_weights_function nsubjpass_set_weights prep_weights_in det_model_the nn_model_loglinear pobj_in_model auxpass_set_are xcomp_set_using amod_algorithm_Ochs nn_algorithm_minium nn_algorithm_error nn_algorithm_rate dobj_using_algorithm ' +o,1752,'As in tile HMM we easily can extend the dependencies in the alignment model of Model 4 easily using the word class of the previous English word E = G -LRB- ci , -RRB- , or the word class of the French word F = G -LRB- Ij -RRB- ',Brown,'dep_in_As nn_HMM_tile pobj_in_HMM nsubj_extend_we advmod_extend_easily aux_extend_can rcmod_HMM_extend det_dependencies_the dobj_extend_dependencies prep_extend_in det_model_the amod_model_alignment nsubj_using_model prep_model_of pobj_of_Model num_Model_4 advmod_using_easily pcomp_in_using det_class_the nn_class_word dobj_using_class prep_extend_of det_E_the amod_E_previous amod_E_English nn_E_word pobj_of_E dep_G_= rcmod_E_G dep_E_ci cc_HMM_or det_class_the nn_class_word conj_HMM_class prep_class_of det_F_the amod_F_French nn_F_word pobj_of_F dep_Ij_= nn_Ij_G rcmod_F_Ij ' +o,1753,'Methods focussing on the use and generation of dictionaries capturing the sentiment of words have ranged from manual approaches of developing domain-dependent lexicons to semi-automated approaches , and even an almost fully automated approach ',Turney,'nsubj_ranged_Methods partmod_Methods_focussing prep_focussing_on det_use_the pobj_on_use cc_use_and conj_use_generation prep_use_of pobj_of_dictionaries partmod_dictionaries_capturing det_sentiment_the dobj_capturing_sentiment prep_sentiment_of pobj_of_words aux_ranged_have prep_ranged_from amod_approaches_manual pobj_from_approaches prep_approaches_of pcomp_of_developing amod__domain-dependent amod__lexicons amod__ prep__to amod_approaches_semi-automated pobj_to_approaches dobj_developing_ cc__and advmod_approach_even det_approach_an amod_approach_almost advmod_almost_fully amod_approach_automated conj__approach ' +o,1754,'Examples of such techniques are Markov Random Fields , and boosting or perceptron approaches to reranking ',Collins,'nsubj__Examples prep_Examples_of amod_techniques_such pobj_of_techniques cop__are nn__Markov nn__Random nn__Fields cc__and conj__boosting cc_approaches_or nn_approaches_perceptron dobj_boosting_approaches prep_approaches_to pobj_to_reranking ' +o,1755,'Inside\\/Outside This representation was first introduced in , and has been applied for base NP chunking ',Ramshaw,'det_representation_This nsubjpass_introduced_representation auxpass_introduced_was advmod_introduced_first ccomp_Inside\\/Outside_introduced prep_introduced_in pobj_in_ cc_introduced_and aux_applied_has auxpass_applied_been conj_introduced_applied prep_applied_for nn_chunking_base nn_chunking_NP pobj_for_chunking ' +o,1756,'We use the n-best generation scheme interleaved with optimization as described in ',Och,'nsubj_use_We det_scheme_the amod_scheme_n-best nn_scheme_generation nsubj_interleaved_scheme ccomp_use_interleaved prep_interleaved_with pobj_with_optimization mark_described_as advcl_interleaved_described prt_described_in ' +o,1757,'We also plan to employ this evaluation metric as feedback in building dialogue coherence models as is done in machine translation ',Och,'nsubj_plan_We advmod_plan_also aux_metric_to dep_metric_employ det_metric_this nn_metric_evaluation xcomp_plan_metric prep_metric_as pobj_as_feedback prep_feedback_in nn_models_building nn_models_dialogue nn_models_coherence pobj_in_models mark_done_as auxpass_done_is advcl_metric_done prep_done_in nn_translation_machine pobj_in_translation ' +o,1758,'In related work , both supervised and unsupervised approaches have been shown to have their pros and cons ',Turney,'prep_shown_In amod__related nn__work pobj_In_ det_approaches_both amod_approaches_supervised cc_supervised_and conj_supervised_unsupervised nsubjpass_shown_approaches aux_shown_have auxpass_shown_been aux_have_to xcomp_shown_have poss_pros_their dobj_have_pros cc_pros_and conj_pros_cons ' +o,1759,'A phrase-based translation model is one of the modern approaches which exploits a phrase , a contiguous sequence of words , as a unit of translation ',Koehn,'det_model_A amod_model_phrase-based nn_model_translation nsubj_one_model cop_one_is prep_one_of det_approaches_the amod_approaches_modern pobj_of_approaches nsubj_exploits_which rcmod_approaches_exploits det_phrase_a dobj_exploits_phrase det_sequence_a amod_sequence_contiguous appos_phrase_sequence prep_sequence_of pobj_of_words prep_exploits_as det_unit_a pobj_as_unit prep_unit_of pobj_of_translation ' +o,1760,' predicates the sentiment orientation of a review by the average semantic orientation of the phrases in the review that contain adjectives or adverbs , which is denoted as the semantic oriented method ',Turney,'advmod_predicates_ det_orientation_the nn_orientation_sentiment dobj_predicates_orientation prep_orientation_of det_review_a pobj_of_review prep_predicates_by det_orientation_the amod_orientation_average amod_orientation_semantic pobj_by_orientation prep_orientation_of det_phrases_the pobj_of_phrases prep_phrases_in det_review_the pobj_in_review nsubj_contain_that dep_predicates_contain dobj_contain_adjectives cc_adjectives_or conj_adjectives_adverbs nsubjpass_denoted_which auxpass_denoted_is rcmod_adjectives_denoted prep_denoted_as det_method_the amod_method_semantic dep_semantic_oriented pobj_as_method ' +o,1761,' for other approaches with an evaluation based on true mentions only -RRB- ',Ponzetto,'nsubj_mentions_ prep__for amod_approaches_other pobj_for_approaches prep_approaches_with det_evaluation_an pobj_with_evaluation partmod_evaluation_based prep_based_on pobj_on_true advmod_mentions_only ' +o,1762,'For evaluation we have selected a set of 8 metric variants corresponding to seven different families : BLEU -LRB- n = 4 -RRB- , NIST -LRB- n = 5 -RRB- , GTM F1-measure -LRB- e = 1,2 -RRB- , 1-WER , 1-PER , ROUGE -LRB- ROUGE-S \\* -RRB- and METEOR3 ',Banerjee,'prep_selected_For pobj_For_evaluation nsubj_selected_we aux_selected_have det_set_a dobj_selected_set prep_set_of num_variants_8 amod_variants_metric pobj_of_variants amod_variants_corresponding prep_variants_to num_families_seven amod_families_different pobj_to_families nn__BLEU nsubj_4_n dep_4_= dep__4 dep_set_ nn__NIST nsubj_5_n dep_5_= dep__5 conj__ nn__GTM nn__F1-measure nsubj_1,2_e dep_1,2_= dep__1,2 conj__ num__1-WER conj__ num__1-PER conj__ nn__ROUGE nn_\\*_ROUGE-S abbrev__\\* conj__ cc__and conj__METEOR3 ' +o,1763,'Previous approaches to processing lnetonymy have used hand-constructed ontologies or semantic networks -LRB- -RRB- ? ass , 1988 ; Iverson and Hehnreich , 1992 ; B ',Brown,'amod_approaches_Previous nsubj_used_approaches prep_approaches_to nn_lnetonymy_processing pobj_to_lnetonymy aux_used_have amod_ontologies_hand-constructed dobj_used_ontologies cc_ontologies_or amod_networks_semantic conj_ontologies_networks dep_networks_ass appos_ass_1988 dep_ass_Iverson cc_Iverson_and conj_Iverson_Hehnreich appos_Iverson_1992 dep_ass_B dep_B_ ' +n,1764,'For the Penn Treebank , reports an accuracy of 966 \% using the Maximum Entropy approach , our much simpler and therefore faster HMM approach delivers 967 \% ',Ratnaparkhi,'prep_reports_For det_Treebank_the nn_Treebank_Penn pobj_For_Treebank nsubj_reports_ det_accuracy_an nsubj_delivers_accuracy prep_accuracy_of num_\%_966 pobj_of_\% partmod_\%_using det_approach_the nn_approach_Maximum nn_approach_Entropy dobj_using_approach poss_approach_our advmod_simpler_much amod_approach_simpler cc_simpler_and advmod_HMM_therefore advmod_HMM_faster conj_simpler_HMM appos_approach_approach ccomp_reports_delivers num_\%_967 dobj_delivers_\% ' +o,1765,'The parameters for each phrase table were tuned separately using minimum error rate training ',Och,'det_parameters_The nsubjpass_tuned_parameters prep_parameters_for det_table_each nn_table_phrase pobj_for_table auxpass_tuned_were advmod_using_separately dep_tuned_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +o,1766,'Most of them were developed for exhaustive parsing , ie , producing all parse results that are given by the grammar ',Marcus,'nsubjpass_developed_Most prep_Most_of pobj_of_them auxpass_developed_were prep_developed_for amod_parsing_exhaustive nsubj_producing_parsing appos_parsing_ie pcomp_for_producing det_results_all amod_results_parse dobj_producing_results nsubjpass_given_that auxpass_given_are rcmod_results_given prep_given_by det__the nn__grammar pobj_by_ ' +o,1767,' and also report results for semi-supervised learning for these domains ',Haghighi,'cc__and conj__ advmod__also nn_results_report nsubj__results prep_results_for amod_learning_semi-supervised pobj_for_learning prep_learning_for det_domains_these pobj_for_domains ' +o,1768,'All 8,907 articles were tagged by the Xerox Part-ofSpeech Tagger 4 ',Cutting,'det_articles_All num_articles_8,907 nsubjpass_tagged_articles auxpass_tagged_were prep_tagged_by det__the nn__Xerox nn__Part-ofSpeech nn__Tagger pobj_by_ num__4 ' +p,1769,' showed that the results for French-English were competitive to state-of-the-art alignment systems ',Macken,'nsubj_showed_ complm_competitive_that det_results_the nsubj_competitive_results prep_results_for pobj_for_French-English cop_competitive_were ccomp_showed_competitive prep_competitive_to amod_systems_state-of-the-art nn_systems_alignment pobj_to_systems ' +o,1770,'Also , even the two-category version of the rating-inference problem for movie reviews has proven quite challenging for many automated classi cation techniques ',Turney,'advmod_proven_Also advmod_version_even det_version_the amod_version_two-category nsubj_proven_version prep_version_of det_problem_the amod_problem_rating-inference pobj_of_problem prep_problem_for nn_reviews_movie pobj_for_reviews aux_proven_has advmod_challenging_quite acomp_proven_challenging prep_proven_for amod_techniques_many amod_techniques_automated nn_techniques_classi nn_techniques_cation pobj_for_techniques ' +o,1771,'31 Data The starting corpus we use is formed by a mix of three different sources of data , namely the Penn Treebank corpus , the Los Angeles Times collection , as provided during TREC conferences1 , and Open Mind Common Sense2 , a collection of about 400,000 commonsense assertions in English as contributed by volunteers over the Web ',Marcus,'num_Data_31 det_corpus_The amod_corpus_starting nsubjpass_formed_corpus nsubj_use_we rcmod_corpus_use auxpass_formed_is dep_Data_formed prep_formed_by det_mix_a pobj_by_mix prep_mix_of num_sources_three amod_sources_different pobj_of_sources prep_sources_of pobj_of_data advmod__namely det__the nn__Penn nn__Treebank nn__corpus appos_data_ det_collection_the nn_collection_Los nn_collection_Angeles nn_collection_Times appos__collection dep_conferences1_as amod_conferences1_provided prep_provided_during pobj_during_TREC appos__conferences1 cc_conferences1_and nn_Sense2_Open nn_Sense2_Mind nn_Sense2_Common conj_conferences1_Sense2 det_collection_a appos_Sense2_collection prep_collection_of quantmod_400,000_about num_assertions_400,000 amod_assertions_commonsense pobj_of_assertions prep_assertions_in pobj_in_English mark_contributed_as dep_Sense2_contributed prep_contributed_by pobj_by_volunteers prep_contributed_over det_Web_the pobj_over_Web ' +o,1772,'2 Related Work There has been extensive research in opinion mining at the document level , for example on product and movie reviews ',Pang,'num_Work_2 amod_Work_Related expl_been_There aux_been_has dep_Work_been amod_research_extensive dobj_been_research prep_research_in nn_mining_opinion pobj_in_mining prep_mining_at det_level_the nn_level_document pobj_at_level prep_been_for pobj_for_example prep_been_on nn_reviews_product cc_product_and conj_product_movie pobj_on_reviews ' +o,1773,'To contrast , concentrated on analyzing human-written summaries in order to determine how professionals construct summaries ',Jing,'dep_concentrated_To pobj_To_contrast advmod_concentrated_ prep_concentrated_on pcomp_on_analyzing amod_summaries_human-written dobj_analyzing_summaries mark_determine_in dep_determine_order aux_determine_to purpcl_analyzing_determine advmod_construct_how nsubj_construct_professionals ccomp_determine_construct dobj_construct_summaries ' +o,1774,'We build sentencespecific zero-cutoff stupid-backoff 5-gram language models , estimated using 47B words of English newswire text , and apply them to rescore either 10000-best lists generated by HCP or word lattices generated by HiFST ',Brants,'nsubj_build_We amod_models_sentencespecific amod_models_zero-cutoff amod_models_stupid-backoff amod_models_ amod_models_5-gram nn_models_language dobj_build_models partmod_models_estimated xcomp_estimated_using num_words_47B dobj_using_words prep_words_of amod_text_English nn_text_newswire pobj_of_text cc_build_and conj_build_apply nsubj_rescore_them aux_rescore_to xcomp_apply_rescore det_lists_either amod_lists_10000-best dobj_rescore_lists partmod_lists_generated prep_generated_by nn_lattices_HCP cc_HCP_or conj_HCP_word pobj_by_lattices partmod_lattices_generated prep_generated_by pobj_by_HiFST ' +o,1775,'The results we obtained on the CoNLL03 test set were consistent with what was reported in ',Finkel,'det_results_The nsubj_consistent_results nsubj_obtained_we rcmod_results_obtained prep_obtained_on det_set_the amod_set_CoNLL03 nn_set_test pobj_on_set cop_consistent_were prep_consistent_with nsubjpass_reported_what auxpass_reported_was pcomp_with_reported prep_reported_in ' +p,1776,'Another kind of popular approaches to dealing with query translation based on corpus-based techniques uses a parallel corpus containing aligned sentences whose translation pairs are corresponding to each other ',Brown,'det_kind_Another nsubj_uses_kind prep_kind_of amod_approaches_popular pobj_of_approaches prep_approaches_to pcomp_to_dealing prep_dealing_with amod_translation_query pobj_with_translation prep_dealing_based dep_based_on amod_techniques_corpus-based pobj_on_techniques det_corpus_a amod_corpus_parallel dobj_uses_corpus partmod_corpus_containing amod_sentences_aligned dobj_containing_sentences poss_corresponding_whose nn_pairs_translation nsubj_corresponding_pairs aux_corresponding_are dep_sentences_corresponding prep_corresponding_to det_other_each pobj_to_other ' +o,1777,'These problems include collocation discovery , smoothing and estimation and question answering ',Brown,'det_problems_These nsubj_include_problems nn__collocation nn__discovery nn__ conj__smoothing cc__and conj__estimation dobj_include_ cc__and nn__question nn__answering conj__ ' +n,1778,'While SCL has been successfully applied to PoS tagging and Sentiment Analysis , its effectiveness for parsing was rather unexplored ',Blitzer,'mark_applied_While nsubjpass_applied_SCL aux_applied_has auxpass_applied_been advmod_applied_successfully advcl_unexplored_applied prep_applied_to nn_tagging_PoS pobj_to_tagging cc_tagging_and nn__Sentiment nn__Analysis conj_tagging_ poss_effectiveness_its nsubj_unexplored_effectiveness prep_effectiveness_for pobj_for_parsing cop_unexplored_was advmod_unexplored_rather ' +o,1779,'K-best suffix arrays have been used in autocomplete applications ',Church,'amod_arrays_K-best nn_arrays_suffix nsubjpass_used_arrays aux_used_have auxpass_used_been prep_used_in amod_applications_autocomplete pobj_in_applications ' +o,1780,'Given the motivations for performing a linguistically-informedextraction whichwere also put forth , among others , by Church and Hanks , and Heid -LRB- 1994 -RRB- and given the recent developmentof linguisticanalysistools , itseemsplausiblethatthe linguisticstructurewill be more and more taken intoaccountbycollocationextractionsystems ',Smadja,'det_motivations_the dep_Given_motivations prep_motivations_for pcomp_for_performing det_whichwere_a amod_whichwere_linguistically-informedextraction nsubj_put_whichwere advmod_put_also ccomp_performing_put advmod_put_forth prep_put_among pobj_among_others prep_put_by pobj_by_Church cc_Church_and nn__Hanks conj_Church_ conj__ cc__and conj__Heid appos_motivations_1994 cc_Given_and conj_Given_given det_developmentof_the amod_developmentof_recent iobj_given_developmentof dobj_given_linguisticanalysistools amod_linguisticanalysistools_itseemsplausiblethatthe dep_itseemsplausiblethatthe_linguisticstructurewill auxpass_taken_be advmod_taken_more cc_more_and conj_more_more dep_linguisticanalysistools_taken dobj_taken_intoaccountbycollocationextractionsystems ' +o,1781,'To avoid this problem we use the concept of class proposed for a word n-gram model ',Brown,'aux_avoid_To det_problem_this dobj_avoid_problem nsubj_use_we rcmod_problem_use det_concept_the dobj_use_concept prep_concept_of pobj_of_class partmod_class_proposed prep_proposed_for det_model_a nn_model_word nn_model_n-gram pobj_for_model ' +o,1782,'Much of the recent work in word alignment has focussed on improving the word alignment quality through better modeling or alternative approaches to training ',Fraser,'nsubj_focussed_Much prep_Much_of det_work_the amod_work_recent pobj_of_work prep_work_in nn_alignment_word pobj_in_alignment aux_focussed_has prep_focussed_on pcomp_on_improving det_quality_the nn_quality_word nn_quality_alignment dobj_improving_quality prep_improving_through amod__better nn__modeling pobj_through_ cc__or amod_approaches_alternative conj__approaches prep_improving_to pobj_to_training ' +o,1783,' report 8715 \% accuracy using a unigram-based SVM classifier combined with subjectivity detection ',Pang,'advmod_report_ num_\%_8715 dobj_report_\% nsubj_report_accuracy partmod_accuracy_using det_classifier_a amod_classifier_unigram-based nn_classifier_SVM dobj_using_classifier prep_using_combined dep_combined_with amod_detection_subjectivity pobj_with_detection ' +o,1784,'In acknowledgment of this fact , a series of conferences like Text Retrieval Conferences -LRB- TREC -RRB- , Message Understanding Conferences -LRB- MUC -RRB- , TIPSTER SUMMAC Text Summarization Evaluation , Document Understanding Conference -LRB- DUC -RRB- , and Text Summar Voorhees , Harman , 1999 Voorhees , E M and Harman , D K , 1999 ',Marcus,'pobj_In_acknowledgment prep_acknowledgment_of det_fact_this pobj_of_fact det_series_a appos_fact_series prep_series_of pobj_of_conferences prep_conferences_like nn__Text nn__Retrieval nn__Conferences abbrev__TREC pobj_like_ nn__Message nn__Understanding nn__Conferences abbrev__MUC conj__ nn__TIPSTER nn__SUMMAC nn__Text nn__Summarization nn__Evaluation conj__ nn__Document nn__Understanding nn__Conference abbrev__DUC conj__ cc__and nn__Text nn__Summar conj__ nn_Voorhees_ nn_Voorhees_ dep_In_Voorhees appos_Voorhees_Harman num_Voorhees_1999 amod_Voorhees_ nn_Voorhees_ appos_Voorhees_Voorhees nn_M_E appos_Voorhees_M cc_M_and conj_M_Harman nn_K_D appos_Voorhees_K appos_Voorhees_1999 ' +n,1785,'V B N P J J R ( a ) ( b ) V 2 V 1 V 2 \' V 1 \' V P V B N P w ill b e J J R Figure 1: Two different binarizations (a) and (b) of the same SCFG rule distinguished by the solid lines and dashed lines ( W e h o p e t h e s i t u a t i o n w i l l b e b e t t e r . ) N P J J R d e c o d i n g m a t c h 8 7 4 r u l e s m a t c h 6 2 r u l e s c o m p e t i n g e d g e s : 8 0 1 c o m p e t i n g e d g e s : 5 7 Figure 2: Edge competitions caused by different binarizations The edge competition problem for SMT decoding is not addressed in previous work (Zhang et al., 2006; Huang, 2007) in which each SCFG rule is binarized in a fixed way.',Zhang,'' +o,1786,'This problem has been considered for instance in for his inversion transduction grammars and has applications in the support of several tasks of automatic annotation of parallel corpora , as for instance segmentation , bracketing , phrasal and word alignment ',Wu,'det_problem_This nsubjpass_considered_problem aux_considered_has auxpass_considered_been prep_considered_for pobj_for_instance prep_instance_in pobj_in_ prep_considered_for poss_grammars_his nn_grammars_inversion nn_grammars_transduction pobj_for_grammars cc_considered_and conj_considered_has dobj_has_applications prep_applications_in det_support_the pobj_in_support prep_support_of amod_tasks_several pobj_of_tasks prep_tasks_of amod_annotation_automatic pobj_of_annotation prep_annotation_of amod_corpora_parallel pobj_of_corpora advmod_corpora_as dep_as_for pobj_for_instance nn_alignment_segmentation conj_segmentation_bracketing conj_segmentation_phrasal cc_segmentation_and conj_segmentation_word dep_corpora_alignment ' +o,1787,'Use of sententially aligned corpora for word alignment has already been recommended in ',Brown,'nsubjpass_recommended_Use prep_Use_of pobj_of_sententially partmod_sententially_aligned dobj_aligned_corpora prep_corpora_for nn_alignment_word pobj_for_alignment aux_recommended_has advmod_recommended_already auxpass_recommended_been prep_recommended_in ' +o,1788,'summarization , paraphrasing , natural language generation , and language modeling ',Collins,'nn__summarization amod__paraphrasing conj__ amod__natural nn__language nn__generation conj__ cc__and nn_modeling_language conj__modeling ' +o,1789,'53 Translation Results For the translation experiments on the BTEC task , we report the two accuracy measures BLEU and NIST as well as the two error rates : word error rate -LRB- WER -RRB- and position-independent word error rate -LRB- PER -RRB- ',Papineni,'num_Results_53 nn_Results_Translation prep_report_For det_experiments_the nn_experiments_translation pobj_For_experiments prep_experiments_on det_task_the nn_task_BTEC pobj_on_task nsubj_report_we dep_Results_report dep__the num__two nn__accuracy nn__measures nn__BLEU dobj_report_ cc__and nn__NIST conj__ dep_well_as cc__well dep_well_as det_rates_the num_rates_two nn_rates_error conj__rates nn_rate_word nn_rate_error dep__rate abbrev_rate_WER cc_rate_and amod_rate_position-independent nn_rate_word nn_rate_error conj_rate_rate abbrev_rate_PER ' +o,1790,'For a detailed description for Model 4 the reader is referred to ',Brown,'prep_referred_For det_description_a amod_description_detailed pobj_For_description prep_description_for pobj_for_Model num_Model_4 det_reader_the nsubjpass_referred_reader auxpass_referred_is xcomp_referred_to ' +o,1791,'Obtained percent agreement of 0988 and coefficient of 0975 suggest high convergence of both annotations ',Carletta,'amod_agreement_Obtained nn_agreement_percent nsubj_suggest_agreement prep_agreement_of pobj_of_0988 cc_agreement_and amod__coefficient conj_agreement_ prep__of pobj_of_0975 amod_convergence_high dobj_suggest_convergence prep_convergence_of det_annotations_both pobj_of_annotations ' +o,1792,'While early machine learning approaches for the task relied on local , discriminative classifiers , more recent approaches use joint and\\/or global models ','Daume III','mark__While amod_machine_early nsubj__machine partmod_machine_learning dobj_learning_approaches prep_learning_for det_task_the pobj_for_task partmod_task_relied prep_relied_on amod_classifiers_local amod_classifiers_discriminative pobj_on_classifiers advcl_use_ advmod_approaches_more amod_approaches_recent nsubj_use_approaches amod_models_joint amod_models_and\\/or amod_models_global dobj_use_models ' +o,1793,'More recent papers , Pereira and Tishby -LRB- 1992 -RRB- proposed to cluster nouns on the basis of a metric derived from the distribution of subject , verb and object in the texts ',Hindle,'amod_papers_More amod_papers_recent nsubj_proposed_ conj__Pereira cc__and conj__Tishby appos__1992 dep_papers_proposed prep_proposed_to nn_nouns_cluster pobj_to_nouns prep_proposed_on det_basis_the pobj_on_basis prep_basis_of det_metric_a pobj_of_metric partmod_metric_derived prep_derived_from det_distribution_the pobj_from_distribution prep_distribution_of pobj_of_subject conj_subject_verb cc_subject_and conj_subject_object prep_derived_in det_texts_the pobj_in_texts ' +o,1794,'A word is considered to be known when it has an ambiguous tag -LRB- henceforth ambitag -RRB- attributed to it in the LEXICON , which is compiled in the same way as for the MBT-tagger ',Berger,'det_word_A nsubjpass_considered_word auxpass_considered_is aux_known_to auxpass_known_be xcomp_considered_known advmod_has_when nsubj_has_it advcl_known_has det_tag_an amod_tag_ambiguous dobj_has_tag amod_ambitag_henceforth appos_tag_ambitag partmod_tag_attributed prep_attributed_to pobj_to_it prep_attributed_in det_LEXICON_the pobj_in_LEXICON nsubjpass_compiled_which auxpass_compiled_is rcmod_LEXICON_compiled prep_compiled_in det_way_the amod_way_same pobj_in_way dep_for_as prep_way_for det__the amod__MBT-tagger pobj_for_ ' +o,1795,'Semantic classification programs use statistical information based on cooccurrence with appropriate marker words to partition a set of words into semantic groups or classes ',Brown,'amod_programs_Semantic nn_programs_classification nsubj__programs nn_information_use amod_information_statistical dobj__information partmod_information_based prep_based_on pobj_on_cooccurrence prep_cooccurrence_with amod_words_appropriate nn_words_marker pobj_with_words aux_partition_to xcomp_based_partition det_set_a dobj_partition_set prep_set_of pobj_of_words prep_partition_into amod_groups_semantic pobj_into_groups cc_groups_or conj_groups_classes ' +o,1796,'317 Citation Observed data Hidden data Treebank tree with head child annotated on each nonterminal No hidden data ',Collins,'num_data_317 nn_data_Citation nn_data_Observed nn_data_data amod_data_Hidden nn_tree_ nn_tree_Treebank dep_data_tree prep_tree_with nn_annotated_head nn_annotated_child pobj_with_annotated prep_annotated_on det_nonterminal_each pobj_on_nonterminal det_data_No amod_data_hidden dep_data_data ' +o,1797,'Word alignments traditionally are based on IBM Models 1-5 or on HMMs ',Brown,'nsubjpass_based_alignments advmod_based_traditionally auxpass_based_are ccomp_Word_based prep_based_on nn__IBM nn__Models nn__1-5 pobj_on_ cc_Word_or conj_Word_on nn__HMMs pobj_on_ ' +o,1798,'32 This problem is also a central concern in the work by ',Bean,'det_problem_This nsubj_concern_problem cop_concern_is advmod_concern_also det_concern_a amod_concern_central dep_32_concern prep_concern_in det_work_the pobj_in_work prep_concern_by ' +p,1799,'The corpus-based statistical parsing community has many fast and accurate automated parsing systems , including systems produced by , Charniak -LRB- 1997 -RRB- and Ratnaparkhi -LRB- 1997 -RRB- ',Collins,'det_community_The amod_community_corpus-based amod_community_statistical amod_community_parsing nsubj_has_community amod_systems_many dep_systems_fast cc_fast_and conj_fast_accurate amod_systems_automated nn_systems_parsing dobj_has_systems prep_systems_including pobj_including_systems partmod_systems_produced prep_produced_by nn_Charniak_ pobj_by_Charniak appos_Charniak_1997 cc_Charniak_and conj_Charniak_Ratnaparkhi appos_Charniak_1997 ' +o,1800,'41 Part-of-speech tagging experiments We split the Penn Treebank corpus into training , development and test sets as in ',Collins,'num_experiments_41 amod_experiments_Part-of-speech nn_experiments_tagging nsubj_split_We dep_experiments_split det__the nn__Penn nn__Treebank nn__corpus dobj_split_ prep_split_into nn_sets_training conj_training_development cc_training_and conj_training_test pobj_into_sets prep_split_as dep_as_in ' +o,1801,'BLEU For all translation tasks , we report caseinsensitive NIST BLEU scores using 4 references per sentence ',Papineni,'dep_For_BLEU prep_report_For det_tasks_all nn_tasks_translation pobj_For_tasks nsubj_report_we amod_scores_caseinsensitive nn_scores_NIST nn_scores_BLEU nsubj_using_scores dep_using_ xcomp_report_using num_references_4 dobj_using_references prep_references_per pobj_per_sentence ' +o,1802,'The success of statistical methods in particular has been quite evident in the area of syntactic parsing , most recently with the outstanding results of and on the now-standard English test set of the Penn Treebank ',Marcus,'det_success_The nsubj_evident_success prep_success_of amod_methods_statistical pobj_of_methods prep_methods_in pobj_in_particular aux_evident_has dep_evident_been advmod_evident_quite prep_evident_in det_area_the pobj_in_area prep_area_of amod_parsing_syntactic pobj_of_parsing advmod_recently_most advmod_parsing_recently prep_evident_with det_results_the amod_results_outstanding pobj_with_results prep_results_of pobj_of_ cc__and conj__ prep_evident_on det_set_the amod_set_now-standard amod_set_English nn_set_test pobj_on_set prep_set_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,1803,'Learning to Disambiguate Word Senses Several recent research projects have taken a corpus-based approach to lexical disambiguation ',Marcus,'amod_projects_Learning prep_Learning_to nn_research_Disambiguate nn_research_Word nn_research_Senses amod_research_Several amod_research_recent pobj_to_research nsubj_taken_projects aux_taken_have det_approach_a amod_approach_corpus-based dobj_taken_approach prep_taken_to amod__lexical nn__disambiguation pobj_to_ ' +o,1804,'Experimental Comparison 41 Experiments on the ATIS corpus For our first comparison , we used I0 splits from the Penn ATIS corpus into training sets of 675 sentences and test sets of 75 sentences ',Marcus,'nsubj_Comparison_Experimental num_Experiments_41 dobj_Comparison_Experiments prep_Experiments_on det_corpus_the nn_corpus_ATIS pobj_on_corpus prep_used_For poss_comparison_our amod_comparison_first pobj_For_comparison nsubj_used_we dep_Comparison_used nsubj_splits_I0 ccomp_used_splits prep_splits_from det__the nn__Penn nn__ATIS nn__corpus pobj_from_ prep_splits_into nn_sets_training pobj_into_sets prep_sets_of amod_sentences_675 pobj_of_sentences cc_sets_and nn_sets_test conj_sets_sets prep_sets_of num_sentences_75 pobj_of_sentences ' +o,1805,'Research in the field of unsupervised and weakly supervised parsing ranges from various forms of EM training over bootstrapping approaches like selftraining to feature-based enhancements of discriminative reranking models and the application of semisupervised SVMs ',McClosky,'nsubj_supervised_Research prep_Research_in det_field_the pobj_in_field prep_field_of pobj_of_unsupervised cc_unsupervised_and conj_unsupervised_weakly amod_ranges_parsing dobj_supervised_ranges prep_supervised_from amod_forms_various pobj_from_forms prep_forms_of nn__EM nn__training pobj_of_ prep_supervised_over pcomp_over_bootstrapping dobj_bootstrapping_approaches prep_bootstrapping_like pcomp_like_selftraining amod__ prep__to amod_enhancements_feature-based pobj_to_enhancements prep_enhancements_of amod_models_discriminative amod_models_reranking pobj_of_models dobj_selftraining_ cc__and det_application_the conj__application prep_application_of amod__semisupervised nn__SVMs pobj_of_ ' +o,1806,' used patterns representing part-of-speech sequences , recognized adjectival phrases , and learned N-grams ',Turney,'nsubj_used_ dobj_used_patterns xcomp_used_representing amod_sequences_part-of-speech dobj_representing_sequences nsubj_recognized_ ccomp_used_recognized amod_phrases_adjectival dobj_recognized_phrases cc_used_and nsubj_learned_ conj_used_learned dobj_learned_N-grams ' +o,1807,'22 Inversion Transduction Grammar s inversion transduction grammar -LRB- ITG -RRB- is a synchronous grammar formalism in which derivations of sentence pairs correspond to alignments ',Wu,'num_grammar_22 nn_grammar_Inversion nn_grammar_Transduction nn_grammar_Grammar nn_grammar_ nn_grammar_s nn_grammar_inversion nn_grammar_transduction nsubj_formalism_grammar abbrev_grammar_ITG cop_formalism_is det_formalism_a amod_formalism_synchronous nn_formalism_grammar rel_correspond_in pobj_in_which nsubj_correspond_derivations prep_derivations_of nn_pairs_sentence pobj_of_pairs rcmod_formalism_correspond prep_correspond_to pobj_to_alignments ' +o,1808,'NER is typically viewed as a sequential prediction problem , the typical models include HMM , CRF , and sequential application of Perceptron or Winnow ',Collins,'nsubjpass_viewed_NER auxpass_viewed_is advmod_viewed_typically prep_include_as det_problem_a amod_problem_sequential nn_problem_prediction pobj_as_problem det_models_the amod_models_typical nsubj_include_models dep_viewed_include nn__HMM dobj_include_ nn__CRF conj__ cc__and amod_application_sequential conj__application prep_application_of pobj_of_Perceptron cc_Perceptron_or conj_Perceptron_Winnow ' +o,1809,'The published F score for voted perceptron is 9353 \% with a different feature set ',Collins,'det_F_The amod_F_published nsubj_score_F mark_voted_for advcl_score_voted nsubj_\%_perceptron cop_\%_is num_\%_9353 ccomp_voted_\% prep_\%_with det_feature_a amod_feature_different pobj_with_feature partmod_feature_set ' +o,1810,'Therefore , the base forms have been introduced manually and the POS tags have been provided partly manually and partly automatically using a statistical maximum-entropy based POS tagger similar to the one described in ',Ratnaparkhi,'advmod_introduced_Therefore det_forms_the nn_forms_base nsubjpass_introduced_forms aux_introduced_have auxpass_introduced_been advmod_introduced_manually cc_introduced_and det_tags_the dep_tags_POS nsubjpass_provided_tags aux_provided_have auxpass_provided_been conj_introduced_provided prep_provided_partly dep_partly_manually cc_partly_and conj_partly_partly dep_partly_automatically xcomp_provided_using det_maximum-entropy_a amod_maximum-entropy_statistical dobj_using_maximum-entropy partmod_maximum-entropy_based dep_tagger_POS nsubj_similar_tagger xcomp_based_similar prep_similar_to det_one_the pobj_to_one partmod_one_described prep_described_in ' +o,1811,'To solve this problem , we will adapt the idea of null generated words from machine translation ',Brown,'aux_solve_To dep_adapt_solve det_problem_this dobj_solve_problem nsubj_adapt_we aux_adapt_will det_idea_the dobj_adapt_idea prep_idea_of pobj_of_null partmod_null_generated dobj_generated_words prep_generated_from nn_translation_machine pobj_from_translation ' +o,1812,'4 Architecture of the SMT system The goal of statistical machine translation (SMT) is to produce a target sentence e from a source sentence f. It is today common practice to use phrases as translation units (Koehn et al., 2003; Och and Ney, 2003) and a log linear framework in order to introduce several models explaining the translation process: e = argmaxp(e|f) = argmaxe {exp(summationdisplay i ihi(e,f))} (1) The feature functions hi are the system models and the i weights are typically optimized to maximize a scoring function on a development set (Och and Ney, 2002).',Koehn,'' +o,1813,'In the training phase , bilingual parallel sentences are preprocessed and aligned using alignment algorithms or tools such as GIZA + + ',Och,'prep_preprocessed_In det_phase_the nn_phase_training pobj_In_phase amod_sentences_bilingual amod_sentences_parallel nsubjpass_preprocessed_sentences auxpass_preprocessed_are cc_preprocessed_and conj_preprocessed_aligned xcomp_preprocessed_using amod_algorithms_alignment dobj_using_algorithms cc_algorithms_or conj_algorithms_tools dep_as_such prep_algorithms_as nn_+_GIZA nn_+_+ pobj_as_+ ' +o,1814,'A natural fit to the existing statistical machine translation framework A metric that ranks a good translation high in an nbest list could be easily integrated in a minimal error rate statistical machine translation training framework ',Och,'det_fit_A amod_fit_natural prep_fit_to det_framework_the amod_framework_existing amod_framework_statistical nn_framework_machine nn_framework_translation pobj_to_framework det_metric_A nsubjpass_integrated_metric nsubj_ranks_that rcmod_metric_ranks det_high_a amod_high_good nn_high_translation dobj_ranks_high prep_high_in det_list_an amod_list_nbest pobj_in_list aux_integrated_could auxpass_integrated_be advmod_integrated_easily dep_fit_integrated prep_integrated_in det_framework_a amod_framework_minimal nn_framework_error nn_framework_rate amod_framework_statistical nn_framework_machine nn_framework_translation nn_framework_training pobj_in_framework ' +p,1815,'Head-lexicalized stochastic grammars have recently become increasingly popular ',Collins,'amod_grammars_Head-lexicalized amod_grammars_stochastic nsubj_become_grammars aux_become_have advmod_become_recently advmod_popular_increasingly dep_become_popular ' +p,1816,'Unsupervised algorit ~ m ~ such as have reported good accuracy that rivals that of supervised algorithms ',Yarowsky,'amod_~_Unsupervised amod_~_algorit amod_~_~ nn_~_m nsubj_reported_~ dep_as_such prep_~_as pobj_as_ aux_reported_have amod_accuracy_good dobj_reported_accuracy nsubj_rivals_that rcmod_accuracy_rivals dobj_rivals_that prep_that_of amod_algorithms_supervised pobj_of_algorithms ' +o,1817,'The set of such ITG alignments , AITG , are a strict subset of A1-1 ',Wu,'det_set_The nsubj_subset_set prep_set_of amod_alignments_such nn_alignments_ITG pobj_of_alignments appos_set_AITG cop_subset_are det_subset_a amod_subset_strict prep_subset_of pobj_of_A1-1 ' +o,1818,'Also , we used Adwait Ratnaparkhis part-of-speech tagger to tag unknown words in the test data ',Ratnaparkhi,'advmod_used_Also nsubj_used_we nn_tagger_Adwait nn_tagger_Ratnaparkhis amod_tagger_part-of-speech nsubj__tagger ccomp_used_ aux_tag_to xcomp__tag amod_words_unknown dobj_tag_words prep_tag_in det_data_the nn_data_test pobj_in_data ' +p,1819,' is one of the most famous work that discussed learning polarity from corpus ',Turney,'nsubj_one_ cop_one_is prep_one_of det_work_the advmod_famous_most amod_work_famous pobj_of_work nsubj_discussed_that rcmod_work_discussed xcomp_discussed_learning advmod_learning_polarity prep_learning_from pobj_from_corpus ' +o,1820,'One of the earliest attempts at extracting interrupted collocations \' -LRB- ie non-contiguous collocations , including VPCs -RRB- , was that of ',Smadja,'nsubj_that_One prep_One_of det_attempts_the amod_attempts_earliest pobj_of_attempts prep_attempts_at pcomp_at_extracting dep_extracting_interrupted dep_interrupted_collocations dep_collocations_ie amod_collocations_non-contiguous dobj_ie_collocations prep_collocations_including pobj_including_VPCs cop_that_was prep_that_of ' +o,1821,'For example , if the lexicon contains an adjective excellent , it matches every adjective phrase that includes excellent such as view-excellent etc As a baseline , we built lexicon similarly by using polarity value of ',Turney,'prep_matches_For pobj_For_example mark_contains_if det_lexicon_the nsubj_contains_lexicon dep_matches_contains det_adjective_an nsubj_excellent_adjective xcomp_contains_excellent nsubj_matches_it det_phrase_every nn_phrase_adjective dobj_matches_phrase nsubj_includes_that rcmod_phrase_includes acomp_includes_excellent dep_as_such prep_excellent_as pobj_as_view-excellent dep_view-excellent_etc prep_built_As det_baseline_a pobj_As_baseline nsubj_built_we dep_matches_built dobj_built_lexicon advmod_built_similarly prep_built_by pcomp_by_using amod_value_polarity dobj_using_value prep_using_of ' +o,1822,'Existing statistical NLG -LRB- i -RRB- uses corpus statistics to inform heuristic decisions in what is otherwise symbolic generation ; -LRB- ii -RRB- applies n-gram models to select the overall most likely realisation after generation -LRB- HALOGEN family -RRB- ; or -LRB- iii -RRB- reuses an existing parsing grammar or treebank for surface realisation ',Cahill,'amod_NLG_Existing amod_NLG_statistical nsubj_uses_NLG appos_NLG_i nn_statistics_corpus nsubj_inform_statistics aux_inform_to xcomp_uses_inform amod_decisions_heuristic dobj_inform_decisions prep_decisions_in nsubj__what cop__is advmod__otherwise amod__symbolic nn__generation pcomp_in_ dep_n-gram_ii amod_n-gram_applies dep_decisions_n-gram dep_n-gram_models aux_select_to infmod_models_select det_realisation_the amod_realisation_overall advmod_likely_most amod_realisation_likely dobj_select_realisation prep_select_after pobj_after_generation nn_family_HALOGEN abbrev_generation_family cc_decisions_or dep_reuses_iii conj_decisions_reuses det_parsing_an amod_parsing_existing dep_for_parsing dep_for_grammar cc_grammar_or conj_grammar_treebank prep_reuses_for nn_realisation_surface pobj_for_realisation ' +o,1823,'GIZA + + refined alignments have been used in state-of-the-art phrase-based statistical MT systems such as ; variations on the refined heuristic have been used by -LRB- diag and diag-and -RRB- and by the phrase-based system Moses -LRB- grow-diag-final -RRB- ',Och,'amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ amod_alignments_refined nsubjpass_used_alignments aux_used_have auxpass_used_been prep_used_in amod_systems_state-of-the-art amod_systems_phrase-based amod_systems_statistical nn_systems_MT pobj_in_systems dep_as_such prep_systems_as pobj_as_ nsubjpass_used_variations prep_variations_on det_heuristic_the amod_heuristic_refined pobj_on_heuristic aux_used_have auxpass_used_been parataxis_used_used prep_used_by pobj_by_ dep__diag cc_diag_and conj_diag_diag-and cc_by_and conj_by_by det__the amod__phrase-based nn__system nn__Moses appos__grow-diag-final pobj_by_ ' +p,1824,'This method , initially proposed by , was successfully evaluated in the context of the SENSEVAL framework ',Yarowsky,'det_method_This nsubjpass_evaluated_method advmod_proposed_initially partmod_method_proposed prep_proposed_by pobj_by_ auxpass_evaluated_was advmod_evaluated_successfully prep_evaluated_in det_context_the pobj_in_context prep_context_of det__the nn__SENSEVAL nn__framework pobj_of_ ' +o,1825,'IC function is a derivative of Fano \'s mutual information formula recently used by to compute word co-occurrence patterns in a 44 million word corpus of Associated Press news stories ',Church,'nn_function_IC nsubj_derivative_function cop_derivative_is det_derivative_a prep_derivative_of poss_formula_Fano possessive_Fano_\'s amod_formula_mutual nn_formula_information pobj_of_formula advmod_used_recently partmod_formula_used prep_used_by pobj_by_ aux_compute_to xcomp_used_compute nn_patterns_word amod_patterns_co-occurrence dobj_compute_patterns prep_compute_in det_corpus_a number_million_44 num_corpus_million nn_corpus_word pobj_in_corpus prep_corpus_of nn_stories_Associated nn_stories_Press nn_stories_news pobj_of_stories ' +o,1826,'This represents the translation probability of a phrase when it is decomposed into a series of independent word-for-word translation steps , and has proven a very effective feature ',Koehn,'nsubj_represents_This det_probability_the nn_probability_translation dobj_represents_probability prep_probability_of det_phrase_a pobj_of_phrase advmod_decomposed_when nsubjpass_decomposed_it auxpass_decomposed_is dep_phrase_decomposed prep_decomposed_into det_series_a pobj_into_series prep_series_of amod__independent amod__word-for-word nn__translation nn__steps pobj_of_ cc_represents_and aux_proven_has conj_represents_proven det__a advmod_effective_very amod__effective nn__feature dobj_proven_ ' +o,1827,'Collins head words finder rules have been modified to extract semantic head word ',Manning,'nn_words_Collins nn_words_head nsubj_finder_words nsubjpass_modified_rules aux_modified_have auxpass_modified_been ccomp_finder_modified aux_extract_to xcomp_modified_extract amod_word_semantic nn_word_head dobj_extract_word ' +o,1828,'For instance , train an independent subjectivity classifier to identify and remove objective sentences from a review prior to polarity classification ',Pang,'prep_train_For pobj_For_instance nsubj_train_ det_classifier_an amod_classifier_independent nn_classifier_subjectivity nsubj_identify_classifier aux_identify_to xcomp_train_identify cc_identify_and conj_identify_remove amod_sentences_objective dobj_identify_sentences prep_sentences_from det_review_a pobj_from_review dep_to_prior prep_identify_to amod_classification_polarity pobj_to_classification ' +o,1829,'Ordinary Prologstyle , backchaining deduction is augmented with the capability of making assumptions and of factoring two goal literals that are unifiable ',Hobbs,'amod_deduction_Ordinary nn_deduction_Prologstyle amod_deduction_backchaining nsubjpass_augmented_deduction auxpass_augmented_is prep_augmented_with det_capability_the pobj_with_capability prep_capability_of pcomp_of_making dobj_making_assumptions cc_with_and conj_with_of pcomp_of_factoring num_literals_two nn_literals_goal dobj_factoring_literals nsubj_unifiable_that cop_unifiable_are rcmod_literals_unifiable ' +o,1830,'Pearsons correlation coefficient is a standard measure of the correlation strength between two distributions; it can be calculated as follows: = E(XY ) E(X)E(Y )radicalbigE(X2) [E(X)]2radicalbigE(Y 2) [E(Y )]2 (1) where X = (x1,,xn) and Y = (y1,,yn) are vectors of numerical scores for each paraphrase provided by the humans and the competing systems, respectively, n is the number of paraphrases to score, and E(X) is the expectation of X. Cosine correlation coefficient is another popular alternative and was used by Nakov and Hearst (2008); it can be seen as an uncentered version of Pearsons correlation coefficient: = X.YbardblXbardblbardblYbardbl (2) Spearmans rank correlation coefficient is suitable for comparing rankings of sets of items; it is a special case of Pearsons correlation, derived by considering rank indices (1,2,) as item scores . It is defined as follows: = n summationtextx iyi ( summationtextx i)( summationtexty i)radicalBig nsummationtextx2i (summationtextxi)2 radicalBig nsummationtexty2i (summationtextyi)2 (3) One problem with using Spearmans rank coefficient for the current task is the assumption that swapping any two ranks has the same effect.',Nakov,'' +o,1831,'3 Methodology Similar to , we use comparison to human assocation datasets as a test bed for the scores produced by computational association measures ',Rapp,'num_Methodology_3 dep_Similar_Methodology aux__to xcomp_Similar_ nsubj_use_we ccomp__use dobj_use_comparison prep_comparison_to amod_datasets_human nn_datasets_assocation pobj_to_datasets prep_use_as det_bed_a nn_bed_test pobj_as_bed prep_bed_for det_scores_the pobj_for_scores partmod_scores_produced prep_produced_by amod_measures_computational nn_measures_association pobj_by_measures ' +p,1832,'23 Classifier Training We chose maximum entropy as our primary classifier because the highest performing systems in both the SemEval-2007 preposition sense disambiguation task and the general word sense disambiguation task used it ',Berger,'num_Training_23 nn_Training_Classifier nsubj_chose_We dep_Training_chose amod__maximum nn__entropy dobj_chose_ prep_chose_as poss_classifier_our amod_classifier_primary pobj_as_classifier prep_chose_because det_systems_the amod_systems_highest amod_systems_performing pobj_because_systems prep_systems_in preconj__both det__the amod__SemEval-2007 nn__preposition nn__sense nn__disambiguation nn__task pobj_in_ cc__and det_sense_the amod_sense_general nn_sense_word conj__sense nn__disambiguation nn__task nsubj_used_ dep_Training_used dobj_used_it ' +o,1833,'A summary of the differences between our proposed approach and that of would include : The reliance of BLEU on the diversity of multiple reference translations in order to capture some of the acceptable alternatives in both word choice and word ordering that we have shown above ',Papineni,'det_summary_A nsubj_include_summary prep_summary_of det_differences_the pobj_of_differences prep_differences_between poss_approach_our amod_approach_proposed pobj_between_approach cc_differences_and conj_differences_that prep_that_of pobj_of_ aux_include_would det_reliance_The dobj_include_reliance prep_reliance_of pcomp_of_BLEU prep_BLEU_on det_diversity_the pobj_on_diversity prep_diversity_of amod_translations_multiple nn_translations_reference pobj_of_translations mark_capture_in dep_capture_order aux_capture_to purpcl_BLEU_capture dobj_capture_some prep_some_of det_alternatives_the amod_alternatives_acceptable pobj_of_alternatives prep_capture_in preconj_choice_both nn_choice_word pobj_in_choice cc_choice_and conj_choice_word partmod_choice_ordering complm_shown_that nsubj_shown_we aux_shown_have ccomp_ordering_shown prep_shown_above ' +o,1834,'As a result , the problem of opinion mining has seen increasing attention over the last three years from and many others ',Turney,'prep_seen_As det_result_a pobj_As_result det_problem_the nsubj_seen_problem prep_problem_of nn_mining_opinion pobj_of_mining aux_seen_has xcomp_seen_increasing dobj_increasing_attention prep_attention_over det_years_the amod_years_last num_years_three pobj_over_years prep_increasing_from amod_others_ cc__and conj__many pobj_from_others ' +p,1835,'Each component model takes the exponential form: a37a55a38a57a56 a51 a42a6a44a59a58a60a56 a61 a51a64a63a65a53a67a66 a53 a45a46a70 a71a16a72a21a73a75a74a77a76a79a78a81a80 a78a16a82a11a78 a38a83a44a59a58a60a56a84a61 a51a64a63a65a53a67a66 a53 a58a60a56 a51 a45a86a85 a87 a38a83a44a59a58a60a56a84a61 a51a64a63a65a53a67a66 a53 a45 a58 (2) where a87 a38a83a44a59a58a60a56 a61 a51a41a63a65a53a67a66 a53 a45 is a normalization term to ensure that a37a55a38a57a56 a51a42a6a44a88a58a60a56a62a61 a51a41a63a65a53a67a66 a53 a45 is a probability, a82a11a78 a38a83a44a59a58a60a56 a61 a51a64a63a65a53a67a66 a53 a58a60a56 a51 a45 is a feature function (often binary) and a80 a78 is the weight ofa82a21a78 . Given a set of features and a corpus of training data, there exist ef cient training algorithms (Darroch and Ratcliff, 1972; Berger et al. , 1996) to nd the optimal parameters a89 a80 a78a14a90 . The art of building a maximum entropy parser then reduces to choosing good features.',Berger,'' +o,1836,'Our learning algorithm stems from Perceptron training in ',Collins,'poss_algorithm_Our nn_algorithm_learning nsubj_stems_algorithm prep_stems_from pobj_from_Perceptron partmod_Perceptron_training prt_training_in ' +o,1837,'4 Relation to Previous Work There is a significant volume of work exploring the use of CRFs for a variety of chunking tasks , including named-entity recognition , gene prediction , shallow parsing and others ',Finkel,'num_Relation_4 dep_Relation_to amod_Work_Previous pobj_to_Work expl_is_There rcmod_Work_is det_volume_a amod_volume_significant nsubj_is_volume prep_volume_of pobj_of_work partmod_work_exploring det_use_the dobj_exploring_use prep_use_of pobj_of_CRFs prep_exploring_for det_variety_a pobj_for_variety prep_variety_of amod_tasks_chunking pobj_of_tasks prep_Work_including amod_recognition_named-entity pobj_including_recognition nn_prediction_gene conj_recognition_prediction nn_parsing_shallow conj_recognition_parsing cc_recognition_and conj_recognition_others ' +o,1838,'Finally we trained model weights by maximizing BLEU and set decoder optimization parameters -LRB- n-best list size , timeouts 14 etc -RRB- on a development test set of 200 held-out sentences each with a single reference translation ',Och,'dep_Finally_we dep_we_trained nn_weights_model dobj_trained_weights prep_trained_by pcomp_by_maximizing amod__BLEU dobj_maximizing_ cc_trained_and conj_trained_set nn_parameters_decoder nn_parameters_optimization dobj_set_parameters amod_size_n-best nn_size_list dep_parameters_size nn_etc_timeouts num_etc_14 appos_size_etc prep_parameters_on det_set_a nn_set_development nn_set_test pobj_on_set prep_set_of num_sentences_200 amod_sentences_held-out pobj_of_sentences dep_sentences_each prep_set_with det_translation_a amod_translation_single nn_translation_reference pobj_with_translation ' +o,1839,'For subproblem -LRB- a -RRB- , we have devised a new method , based on LPR , which has some good properties not shared by the methods proposed so far ',Marcus,'prep_devised_For pobj_For_subproblem dep_-LRB-_a nsubj_devised_we aux_devised_have det_method_a amod_method_new dobj_devised_method partmod_method_based prep_based_on pobj_on_LPR nsubj_shared_which aux_shared_has det_properties_some amod_properties_good dep_not_properties neg_shared_not rcmod_method_shared prep_shared_by det_methods_the pobj_by_methods partmod_methods_proposed advmod__so advmod__far acomp_proposed_ ' +o,1840,'Recent computational work either focuses on sentence subjectivity , concentrates just on explicit statements of evaluation , such as of films , or focuses on just one aspect of opinion , eg , on adjectives ',Pang,'amod_work_Recent amod_work_computational nsubj_focuses_work advmod_focuses_either prep_focuses_on nn__sentence nn__subjectivity pobj_on_ conj_focuses_concentrates advmod_concentrates_just prep_concentrates_on amod_statements_explicit pobj_on_statements prep_statements_of pobj_of_evaluation dep_as_such prep_statements_as dep_as_of nn__films pobj_of_ cc_focuses_or conj_focuses_focuses prep_focuses_on advmod_aspect_just num_aspect_one pobj_on_aspect prep_aspect_of nn__opinion dep__eg pobj_of_ prep_focuses_on pobj_on_adjectives ' +o,1841,'NP chunks in the shared task data are BaseNPs , which are non-recursive NPs , a definition first proposed by ',Ramshaw,'nn_chunks_NP nsubj_BaseNPs_chunks prep_chunks_in det_data_the amod_data_shared nn_data_task pobj_in_data cop_BaseNPs_are nsubj_NPs_which cop_NPs_are amod_NPs_non-recursive rcmod_BaseNPs_NPs det_definition_a appos_NPs_definition advmod_proposed_first partmod_definition_proposed prep_proposed_by ' +o,1842,'We adopted an N-best hypothesis approach to train ',Och,'nsubj_adopted_We det_approach_an amod_approach_N-best nn_approach_hypothesis nsubj__approach ccomp_adopted_ aux_train_to xcomp__train ' +o,1843,'Above the phrase level , some models perform no reordering , some have a simple distortion model that reorders phrases independently of their content , and some , for example , the Alignment Template System , hereafter ATS , and the IBM phrase-based system , have phrase-reordering models that add some lexical sensitivity ',Och,'dep_have_Above det_level_the nn_level_phrase pobj_Above_level det_models_some nsubj_perform_models dep_have_perform det__no nn__reordering dobj_perform_ nsubj_have_some det_model_a amod_model_simple nn_model_distortion dobj_have_model nsubj_reorders_that rcmod_model_reorders dobj_reorders_phrases dep_of_independently prep_phrases_of poss__their nn__content pobj_of_ cc_have_and nsubj_have_some prep_have_for pobj_for_example det__the nn__Alignment nn__Template nn__System conj_example_ nn_ATS_hereafter conj_example_ATS cc_example_and det__the nn__IBM amod__phrase-based nn__system conj_example_ conj_have_have amod_models_phrase-reordering dobj_have_models nsubj_add_that rcmod_models_add det_sensitivity_some amod_sensitivity_lexical dobj_add_sensitivity ' +o,1844,'Comparison With Previous Work Most of the recent corpus-based POS taggers in the literature are either statistically based , and use Markov Model or Statistical Decision Tree -LRB- SDT -RRB- techniques , or are primarily rule based , such as Drill \'s Transformation Based Learner -LRB- TBL -RRB- ',Marcus,'dep_Comparison_With amod_Work_Previous pobj_With_Work nsubjpass_based_Most prep_Most_of det_taggers_the amod_taggers_recent amod_taggers_corpus-based dep_taggers_POS pobj_of_taggers prep_taggers_in det_literature_the pobj_in_literature auxpass_based_are advmod_based_either advmod_based_statistically rcmod_Work_based cc_based_and conj_based_use nn__Markov nn__Model dobj_use_ cc__or nn_techniques_Statistical nn_techniques_Decision nn_techniques_Tree nn_techniques_ abbrev_techniques_SDT conj__techniques cc_based_or cop_rule_are advmod_rule_primarily conj_based_rule amod_Work_based dep_as_such prep_Work_as poss__Drill possessive_Drill_\'s nn__Transformation nn__Based nn__Learner pobj_as_ abbrev__TBL ' +o,1845,'Previous studies have shed light on the predictability of the next unix command that a user will enter , the next keystrokes on a small input device such as a PDA , and of the translation that a human translator will choose for a given foreign sentence ',Brown,'amod_studies_Previous nsubj_light_studies aux_light_have dep_light_shed prep_light_on det_predictability_the pobj_on_predictability prep_predictability_of det_command_the amod_command_next nn_command_unix pobj_of_command complm_enter_that det_user_a nsubj_enter_user aux_enter_will ccomp_light_enter dobj_enter_ det_keystrokes_the amod_keystrokes_next appos__keystrokes prep_keystrokes_on det_device_a amod_device_small nn_device_input pobj_on_device dep_as_such prep_device_as det__a nn__PDA pobj_as_ cc_on_and conj_on_of det_translation_the pobj_of_translation complm_choose_that det_translator_a amod_translator_human nsubj_choose_translator aux_choose_will ccomp_enter_choose prep_choose_for det__a amod__given amod__foreign nn__sentence pobj_for_ ' +o,1846,'5 Effectiveness Comparison 51 English-Chinese ATIS Models Both the transfer and transducer systems were trained and evaluated on English-to-Mandarin Chinese translation of transcribed utterances from the ATIS corpus ',Dunning,'num_Comparison_5 nn_Comparison_Effectiveness dep_transfer_Comparison num_Models_51 amod_Models_English-Chinese nn_Models_ATIS dep_Comparison_Models dep_transfer_Both det_transfer_the cc_transfer_and nn_systems_transducer conj_transfer_systems auxpass_trained_were dep_transfer_trained cc_trained_and conj_trained_evaluated prep_trained_on amod_translation_English-to-Mandarin amod_translation_Chinese pobj_on_translation prep_translation_of amod_utterances_transcribed pobj_of_utterances prep_utterances_from det__the nn__ATIS nn__corpus pobj_from_ ' +o,1847,'Using techniques described in , Church and Hanks -LRB- 1990 -RRB- , and Hindle and Rooth -LRB- 1991 -RRB- , below are some examples of the most frequent V-O pairs from the AP corpus ',Church,'dep_are_Using dobj_Using_techniques partmod_techniques_described prep_described_in pobj_in_ conj__Church cc__and conj__Hanks appos_techniques_1990 cc_techniques_and conj_techniques_Hindle cc_Hindle_and conj_Hindle_Rooth appos_Hindle_1991 advmod_are_below det_examples_some nsubj_are_examples prep_examples_of det_pairs_the advmod_frequent_most amod_pairs_frequent amod_pairs_V-O pobj_of_pairs prep_pairs_from det_corpus_the nn_corpus_AP pobj_from_corpus ' +o,1848,'First , the graph-based models have better precision than the transition-based models when predicting long arcs , which is compatible with the results of ',McDonald,'advmod_have_First det_models_the amod_models_graph-based nsubj_have_models amod_precision_better dobj_have_precision prep_precision_than det_models_the amod_models_transition-based pobj_than_models advmod_predicting_when advcl_have_predicting amod_arcs_long dobj_predicting_arcs nsubj_compatible_which cop_compatible_is rcmod_arcs_compatible prep_compatible_with det_results_the pobj_with_results prep_results_of ' +o,1849,'Brown , uses the same bigrams and by means of a greedy algorithm forms the hierarchical clusters of words ',Brown,'nsubj_forms_Brown amod_uses_ nsubj_forms_uses det_bigrams_the amod_bigrams_same dep_uses_bigrams cc_bigrams_and conj_bigrams_by pobj_by_means prep_means_of det_algorithm_a amod_algorithm_greedy pobj_of_algorithm det_clusters_the amod_clusters_hierarchical dobj_forms_clusters prep_clusters_of pobj_of_words ' +o,1850,' ; -RRB- ',Och,'dep__ ' +o,1851,'Lexical collocation functions , especially those determined statistically , have recently attracted considerable attention in computational linguistics mainly , though not exclusively , for use in disambiguation ',Dunning,'amod_functions_Lexical nn_functions_collocation nsubj_attracted_functions advmod_statistically_especially det_statistically_those amod_statistically_determined appos_functions_statistically aux_attracted_have advmod_attracted_recently amod_attention_considerable dobj_attracted_attention prep_attention_in amod_linguistics_computational pobj_in_linguistics partmod_linguistics_ advmod__mainly mark_exclusively_though neg_exclusively_not parataxis__exclusively prep_attracted_for pobj_for_use prep_use_in pobj_in_disambiguation ' +o,1852,'As a first step , SemEval2007 Task 4 offered many useful insights into the performance of different approaches to semantic relation classification ; it has also motivated followup research ',Nakov,'prep_offered_As det_step_a amod_step_first pobj_As_step nn_Task_SemEval2007 nsubj_offered_Task num_Task_4 amod_insights_many amod_insights_useful dobj_offered_insights prep_offered_into det_performance_the pobj_into_performance prep_performance_of amod_approaches_different pobj_of_approaches prep_offered_to amod_classification_semantic nn_classification_relation pobj_to_classification nsubj_motivated_it aux_motivated_has advmod_motivated_also parataxis_offered_motivated nn_research_followup dobj_motivated_research ' +o,1853,'Our evaluation metric is BLEU with caseinsensitive matching from unigram to four-gram ',Papineni,'poss_metric_Our nn_metric_evaluation nsubj__metric cop__is nn__BLEU prep__with pobj_with_caseinsensitive partmod_caseinsensitive_matching prep_matching_from pobj_from_unigram prep_matching_to pobj_to_four-gram ' +o,1854,'This is in contrast to work by researchers such as Schiitze and Pedersen , et al and Futrelle and Gauch -LRB- 1995 -RRB- , where it is often the most frequent words in the lexicon which are clustered , predominantly with the purpose of determining their grammatical classes ',Brown,'nsubj_is_This prep_is_in pobj_in_contrast aux_work_to xcomp_is_work prep_work_by pobj_by_researchers dep_as_such prep_researchers_as pobj_as_Schiitze cc_Schiitze_and nn__Pedersen nn__ conj__ cc__et conj__al conj_Schiitze_ cc_researchers_and conj_researchers_Futrelle cc_Futrelle_and conj_Futrelle_Gauch appos_Futrelle_1995 advmod_words_where nsubj_words_it cop_words_is advmod_words_often det_words_the advmod_words_most amod_words_frequent rcmod_Futrelle_words prep_words_in det_lexicon_the pobj_in_lexicon nsubjpass_clustered_which auxpass_clustered_are rcmod_lexicon_clustered dep_by_predominantly dep_by_with det_purpose_the pobj_with_purpose prep_purpose_of pcomp_of_determining poss_classes_their amod_classes_grammatical dobj_determining_classes ' +o,1855,'Then , we used the refinement technique grow-diag-final-and to all 50 50 bidirectional alignment pairs ',Koehn,'advmod_used_Then nsubj_used_we det__the amod__refinement nn__technique amod__grow-diag-final-and dobj_used_ prep_used_to det_pairs_all amod_pairs_50 dep_50_50 amod_pairs_bidirectional nn_pairs_alignment pobj_to_pairs ' +p,1856,'A variety of classifiers have been employed for this task , the most popular being decision lists and naive Bayesian classifiers ',Yarowsky,'det_variety_A nsubjpass_employed_variety prep_variety_of pobj_of_classifiers aux_employed_have auxpass_employed_been prep_employed_for det__this nn__task pobj_for_ dep_popular_the dep_popular_most nsubj_lists_popular cop_lists_being nn_lists_decision rcmod__lists amod_lists_ cc__and conj__naive nsubj_classifiers_Bayesian dep_employed_classifiers ' +o,1857,'The significance of G 2 based on the exact conditional distribution does not rely on an asymptotic approximation and is accurate for sparse and skewed data samples 42 Information criteria The family of model evaluation criteria known as information criteria have the following expression : IC , ~ = G 2 ~ x dof -LRB- 3 -RRB- where G ~ and dof are defined above ',Berger,'det_significance_The nsubj_rely_significance prep_significance_of pobj_of_G num_G_2 partmod_significance_based prep_based_on det_distribution_the amod_distribution_exact amod_distribution_conditional pobj_on_distribution aux_rely_does neg_rely_not prep_rely_on det_approximation_an amod_approximation_asymptotic pobj_on_approximation cc_rely_and cop_accurate_is conj_rely_accurate prep_accurate_for amod_samples_sparse cc_sparse_and conj_sparse_skewed nn_samples_data pobj_for_samples number_42_ num_criteria_42 nn_criteria_Information prep_samples_criteria det_family_The dep_criteria_family prep_family_of nn_criteria_model nn_criteria_evaluation pobj_of_criteria partmod_criteria_known prep_known_as nn_criteria_information pobj_as_criteria dep_rely_have det_expression_the amod_expression_following dobj_have_expression advmod_have_IC advmod_=_~ conj_rely_= nn_~_G num_~_2 dobj_=_~ nsubj_dof_x dep_=_dof dep_dof_3 advmod_defined_where nn_~_G nsubjpass_defined_~ cc_~_and conj_~_dof auxpass_defined_are advcl_dof_defined prep_defined_above ' +o,1858,'2 Overview 21 The word segmentation problem As statistical machine translation systems basically rely on the notion of words through their lexicon models , they are usually capable of outputting sentences already segmented into words when they translate into languages like Chinese or Japanese ',Brown,'num_Overview_2 num_Overview_21 det_problem_The nn_problem_word nn_problem_segmentation nsubj_segmented_problem mark_rely_As amod_systems_statistical nn_systems_machine nn_systems_translation nsubj_rely_systems advmod_rely_basically advcl_capable_rely prep_rely_on det_notion_the pobj_on_notion prep_notion_of pobj_of_words prep_rely_through poss__their amod__lexicon nn__models pobj_through_ nsubj_capable_they cop_capable_are advmod_capable_usually rcmod_problem_capable prep_capable_of pcomp_of_outputting dobj_outputting_sentences advmod_segmented_already dep_Overview_segmented prep_segmented_into pobj_into_words advmod_translate_when nsubj_translate_they advcl_segmented_translate prep_translate_into pobj_into_languages prep_languages_like pobj_like_Chinese cc_Chinese_or conj_Chinese_Japanese ' +o,1859,'WSD has received increasing attention in recent literature on computational linguistics ',Yarowsky,'nsubj_received_WSD aux_received_has xcomp_received_increasing dobj_increasing_attention prep_increasing_in amod_literature_recent pobj_in_literature prep_literature_on amod_linguistics_computational pobj_on_linguistics ' +o,1860,'35 Maximum Entropy Model In order to build a unified probabilistic query alteration model , we used the maximum entropy approach of , which Li et al ',Berger,'number_Maximum_35 num_Model_Maximum nn_Model_Entropy mark_build_In dep_build_order aux_build_to dep_used_build det_model_a amod_model_unified amod_model_probabilistic nn_model_query nn_model_alteration dobj_build_model nsubj_used_we dep_Model_used det_approach_the amod_approach_maximum nn_approach_entropy dobj_used_approach prep_approach_of pobj_of_ dep_Li_which rcmod__Li cc_Li_et conj_Li_al ' +o,1861,'21 Data representation We have compared four complete and three partial data representation formats for the baseNP recognition task presented in ',Ramshaw,'num_representation_21 nn_representation_Data nsubj_compared_We aux_compared_have dep_representation_compared dep_complete_four dobj_compared_complete cc_complete_and num_formats_three amod_formats_partial nn_formats_data nn_formats_representation conj_complete_formats prep_compared_for det_task_the nn_task_baseNP nn_task_recognition pobj_for_task partmod_task_presented prep_presented_in ' +p,1862,'2 Related Work Recently , several successful attempts have been made at using supervised machine learning for word alignment ',Fraser,'num_Work_2 amod_Work_Related advmod_made_Recently amod_attempts_several amod_attempts_successful nsubjpass_made_attempts aux_made_have auxpass_made_been dep_Work_made prep_made_at pcomp_at_using amod_machine_supervised dobj_using_machine partmod_machine_learning prep_learning_for nn_alignment_word pobj_for_alignment ' +o,1863,'Importantly , this Bayesian approach facilitates the incorporation of sparse priors that result in a more practical distribution of tokens to lexical categories ',Johnson,'advmod_facilitates_Importantly det_approach_this nn_approach_Bayesian nsubj_facilitates_approach det_incorporation_the dobj_facilitates_incorporation prep_incorporation_of amod_priors_sparse pobj_of_priors nsubj_result_that rcmod_incorporation_result prep_result_in det_distribution_a advmod_practical_more amod_distribution_practical pobj_in_distribution prep_distribution_of pobj_of_tokens prep_result_to amod_categories_lexical pobj_to_categories ' +o,1864,'The last important fact is that it is possible to demonstrate that -LRB- Ei , j -RRB- = k P -LRB- Ri , jT ei , j -RRB- 1P -LRB- Ri , jT ei , j -RRB- = = kodds -LRB- Ri , j -RRB- where k is a constant -LRB- see -RRB- that will be neglected in the maximization process ',Snow,'det_fact_The amod_fact_last amod_fact_important nsubj_is_fact complm_possible_that nsubj_possible_it cop_possible_is ccomp_is_possible aux_demonstrate_to xcomp_possible_demonstrate nsubj_=_that appos_that_Ei dep_Ei_j ccomp_demonstrate_= nn_P_k dobj_=_P appos_P_Ri nn_ei_jT dep_Ri_ei dep_Ri_j nsubj_=_1P appos_1P_Ri nn_ei_jT dep_Ri_ei dep_Ri_j rcmod_P_= amod_kodds_= dobj_=_kodds appos_kodds_Ri dep_Ri_j advmod_constant_where nsubj_constant_k cop_constant_is det_constant_a rcmod_kodds_constant dep_constant_see acomp_see_ nsubjpass_neglected_that aux_neglected_will auxpass_neglected_be rcmod_constant_neglected prep_neglected_in det_process_the nn_process_maximization pobj_in_process ' +p,1865,'Pr -LRB- cJ1 , aJ1 eI1 -RRB- = p -LRB- J I -RRB- -LRB- I + 1 -RRB- J Jproductdisplay j = 1 p -LRB- cj eaj -RRB- -LRB- 8 -RRB- 312 Log-likelihood ratio The log-likelihood ratio statistic has been found to be accurate for modeling the associations between rare events ',Dunning,'appos_p_cJ1 amod_eI1_aJ1 dep_cJ1_eI1 amod_p_= dobj_Pr_p dep_p_J dep_J_I nsubj_1_I dep_1_+ dep_p_1 nn_Jproductdisplay_J nsubj_j_Jproductdisplay dep_Pr_j dep_j_= iobj_=_1 dobj_=_p nn_eaj_cj appos_p_eaj appos_p_8 num_ratio_312 amod_ratio_Log-likelihood nsubj_Pr_ratio det_statistic_The amod_statistic_log-likelihood nn_statistic_ratio nsubjpass_found_statistic aux_found_has auxpass_found_been rcmod_ratio_found aux_accurate_to cop_accurate_be xcomp_found_accurate prep_accurate_for pcomp_for_modeling det_associations_the dobj_modeling_associations prep_associations_between amod_events_rare pobj_between_events ' +p,1866,'In contrast , the idea of bootstrapping for relation and information extraction was first proposed in , and successfully applied to the construction of semantic lexicons , named entity recognition , extraction of binary relations , and acquisition of structured data for tasks such as Question Answering ',Fleischman,'prep_proposed_In pobj_In_contrast det_idea_the nsubjpass_proposed_idea prep_idea_of pcomp_of_bootstrapping prep_bootstrapping_for nn_extraction_relation cc_relation_and conj_relation_information pobj_for_extraction auxpass_proposed_was advmod_proposed_first prep_proposed_in pobj_in_ cc_proposed_and advmod_proposed_successfully conj_proposed_applied prep_applied_to det_construction_the pobj_to_construction prep_construction_of amod__semantic nn__lexicons pobj_of_ partmod_construction_named nn__entity nn__recognition dobj_named_ conj__extraction prep_extraction_of amod__binary nn__relations pobj_of_ cc__and conj__acquisition prep_acquisition_of amod_data_structured pobj_of_data prep_named_for pobj_for_tasks dep_as_such prep_tasks_as nn_Answering_Question pobj_as_Answering ' +o,1867,'Reranking methods have also been proposed as a method for using syntactic information ',Koehn,'amod_methods_Reranking nsubjpass_proposed_methods aux_proposed_have advmod_proposed_also auxpass_proposed_been prep_proposed_as det_method_a pobj_as_method prep_proposed_for pcomp_for_using amod_information_syntactic dobj_using_information ' +o,1868,'For the efficiency of minimum-errorrate training , we built our development set -LRB- 580 sentences -RRB- using sentences not exceeding 50 characters from the NIST MT-02 evaluation test data ',Och,'prep_built_For det_efficiency_the pobj_For_efficiency prep_efficiency_of amod__minimum-errorrate nn__training pobj_of_ nsubj_built_we poss_set_our nn_set_development dobj_built_set number_sentences_580 dep_set_sentences partmod_set_using dobj_using_sentences neg_exceeding_not xcomp_using_exceeding num_characters_50 dobj_exceeding_characters prep_exceeding_from det_data_the nn_data_NIST nn_data_MT-02 nn_data_evaluation nn_data_test pobj_from_data ' +p,1869,'22 Perceptron-based training To tune the parameters w of the model , we use the averaged perceptron algorithm because of its efficiency and past success on various NLP tasks ',Collins,'num_training_22 amod_training_Perceptron-based aux_tune_To dep_use_tune det_w_the nn_w_parameters dobj_tune_w prep_w_of det_model_the pobj_of_model nsubj_use_we dep_training_use det__the amod__averaged nn__perceptron nn__algorithm dobj_use_ dep_of_because prep__of poss_efficiency_its pobj_of_efficiency cc__and amod_success_past conj__success prep_success_on amod_tasks_various nn_tasks_NLP pobj_on_tasks ' +n,1870,'Although such approaches have been employed effectively , there appears to remain considerable room for improvement ',Pang,'mark_employed_Although amod_approaches_such nsubjpass_employed_approaches aux_employed_have auxpass_employed_been advcl_appears_employed advmod_employed_effectively advmod_employed_ expl_appears_there aux_room_to cop_room_remain amod_room_considerable xcomp_appears_room prep_room_for pobj_for_improvement ' +o,1871,'Therefore , having correct transliterations would give only small improvements in terms of BLEU and NIST scores ',Papineni,'nsubj_give_Therefore partmod_Therefore_having amod_transliterations_correct dobj_having_transliterations aux_give_would advmod_improvements_only amod_improvements_small dobj_give_improvements prep_give_in pobj_in_terms prep_terms_of amod__BLEU pobj_of_ cc__and nn_scores_NIST conj__scores ' +p,1872,'But in fact , the issue of editing in text summarization has usually been neglected , notable exceptions being the works by and Mani , Gates , and Bloedorn -LRB- 1999 -RRB- ',Jing,'cc_neglected_But prep_neglected_in pobj_in_fact det_issue_the nsubjpass_neglected_issue prep_issue_of pobj_of_editing prep_editing_in nn_summarization_text pobj_in_summarization aux_neglected_has advmod_neglected_usually auxpass_neglected_been amod_exceptions_notable dobj_neglected_exceptions aux_works_being det_works_the partmod_exceptions_works prep_works_by pobj_by_ cc__and conj__Mani conj__Gates cc__and conj__Bloedorn appos__1999 ' +o,1873,'Preparing an aligned abbreviation corpus , we obtain the optimal combination of the features by using the maximum entropy framework ',Berger,'dep_obtain_Preparing det_corpus_an amod_corpus_aligned nn_corpus_abbreviation dobj_Preparing_corpus nsubj_obtain_we det_combination_the amod_combination_optimal dobj_obtain_combination prep_combination_of det_features_the pobj_of_features prep_obtain_by pcomp_by_using det_framework_the amod_framework_maximum nn_framework_entropy dobj_using_framework ' +o,1874,'8This result is presented as 0053 with the official ROUGE scorer ',Lin,'amod_result_8This nsubjpass_presented_result auxpass_presented_is prep_presented_as pobj_as_0053 prep_presented_with det_scorer_the amod_scorer_official amod_scorer_ROUGE pobj_with_scorer ' +o,1875,'After line 17 , we can employ the one-sense-per-discourse heuristic to further classify unclassified data , as proposed in ',Yarowsky,'prep_employ_After pobj_After_line num_line_17 nsubj_employ_we aux_employ_can det_heuristic_the amod_heuristic_one-sense-per-discourse dobj_employ_heuristic aux_classify_to advmod_classify_further xcomp_employ_classify amod_data_unclassified dobj_classify_data advmod_proposed_as amod_data_proposed prep_classify_in ' +o,1876,'Many studies on collocation extraction are carried out based on co-occurring frequencies of the word pairs in texts ',Dunning,'amod_studies_Many nsubjpass_carried_studies prep_studies_on nn_extraction_collocation pobj_on_extraction auxpass_carried_are prt_carried_out prep_carried_based dep_based_on amod_frequencies_co-occurring pobj_on_frequencies prep_frequencies_of det_pairs_the nn_pairs_word pobj_of_pairs prep_pairs_in pobj_in_texts ' +o,1877,'1 Introduction Parsers have been developed for a variety of grammar formalisms , for example HPSG , LFG , TAG , CCG , and variants of phrase-structure grammar , including the phrase-structure grammar implicit in the Penn Treebank ',Cahill,'num_Parsers_1 nn_Parsers_Introduction nsubjpass_developed_Parsers aux_developed_have auxpass_developed_been prep_developed_for det_variety_a pobj_for_variety prep_variety_of nn_formalisms_grammar pobj_of_formalisms dep__for pobj_for_example nn__HPSG dep_variety_ nn__LFG conj__ nn__TAG conj__ nn__CCG conj__ cc__and conj__variants prep_variants_of amod__phrase-structure nn__grammar pobj_of_ dep__including det_grammar_the amod_grammar_phrase-structure pobj_including_grammar amod_grammar_implicit prep_implicit_in det__the nn__Penn nn__Treebank pobj_in_ ' +o,1878,'Decoding weights are optimized using Ochs algorithm to set weights for the four components of the log-linear model : language model , phrase translation model , distortion model , and word-length feature ',Och,'amod_weights_Decoding auxpass_optimized_are dep_weights_optimized xcomp_optimized_using nn__Ochs nn__algorithm nsubj_set_ aux_set_to xcomp_using_set dobj_set_weights prep_set_for det_components_the num_components_four pobj_for_components prep_components_of det_model_the amod_model_log-linear pobj_of_model nn_model_language dep_weights_model nn_model_phrase nn_model_translation conj_model_model nn_model_distortion conj_model_model cc_model_and amod_feature_word-length conj_model_feature ' +o,1879,'The way a decoder constructs translation hypotheses is directly related to the weights for different model features in a SMT system , which are usually optimized for a given set of models with minimum error rate training -LRB- MERT -RRB- to achieve better translation performance ',Och,'det_way_The det_constructs_a nn_constructs_decoder nsubj__constructs nn_hypotheses_translation nsubjpass_related_hypotheses auxpass_related_is advmod_related_directly rcmod_constructs_related prep_related_to det_weights_the pobj_to_weights prep_weights_for amod_features_different nn_features_model pobj_for_features prep_features_in det_system_a nn_system_SMT pobj_in_system nsubjpass_optimized_which auxpass_optimized_are advmod_optimized_usually rcmod_weights_optimized prep_optimized_for det_set_a amod_set_given pobj_for_set prep_set_of pobj_of_models prep_models_with amod_training_minimum nn_training_error nn_training_rate pobj_with_training abbrev_constructs_MERT dep_way_ aux_achieve_to xcomp__achieve amod_performance_better nn_performance_translation dobj_achieve_performance ' +o,1880,'Null productions are also a source of double counting, as there are many possible orders in 926 N I 2+ N IN N I } N IN I I I N N N (a) Normal Domain Rules } I squigglerightN 2+ I squigglerightNI I squigglerightNI I squigglerightN N N N I I I (b) Inverted Domain Rules N 11 ,fN 11 N 11 N 10 N 10 N 10 e, N 10 N 00 } N 11 ,f N 10 } N 10 N 00 e, } N 00 I 11 N NI 11 N NI 00 N 00 I + 11 I 00 N 00 N 10 N 10 N 11 N N I 11 I 11 I 00 N 00 N 11 (c) Normal Domain with Null Rules } } } I 11 squiggleright ,fI 11 I 11 squigglerightI 10 I 11 squiggleright ,f I 10 I 10 squiggleright I 10 e, I 10 squigglerightI 00 I 10 squiggleright I 00 e, I 00 squigglerightN + 11 N 00 I I N 00 N 11 N 11 I 00 squigglerightN 11 I I squigglerightN 11 I I squigglerightN 00 I 00 I 00 I 10 I 10 I 11 I 11 (d) Inverted Domain with Null Rules Figure 2: Illustration of two unambiguous forms of ITG grammars: In (a) and (b), we illustrate the normal grammar without nulls (presented in Wu (1997) and Zens and Ney (2003)).',Wu,'' +o,1881,'Models of this kind assume that an input word is generated by only one output word ',Brown,'nsubj_assume_Models prep_Models_of det_kind_this pobj_of_kind complm_generated_that det_word_an nn_word_input nsubjpass_generated_word auxpass_generated_is ccomp_assume_generated prep_generated_by quantmod_one_only num_word_one nn_word_output pobj_by_word ' +p,1882,'We also use Cube Pruning algorithm to speed up the translation process ',Huang,'nsubj_use_We advmod_use_also nn__Cube nn__Pruning nn__algorithm dobj_use_ aux_speed_to xcomp_use_speed prt_speed_up det_process_the nn_process_translation dobj_speed_process ' +o,1883,'N-gram language models have also been used in Statistical Machine Translation -LRB- SMT -RRB- as proposed by ',Brown,'amod_models_N-gram nn_models_language nsubjpass_used_models aux_used_have advmod_used_also auxpass_used_been prep_used_in nn_Machine_Statistical pobj_in_Machine dep_as_Translation abbrev_Translation_SMT advmod_proposed_as partmod_Machine_proposed prep_proposed_by ' +o,1884,'In natural language processing , recent years have seen ME techniques used for sentence boundary detection , part of speech tagging , parse selection and ambiguity resolution , and stochastic attribute-value grammars , to name just a few applications ',Berger,'prep_seen_In amod_processing_natural nn_processing_language pobj_In_processing amod_years_recent nsubj_seen_years aux_seen_have nn_techniques_ME dobj_seen_techniques partmod_techniques_used prep_used_for nn_detection_sentence nn_detection_boundary pobj_for_detection conj_detection_part prep_part_of nn_tagging_speech pobj_of_tagging amod_selection_parse conj_tagging_selection cc_tagging_and amod_resolution_ambiguity conj_tagging_resolution cc_detection_and amod_grammars_stochastic amod_grammars_attribute-value conj_detection_grammars aux_name_to xcomp_seen_name quantmod_few_just quantmod_few_a num_applications_few dobj_name_applications ' +p,1885,'An important contribution to interactive CAT technology was carried out around the TransType -LRB- TT -RRB- project ',Och,'det_contribution_An amod_contribution_important nsubjpass_carried_contribution prep_contribution_to amod_technology_interactive nn_technology_CAT pobj_to_technology auxpass_carried_was prt_carried_out prep_carried_around det_project_the amod_project_TransType abbrev_project_TT pobj_around_project ' +o,1886,'Due to the parameter interdependencies introduced by the one-to-one assumption , we are unlikely to find a method for decomposing the assignments into parameters that can be estimated independently of each other as in et al -RRB- ',Brown,'dep_to_Due prep_unlikely_to det_interdependencies_the nn_interdependencies_parameter pobj_to_interdependencies partmod_interdependencies_introduced prep_introduced_by det_assumption_the amod_assumption_one-to-one pobj_by_assumption nsubj_unlikely_we cop_unlikely_are aux_find_to xcomp_unlikely_find det_method_a dobj_find_method prep_find_for pcomp_for_decomposing det_assignments_the dobj_decomposing_assignments prep_decomposing_into pobj_into_parameters nsubjpass_estimated_that aux_estimated_can auxpass_estimated_be rcmod_parameters_estimated advmod_estimated_independently prep_estimated_of det_other_each pobj_of_other prep_estimated_as dep_as_in nn__ cc__et conj__al pobj_in_ ' +o,1887,'This tolerant search uses the well known concept of Levenshtein distance in order to obtain the most similar string for the given prefix -LRB- see for more details -RRB- ',Och,'det_search_This nn_search_tolerant nsubj_uses_search det_concept_the advmod_known_well amod_concept_known dobj_uses_concept prep_concept_of nn_distance_Levenshtein pobj_of_distance mark_obtain_in dep_obtain_order aux_obtain_to purpcl_uses_obtain det_string_the advmod_similar_most amod_string_similar dobj_obtain_string prep_string_for det_prefix_the amod_prefix_given pobj_for_prefix dep_prefix_see dobj_see_ prep__for amod_details_more pobj_for_details ' +o,1888,'This is related to the wellstudied problem of identifying paraphrases and the more general variant of recognizing textual entailment , which explores whether information expressed in a hypothesis can be inferred from a given premise ',Barzilay,'nsubjpass_related_This auxpass_related_is prep_related_to det_problem_the amod_problem_wellstudied pobj_to_problem prep_problem_of pcomp_of_identifying amod__paraphrases dobj_identifying_ cc_problem_and det_variant_the advmod_general_more amod_variant_general conj_problem_variant prep_variant_of pcomp_of_recognizing amod_entailment_textual dobj_recognizing_entailment nsubj_explores_which rcmod_entailment_explores complm_inferred_whether nsubjpass_inferred_information partmod_information_expressed prep_expressed_in det_hypothesis_a pobj_in_hypothesis aux_inferred_can auxpass_inferred_be ccomp_explores_inferred prep_inferred_from det_premise_a amod_premise_given pobj_from_premise ' +o,1889,'A systematic exploration of a set of such features for proteinprotein interaction extraction was recently provided by , who also used features derived from the Collins parser ',Jiang,'det_exploration_A amod_exploration_systematic nsubjpass_provided_exploration prep_exploration_of det_set_a pobj_of_set prep_set_of amod_features_such pobj_of_features prep_features_for nn_extraction_proteinprotein nn_extraction_interaction pobj_for_extraction auxpass_provided_was advmod_provided_recently dep_used_provided prep_provided_by pobj_by_ nsubj_used_who advmod_used_also dobj_used_features partmod_features_derived prep_derived_from det_parser_the nn_parser_Collins pobj_from_parser ' +o,1890,'The feature weights i in the log-linear model are determined using a minimum error rate training method , typically Powells method ',Och,'det_weights_The nn_weights_feature measure_i_weights dep_determined_i prep_i_in det_model_the amod_model_log-linear pobj_in_model auxpass_determined_are xcomp_determined_using det_rate_a amod_rate_minimum nn_rate_error dobj_using_rate nn_method_training nsubjpass_determined_method advmod_Powells_typically amod_method_Powells appos_method_method ' +o,1891,'We use the version extracted and preprocessed by ','Daume III','nsubj_use_We det_version_the dobj_use_version partmod_version_extracted cc_extracted_and conj_extracted_preprocessed prep_extracted_by ' +p,1892,'We conclude by noting that English language models currently used in speech recognition and automated language translation are much more powerful , employing , for example , 7-gram word models -LRB- not letter models -RRB- trained on trillions of words ',Brants,'nsubj_conclude_We prep_conclude_by pcomp_by_noting complm_used_that amod_models_English nn_models_language nsubj_used_models advmod_used_currently ccomp_noting_used prep_used_in nn__speech nn__recognition pobj_in_ cc_used_and amod__automated nn__language nn__translation nsubj_powerful_ cop_powerful_are advmod_powerful_much advmod_powerful_more conj_used_powerful dep_noting_employing prep_employing_for pobj_for_example amod_models_7-gram nn_models_word dobj_employing_models dep_models_not nn_models_letter appos_models_models partmod_models_trained prep_trained_on pobj_on_trillions prep_trillions_of pobj_of_words ' +o,1893,'This problem will be solved by incorporating other resources such as thesaurus or a dictionary , orcombiningourmethodwithothermethods using external wider contexts ',Turney,'det_problem_This nsubjpass_solved_problem aux_solved_will auxpass_solved_be prep_solved_by pcomp_by_incorporating amod_resources_other dobj_incorporating_resources dep_as_such prep_resources_as pobj_as_thesaurus cc_thesaurus_or det_dictionary_a conj_thesaurus_dictionary appos_resources_orcombiningourmethodwithothermethods partmod_orcombiningourmethodwithothermethods_using amod_contexts_external amod_contexts_wider dobj_using_contexts ' +p,1894,'22 ITG Space Inversion Transduction Grammars , or ITGs provide an efficient formalism to synchronously parse bitext ',Wu,'num_Grammars_22 nn_Grammars_ITG nn_Grammars_Space nn_Grammars_Inversion nn_Grammars_Transduction nsubj_provide_Grammars cc_Grammars_or nn__ITGs conj_Grammars_ det_formalism_an amod_formalism_efficient dobj_provide_formalism prep_provide_to advmod_parse_synchronously amod_bitext_parse pobj_to_bitext ' +o,1895,' , , ',Brants,'nn__ appos__ ' +o,1896,'(Collins, 2002).',Collins,'' +o,1897,'The first is to align the words using a standard word alignement technique , such as the Refined Method described in -LRB- the intersection of two IBM Viterbi alignments , forward and reverse , enriched with alignments from the union -RRB- and then generate bi-phrases by combining together individual alignments that co-occur in the same pair of sentences ',Och,'det_first_The nsubj_is_first aux_align_to xcomp_is_align det_words_the nsubj_using_words dep_align_using det_technique_a amod_technique_standard nn_technique_word nn_technique_alignement dobj_using_technique dep_as_such prep_technique_as det_Method_the nn_Method_Refined pobj_as_Method partmod_Method_described prep_described_in pobj_in_ det_intersection_the dep__intersection prep_intersection_of num_alignments_two nn_alignments_IBM nn_alignments_Viterbi pobj_of_alignments advmod_alignments_forward cc_alignments_and conj_alignments_reverse partmod_reverse_enriched prep_enriched_with pobj_with_alignments prep_alignments_from det_union_the pobj_from_union cc_align_and advmod_align_then conj_align_generate dobj_generate_bi-phrases prep_generate_by pcomp_by_combining prt_combining_together amod_alignments_individual dobj_combining_alignments det_co-occur_that dobj_align_co-occur prep_align_in det_pair_the amod_pair_same pobj_in_pair prep_pair_of pobj_of_sentences ' +o,1898,'The simplest ` period-space-capital_letter \' approach works well for simple texts but is rather unreliable for texts with many proper names and abbreviations at the end of sentence as , for instance , the Wall Street Journal -LRB- WSJ -RRB- corpus -LRB- -RRB- ',Marcus,'det_approach_The amod_approach_simplest nn_approach_period-space-capital_letter nsubj_works_approach advmod_works_well prep_works_for amod_texts_simple pobj_for_texts cc_works_but cop_unreliable_is advmod_unreliable_rather conj_works_unreliable prep_unreliable_for pobj_for_texts prep_texts_with amod_names_many amod_names_proper pobj_with_names cc_names_and conj_names_abbreviations prep_names_at det_end_the pobj_at_end prep_end_of pobj_of_sentence prep_unreliable_as dep_as_for pobj_for_instance det_corpus_the nn_Journal_Wall nn_Journal_Street nn_corpus_Journal abbrev_Journal_WSJ pobj_as_corpus appos_corpus_ ' +o,1899,'In addition , since word senses are often associated with domains , word senses can be consequently distinguished by way of determining the domain of each description ',Yarowsky,'prep_distinguished_In pobj_In_addition mark_associated_since nn_senses_word nsubjpass_associated_senses auxpass_associated_are advmod_associated_often dep_distinguished_associated prep_associated_with nn__domains pobj_with_ nn_senses_word nsubj_distinguished_senses aux_distinguished_can cop_distinguished_be advmod_distinguished_consequently prep_distinguished_by pobj_by_way prep_way_of pcomp_of_determining det_domain_the dobj_determining_domain prep_domain_of det_description_each pobj_of_description ' +p,1900,'18 More recently , have proposed methods for automatically extracting from a corpus heads that correlate well with discourse novelty ',Bean,'dep_More_18 advmod_proposed_More advmod_proposed_recently nsubj_proposed_ aux_proposed_have dobj_proposed_methods prep_proposed_for advmod_extracting_automatically pcomp_for_extracting prep_extracting_from det_heads_a nn_heads_corpus pobj_from_heads nsubj_correlate_that rcmod_heads_correlate advmod_correlate_well prep_correlate_with nn_novelty_discourse pobj_with_novelty ' +o,1901,'Smadja , which is the classic work on collocation extraction , uses a two-stage filtering model in which , in the first step , n-gram statistics determine possible collocations and , in the second step , these candidates are submitted to a syntactic valida7Of course , lexical material is always at least partially dependent on the domain in question ',Dunning,'nn__Smadja nsubj_uses_ nsubj_work_which cop_work_is det_work_the amod_work_classic rcmod__work prep_work_on nn_extraction_collocation pobj_on_extraction det_model_a amod_model_two-stage amod_model_filtering dobj_uses_model rel_determine_in pobj_in_which dep_determine_in det_step_the amod_step_first pobj_in_step amod_statistics_n-gram nsubj_determine_statistics rcmod_model_determine amod_collocations_possible dobj_determine_collocations cc_collocations_and prep_,_in det_step_the amod_step_second pobj_in_step det_candidates_these nsubjpass_submitted_candidates auxpass_submitted_are ccomp_determine_submitted prep_submitted_to det_course_a amod_course_syntactic amod_course_valida7Of pobj_to_course amod_material_lexical nsubj_dependent_material cop_dependent_is advmod_dependent_always advmod_dependent_at dep_at_least advmod_dependent_partially dep_determine_dependent prep_dependent_on det_domain_the pobj_on_domain prep_domain_in pobj_in_question ' +o,1902,'Previously published approaches to reducing the rule set include : enforcing a minimum span of two words per non-terminal , which would reduce our set to 115M rules ; or a minimum count -LRB- mincount -RRB- threshold , which would reduce our set to 78M -LRB- mincount = 2 -RRB- or 57M -LRB- mincount = 3 -RRB- rules ',Lopez,'advmod_published_Previously dobj_published_approaches prep_published_to pcomp_to_reducing det_set_the nn_set_rule dobj_reducing_set rcmod_set_include parataxis_reducing_enforcing det_span_a amod_span_minimum dobj_enforcing_span prep_span_of num_words_two pobj_of_words prep_words_per amod__non-terminal pobj_per_ nsubj_reduce_which aux_reduce_would rcmod__reduce poss_set_our dobj_reduce_set prep_reduce_to num_rules_115M pobj_to_rules cc_rules_or det_count_a amod_count_minimum conj_rules_count appos_rules_mincount nn__threshold nsubj_published_ nsubj_reduce_which aux_reduce_would rcmod__reduce poss_set_our dobj_reduce_set prep_reduce_to pobj_to_78M nsubj_2_mincount dep_2_= dep_78M_2 cc_78M_or amod_rules_57M nsubj_3_mincount dep_3_= dep_57M_3 conj_78M_rules ' +o,1903,'We also tested other automatic methods : content-based evaluation , BLEU and ROUGE-1 , and compared their results with that of evaluation by revision as reference ',Lin,'nsubj_tested_We advmod_tested_also amod_methods_other amod_methods_automatic dobj_tested_methods amod_evaluation_content-based dep_methods_evaluation nn__BLEU conj_evaluation_ cc_evaluation_and nn__ROUGE-1 conj_evaluation_ cc_tested_and conj_tested_compared poss_results_their dobj_compared_results prep_compared_with pobj_with_that prep_that_of pobj_of_evaluation prep_compared_by pobj_by_revision prep_revision_as pobj_as_reference ' +o,1904,'From a theoretical point of view , it is difficult to find motivation for the parameter estimation methods used by see for discussion ',Collins,'prep_difficult_From det_point_a amod_point_theoretical pobj_From_point prep_point_of pobj_of_view nsubj_difficult_it cop_difficult_is aux_find_to xcomp_difficult_find dobj_find_motivation prep_find_for det_parameter_the pobj_for_parameter nn_methods_estimation nsubj_see_methods partmod_methods_used prep_used_by pobj_by_ dep_difficult_see dobj_see_ prep__for pobj_for_discussion ' +p,1905,'A key component of the parsing system is a Maximum Entropy CCG supertagger which assigns lexical categories to words in a sentence ',Ratnaparkhi,'det_component_A amod_component_key nsubj__component prep_component_of det_system_the amod_system_parsing pobj_of_system cop__is det__a nn__Maximum nn__Entropy nn__CCG nn__supertagger nsubj_assigns_which rcmod__assigns amod_categories_lexical dobj_assigns_categories prep_assigns_to pobj_to_words prep_words_in det_sentence_a pobj_in_sentence ' +n,1906,'In terms of alignment , this wordnumber difference means that multiword connections must be considered , a task which 334 Sue J Ker and Jason S Chang Word Alignment is beyond the reach of methods proposed in recent alignment works based on Model 1 and 2 ',Brown,'prep_means_In pobj_In_terms prep_terms_of pobj_of_alignment det_difference_this nn_difference_wordnumber nsubj_means_difference complm_considered_that amod_connections_multiword nsubjpass_considered_connections aux_considered_must auxpass_considered_be ccomp_means_considered det_task_a dobj_considered_task rel_is_which nn_Ker_334 nn_Ker_Sue nn_Ker_J nsubj_is_Ker cc_Ker_and nn_Alignment_Jason nn_Alignment_S nn_Alignment_Chang nn_Alignment_Word conj_Ker_Alignment rcmod_task_is prep_is_beyond det_reach_the pobj_beyond_reach prep_reach_of pobj_of_methods partmod_methods_proposed prep_proposed_in amod_works_recent nn_works_alignment pobj_in_works prep_proposed_based dep_based_on dep_on_ pobj_on_Model num_Model_1 cc_Model_and conj_Model_2 ' +o,1907,'3 Probability Model This paper takes a ` history-based \' approach where each tree-building procedure uses a probability model p -LRB- alb -RRB- , derived from p -LRB- a , b -RRB- , to weight any action a based on the available context , or history , b First , we present a few simple categories of contextual predicates that capture any information in b that is useful for predicting a Next , the predicates are used to extract a set of features from a corpus of manually parsed sentences ',Marcus,'num_Model_3 nn_Model_Probability det_paper_This nsubj_takes_paper rcmod_Model_takes det_approach_a amod_approach_history-based dobj_takes_approach advmod_takes_ advmod_uses_where det_procedure_each amod_procedure_tree-building nsubj_uses_procedure dep_First_uses det_p_a nn_p_probability nn_p_model nsubj_weight_p appos_p_alb partmod_p_derived prep_derived_from pobj_from_p dep_p_a appos_a_b aux_weight_to xcomp_uses_weight det_action_any iobj_weight_action det_context_a amod_context_based prep_based_on det_available_the pobj_on_available dobj_weight_context cc_context_or conj_context_history dep_First_b dep_Model_First nsubj_present_we ccomp_used_present det_categories_a amod_categories_few amod_categories_simple dobj_present_categories prep_categories_of amod_predicates_contextual pobj_of_predicates nsubj_capture_that rcmod_categories_capture det_information_any dobj_capture_information prep_information_in pobj_in_b nsubj_useful_that cop_useful_is rcmod_b_useful prep_useful_for pcomp_for_predicting det_Next_a dobj_predicting_Next det_predicates_the nsubjpass_used_predicates auxpass_used_are rcmod_First_used aux_extract_to xcomp_used_extract det_set_a dobj_extract_set prep_set_of pobj_of_features prep_extract_from det_corpus_a pobj_from_corpus prep_First_of amod_sentences_manually amod_sentences_parsed pobj_of_sentences ' +o,1908,'The algorithms were trained and tested using version 3 of the Penn Treebank , using the training , development , and test split described in and also employed by Toutanova et al ',Collins,'det_algorithms_The nsubjpass_trained_algorithms auxpass_trained_were cc_trained_and conj_trained_tested ccomp_trained_using dobj_using_version num_version_3 prep_version_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank dep_using_using det_training_the dobj_using_training conj_training_development cc_training_and nn_split_test conj_training_split dep_split_described prep_described_in pobj_in_ cc_described_and advmod_employed_also conj_described_employed prep_employed_by pobj_by_Toutanova cc_Toutanova_et conj_Toutanova_al ' +o,1909,'Other possibilities for the weighting include assigning constant one or the exponential of the final score etc One of the advantages of the proposed phrase training algorithm is that it is a parameterized procedure that can be optimized jointly with the trans82 lation engine to minimize the final translation errors measured by automatic metrics such as BLEU ',Papineni,'amod_possibilities_Other nsubj_include_possibilities prep_possibilities_for det_weighting_the pobj_for_weighting xcomp_include_assigning amod_one_constant dep_etc_one cc_etc_or det_exponential_the conj_etc_exponential prep_exponential_of det_score_the amod_score_final pobj_of_score dobj_assigning_etc nsubj_is_One prep_One_of det_advantages_the pobj_of_advantages prep_advantages_of det_algorithm_the amod_algorithm_proposed nn_algorithm_phrase nn_algorithm_training pobj_of_algorithm dep_include_is complm_procedure_that nsubj_procedure_it cop_procedure_is det_procedure_a amod_procedure_parameterized ccomp_is_procedure nsubjpass_optimized_that aux_optimized_can auxpass_optimized_be rcmod_procedure_optimized advmod_optimized_jointly prep_optimized_with det_engine_the amod_engine_trans82 nn_engine_lation pobj_with_engine aux_minimize_to purpcl_optimized_minimize det_errors_the amod_errors_final nn_errors_translation dobj_minimize_errors partmod_errors_measured prep_measured_by amod_metrics_automatic pobj_by_metrics dep_as_such prep_metrics_as pobj_as_BLEU ' +o,1910,'Indeed , the result of that including low support features helps a voted perceptron model but harms a maximum entropy model is undone once the weights of the maximum entropy model are regularized ',Collins,'advmod_helps_Indeed det_result_the nsubj_helps_result prep_result_of pcomp_of_ dobj__that prep_that_including amod_features_low nn_features_support pobj_including_features det_model_a amod_model_voted nn_model_perceptron dobj_helps_model cc_helps_but conj_helps_harms det_model_a amod_model_maximum nn_model_entropy nsubjpass_undone_model auxpass_undone_is ccomp_harms_undone mark_regularized_once det_weights_the nsubjpass_regularized_weights prep_weights_of det_model_the amod_model_maximum nn_model_entropy pobj_of_model auxpass_regularized_are advcl_undone_regularized ' +o,1911,'Of the 1600 IBM sentences that have been parsed , only 67 overlapped with the IBM-manual treebank that was bracketed by University of Lancaster ',Marcus,'prep_overlapped_Of det_sentences_the num_sentences_1600 nn_sentences_IBM pobj_Of_sentences nsubjpass_parsed_that aux_parsed_have auxpass_parsed_been rcmod_sentences_parsed dep_parsed_ advmod_67_only nsubj_overlapped_67 prep_overlapped_with det_treebank_the amod_treebank_IBM-manual pobj_with_treebank nsubjpass_bracketed_that auxpass_bracketed_was rcmod_treebank_bracketed prep_bracketed_by pobj_by_University prep_University_of pobj_of_Lancaster ' +o,1912,'The goal of integrating syntactic information into the translation model has prompted many researchers to pursue tree-based transfer models , with increasingly encouraging results ',Wu,'det_goal_The nsubj_prompted_goal prep_goal_of pcomp_of_integrating amod_information_syntactic dobj_integrating_information prep_integrating_into det_model_the nn_model_translation pobj_into_model aux_prompted_has amod_researchers_many dobj_prompted_researchers aux_pursue_to xcomp_prompted_pursue amod_models_tree-based nn_models_transfer dobj_pursue_models advmod_pursue_ prep_prompted_with advmod_encouraging_increasingly amod_results_encouraging pobj_with_results ' +o,1913,'The syntactic parameters are the same as in Section 51 and are smoothed as in ',Collins,'det_parameters_The amod_parameters_syntactic nsubjpass_as_parameters cop_as_are det_same_the attr_as_same prep_as_in pobj_in_Section num_Section_51 cc_as_and auxpass_smoothed_are conj_as_smoothed prep_smoothed_as dep_as_in ' +o,1914,'The sets obtainedare then ranked usingthe loglikelihoodratiostes ',Dunning,'det_obtainedare_The nn_obtainedare_sets nsubj_ranked_obtainedare advmod_ranked_then amod_loglikelihoodratiostes_usingthe dobj_ranked_loglikelihoodratiostes ' +o,1915,'with parse action sequences for 40,000 Wall Street Journal sentences derived from the Penn Treebank ',Marcus,'amod_sequences_parse nn_sequences_action pobj_with_sequences prep_sequences_for num_sentences_40,000 nn_sentences_Wall nn_sentences_Street nn_sentences_Journal pobj_for_sentences partmod_sentences_derived prep_derived_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,1916,'Supertags Part-of-speech disambiguation techniques -LRB- POS taggers -RRB- are often used prior to parsing to eliminate -LRB- or substantially reduce -RRB- the part-of-speech ambiguity ',Marcus,'amod_techniques_Supertags amod_techniques_Part-of-speech nn_techniques_disambiguation nn__techniques dep_taggers_POS appos__taggers nsubjpass_used_ auxpass_used_are advmod_used_often advmod_used_prior dep_prior_to pcomp_to_parsing aux_eliminate_to xcomp_parsing_eliminate dep_reduce_or advmod_reduce_substantially dep_eliminate_reduce det_ambiguity_the amod_ambiguity_part-of-speech dobj_eliminate_ambiguity ' +p,1917,'They provide pairs of phrases that are used to construct a large set of potential translations for each input sentence , along with feature values associated with each phrase pair that are used to select the best translation from this set1 The most widely used method for building phrase translation tables selects , from a word alignment of a parallel bilingual training corpus , all pairs of phrases -LRB- up to a given length -RRB- that are consistent with the alignment ',Koehn,'nsubj_provide_They dobj_provide_pairs prep_pairs_of pobj_of_phrases nsubjpass_used_that auxpass_used_are rcmod_phrases_used aux_construct_to xcomp_used_construct det_set_a amod_set_large dobj_construct_set prep_set_of amod_translations_potential pobj_of_translations prep_translations_for det_sentence_each nn_sentence_input pobj_for_sentence advmod_translations_along dep_along_with nn_values_feature pobj_with_values partmod_values_associated prep_associated_with det_pair_each nn_pair_phrase pobj_with_pair nsubjpass_used_that auxpass_used_are rcmod_pair_used aux_select_to xcomp_used_select det_translation_the amod_translation_best dobj_select_translation prep_select_from det_set1_this pobj_from_set1 det_method_The advmod_used_most advmod_used_widely amod_method_used nsubj_selects_method prep_method_for nn_tables_building nn_tables_phrase nn_tables_translation pobj_for_tables advmod_selects_ dep_provide_selects prep_selects_from det_alignment_a nn_alignment_word pobj_from_alignment prep_alignment_of det_corpus_a amod_corpus_parallel amod_corpus_bilingual nn_corpus_training pobj_of_corpus det_pairs_all dobj_selects_pairs prep_pairs_of pobj_of_phrases dep_phrases_up dep_up_to det_length_a amod_length_given pobj_to_length nsubj_consistent_that cop_consistent_are rcmod_phrases_consistent prep_consistent_with det_alignment_the pobj_with_alignment ' +o,1918,' and Liang et al ',Tillmann,'cc__and conj__Liang cc_Liang_et conj_Liang_al ' +n,1919,'2 Motivation and Prior Work While several authors have looked at the supervised adaptation case , there are less -LRB- and especially less successful -RRB- studies on semi-supervised domain adaptation ',Blitzer,'dep_Motivation_2 cc_Motivation_and amod_Work_Prior conj_Motivation_Work mark_looked_While amod_authors_several nsubj_looked_authors aux_looked_have advcl_are_looked prep_looked_at det_case_the amod_case_supervised nn_case_adaptation pobj_at_case expl_are_there dep_Motivation_are acomp_are_less dep_-LRB-_and advmod_less_especially dep_-LRB-_less dep_less_successful dep_on_studies prep_are_on amod_adaptation_semi-supervised nn_adaptation_domain pobj_on_adaptation ' +o,1920,'As association measure we apply log-likelihood ratio to normalized frequency ',Dunning,'prep_apply_As nn_measure_association pobj_As_measure nsubj_apply_we amod__log-likelihood nn__ratio dobj_apply_ prep_apply_to amod_frequency_normalized pobj_to_frequency ' +o,1921,'item form : -LRB- i , j , ueve -RRB- goal : -LRB- I , j , ue -RRB- rules : -LRB- i , j , ue -RRB- R -LRB- fifiprime\\/ejejprime -RRB- -LRB- iprime , j , ejejprime -RRB- -LRB- i , j , ueejve -RRB- -LRB- i , j + 1 , ueejve -RRB- ej +1 = rj +1 -LRB- Logic MONOTONE-ALIGN -RRB- Under the boolean semiring , this -LRB- minimal -RRB- logic decides if a training example is reachable by the model , which is required by some discriminative training regimens ',Liang,'nn_form_item nn_R_form appos_goal_i dep_i_j dep_i_ueve dep_form_goal appos_rules_I dep_I_j dep_I_ue dep_goal_rules appos_form_i dep_i_j dep_i_ue nsubj_ej_R appos_R_fifiprime\\/ejejprime nn_ejejprime_iprime dep_ejejprime_j appos_R_ejejprime appos_R_i dep_i_j dep_i_ueejve nn_+_i dep_+_j appos_R_+ num_+_1 dep_+_ueejve dobj_ej_+1 dep_+1_= number_+1_rj dobj_=_+1 nn_MONOTONE-ALIGN_Logic abbrev_+1_MONOTONE-ALIGN prep_decides_Under det_semiring_the nn_semiring_boolean pobj_Under_semiring det_logic_this dep_logic_minimal nsubj_decides_logic dep_ej_decides mark_reachable_if det_example_a nn_example_training nsubj_reachable_example cop_reachable_is advcl_decides_reachable prep_reachable_by det_model_the pobj_by_model nsubjpass_required_which auxpass_required_is rcmod_model_required prep_required_by det_regimens_some amod_regimens_discriminative nn_regimens_training pobj_by_regimens ' +o,1922,'5 Related Work Although there have been many studies on collocation extraction and mining using only statistical approaches , there has been much less work on collocation acquisition which takes into account the linguistic properties typically associated with collocations ',Church,'num_Work_5 amod_Work_Related mark_been_Although expl_been_there aux_been_have advcl_been_been amod_studies_many dobj_been_studies prep_been_on nn_extraction_collocation pobj_on_extraction cc_extraction_and conj_extraction_mining xcomp_been_using advmod__only amod__statistical nn__approaches dobj_using_ expl_been_there aux_been_has dep_Work_been advmod_less_much amod_work_less dobj_been_work prep_work_on nn_acquisition_collocation pobj_on_acquisition nsubj_takes_which rcmod_acquisition_takes prep_takes_into pobj_into_account det_properties_the amod_properties_linguistic dobj_takes_properties advmod_associated_typically partmod_properties_associated prep_associated_with pobj_with_collocations ' +p,1923,'In the SMT research community , the second step has been well studied and many methods have been proposed to speed up the decoding process , such as node-based or span-based beam search with different pruning strategies and cube pruning ',Huang,'prep_been_In det_community_the nn_community_SMT nn_community_research pobj_In_community det_step_the amod_step_second nsubj_been_step aux_been_has advmod_studied_well amod__studied cc_studied_and conj_studied_many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been ccomp_many_proposed aux_speed_to xcomp_proposed_speed prt_speed_up det_process_the amod_process_decoding dobj_speed_process dep_as_such prep_process_as amod_search_node-based cc_node-based_or conj_node-based_span-based nn_search_beam pobj_as_search prep_search_with amod_pruning_different pobj_with_pruning nn__strategies dobj_been_ cc__and nn_pruning_cube conj__pruning ' +o,1924,' and Alshawi et al ',Wu,'cc__and conj__Alshawi cc_Alshawi_et conj_Alshawi_al ' +o,1925,'The noun phrases in this data set are the same as in the Treebank and therefore the baseNPs in this data set are slightly different from the ones in the data sets ',Ramshaw,'det_phrases_The nn_phrases_noun nsubj_are_phrases prep_phrases_in det_set_this nn_set_data pobj_in_set det_same_the dep_as_same dep_in_as dep_are_in det_Treebank_the pobj_in_Treebank cc_in_and conj_in_therefore det_baseNPs_the nsubj_different_baseNPs prep_baseNPs_in det_set_this nn_set_data pobj_in_set cop_different_are advmod_different_slightly dep_are_different prep_different_from det_ones_the pobj_from_ones prep_ones_in det_sets_the amod_sets_ nn_sets_data pobj_in_sets ' +o,1926,'This second expression is similar to that used in ',Ramshaw,'det_expression_This amod_expression_second nsubj_similar_expression cop_similar_is prep_similar_to pobj_to_that partmod_that_used prep_used_in pobj_in_ ' +o,1927,'The result in implies that for the special case of Bracketing ITGs , the time complexity of the algorithm is parenleftbigT3V 3parenrightbig where T and V are the lengths of the two sentences ',Wu,'det_result_The nsubj_implies_result prep_result_in pobj_in_ complm_3parenrightbig_that prep_3parenrightbig_for det_case_the amod_case_special pobj_for_case prep_case_of nn_ITGs_Bracketing pobj_of_ITGs det_complexity_the nn_complexity_time nsubj_3parenrightbig_complexity prep_complexity_of det_algorithm_the pobj_of_algorithm cop_3parenrightbig_is amod_3parenrightbig_parenleftbigT3V ccomp_implies_3parenrightbig advmod_lengths_where nsubj_lengths_T cc_T_and conj_T_V cop_lengths_are det_lengths_the rcmod_3parenrightbig_lengths prep_lengths_of det_sentences_the num_sentences_two pobj_of_sentences ' +o,1928,' used a likelihood ratio to test word similarity under the assumption that the words in text have a binomial distribution ',Dunning,'advmod_used_ det_ratio_a nn_ratio_likelihood dobj_used_ratio aux_test_to xcomp_used_test nn_similarity_word dobj_test_similarity prep_test_under det_assumption_the pobj_under_assumption dep_have_that det_words_the nsubj_have_words prep_words_in pobj_in_text dep_used_have det_distribution_a amod_distribution_binomial dobj_have_distribution ' +o,1929,'We use eight similarity measures implemented within the WordNet : : Similarity package5 , described in ; these include three measures derived from the paths between the synsets in WordNet : HSO , LCH , and WUP ; three measures based on information content : RES , LIN , and JCN ; the gloss-based Extended Lesk Measure LESK , , and finally the gloss vector similarity measure VECTOR ',Pedersen,'nsubj_use_We num_measures_eight nn_measures_similarity dobj_use_measures partmod_measures_implemented prep_implemented_within det_WordNet_the pobj_within_WordNet nn_package5_Similarity dep_WordNet_package5 partmod_package5_described prep_described_in pobj_in_ dep_WordNet_these rcmod_these_include num_measures_three dobj_include_measures partmod_measures_derived prep_derived_from det_paths_the pobj_from_paths prep_paths_between det_synsets_the pobj_between_synsets prep_synsets_in pobj_in_WordNet nn__HSO dep_WordNet_ nn__LCH conj__ cc__and nn__WUP conj__ num_measures_three dep_WordNet_measures partmod_measures_based prep_based_on nn_content_information pobj_on_content nn__RES dep_WordNet_ nn__LIN conj__ cc__and nn__JCN conj__ det_LESK_the amod_LESK_gloss-based nn_LESK_Extended nn_LESK_Lesk nn_LESK_Measure dep_WordNet_LESK appos_LESK_ cc_WordNet_and advmod_measure_finally det_measure_the amod_measure_gloss nn_measure_vector nn_measure_similarity conj_WordNet_measure nn__VECTOR dep_measure_ ' +o,1930,'Following this idea , there have been introduced a parameter estimation approach for non-generative approaches that can effectively incorporate unlabeled data ',Suzuki,'prep_introduced_Following det_idea_this pobj_Following_idea expl_introduced_there aux_introduced_have auxpass_introduced_been det_approach_a amod_approach_parameter nn_approach_estimation dobj_introduced_approach prep_introduced_for amod_approaches_non-generative pobj_for_approaches nsubj_incorporate_that aux_incorporate_can advmod_incorporate_effectively rcmod_approaches_incorporate amod_data_unlabeled dobj_incorporate_data ' +o,1931,'First , hierarchical word clusters are derived from unlabeled data using the Brown et al clustering algorithm ',Brown,'advmod_derived_First amod_clusters_hierarchical nn_clusters_word nsubjpass_derived_clusters auxpass_derived_are prep_derived_from amod_data_unlabeled pobj_from_data xcomp_derived_using det_algorithm_the nn_algorithm_Brown nn_algorithm_et nn_algorithm_al amod_algorithm_clustering dobj_using_algorithm ' +o,1932,'Some improvements on BOW are given by the use of dependency trees and syntactic parse trees , , , but these , too are not adequate when dealing with complex questions whose answers are expressed by long and articulated sentences or even paragraphs ',Lin,'det_improvements_Some nsubjpass_given_improvements prep_improvements_on pobj_on_BOW auxpass_given_are prep_given_by det_use_the pobj_by_use prep_use_of amod_trees_dependency pobj_of_trees cc_use_and amod_trees_syntactic nn_trees_parse conj_use_trees nn__ appos__ dep_trees_ cc_given_but nsubj_adequate_these advmod_adequate_too cop_adequate_are neg_adequate_not conj_given_adequate advmod_dealing_when ccomp_adequate_dealing prep_dealing_with amod_questions_complex pobj_with_questions poss_answers_whose nsubjpass_expressed_answers auxpass_expressed_are rcmod_questions_expressed prep_expressed_by amod_sentences_long cc_long_and conj_long_articulated pobj_by_sentences cc_adequate_or advmod_adequate_even conj_adequate_paragraphs ' +o,1933,'Following , we only include features which occur 5 times or more in training data ',Ratnaparkhi,'prep_include_Following pobj_Following_ nsubj_include_we advmod_include_only dobj_include_features nsubj_occur_which rcmod_features_occur num_times_5 advmod_occur_times cc_times_or conj_times_more prep_occur_in nn_data_training pobj_in_data ' +o,1934,'1 Introduction The Inversion Transduction Grammar -LRB- ITG -RRB- of is a syntactically motivated algorithm for producing word-level alignments of pairs of translationally equivalent sentences in two languages ',Wu,'num_Introduction_1 det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction nsubj_algorithm_Grammar abbrev_Grammar_ITG prep_Grammar_of pobj_of_ cop_algorithm_is det_algorithm_a advmod_motivated_syntactically amod_algorithm_motivated dep_Introduction_algorithm prep_algorithm_for pcomp_for_producing amod_alignments_word-level dobj_producing_alignments prep_alignments_of pobj_of_pairs prep_pairs_of advmod_equivalent_translationally amod_sentences_equivalent pobj_of_sentences prep_sentences_in num_languages_two pobj_in_languages ' +o,1935,'Dependency models have recently gained considerable interest in many NLP applications , including machine translation ',Shen,'amod_models_Dependency nsubj_gained_models aux_gained_have advmod_gained_recently amod_interest_considerable dobj_gained_interest prep_gained_in amod_applications_many nn_applications_NLP pobj_in_applications prep_gained_including nn_translation_machine pobj_including_translation ' +o,1936,'The classical Bayes relation is used to introduce a target language model : e = argmaxe Pr -LRB- e f -RRB- = argmaxe Pr -LRB- f e -RRB- Pr -LRB- e -RRB- where Pr -LRB- f e -RRB- is the translation model and Pr -LRB- e -RRB- is the target language model ',Brown,'det_relation_The amod_relation_classical nn_relation_Bayes nsubjpass_used_relation auxpass_used_is aux_introduce_to xcomp_used_introduce det__a nn__target nn__language nn__model dobj_introduce_ advmod_=_e dep_introduce_= dobj_=_argmaxe nsubj_=_Pr dep_f_e dep_-LRB-_f dep_=_= nn_Pr_argmaxe nn_Pr_Pr dep_-LRB-_f dep_f_e dobj_=_Pr dep_-LRB-_e advmod_model_where nsubj_model_Pr dep_-LRB-_f dep_f_e cop_model_is det_model_the nn_model_translation rcmod_Pr_model cc_model_and nsubj_model_Pr dep_-LRB-_e cop_model_is det_model_the nn_model_target nn_model_language conj_model_model ' +p,1937,'A promising approach may be to use aligned bilingual corpora , especially for augmenting existing lexicons with domain-specific terminology ',Dagan,'det_approach_A amod_approach_promising nsubj_be_approach aux_be_may aux_use_to xcomp_be_use dep_use_aligned amod_corpora_bilingual dobj_aligned_corpora advmod_aligned_especially prep_aligned_for pcomp_for_augmenting amod_lexicons_existing dobj_augmenting_lexicons prep_augmenting_with amod_terminology_domain-specific pobj_with_terminology ' +o,1938,'Note that Row 3 of Table 3 corresponds to s system which applies only word pair features ',Marcu,'complm_corresponds_that nsubj_corresponds_Row num_Row_3 prep_Row_of pobj_of_Table num_Table_3 ccomp_Note_corresponds aux__to xcomp_corresponds_ dobj__s nsubj_Note_system nsubj_applies_which rcmod_system_applies amod_features_only nn_features_word nn_features_pair dobj_applies_features ' +o,1939,'Since the advent of manually tagged corpora such as the Brown Corpus and the Penn Treebank -LRB- Francis , Marcus -RRB- , the efficacy of machine learning for training a tagger has been demonstrated using a wide array of techniques , including : Markov models , decision trees , connectionist machines , transformations , nearest-neighbor algorithms , and maximum entropy -LRB- Weischedel , Black , Schmid , Brill , Daelemans , -RRB- ',Ratnaparkhi,'prep_demonstrated_Since det_advent_the pobj_Since_advent prep_advent_of advmod_corpora_manually amod_corpora_tagged pobj_of_corpora dep_as_such prep_corpora_as det_Corpus_the nn_Corpus_Brown pobj_as_Corpus cc_corpora_and det_Treebank_the nn_Treebank_Penn conj_corpora_Treebank nn__Francis appos_Treebank_ nn__Marcus dep__ det_efficacy_the nsubjpass_demonstrated_efficacy prep_efficacy_of pobj_of_machine partmod_machine_learning prep_learning_for pcomp_for_training det_tagger_a dobj_training_tagger aux_demonstrated_has auxpass_demonstrated_been xcomp_demonstrated_using det_array_a amod_array_wide dobj_using_array prep_array_of pobj_of_techniques prep_demonstrated_including nn_models_Markov pobj_including_models nn_trees_decision conj_models_trees amod_machines_connectionist conj_models_machines conj_models_transformations amod_algorithms_nearest-neighbor conj_models_algorithms cc_models_and amod_entropy_maximum conj_models_entropy nn__Weischedel dep_entropy_ nn__Black appos__ nn__Schmid appos__ nn__Brill appos__ nn__Daelemans appos__ appos__ ' +p,1940,'It was found to produce automated scores , which strongly correlate with human judgements about translation fluency ',Papineni,'nsubjpass_found_It auxpass_found_was aux_produce_to xcomp_found_produce amod_scores_automated dobj_produce_scores nsubj_correlate_which advmod_correlate_strongly rcmod_scores_correlate prep_correlate_with amod_judgements_human pobj_with_judgements prep_judgements_about nn_fluency_translation pobj_about_fluency ' +o,1941,'With our best performing features , we get ROUGE-2 scores of 011 and 00925 on 2007 and 2006 5This threshold was derived experimentally with previous data ',Lin,'prep_get_With poss_features_our dep_performing_best amod_features_performing pobj_With_features nsubj_get_we nn_scores_ROUGE-2 nn_scores_ dobj_get_scores prep_scores_of pobj_of_011 cc_011_and conj_011_00925 prep_get_on pobj_on_2007 cc_2007_and conj_2007_2006 amod_threshold_5This nsubjpass_derived_threshold auxpass_derived_was dep_get_derived advmod_derived_experimentally prep_derived_with amod_data_previous pobj_with_data ' +o,1942,'For the future , the joint model would benefit from lexical weighting like that used in the standard model ',Koehn,'prep_benefit_For det_future_the pobj_For_future det_model_the amod_model_joint nsubj_benefit_model aux_benefit_would prep_benefit_from amod_weighting_lexical pobj_from_weighting prep_benefit_like pobj_like_that partmod_that_used prep_used_in det_model_the amod_model_standard pobj_in_model ' +o,1943,'All model weights were trained on development sets via minimum-error rate training -LRB- MERT -RRB- with 200 unique n-best lists and optimizing toward BLEU ',Och,'det_weights_All nn_weights_model nsubjpass_trained_weights auxpass_trained_were prep_trained_on nn_sets_development pobj_on_sets prep_sets_via amod_training_minimum-error nn_training_rate pobj_via_training abbrev_training_MERT ccomp_trained_ prep__with num_lists_200 amod_lists_unique amod_lists_n-best pobj_with_lists cc__and conj__optimizing prep_optimizing_toward pobj_toward_BLEU ' +o,1944,'Seen from Table 2 , our result about SCL is in accord with that in on the whole ',Blitzer,'nsubj_is_Seen prep_Seen_from pobj_from_Table num_Table_2 poss_result_our nsubj_is_result prep_result_about pobj_about_SCL prep_is_in pobj_in_accord prep_accord_with pobj_with_that prep_that_in pobj_in_ prep__on det_whole_the pobj_on_whole ' +o,1945,'71 Interand Intra-annotator agreement We measured pairwise agreement among annotators usingthekappacoefficient -LRB- K -RRB- whichiswidelyused in computational linguistics for measuring agreement in category judgments ',Carletta,'num_agreement_71 nn_agreement_Interand nn_agreement_Intra-annotator nsubj_measured_We dep_agreement_measured amod_agreement_pairwise dobj_measured_agreement prep_measured_among amod_usingthekappacoefficient_annotators pobj_among_usingthekappacoefficient appos_usingthekappacoefficient_K partmod_usingthekappacoefficient_whichiswidelyused prep_whichiswidelyused_in amod_linguistics_computational pobj_in_linguistics prep_whichiswidelyused_for pcomp_for_measuring dobj_measuring_agreement prep_measuring_in nn_judgments_category pobj_in_judgments ' +n,1946,'In particular , the model in failed to generate punctuation , a deficiency of the model ',Collins,'prep_failed_In pobj_In_particular det_model_the nsubj_failed_model prep_model_in pobj_in_ aux_generate_to xcomp_failed_generate dobj_generate_punctuation det_deficiency_a appos_punctuation_deficiency prep_deficiency_of det_model_the pobj_of_model ' +o,1947,'Online baselines include Top-1 Perceptron , Top-1 Passive-Aggressive -LRB- PA -RRB- , and k-best PA ',Collins,'amod_baselines_Online nsubj_include_baselines nn__Top-1 nn__Perceptron dobj_include_ amod_Passive-Aggressive_Top-1 conj__Passive-Aggressive abbrev_Passive-Aggressive_PA cc__and amod__k-best nn__PA conj__ ' +o,1948,'For the MUC6 data set , we extract noun phrases -LRB- mentions -RRB- automatically , but for MPQA , we assume mentions for coreference resolution are given as in ',Stoyanov,'prep_extract_For det_set_the nn_set_MUC6 nn_set_data pobj_For_set nsubj_extract_we nn_phrases_noun dobj_extract_phrases appos_phrases_mentions advmod_extract_automatically cc_extract_but prep_assume_for pobj_for_MPQA nsubj_assume_we conj_extract_assume ccomp_assume_mentions mark_given_for amod_resolution_coreference nsubjpass_given_resolution auxpass_given_are advcl_mentions_given prep_given_as dep_as_in ' +o,1949,'MET iterative parameter estimation under IBM BLEU is performed on the development set ',Och,'nn_estimation_MET nn_estimation_ amod_estimation_iterative nn_estimation_parameter nsubjpass_performed_estimation prep_estimation_under nn_BLEU_IBM pobj_under_BLEU auxpass_performed_is prep_performed_on det_set_the nn_set_development pobj_on_set ' +o,1950,'The only requirement will be that a parallel corpus exist for the language under consideration and one or more other languages.2 Induction of grammars from parallel corpora is rarely viewed as a promising task in its own right; in work that has addressed the issue directly (Wu, 1997; Melamed, 2003; Melamed, 2004), the synchronous grammar is mainly viewed as instrumental in the process of improving the translation model in a noisy channel approach to statistical MT.3 In the present paper, we provide an important prerequisite for parallel corpus-based grammar induction work: an efficient algorithm for synchronous parsing of sentence pairs, given a word alignment.',Wu,'' +o,1951,'We use the averaged perceptron algorithm , as presented in , to train the parser ',Collins,'nsubj_use_We det_algorithm_the amod_algorithm_averaged nn_algorithm_perceptron nsubj_train_algorithm mark_presented_as dep_algorithm_presented prep_presented_in pobj_in_ aux_train_to xcomp_use_train det_parser_the dobj_train_parser ' +o,1952,'IBM constraints , lexical word reordering model , and inversion transduction grammar -LRB- ITG -RRB- constraints belong to this type of approach ',Wu,'nn_constraints_IBM nsubj__constraints amod_model_lexical nn_model_word nn_model_reordering nsubj__model ccomp__ cc__and nn_grammar_inversion nn_grammar_transduction nn_constraints_grammar abbrev_constraints_ITG nsubj_belong_constraints advmod_belong_ conj__belong prep_belong_to det_type_this pobj_to_type prep_type_of pobj_of_approach ' +o,1953,'We evaluated annotation reliability by using the Kappa statistic ',Carletta,'nsubj_evaluated_We nn_reliability_annotation dobj_evaluated_reliability prep_evaluated_by pcomp_by_using det_statistic_the nn_statistic_Kappa dobj_using_statistic ' +p,1954,'1 Introduction The most widely used alignment model is IBM Model 4 ',Brown,'num_Introduction_1 det_model_The advmod_used_most advmod_used_widely amod_model_used nn_model_alignment nsubj_Model_model cop_Model_is nn_Model_IBM dep_Introduction_Model num_Model_4 ' +o,1955,'Direct feedback loops that copy a predicted output label to the input representation of the next example have been used in symbolic machine-learning architectures such as the the maximum-entropy tagger described by and the memory-based tagger -LRB- MBT -RRB- proposed by Daelemans et al ',Ratnaparkhi,'amod_loops_Direct nn_loops_feedback nsubjpass_used_loops nsubj_copy_that rcmod_loops_copy det_label_a amod_label_predicted nn_label_output dobj_copy_label prep_copy_to det_representation_the nn_representation_input pobj_to_representation prep_representation_of det_example_the amod_example_next pobj_of_example aux_used_have auxpass_used_been prep_used_in amod_architectures_symbolic amod_architectures_machine-learning pobj_in_architectures dep_as_such prep_architectures_as dep_tagger_the det_tagger_the amod_tagger_maximum-entropy pobj_as_tagger partmod_tagger_described prep_described_by pobj_by_ cc_tagger_and det_tagger_the amod_tagger_memory-based conj_tagger_tagger abbrev_tagger_MBT partmod_tagger_proposed prep_proposed_by pobj_by_Daelemans cc_Daelemans_et conj_Daelemans_al ' +o,1956,'This is important when LARGE CUT-OFF 0 5 100 NAIVE 541,721 184,493 35,617 SASH 10,599 8,796 6,231 INDEX 5,844 13,187 32,663 Table 4 : Average number of comparisons per term considering that different tasks may require different weights and measures ',Weeds,'nsubj_important_This cop_important_is advmod_CUT-OFF_when nsubj_CUT-OFF_LARGE advcl_important_CUT-OFF num_Table_0 num_Table_5 num_Table_100 num_Table_NAIVE num_Table_541,721 num_Table_184,493 num_Table_35,617 num_Table_SASH num_Table_10,599 num_Table_8,796 num_Table_6,231 num_Table_INDEX num_Table_5,844 num_Table_13,187 nn_Table_32,663 dobj_CUT-OFF_Table dep_Table_4 amod_number_Average dep_Table_number prep_number_of pobj_of_comparisons prep_comparisons_per pobj_per_term partmod_term_considering complm_require_that amod_tasks_different nsubj_require_tasks aux_require_may ccomp_considering_require amod_weights_different dobj_require_weights cc_weights_and conj_weights_measures ' +o,1957,'264-285. T Fukushima M Okumura Text summarization challenge: text summarization in Japan 2001 in Proceedings of NAACL 2001 Workshop Automatic Summarization 51--59 Conferences (MUC) (Chinchor et al, 1993), TIPSTER SUMMAC Text Summarization Evaluation (Mani et al, 1998), Document Understanding Conference (DUC) (DUC, 2004), and Text Summarization Challenge (TSC) (Fukushima and Okumura, 2001), have attested the importance of this topic.',Marcus,'' +p,1958,'We view this as a particularly promising aspect of our work , given that phrase-based systems such as Pharaoh perform better with higher recall alignments ',Koehn,'nsubj_view_We dobj_view_this prep_this_as det_aspect_a advmod_aspect_particularly amod_aspect_promising pobj_as_aspect prep_aspect_of poss_work_our pobj_of_work prep_view_given complm_perform_that amod_systems_phrase-based nsubj_perform_systems dep_as_such prep_systems_as nn__Pharaoh pobj_as_ pcomp_given_perform advmod_perform_better prep_perform_with amod_alignments_higher nn_alignments_recall pobj_with_alignments ' +o,1959,'Thus , we are lead to an ` ontologically promiscuous \' semantics ',Hobbs,'advmod_lead_Thus nsubj_lead_we cop_lead_are prep_lead_to det_semantics_an nn_semantics_ontologically nn_semantics_promiscuous pobj_to_semantics ' +p,1960,'To facilitate comparisons with previous work , we used the training\\/development\\/test partition defined in the corpus and we also used the automatically-assigned part of speech tags provided in the corpus10 Czech word clusters were derived from the raw text section of the PDT 10 , which contains about 39 million words of newswire text11 We trained the parsers using the averaged perceptron , which represents a balance between strong performance and fast training times ',Collins,'aux_facilitate_To dep_used_facilitate dobj_facilitate_comparisons prep_facilitate_with amod__previous nn__work pobj_with_ nsubj_used_we det_partition_the amod_partition_training\\/development\\/test dobj_used_partition partmod_partition_defined prep_defined_in det_corpus_the pobj_in_corpus cc_used_and nsubj_used_we advmod_used_also conj_used_used det_part_the amod_part_automatically-assigned nsubjpass_derived_part prep_part_of nn_tags_speech pobj_of_tags partmod_tags_provided prep_provided_in det_clusters_the amod_clusters_corpus10 amod_clusters_Czech nn_clusters_word pobj_in_clusters auxpass_derived_were ccomp_used_derived prep_derived_from det_section_the amod_section_raw nn_section_text pobj_from_section prep_section_of det_PDT_the pobj_of_PDT num_PDT_10 nsubj_contains_which rcmod_PDT_contains quantmod_million_about number_million_39 num_words_million dobj_contains_words prep_words_of nn_text11_newswire pobj_of_text11 nsubj_trained_We dep_used_trained det_parsers_the nsubj_using_parsers dep_trained_using det__the amod__averaged nn__perceptron dobj_using_ nsubj_represents_which rcmod__represents det_balance_a dobj_represents_balance prep_balance_between amod_performance_strong pobj_between_performance cc_performance_and amod_times_fast nn_times_training conj_performance_times ' +p,1961,'This kind of corpus has served as an extremely valuable resource for computational linguistics applications such as machine translation and question answering , and has also proved useful in theoretical linguistics research ',Marcus,'det_kind_This nsubj_served_kind prep_kind_of pobj_of_corpus aux_served_has prep_served_as det_resource_an advmod_valuable_extremely amod_resource_valuable pobj_as_resource prep_resource_for amod_applications_computational nn_applications_linguistics pobj_for_applications dep_as_such prep_applications_as nn_translation_machine pobj_as_translation cc_translation_and nn__question nn__answering conj_translation_ cc_served_and aux_proved_has advmod_proved_also conj_served_proved amod_research_useful prep_useful_in amod_linguistics_theoretical pobj_in_linguistics dobj_proved_research ' +o,1962,'Movie-review dataset consists of positive and negative reviews from the Internet Movie Database -LRB- IMDb -RRB- archive ',Pang,'amod_consists_dataset dobj_Movie-review_consists prep_consists_of amod_reviews_positive cc_positive_and conj_positive_negative pobj_of_reviews prep_Movie-review_from det_Internet_the pobj_from_Internet nn_archive_Movie nn_archive_Database abbrev_archive_IMDb nsubj_Movie-review_archive ' +o,1963,'The idea of topic signature terms was introduced by Lin and Hovy in the context of single document summarization , and was later used in several multi-document summarization systems ',Lin,'det_idea_The nsubjpass_introduced_idea prep_idea_of nn_terms_topic nn_terms_signature pobj_of_terms auxpass_introduced_was prep_introduced_by nn__Lin cc_Lin_and conj_Lin_Hovy pobj_by_ prep_introduced_in det_context_the pobj_in_context prep_context_of amod_summarization_single nn_summarization_document pobj_of_summarization cc_introduced_and auxpass_used_was advmod_used_later conj_introduced_used prep_used_in amod__several amod__multi-document nn__summarization nn__systems pobj_in_ ' +o,1964,'421 Teufel and Moens Summarizing Scientific Articles We use the kappa coefficient K to measure stability and reproducibility , following ',Carletta,'aux_Summarizing_421 nsubj_Summarizing_Teufel cc_Teufel_and conj_Teufel_Moens nn_Articles_Scientific dobj_Summarizing_Articles nsubj_use_We rcmod_Articles_use det__the dep_coefficient_kappa amod__coefficient nn__K dobj_use_ aux_measure_to xcomp_use_measure dobj_measure_stability cc_stability_and conj_stability_reproducibility xcomp_use_following ' +o,1965,'Classes were identified using a POS tagger trained on the tagged Switchboard corpus ',Ratnaparkhi,'nsubjpass_identified_Classes auxpass_identified_were xcomp_identified_using det__a dep__POS nn__tagger dobj_using_ partmod__trained prep_trained_on det_corpus_the nn_corpus_tagged nn_corpus_Switchboard pobj_on_corpus ' +o,1966,'303 Wiebe , Wilson , Bruce , Bell , and Martin Learning Subjective Language While it is common in studies of collocations to omit low-frequency words and expressions from analysis , because they give rise to invalid or unrealistic statistical measures , we are able to identify higher-precision collocations by including placeholders for unique words -LRB- ie , the ugen-n-grams -RRB- ',Church,'nn_Bell_303 nn_Bell_Wiebe appos_Bell_Wilson appos_Bell_Bruce cc_Bell_and nn_Learning_Martin conj_Bell_Learning nn_Language_Subjective dep_common_Language dep_common_While nsubj_common_it cop_common_is advcl_able_common prep_common_in pobj_in_studies prep_studies_of pobj_of_collocations aux_omit_to xcomp_common_omit amod_words_low-frequency dobj_omit_words cc_words_and conj_words_expressions prep_omit_from pobj_from_analysis mark_give_because nsubj_give_they advcl_common_give iobj_give_rise prep_rise_to amod_measures_invalid cc_invalid_or conj_invalid_unrealistic amod_measures_statistical pobj_to_measures dobj_give_ nsubj_able_we cop_able_are dep_Bell_able aux_identify_to xcomp_able_identify amod_collocations_higher-precision dobj_identify_collocations prep_identify_by pcomp_by_including dobj_including_placeholders prep_placeholders_for amod_words_unique pobj_for_words dep_words_ie det_ugen-n-grams_the appos_ie_ugen-n-grams ' +o,1967,'In showing how DLTAG and an interpretative process on its derivations operate , we must , of necessity , gloss over how inference triggered by adjacency or associated with a structural connective provides the intended relation between adjacent discourse 578 Computational Linguistics Volume 29 , Number 4 units : It may be a matter simply of statistical inference , as in , or of more complex inference , as in Hobbs et al ',Marcu,'prep_gloss_In pcomp_In_showing advmod_operate_how nsubj_operate_DLTAG cc_DLTAG_and det_process_an amod_process_interpretative conj_DLTAG_process prep_process_on poss_derivations_its pobj_on_derivations ccomp_showing_operate nsubj_gloss_we aux_gloss_must prep_gloss_of pobj_of_necessity prt_gloss_over advmod_provides_how nsubj_provides_inference dep_inference_triggered prep_triggered_by pobj_by_adjacency cc_triggered_or conj_triggered_associated prep_associated_with det_connective_a amod_connective_structural pobj_with_connective ccomp_gloss_provides det_relation_the amod_relation_intended dobj_provides_relation prep_relation_between amod_units_adjacent nn_units_discourse num_units_578 nn_units_Computational nn_units_Linguistics nn_units_Volume num_units_29 dep_units_Number num_units_4 pobj_between_units nsubj_matter_It aux_matter_may cop_matter_be det_matter_a parataxis_gloss_matter dep_of_simply prep_matter_of amod_inference_statistical pobj_of_inference advmod_matter_as dep_as_in pobj_in_ cc_in_or conj_in_of advmod_complex_more amod_inference_complex pobj_of_inference advmod_inference_as dep_as_in pobj_in_Hobbs cc_Hobbs_et conj_Hobbs_al ' +o,1968,'An alternative would be using a vector space model for classi cation where calltypes and utterances are represented as vectors including word a2 - grams ',Chu-Carroll,'det_alternative_An nsubj_model_alternative aux_model_would aux_model_be dep_model_using det_model_a nn_model_vector nn_model_space prep_model_for nn_cation_classi pobj_for_cation advmod_represented_where nsubjpass_represented_calltypes cc_calltypes_and conj_calltypes_utterances auxpass_represented_are advcl_model_represented prep_represented_as pobj_as_vectors prep_vectors_including nn_a2_word pobj_including_a2 dep_a2_grams ' +o,1969,'2 IBM Model 4 Various statistical alignment models of the form Pr -LRB- fJ1 ; aJ1jeI1 -RRB- have been introduced in ',Brown,'num_Model_2 nn_Model_IBM num_models_4 amod_models_Various amod_models_statistical nn_models_alignment nsubjpass_introduced_models prep_models_of det_Pr_the nn_Pr_form pobj_of_Pr nn_aJ1jeI1_fJ1 appos_Pr_aJ1jeI1 aux_introduced_have auxpass_introduced_been dep_Model_introduced prep_introduced_in ' +o,1970,'1 Introduction Recent work in statistical machine translation -LRB- MT -RRB- has sought to overcome the limitations of phrasebased models by making use of syntactic information ',Koehn,'num_work_1 amod_work_Introduction amod_work_Recent nsubj_sought_work prep_work_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_MT aux_sought_has aux_overcome_to xcomp_sought_overcome det_limitations_the dobj_overcome_limitations prep_limitations_of amod_models_phrasebased pobj_of_models partmod_models_ prep__by pcomp_by_making dobj_making_use prep_use_of amod_information_syntactic pobj_of_information ' +o,1971,'Several authors have used mutual information and similar statistics as an objective function for word clustering , for automatic determination of phonemic baseforms , and for language modeling for speech recognition ',Brown,'amod_authors_Several nsubj_used_authors aux_used_have amod_information_mutual dobj_used_information cc_information_and amod_statistics_similar conj_information_statistics prep_information_as det_function_an amod_function_objective pobj_as_function prep_used_for nn__word nn__clustering pobj_for_ prep_used_for amod_determination_automatic pobj_for_determination prep_determination_of amod__phonemic nn__baseforms pobj_of_ cc_for_and conj_for_for nn_modeling_language pobj_for_modeling prep_modeling_for nn__speech nn__recognition pobj_for_ ' +p,1972,'1 Motivation Statistical part-of-speech disambiguation can be efficiently done with n-gram models ',Cutting,'num_disambiguation_1 nn_disambiguation_Motivation nn_disambiguation_Statistical amod_disambiguation_part-of-speech nsubjpass_done_disambiguation aux_done_can auxpass_done_be advmod_done_efficiently prep_done_with amod_models_n-gram pobj_with_models ' +o,1973,'2 The IBM Model 4 For the work described in this paper we used a modified version of the statistical machine translation tool developed in the context of the 1999 Johns HopkinsSummer Workshop , which implements IBM translation model 4 ',Brown,'dep_described_2 det_Model_The nn_Model_IBM nsubj_described_Model num_Model_4 prep_Model_For det_work_the pobj_For_work prep_described_in det_paper_this pobj_in_paper nsubj_used_we dep_described_used det_version_a amod_version_modified dobj_used_version prep_version_of det_tool_the amod_tool_statistical nn_tool_machine nn_tool_translation pobj_of_tool partmod_tool_developed prep_developed_in det_context_the pobj_in_context prep_context_of det__the num__1999 nn__Johns nn__HopkinsSummer nn__Workshop pobj_of_ nsubj_implements_which rcmod_tool_implements nn_translation_IBM nsubj_model_translation ccomp_implements_model dobj_model_4 ' +o,1974,'For this we used two resources : CELEX a linguistically annotated dictionary of English , Dutch and German , and the Dutch snowball stemmer implementing a suf x stripping algorithm based on the Porter stemmer ',Brown,'prep_used_For pobj_For_this nsubj_used_we num_resources_two dobj_used_resources dep_resources_CELEX det_dictionary_a advmod_annotated_linguistically amod_dictionary_annotated dep_CELEX_dictionary prep_resources_of amod__English conj_English_Dutch cc_English_and conj_English_German pobj_of_ cc_resources_and det_stemmer_the amod_stemmer_Dutch amod_stemmer_snowball conj_resources_stemmer partmod_stemmer_implementing det_suf_a dobj_implementing_suf nsubj_stripping_x dep_used_stripping dobj_stripping_algorithm prep_stripping_based dep_based_on det_stemmer_the nn_stemmer_Porter pobj_on_stemmer ' +o,1975,'32 Word Order Differences Another problem that has been noticed as early as 1993 with the first research on word alignment concerns the differences in word order between source and target language ',Brown,'num_Differences_32 nn_Differences_Word nn_Differences_Order det_problem_Another nsubj_concerns_problem nsubjpass_noticed_that aux_noticed_has auxpass_noticed_been rcmod_problem_noticed advmod_early_as advmod_noticed_early prep_noticed_as pobj_as_1993 prep_noticed_with det_research_the amod_research_first pobj_with_research prep_research_on nn__word nn__alignment pobj_on_ dep_Differences_concerns det_differences_the dobj_concerns_differences prep_concerns_in nn_order_word pobj_in_order prep_order_between nn_language_source cc_source_and conj_source_target pobj_between_language ' +p,1976,'For the IBM models defined by a pioneering paper , a decoding algorithm based on a left-to-right search was described in ',Brown,'mark_defined_For det_models_the nn_models_IBM nsubj_defined_models advcl_described_defined prep_defined_by det__a amod__pioneering nn__paper pobj_by_ det_algorithm_a amod_algorithm_decoding nsubjpass_described_algorithm prep_algorithm_based dep_based_on det_search_a amod_search_left-to-right pobj_on_search auxpass_described_was prep_described_in pobj_in_ ' +p,1977,'1 Introduction The emergence of phrase-based statistical machine translation -LRB- PSMT -RRB- has been one of the major developments in statistical approaches to translation ',Koehn,'nsubj_been_1 partmod_1_Introduction det_emergence_The dobj_Introduction_emergence prep_emergence_of amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_PSMT advmod_Introduction_ aux_been_has dobj_been_one prep_one_of det_developments_the amod_developments_major pobj_of_developments prep_developments_in amod_approaches_statistical pobj_in_approaches prep_been_to pobj_to_translation ' +p,1978,'In the supervised setting , a recent paper by shows that , using a very simple feature augmentation method coupled with Support Vector Machines , he is able to effectively use both labeled target and source data to provide the best results in a number of NLP tasks ','Daume III','prep_shows_In det_setting_the amod_setting_supervised pobj_In_setting det_paper_a amod_paper_recent nsubj_shows_paper prep_paper_by pobj_by_ complm_able_that dep_able_using det_method_a advmod_simple_very amod_method_simple nn_method_feature nn_method_augmentation dobj_using_method prep_method_coupled dep_coupled_with nn_Machines_Support nn_Machines_Vector pobj_with_Machines nsubj_able_he cop_able_is ccomp_shows_able aux_use_to advmod_use_effectively xcomp_able_use dobj_use_both partmod_both_labeled nn_data_target cc_target_and conj_target_source dobj_labeled_data aux_provide_to xcomp_labeled_provide det_results_the amod_results_best dobj_provide_results prep_results_in det_number_a pobj_in_number prep_number_of nn_tasks_NLP pobj_of_tasks ' +o,1979,' , and Lee , Wilson et al ',Pang,'nn__ conj__ cc__and conj__Lee appos__Wilson cc_Wilson_et conj_Wilson_al ' +o,1980,'As suggested in , we use the averaged perceptron when applying the model to held-out or test data ',Collins,'mark_suggested_As advcl_use_suggested prep_suggested_in pobj_in_ nsubj_use_we det_perceptron_the amod_perceptron_averaged dobj_use_perceptron advmod_applying_when advcl_use_applying det_model_the dobj_applying_model prep_applying_to nn_data_held-out cc_held-out_or conj_held-out_test pobj_to_data ' +o,1981,'Opinion forecasting differs from that of opinion analysis , such as extracting opinions , evaluating sentiment , and extracting predictions ',Kim,'nn_forecasting_Opinion nsubj_differs_forecasting prep_differs_from pobj_from_that prep_that_of nn_analysis_opinion pobj_of_analysis dep_as_such prep_that_as pcomp_as_extracting dobj_extracting_opinions conj_extracting_evaluating dobj_evaluating_sentiment cc_extracting_and conj_extracting_extracting dobj_extracting_predictions ' +o,1982,', Models 2 and 3 of ',Collins,'dobj_Models_2 cc_2_and conj_2_3 prep_Models_of ' +o,1983,'The approach is in the spirit of Smadja on retrieving collocations from text corpora , but is more integrated with parsing ',Dunning,'det_approach_The nsubj_is_approach prep_is_in det_spirit_the pobj_in_spirit prep_spirit_of nn__Smadja pobj_of_ prep_is_on pcomp_on_retrieving dobj_retrieving_collocations prep_retrieving_from nn_corpora_text pobj_from_corpora cc_is_but cop_integrated_is advmod_integrated_more conj_is_integrated prep_integrated_with pcomp_with_parsing ' +p,1984,'Constraining learning by using document boundaries has been used quite effectively in unsupervised word sense disambiguation ',Yarowsky,'xcomp_Constraining_learning prep_learning_by pcomp_by_using nn_boundaries_document nsubjpass_used_boundaries aux_used_has auxpass_used_been ccomp_using_used advmod_effectively_quite advmod_used_effectively prep_used_in amod_disambiguation_unsupervised nn_disambiguation_word nn_disambiguation_sense pobj_in_disambiguation ' +o,1985,' indicated that their results can not directly compare to the results of Shi and Wang due to different experimental settings ',Zhang,'nsubj_indicated_ complm_compare_that poss_results_their nsubj_compare_results aux_compare_can neg_compare_not advmod_compare_directly ccomp_indicated_compare prep_compare_to det_results_the pobj_to_results prep_results_of pobj_of_Shi cc_compare_and nn__Wang dep_due_ conj_compare_due prep_due_to amod_settings_different amod_settings_experimental pobj_to_settings ' +n,1986,'1 Introduction Phrase-based translation models , which go beyond the original IBM translation models 1 by modeling translations of phrases rather than individual words , have been suggested to be the state-of-theart in statistical machine translation by empirical evaluations ',Brown,'num__1 nn__Introduction nn__Phrase-based nn__translation nn__models nsubjpass_suggested_ nsubj_go_which rcmod__go prep_go_beyond det_models_the amod_models_original nn_models_IBM nn_models_translation nsubj__models pcomp_beyond_ dobj__1 prep__by nn_translations_modeling pobj_by_translations prep_translations_of pobj_of_phrases cc_phrases_rather dep_rather_than amod_words_individual conj_phrases_words aux_suggested_have auxpass_suggested_been aux_state-of-theart_to cop_state-of-theart_be det_state-of-theart_the xcomp_suggested_state-of-theart prep_state-of-theart_in amod_translation_statistical nn_translation_machine pobj_in_translation prep_state-of-theart_by amod_evaluations_empirical pobj_by_evaluations ' +o,1987,' found that direct annotation takes twice as long as automatic tagging plus correction , for partof-speech annotation -RRB- ; and the output quality reflects the difficulty of the task -LRB- inter-annotator disagreement is on the order of 10 \% , as contrasted with the approximately 3 \% error rate reported for part-of-speech annotation by et al -RRB- ',Marcus,'nsubj_found_ complm_takes_that amod_annotation_direct nsubj_takes_annotation ccomp_found_takes advmod_long_twice advmod_long_as advmod_takes_long prep_takes_as amod_tagging_automatic pobj_as_tagging prep_tagging_plus dep_plus_correction prep_takes_for amod_annotation_partof-speech pobj_for_annotation cc_takes_and det_quality_the nn_quality_output nsubj_reflects_quality conj_takes_reflects det_difficulty_the dobj_reflects_difficulty prep_difficulty_of det_task_the pobj_of_task amod_disagreement_inter-annotator nsubj_is_disagreement dep_task_is prep_is_on det_order_the pobj_on_order prep_order_of num_\%_10 pobj_of_\% mark_reported_as csubj_reported_contrasted prep_contrasted_with det_rate_the quantmod_3_approximately dep_\%_3 dep_rate_\% nn_rate_error pobj_with_rate advcl_is_reported prep_reported_for amod_annotation_part-of-speech pobj_for_annotation prep_reported_by pobj_by_ cc__et conj__al ' +p,1988,'His results may be improved if more sophisticated methods and larger corpora are used to establish similarity between words ',Hindle,'poss_results_His nsubjpass_improved_results aux_improved_may auxpass_improved_be mark_used_if advmod_sophisticated_more amod_methods_sophisticated nsubjpass_used_methods cc_methods_and amod_corpora_larger conj_methods_corpora auxpass_used_are advcl_improved_used aux_establish_to xcomp_used_establish dobj_establish_similarity prep_similarity_between pobj_between_words ' +o,1989,'1 Introduction In the community of sentiment analysis , transferring a sentiment classifier from one source domain to another target domain is still far from a trivial work , because sentiment expression often behaves with strong domain-specific nature ',Pang,'num_Introduction_1 dep_Introduction_In det_community_the pobj_In_community prep_community_of nn__sentiment nn__analysis pobj_of_ partmod_community_transferring det_classifier_a nn_classifier_sentiment dobj_transferring_classifier prep_transferring_from num_domain_one nn_domain_source pobj_from_domain prep_domain_to det_domain_another nn_domain_target pobj_to_domain rcmod_domain_is advmod_is_still advmod_is_far dep_far_from det_work_a amod_work_trivial pobj_from_work mark_behaves_because nn_expression_sentiment nsubj_behaves_expression advmod_behaves_often advcl_is_behaves prep_behaves_with amod_nature_strong amod_nature_domain-specific pobj_with_nature ' +p,1990,'1 Introduction Statistical parsing models have been shown to be successful in recovering labeled constituencies and have also been shown to be adequate in recovering dependency relationships ',Collins,'num_models_1 nn_models_Introduction nn_models_Statistical nn_models_parsing nsubjpass_shown_models aux_shown_have auxpass_shown_been aux_successful_to cop_successful_be xcomp_shown_successful prep_successful_in pcomp_in_recovering amod_constituencies_labeled dobj_recovering_constituencies advmod_recovering_ cc_successful_and aux_shown_have advmod_shown_also auxpass_shown_been conj_successful_shown aux_adequate_to cop_adequate_be xcomp_shown_adequate prep_adequate_in pcomp_in_recovering amod_relationships_dependency dobj_recovering_relationships ' +p,1991,'1 Introduction Recent works in statistical machine translation -LRB- SMT -RRB- shows how phrase-based modeling significantly outperform the historical word-based modeling ',Koehn,'num_works_1 nn_works_Introduction amod_works_Recent nsubj_shows_works prep_works_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT advmod_outperform_how amod__phrase-based nn__modeling nsubj_outperform_ advmod_outperform_significantly ccomp_shows_outperform det__the amod__historical amod__word-based nn__modeling dobj_outperform_ ' +o,1992,'We view L2P as a tagging task that can be performed with a discriminative learning method , such as the Perceptron HMM ',Collins,'nsubj_view_We dobj_view_L2P prep_view_as det_task_a amod_task_tagging pobj_as_task nsubjpass_performed_that aux_performed_can auxpass_performed_be rcmod_task_performed prep_performed_with det_method_a amod_method_discriminative nn_method_learning pobj_with_method dep_as_such prep_task_as det_HMM_the nn_HMM_Perceptron pobj_as_HMM ' +o,1993,'Using a variant of the voted perceptron , we discriminatively trained our parser in an on-line fashion ',Collins,'dep_discriminatively_Using det_variant_a dobj_Using_variant prep_variant_of det__the amod__voted nn__perceptron pobj_of_ nsubj_discriminatively_we dep_discriminatively_trained poss_parser_our dobj_trained_parser prep_trained_in det_fashion_an amod_fashion_on-line pobj_in_fashion ' +o,1994,'Perhaps the most widely accepted convention is that of ignoring punctuation for the purposes of assigning constituent span, under the perspective that, fun788 Phrase Evaluation Scenario System Type (a) (b) (c) Modified All 98.37 99.72 99.72 Truth VP 92.14 98.70 98.70 Li and Roth All 94.64 (2001) VP 95.28 Collins (1997) All 92.16 93.42 94.28 VP 88.15 94.31 94.42 Charniak All 93.88 95.15 95.32 (2000) VP 88.92 95.11 95.19 Table 1: F-measure shallow bracketing accuracy under three different evaluation scenarios: (a) baseline, used in Li and Roth (2001), with original chunklink script converting treebank trees and context-free parser output; (b) same as (a), except that empty subject NPs are inserted into every unary SVP production; and (c) same as (b), except that punctuation is ignored for setting constituent span.',Collins,'' +o,1995,'32-39 Proceedings of HLT-NAACL 2003 similar distribution patterns ',Hindle,'dep_32-39_Proceedings prep_Proceedings_of pobj_of_HLT-NAACL num_HLT-NAACL_2003 amod_patterns_similar nn_patterns_distribution dep_32-39_patterns ' +o,1996,'31 System Tuning Minimum error training under BLEU was used to optimise the feature weights of the decoder with respect to the dev2006 development set ',Och,'num__31 nn__System nn__Tuning nn__Minimum nn__error nn__training nsubjpass_used_ prep__under nn__BLEU pobj_under_ auxpass_used_was aux_optimise_to xcomp_used_optimise det_weights_the nn_weights_feature dobj_optimise_weights prep_weights_of det_decoder_the pobj_of_decoder prep_decoder_with pobj_with_respect prep_optimise_to det_set_the amod_set_dev2006 nn_set_development pobj_to_set ' +o,1997,'The interpolation weights a65 -LRB- Equation 2 -RRB- are trained using discriminative training using ROUGEa129 as the objective function , on the development set ',Och,'det_a65_The nn_a65_interpolation nn_a65_weights nsubjpass_trained_a65 appos_a65_Equation num_Equation_2 auxpass_trained_are xcomp_trained_using amod__discriminative nn__training dobj_using_ xcomp_using_using dobj_using_ROUGEa129 prep_using_as det_function_the amod_function_objective pobj_as_function prep_trained_on det_development_the pobj_on_development partmod_development_set ' +p,1998,'Tools like Xtract were based on the work of Church and others , but made a step forward by incorporating various statistical measurements like z-score and variance of distribution , as well as shallow linguistic techniques like part-of-speech tagging and lemmatization of input data and partial parsing of raw output ',Smadja,'nsubjpass_based_Tools prep_Tools_like nn__Xtract pobj_like_ auxpass_based_were prep_based_on det_work_the pobj_on_work prep_work_of pobj_of_Church cc_Church_and conj_Church_others cc_based_but conj_based_made det_step_a dobj_made_step advmod_made_forward prep_made_by pcomp_by_incorporating amod_measurements_various amod_measurements_statistical dobj_incorporating_measurements prep_incorporating_like pobj_like_z-score cc_z-score_and conj_z-score_variance prep_z-score_of pobj_of_distribution dep_well_as cc_z-score_well dep_well_as nn_techniques_shallow amod_techniques_linguistic conj_z-score_techniques prep_techniques_like amod_tagging_part-of-speech pobj_like_tagging cc_tagging_and conj_tagging_lemmatization prep_techniques_of nn_data_input pobj_of_data cc_z-score_and amod_parsing_partial conj_z-score_parsing prep_parsing_of amod_output_raw pobj_of_output ' +o,1999,'53 Evaluation Metric This paper focuses on the BLEU metric as presented in ',Papineni,'num_Metric_53 nn_Metric_Evaluation det_paper_This nsubj_focuses_paper dep_Metric_focuses prep_focuses_on det_metric_the amod_metric_BLEU pobj_on_metric mark_presented_as advcl_focuses_presented prt_presented_in ' +o,2000,'where mk is one mention in entity e , and the basic model building block PL -LRB- L = 1je , mk , m -RRB- is an exponential or maximum entropy model ',Berger,'advmod_mention_where nsubj_mention_mk cop_mention_is num_mention_one prep_mention_in pobj_in_entity dep_mention_e cc_mention_and det_PL_the amod_PL_basic nn_PL_model nn_PL_building nn_PL_block nsubj_exponential_PL nn_mk_L amod_mk_= nn_mk_1je appos_PL_mk dep_mk_m cop_exponential_is dep_exponential_an conj_mention_exponential cc_exponential_or amod_model_maximum nn_model_entropy conj_exponential_model ' +o,2001,'Word alignment models were first introduced in statistical machine translation ',Brown,'amod_models_Word nn_models_alignment nsubjpass_introduced_models auxpass_introduced_were advmod_introduced_first prep_introduced_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,2002,'Applying the projection WTx -LRB- where x is a training instance -RRB- would give us m new features , however , for both computational and statistical reasons a low-dimensional approximation of the original feature space is computed by applying Singular Value Decomposition -LRB- SVD -RRB- on W -LRB- step 4 -RRB- ',Blitzer,'det_WTx_the nn_WTx_projection dobj_Applying_WTx advmod_instance_where nsubj_instance_x cop_instance_is det_instance_a nn_instance_training dep_WTx_instance aux_give_would dep_Applying_give iobj_give_us dobj_give_m amod_features_new nsubj_Applying_features appos_features_however mark__for det_reasons_both amod_reasons_computational cc_computational_and conj_computational_statistical nsubj__reasons dep_however_ det_approximation_a amod_approximation_low-dimensional nsubjpass_computed_approximation prep_approximation_of det_space_the amod_space_original nn_space_feature pobj_of_space auxpass_computed_is ccomp__computed prep_computed_by pcomp_by_applying nn_Decomposition_Singular nn_Decomposition_Value dobj_applying_Decomposition abbrev_Decomposition_SVD prep_Decomposition_on pobj_on_W appos_W_step num_step_4 ' +o,2003,'One of the main directions is sentiment classification , which classifies the whole opinion document -LRB- eg , a product review -RRB- as positive or negative ',Pang,'nsubj_classification_One prep_One_of det_directions_the amod_directions_main pobj_of_directions cop_classification_is nn_classification_sentiment nsubj_classifies_which rcmod_classification_classifies det_document_the amod_document_whole nn_document_opinion nsubj_positive_document appos_document_eg det_review_a nn_review_product dep_eg_review advmod_positive_as xcomp_classifies_positive cc_positive_or conj_positive_negative ' +o,2004,' all use multiple context words as discriminating features ',Yarowsky,'advmod__all dep__use amod_words_multiple nn_words_context dobj_use_words prep_use_as pcomp_as_discriminating dobj_discriminating_features ' +o,2005,'2Note that sentence extraction does not solve the problem of selecting and ordering summary sentences to form a coherent There are several approaches to modeling document content : simple word frequency-based methods , graph-based approaches , as well as more linguistically motivated techniques ',Lin,'complm_solve_that nn_extraction_sentence nsubj_solve_extraction aux_solve_does neg_solve_not dep_2Note_solve det_problem_the dobj_solve_problem prep_problem_of pcomp_of_selecting cc_selecting_and conj_selecting_ordering nn_sentences_summary dobj_selecting_sentences aux_form_to xcomp_solve_form det_coherent_a dobj_form_coherent expl_are_There rcmod_coherent_are amod_approaches_several nsubj_are_approaches prep_approaches_to nn_content_modeling nn_content_document pobj_to_content amod__simple nn__word amod__frequency-based nn__methods dep_approaches_ amod__graph-based nn__approaches appos__ dep_well_as cc_approaches_well dep_well_as advmod_techniques_more advmod_motivated_linguistically amod_techniques_motivated conj_approaches_techniques dep_techniques_ ' +o,2006,'By introducing the hidden word alignment variable a, the following approximate optimization criterion can be applied for that purpose: e = argmaxe Pr(e | f) = argmaxe summationdisplay a Pr(e,a | f) argmaxe,a Pr(e,a | f) Exploiting the maximum entropy (Berger et al. , 1996) framework, the conditional distribution Pr(e,a | f) can be determined through suitable real valued functions (called features) hr(e,f,a),r = 1R, and takes the parametric form: p(e,a | f) exp Rsummationdisplay r=1 rhr(e,f,a)} The ITC-irst system (Chen et al. , 2005) is based on a log-linear model which extends the original IBM Model 4 (Brown et al. , 1993) to phrases (Koehn et al. , 2003; Federico and Bertoldi, 2005).',Koehn,'' +o,2007,'Formally , by distributional similarity -LRB- or co-occurrence similarity -RRB- of two words w 1 and w 2 , we mean that they tend to occur in similar contexts , for some definition of context ; or that the set of words that w 1 tends to co-occur with is similar to the set that w 2 tends to co-occur with ; or that if w 1 is substituted for w 2 in a context , its plausibility is unchanged ',Weeds,'advmod_mean_Formally prep_mean_by amod_similarity_distributional pobj_by_similarity cc_similarity_or amod_similarity_co-occurrence conj_similarity_similarity prep_similarity_of num_words_two pobj_of_words number_2_w number_2_1 cc_2_and conj_2_w num_words_2 nsubj_mean_we dep_tend_that nsubj_tend_they dep_mean_tend aux_occur_to xcomp_tend_occur prep_occur_in amod_contexts_similar pobj_in_contexts prep_occur_for det_definition_some pobj_for_definition prep_definition_of pobj_of_context cc_tend_or dep_similar_that det_set_the nsubj_similar_set prep_set_of pobj_of_words complm_tends_that nsubj_tends_w num_w_1 dep_words_tends aux_co-occur_to xcomp_tends_co-occur prep_co-occur_with cop_similar_is conj_tend_similar prep_similar_to det_set_the pobj_to_set dep_tends_that nsubj_tends_w num_w_2 dep_similar_tends aux_co-occur_to xcomp_tends_co-occur prep_co-occur_with cc_tends_or dep_unchanged_that mark_substituted_if nsubjpass_substituted_w num_w_1 auxpass_substituted_is advcl_unchanged_substituted prep_substituted_for pobj_for_w num_w_2 prep_w_in det_context_a pobj_in_context poss__its amod__plausibility nsubj_unchanged_ cop_unchanged_is conj_tends_unchanged ' +o,2008,'(Note that conditioning on the rules parent is needed to disallow the structure [NP [NP PP] PP]; see Johnson [1997] for further discussion.)',Collins,'' +o,2009,'Recent work has explored two-stage decoding , which explicitly decouples decoding into a source parsing stage and a target language model integration stage ',Huang,'amod_work_Recent nsubj_explored_work aux_explored_has amod_decoding_two-stage dobj_explored_decoding nsubj_decouples_which advmod_decouples_explicitly rcmod_decoding_decouples xcomp_decouples_decoding prep_decoding_into det_stage_a nn_stage_source amod_stage_parsing pobj_into_stage cc_stage_and det_stage_a nn_stage_target nn_stage_language nn_stage_model nn_stage_integration conj_stage_stage ' +o,2010,'Machine translation has code-like characteristics , and indeed , the initial models of took a word-substitution\\/transposition approach , trained on a parallel text ',Brown,'nn_translation_Machine nsubj_has_translation amod_characteristics_code-like dobj_has_characteristics cc_has_and advmod_took_indeed det_models_the amod_models_initial nsubj_took_models prep_models_of pobj_of_ conj_has_took det_approach_a amod_approach_word-substitution\\/transposition dobj_took_approach partmod_approach_trained prep_trained_on det_text_a amod_text_parallel pobj_on_text ' +p,2011,'73 224 Minimum Error Rate Training A good way of training is to minimize empirical top-1 error on training data ',Och,'num_Rate_73 num_Rate_224 nn_Rate_Minimum nn_Rate_Error nsubj_is_Rate partmod_Rate_Training det_way_A amod_way_good dobj_Training_way prep_way_of pobj_of_training aux_minimize_to xcomp_is_minimize amod_error_empirical amod_error_top-1 dobj_minimize_error prep_minimize_on nn_data_training pobj_on_data ' +o,2012,'Training Procedure Our algorithm is a modification of the perceptron ranking algorithm , which allows for joint learning across several ranking problems ','Daume III','xcomp_Training_Procedure poss_algorithm_Our dobj_Procedure_algorithm cop_modification_is det_modification_a dep_Training_modification prep_modification_of det_perceptron_the pobj_of_perceptron amod__ranking nn__algorithm nsubj_Training_ nsubj_allows_which rcmod__allows prep_allows_for amod_learning_joint pobj_for_learning prep_learning_across amod_problems_several amod_problems_ranking pobj_across_problems ' +o,2013,'We are currently investigating more challenging problems like multiple category classification using the Reuters-21578 data set and subjective sentiment classification ',Turney,'nsubj_investigating_We aux_investigating_are advmod_investigating_currently advmod_challenging_more amod_problems_challenging dobj_investigating_problems prep_investigating_like amod_classification_multiple nn_classification_category pobj_like_classification partmod_classification_using det_data_the amod_data_Reuters-21578 dobj_using_data partmod_data_set amod_classification_ cc__and conj__subjective nn_classification_sentiment dobj_set_classification ' +o,2014,'We used statistical parser trained on examples from the Penn Treebank to generate parses of the same format for the sentences in our data ',Collins,'nsubj_used_We amod_parser_ amod_parser_statistical dobj_used_parser partmod_parser_trained prep_trained_on pobj_on_examples prep_examples_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank aux_generate_to xcomp_trained_generate dobj_generate_parses prep_parses_of det_format_the amod_format_same pobj_of_format prep_format_for det_sentences_the pobj_for_sentences prep_sentences_in poss_data_our pobj_in_data ' +p,2015,'2 Parsing Model The Berkeley parser is an efficient and effective parser that introduces latent annotations to refine syntactic categories to learn better PCFG grammars ',Matsuzaki,'num_Model_2 nn_Model_Parsing det__The nn__Berkeley nn__parser nsubj_parser_ cop_parser_is det_parser_an amod_parser_efficient cc_efficient_and conj_efficient_effective dep_Model_parser nsubj_introduces_that rcmod_parser_introduces amod_annotations_latent dobj_introduces_annotations partmod_annotations_ aux_refine_to xcomp__refine amod_categories_syntactic dobj_refine_categories aux_learn_to xcomp_refine_learn amod_grammars_better nn_grammars_PCFG dobj_learn_grammars ' +o,2016,'They constructed word clusters by using HMMs or Browns clustering algorithm , which utilize only information from neighboring words ',Brown,'nsubj_constructed_They nn_clusters_word dobj_constructed_clusters prep_constructed_by pcomp_by_using dobj_using_HMMs cc_HMMs_or conj_HMMs_Browns partmod_HMMs_clustering nn__algorithm dobj_clustering_ nsubj_utilize_which rcmod__utilize amod_information_only dobj_utilize_information prep_utilize_from amod_words_neighboring pobj_from_words ' +o,2017,'An additional translation set called the Maximum BLEU set is employed by the SMT system to train the weights associated with the components of its log-linear model ',Och,'det_set_An amod_set_additional nn_set_translation nsubj_called_set det_set_the amod_set_Maximum nn_set_BLEU nsubjpass_employed_set auxpass_employed_is ccomp_called_employed prep_employed_by det_system_the nn_system_SMT pobj_by_system aux_train_to xcomp_employed_train det_weights_the dobj_train_weights partmod_weights_associated prep_associated_with det_components_the pobj_with_components prep_components_of poss_model_its amod_model_log-linear pobj_of_model ' +o,2018,'In our SRL system , we select maximum entropy as a classi er to implement the semantic role labeling system ',Berger,'prep_select_In poss_system_our nn_system_SRL pobj_In_system nsubj_select_we amod__maximum nn__entropy dobj_select_ prep_select_as det_er_a nn_er_classi pobj_as_er aux_implement_to xcomp_select_implement det_system_the amod_system_semantic nn_system_role nn_system_labeling dobj_implement_system ' +o,2019,'5 Related Work There has not been much previous work on graphical models for full parsing , although recently several latent variable models for parsing have been proposed ',Matsuzaki,'num_Work_5 amod_Work_Related expl_been_There aux_been_has neg_been_not dep_Work_been amod_work_much amod_work_previous dobj_been_work prep_work_on amod_models_graphical pobj_on_models prep_models_for amod_parsing_full pobj_for_parsing mark_proposed_although advmod_proposed_recently amod_models_several amod_models_latent amod_models_variable nsubjpass_proposed_models prep_models_for pobj_for_parsing aux_proposed_have auxpass_proposed_been advcl_been_proposed ' +o,2020,'Although some work has been done on syllabifying orthographic forms , syllables are , technically speaking , phonological entities that can only be composed of strings of phonemes ',Bartlett,'mark_done_Although det_work_some nsubjpass_done_work aux_done_has auxpass_done_been advcl_entities_done prep_done_on pcomp_on_syllabifying amod_forms_orthographic dobj_syllabifying_forms advmod_syllabifying_ nsubj_entities_syllables cop_entities_are advmod_speaking_technically amod_entities_speaking amod_entities_phonological nsubjpass_composed_that aux_composed_can advmod_composed_only auxpass_composed_be rcmod_entities_composed prep_composed_of pobj_of_strings prep_strings_of pobj_of_phonemes ' +o,2021,'Also , on WS-353 , our hybrid sense-filtered variants and word-cos-ll obtained a correlation score higher than published results using WordNet-based measures -LRB- 33 to 35 -RRB- and Wikipediabased methods -LRB- 19 to 48 -RRB- ; and very close to the results obtained by thesaurus-based -LRB- 55 -RRB- and LSA-based methods -LRB- 56 -RRB- ',Ponzetto,'advmod_obtained_Also prep_obtained_on pobj_on_WS-353 poss_variants_our amod_variants_hybrid amod_variants_sense-filtered nsubj_obtained_variants cc_variants_and conj_variants_word-cos-ll det_correlation_a nsubj_score_correlation ccomp_obtained_score dep_score_higher prep_higher_than amod_results_published pobj_than_results partmod_results_using amod__WordNet-based nn__measures dobj_using_ number_35_33 dep_35_to dep__35 cc__and nn__Wikipediabased nn__methods conj__ number_48_19 dep_48_to dep__48 cc_higher_and advmod_close_very conj_higher_close prep_close_to det_results_the pobj_to_results partmod_results_obtained prep_obtained_by amod__thesaurus-based pobj_by_ appos__55 cc__and amod__LSA-based nn__methods conj__ appos__56 ' +o,2022,'Also in the Penn Treebank -LRB- , -RRB- a limited set of relations is placed over the constituencybased annotation in order to make explicit the -LRB- morpho-syntactic or semantic -RRB- roles that the constituents play ',Marcus,'dep_in_Also det_Treebank_the nn_Treebank_Penn pobj_in_Treebank appos_Treebank_ dep__ det_set_a amod_set_limited nsubjpass_placed_set prep_set_of pobj_of_relations auxpass_placed_is rcmod_Treebank_placed prep_placed_over det_annotation_the amod_annotation_constituencybased pobj_over_annotation prep_annotation_in pobj_in_order aux_make_to xcomp_placed_make amod_roles_explicit amod_roles_the dep_the_morpho-syntactic cc_morpho-syntactic_or conj_morpho-syntactic_semantic dobj_make_roles complm_play_that det_constituents_the nsubj_play_constituents ccomp_make_play ' +o,2023,'1 Introduction Syntax-based translation models are usually built directly from Penn Treebank -LRB- PTB -RRB- style parse trees by composing treebank grammar rules ',Marcus,'num__1 nn__Introduction amod__Syntax-based nn__translation nn__models nsubjpass_built_ auxpass_built_are advmod_built_usually advmod_built_directly prep_built_from nn_Treebank_Penn pobj_from_Treebank abbrev_Treebank_PTB xcomp_built_ nn_trees_style nn_trees_parse dobj__trees prep__by pcomp_by_composing nn_rules_treebank nn_rules_grammar dobj_composing_rules ' +o,2024,'As an alternative to linear interpolation , we also employ a weighted product for phrase-table combination : p -LRB- s t -RRB- productdisplay j pj -LRB- s t -RRB- j -LRB- 3 -RRB- This has the same form used for log-linear training of SMT decoders , which allows us to treateachdistributionasafeature , andlearnthemixing weights automatically ',Och,'prep_employ_As det_alternative_an pobj_As_alternative prep_alternative_to amod_interpolation_linear pobj_to_interpolation nsubj_employ_we advmod_employ_also det_product_a amod_product_weighted dobj_employ_product prep_product_for amod_p_phrase-table nn_p_combination pobj_for_p nsubj_t_s dep_product_t amod_j_productdisplay nn_j_j nn_j_pj amod_t_s appos_j_t dep_product_j dep_has_3 nsubj_has_This dep_employ_has det_form_the amod_form_same dobj_has_form partmod_form_used prep_used_for amod_training_log-linear pobj_for_training prep_training_of nn__SMT nn__decoders pobj_of_ nsubj_allows_which rcmod__allows nsubj_treateachdistributionasafeature_us aux_treateachdistributionasafeature_to xcomp_allows_treateachdistributionasafeature xcomp_treateachdistributionasafeature_andlearnthemixing dobj_andlearnthemixing_weights advmod_andlearnthemixing_automatically ' +o,2025,'This second point is emphasized by the second paper on self-training for adaptation ',McClosky,'det_point_This amod_point_second nsubjpass_emphasized_point auxpass_emphasized_is prep_emphasized_by det_paper_the amod_paper_second pobj_by_paper prep_paper_on pobj_on_self-training prep_self-training_for pobj_for_adaptation ' +o,2026,'2 Background Several graph-based learning techniques have recently been developed and applied to NLP problems : minimum cuts , random walks , graph matching , and label propagation ',Pang,'num_techniques_2 amod_techniques_Background amod_techniques_Several amod_techniques_graph-based nn_techniques_learning aux_developed_have advmod_developed_recently auxpass_developed_been dep_techniques_developed cc_developed_and conj_developed_applied prep_developed_to nn_problems_NLP pobj_to_problems amod__minimum nn__cuts dep_techniques_ amod__random amod__walks conj__ nn__graph nn__matching conj__ cc__and nn__label nn__propagation conj__ ' +o,2027,'The main reason behind this lies in the difference between the two corpora used : Penn Treebank and EDR corpus ',Marcus,'det_reason_The amod_reason_main advmod_lies_behind dep_behind_this dep_reason_lies prep_lies_in det_difference_the pobj_in_difference prep_difference_between det_corpora_the num_corpora_two pobj_between_corpora partmod_corpora_used nn__Penn nn__Treebank dep_reason_ cc__and nn__EDR nn__corpus conj__ ' +o,2028,'31 Model-based Phrase Pair Posterior In a statistical generative word alignment model , it is assumed that -LRB- i -RRB- a random variable a specifies how each target word fj is generated by -LRB- therefore aligned to -RRB- a source 1 word eaj ; and -LRB- ii -RRB- the likelihood function f -LRB- f , a e -RRB- specifies a generativeprocedurefromthesourcesentencetothe target sentence ',Brown,'measure_Model-based_31 amod_Posterior_Model-based nn_Posterior_Phrase nn_Posterior_Pair nsubjpass_assumed_Posterior prep_assumed_In det__a amod__statistical nn__generative nn__word nn__alignment nn__model pobj_In_ nsubjpass_assumed_it auxpass_assumed_is complm_specifies_that dep_specifies_i det_variable_a amod_variable_random nsubj_specifies_variable prep_variable_a ccomp_assumed_specifies advmod_generated_how det_fj_each nn_fj_target nn_fj_word nsubjpass_generated_fj auxpass_generated_is ccomp_specifies_generated prep_generated_by advmod_aligned_therefore dep_by_aligned xcomp_aligned_to det_eaj_a nn_eaj_source num_eaj_1 nn_eaj_word pobj_by_eaj cc_specifies_and dep_function_ii det_likelihood_the nsubj_function_likelihood conj_specifies_function nsubj_specifies_f dep_f_f det_e_a appos_f_e ccomp_function_specifies det_sentence_a amod_sentence_generativeprocedurefromthesourcesentencetothe nn_sentence_target dobj_specifies_sentence ' +o,2029,'Unfortunately , as shown in , with the represetation of sentences that we use , linear classifiers can not discriminate real sentences from sentences sampled from a trigram , which is the model we use as a baseline , so here we resort to a non-linear large-margin classifier -LRB- see section 3 for details -RRB- ',Okanohara,'advmod_discriminate_Unfortunately mark_shown_as dep_discriminate_shown prep_shown_in pobj_in_ prep_shown_with det_represetation_the pobj_with_represetation prep_represetation_of pobj_of_sentences complm_use_that nsubj_use_we dep_sentences_use amod_classifiers_linear nsubj_discriminate_classifiers aux_discriminate_can neg_discriminate_not amod_sentences_real dobj_discriminate_sentences prep_discriminate_from pobj_from_sentences partmod_sentences_sampled prep_sampled_from det_trigram_a pobj_from_trigram nsubj_model_which cop_model_is det_model_the rcmod_trigram_model nsubj_use_we rcmod_model_use prep_use_as det_baseline_a pobj_as_baseline dep_use_so advmod_resort_here nsubj_resort_we dep_use_resort prep_resort_to det_classifier_a amod_classifier_non-linear amod_classifier_large-margin pobj_to_classifier dep_classifier_see dobj_see_section num_section_3 prep_section_for pobj_for_details ' +o,2030,'Tuning is done for each experimental condition using Ochs Minimum Error Training ',Och,'nsubjpass_done_Tuning auxpass_done_is prep_done_for det_condition_each amod_condition_experimental pobj_for_condition partmod_condition_using nn_Training_Ochs nn_Training_Minimum nn_Training_Error dobj_using_Training ' +o,2031,'Maximum entropy estimation for translation of individual words dates back to , and the idea of using multi-class classifiers to sharpen predictions normally made through relative frequency estimates has been recently reintroducedundertherubricofwordsensedisambiguation and generalized to substrings ',Berger,'amod_estimation_Maximum nn_estimation_entropy nsubj_dates_estimation prep_estimation_for pobj_for_translation prep_translation_of amod_words_individual pobj_of_words prt_dates_back prep_dates_to pobj_to_ cc_dates_and det_idea_the nsubjpass_reintroducedundertherubricofwordsensedisambiguation_idea prep_idea_of pcomp_of_using amod_classifiers_multi-class dobj_using_classifiers aux_sharpen_to xcomp_using_sharpen dobj_sharpen_predictions advmod_made_normally partmod_predictions_made prep_made_through amod_estimates_relative nn_estimates_frequency pobj_through_estimates aux_reintroducedundertherubricofwordsensedisambiguation_has auxpass_reintroducedundertherubricofwordsensedisambiguation_been advmod_reintroducedundertherubricofwordsensedisambiguation_recently conj_dates_reintroducedundertherubricofwordsensedisambiguation cc_reintroducedundertherubricofwordsensedisambiguation_and conj_reintroducedundertherubricofwordsensedisambiguation_generalized prep_reintroducedundertherubricofwordsensedisambiguation_to amod__substrings pobj_to_ ' +p,2032,'However , to be more expressive and flexible , it is often easier to start with a general SCFG or tree-transducer ',Galley,'advmod_easier_However aux_expressive_to cop_expressive_be advmod_expressive_more dep_easier_expressive cc_expressive_and conj_expressive_flexible nsubj_easier_it cop_easier_is advmod_easier_often aux_start_to xcomp_easier_start prep_start_with det_SCFG_a amod_SCFG_general pobj_with_SCFG cc_SCFG_or conj_SCFG_tree-transducer ' +o,2033,'Experimental results are reported in Table 2 : here cased BLEU results are reported on MT03 Arabic-English test set ',Papineni,'amod_results_Experimental nsubjpass_reported_results auxpass_reported_are prep_reported_in pobj_in_Table num_Table_2 advmod_cased_here parataxis_reported_cased amod_results_BLEU nsubjpass_reported_results auxpass_reported_are ccomp_cased_reported prep_reported_on nn_test_MT03 amod_test_Arabic-English pobj_on_test partmod_test_set ' +o,2034,'For this study, the Levenshtein edit-distance score (where a perfect match scores zero) is Roman Chinese (Pinyin) Alignment Score LEV ashburton ashenbodu | a s h b u r t o n | | a s h e n b o d u | 0.67 MLEV ashburton ashenbodu | a s h b u r t o n | | a s h e n b o d u | 0.72 MALINE asVburton aseCnpotu | a sV b < u r t o | n | a s eC n p o t u | 0.48 3 normalized to a similarity score as in (Freeman et al. 2006), where the score ranges from 0 to 1, with 1 being a perfect match.',Freeman,'' +o,2035,'To find the optimal coefficients for a loglinear combination of these experts , we use separate development data , using the following procedure due to : 1 ',Och,'aux_find_To dep_use_find det_coefficients_the amod_coefficients_optimal dobj_find_coefficients prep_find_for det_combination_a nn_combination_loglinear pobj_for_combination prep_combination_of det_experts_these pobj_of_experts nsubj_use_we amod_data_separate nn_data_development dobj_use_data xcomp_use_using det_procedure_the amod_procedure_following dobj_using_procedure amod_procedure_due aux__to xcomp_due_ dobj__1 ' +o,2036,'Discriminatively trained parsers that score entire trees for a given sentence have only recently been investigated ',Collins,'advmod_trained_Discriminatively amod_parsers_trained nsubjpass_investigated_parsers nsubj_score_that rcmod_parsers_score amod_trees_entire dobj_score_trees prep_trees_for det_sentence_a amod_sentence_given pobj_for_sentence aux_investigated_have advmod_recently_only advmod_investigated_recently auxpass_investigated_been ' +o,2037,'have been proposed ',Brown,'aux_proposed_have auxpass_proposed_been ' +n,2038,'Table 2 : Figures about clustering algorithms Algorithm # Sentences \\/ # Clusters S-HAC 6,23 C-HAC 2,17 QT 2,32 EM 4,16 In fact , table 2 shows that most of the clusters have less than 6 sentences which leads to question the results presented by who only keep the clusters that contain more than 10 sentences ',Barzilay,'num_Table_2 nsubj_Algorithm_Figures prep_Figures_about amod_algorithms_clustering pobj_about_algorithms dep_Table_Algorithm dep_\\/_# number_\\/_Sentences dobj_Algorithm_\\/ dep_S-HAC_# number_S-HAC_Clusters num_C-HAC_S-HAC number_C-HAC_6,23 prep_\\/_C-HAC num_QT_2,17 dep_C-HAC_QT num_QT_2,32 nsubj_shows_EM num_EM_4,16 prep_EM_In pobj_In_fact appos_fact_table num_table_2 rcmod_QT_shows complm_have_that nsubj_have_most prep_most_of det_clusters_the pobj_of_clusters ccomp_shows_have dep_than_less quantmod_6_than num_sentences_6 dobj_have_sentences nsubj_leads_which rcmod_sentences_leads aux_question_to xcomp_leads_question det_results_the dobj_question_results partmod_results_presented prep_presented_by pobj_by_ nsubj_keep_who advmod_keep_only rcmod_results_keep det_clusters_the dobj_keep_clusters nsubj_contain_that rcmod_clusters_contain dep_than_more quantmod_10_than num_sentences_10 dobj_contain_sentences ' +o,2039,'1 Introduction NLP researchers have developed many algorithms for mining knowledge from text and the Web , including facts , semantic lexicons , concept lists , and word similarity lists ',Hindle,'num_researchers_1 nn_researchers_Introduction nn_researchers_NLP nsubj_developed_researchers aux_developed_have amod_algorithms_many dobj_developed_algorithms prep_algorithms_for nn_knowledge_mining pobj_for_knowledge prep_developed_from pobj_from_text cc_text_and det_Web_the conj_text_Web prep_developed_including nn__facts pobj_including_ amod__semantic nn__lexicons conj__ nn__concept nn__lists conj__ cc__and nn_lists_word nn_lists_similarity conj__lists ' +o,2040,'The later IBM models are formulated to prefer collocations ',Brown,'det_models_The amod_models_later nn_models_IBM nsubjpass_formulated_models auxpass_formulated_are aux_prefer_to xcomp_formulated_prefer dobj_prefer_collocations ' +o,2041,'Since the lexical translations and dependency paths are typically not labeled in the English corpus , a given pair must be counted fractionally according to its posterior probability of satisfying these conditions , given models of contextual translation and English parsing3 3Similarly , imputes missing trees by using comparable corpora ',Jansche,'mark_labeled_Since det_paths_the amod_paths_lexical amod_paths_translations cc_translations_and conj_translations_dependency nsubjpass_labeled_paths auxpass_labeled_are advmod_labeled_typically neg_labeled_not advcl_imputes_labeled prep_labeled_in det_corpus_the amod_corpus_English pobj_in_corpus det_pair_a amod_pair_given nsubjpass_counted_pair aux_counted_must auxpass_counted_be parataxis_imputes_counted advmod_counted_fractionally prep_counted_according dep_according_to poss_probability_its nn_probability_posterior pobj_to_probability prep_probability_of pcomp_of_satisfying det_conditions_these dobj_satisfying_conditions prep_imputes_given dep_given_models prep_models_of amod_translation_contextual pobj_of_translation cc_translation_and amod_3Similarly_English nn_3Similarly_parsing3 conj_translation_3Similarly nsubj_imputes_ amod_trees_missing dobj_imputes_trees prep_imputes_by pcomp_by_using amod_corpora_comparable dobj_using_corpora ' +p,2042,'This is a common technique in machine translation for which the IBM translation models are popular methods ',Brown,'nsubj_technique_This cop_technique_is det_technique_a amod_technique_common prep_technique_in nn_translation_machine pobj_in_translation rel_methods_for pobj_for_which det_models_the nn_models_IBM nn_models_translation nsubj_methods_models cop_methods_are amod_methods_popular rcmod_technique_methods ' +o,2043,'While this heuristic estimator gives good empirical results , it does not seem to optimize any intuitively reasonable objective function of the -LRB- wordaligned -RRB- parallel corpus -LRB- see eg , -RRB- The mounting number of efforts attacking this problem over the last few years exhibits its difficulty ',Zhang,'mark_gives_While det_estimator_this amod_estimator_heuristic nsubj_gives_estimator advcl_number_gives amod_results_good amod_results_empirical dobj_gives_results nsubj_number_it aux_number_does neg_number_not cop_number_seem aux_optimize_to xcomp_number_optimize det_function_any advmod_reasonable_intuitively amod_function_reasonable nn_function_objective dobj_optimize_function prep_function_of det_corpus_the appos_corpus_wordaligned amod_corpus_parallel pobj_of_corpus parataxis_number_see dobj_see_eg dep_see_ det_number_The amod_number_mounting dep_exhibits_number prep_number_of pobj_of_efforts partmod_efforts_attacking det_problem_this dobj_attacking_problem prep_number_over det__the amod__last amod__few nn__years pobj_over_ poss_difficulty_its nsubj_exhibits_difficulty ' +o,2044,'Other work aims to do truly unsupervised learning of taggers , such as Goldwater and Griffiths and ',Johnson,'amod_work_Other nsubj_aims_work aux_do_to xcomp_aims_do advmod_unsupervised_truly amod_learning_unsupervised dobj_do_learning prep_learning_of pobj_of_taggers dep_as_such prep_learning_as nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths pobj_as_ cc__and nn__ conj__ ' +o,2045,'The LFG annotation algorithm of was used to produce the f-structures for development , test and training sets ',Cahill,'det_algorithm_The nn_algorithm_LFG nn_algorithm_annotation nsubjpass_used_algorithm prep_algorithm_of pobj_of_ auxpass_used_was aux_produce_to xcomp_used_produce det_f-structures_the dobj_produce_f-structures prep_produce_for nn_sets_development conj_development_test cc_development_and conj_development_training pobj_for_sets ' +o,2046,'Decoding is carried-out using the Moses decoder ',Koehn,'nsubj_carried-out_Decoding cop_carried-out_is xcomp_carried-out_using det_decoder_the nn_decoder_Moses dobj_using_decoder ' +o,2047,'Statistical techniques , both supervised learning from tagged corpora , , and unsupervised learning , , have been investigated ',Yarowsky,'nn_techniques_Statistical nsubj__techniques det_learning_both amod_learning_supervised appos_techniques_learning prep_learning_from amod__tagged nn__corpora pobj_from_ cc__and conj__unsupervised xcomp_unsupervised_learning nsubjpass_investigated_ appos__ aux_investigated_have auxpass_investigated_been ccomp_learning_investigated ' +o,2048,'A variety of approaches have been investigated for speech summarization , for example , maximum entropy , conditional random fields , latent semantic analysis , support vector machines , maximum marginal relevance ',Galley,'det_variety_A nsubjpass_investigated_variety prep_variety_of pobj_of_approaches aux_investigated_have auxpass_investigated_been prep_investigated_for nn_summarization_speech pobj_for_summarization prep_investigated_for pobj_for_example amod_entropy_maximum appos_example_entropy amod_fields_conditional amod_fields_random appos_example_fields amod_analysis_latent amod_analysis_semantic appos_example_analysis nn_machines_support nn_machines_vector appos_example_machines amod_relevance_maximum amod_relevance_marginal appos_example_relevance ' +o,2049,' The heuristics in Section 6 are designed specifically to find the interesting features in that featureless desert ',Brown,'det_heuristics_The nsubjpass_designed_heuristics prep_heuristics_in pobj_in_Section num_Section_6 auxpass_designed_are ccomp__designed advmod_designed_specifically aux_find_to xcomp_designed_find det_features_the amod_features_interesting dobj_find_features prep_features_in det_desert_that amod_desert_featureless pobj_in_desert ' +o,2050,'During the SRC stage, a Maximum entropy (Berger et al., 1996) classifier is used to predict the probabilities of a word in the sentence Language No-duplicated-roles Catalan arg0-agt, arg0-cau, arg1-pat, arg2-atr, arg2-loc Chinese A0, A1, A2, A3, A4, A5, Czech ACT, ADDR, CRIT, LOC, PAT, DIR3, COND English A0, A1, A2, A3, A4, A5, German A0, A1, A2, A3, A4, A5, Japanese DE, GA, TMP, WO Spanish arg0-agt, arg0-cau, arg1-pat, arg1-tem, arg2-atr, arg2-loc, arg2-null, arg4-des, argL-null, argMcau, argM-ext, argM-fin Table 1: No-duplicated-roles for different languages to be each semantic role.',Berger,'' +o,2051,' proposed a method to identify discourse relations between text segments using Nave Bayes classifiers trained on a huge corpus ',Marcu,'nsubj_proposed_ det_method_a dobj_proposed_method aux_identify_to infmod_method_identify nn_relations_discourse dobj_identify_relations prep_relations_between nn_segments_text pobj_between_segments xcomp_identify_using nn_classifiers_Nave nn_classifiers_Bayes nsubj_trained_classifiers ccomp_using_trained prep_trained_on det_corpus_a amod_corpus_huge pobj_on_corpus ' +o,2052,'Note that the predicate language representation utilized by Carmel-Tools is in the style of Davidsonian event based semantics ',Hobbs,'complm_is_that det_representation_the amod_representation_predicate nn_representation_language nsubj_is_representation partmod_representation_utilized prep_utilized_by pobj_by_Carmel-Tools ccomp_Note_is prep_is_in det_style_the pobj_in_style prep_style_of amod_event_Davidsonian pobj_of_event partmod_event_based dobj_based_semantics ' +o,2053,'Alignment performance is measured by the Alignment Error Rate -LRB- AER -RRB- AER -LRB- B ; B -RRB- = 12 B B \\/ -LRB- B + B -RRB- where B is a set reference word links , and B are the word links generated automatically ',Och,'amod_performance_Alignment nsubjpass_measured_performance auxpass_measured_is prep_measured_by det_Rate_the nn_Rate_Alignment nn_Rate_Error pobj_by_Rate abbrev_Rate_AER nn_AER_ nsubj_=_AER nn_B_B appos_AER_B rcmod_Rate_= num_\\/_12 nn_\\/_B nn_\\/_B dobj_=_\\/ nn_B_B nn_B_+ appos_\\/_B advmod_links_where nsubj_links_B dep_links_is det_links_a amod_links_set nn_links_reference nn_links_word rcmod_\\/_links cc_measured_and nsubj_links_B cop_links_are det_links_the nn_links_word conj_measured_links partmod_links_generated advmod_generated_automatically ' +o,2054,'43 Corpora The evaluations of the different models were carried out on the Penn Wall Street Journal corpus for English , and the Tiger treebank for German ',Marcus,'num_evaluations_43 nn_evaluations_Corpora nn_evaluations_The nsubjpass_carried_evaluations prep_evaluations_of det_models_the amod_models_different pobj_of_models auxpass_carried_were prt_carried_out prep_carried_on det__the nn__Penn nn__Wall nn__Street nn__Journal nn__corpus pobj_on_ prep__for pobj_for_English cc__and det__the nn__Tiger nn__treebank conj__ prep__for pobj_for_German ' +o,2055,'Open-domain opinion extraction is another trend of research on opinion extraction , which aims to extract a wider range of opinions from such texts as newspaper articles ',Kim,'amod_extraction_Open-domain nn_extraction_opinion nsubj_trend_extraction cop_trend_is det_trend_another prep_trend_of pobj_of_research prep_research_on nn_extraction_opinion pobj_on_extraction nsubj_aims_which rcmod_extraction_aims aux_extract_to xcomp_aims_extract det_range_a amod_range_wider dobj_extract_range prep_range_of pobj_of_opinions prep_extract_from amod_texts_such pobj_from_texts prep_extract_as nn_articles_newspaper pobj_as_articles ' +o,2056,'Alignment , whether for training a translation model using EM or for nding the Viterbi alignment of test data , is O -LRB- n6 -RRB- , while translation -LRB- decoding -RRB- is O -LRB- n7 -RRB- using a bigram language model , and O -LRB- n11 -RRB- with trigrams ',Wu,'nsubj__Alignment appos_Alignment_whether prep_whether_for pcomp_for_training det_model_a nn_model_translation dobj_training_model xcomp_training_using dobj_using_EM cc_for_or conj_for_for pcomp_for_nding det_alignment_the nn_alignment_Viterbi dobj_nding_alignment prep_alignment_of nn_data_test pobj_of_data cop__is nn__O appos__n6 mark_n7_while nsubj_n7_translation appos_translation_decoding cop_n7_is nn_n7_O dep__n7 partmod_n7_using det_model_a nn_model_bigram nn_model_language dobj_using_model cc_n7_and conj_n7_O appos_O_n11 prep_O_with pobj_with_trigrams ' +n,2057,'These methods go beyond the original IBM machine translation models , by allowing multi-word units -LRB- phrases -RRB- in one language to be translated directly into phrases in another language ',Brown,'det_methods_These nsubj_go_methods prep_go_beyond det__the amod__original nn__IBM nn__machine nn__translation nn__models pobj_beyond_ prep_go_by pcomp_by_allowing amod_units_multi-word dobj_allowing_units appos_units_phrases prep_units_in num_language_one pobj_in_language aux_translated_to auxpass_translated_be xcomp_go_translated advmod_translated_directly prep_translated_into pobj_into_phrases prep_phrases_in det_language_another pobj_in_language ' +o,2058,'The data sets used are the standard data sets for this problem taken from the Wall Street Journal corpus in the Penn Treebank ',Ramshaw,'det_data_The nsubj_sets_data csubj_sets_used cop_sets_are det_sets_the amod_sets_standard nn_sets_data ccomp_sets_sets prep_sets_for det__this nn__problem pobj_for_ partmod__taken prep_taken_from det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_from_corpus prep_corpus_in det__the nn__Penn nn__Treebank pobj_in_ ' +o,2059,'2 Phrase-based SMT We use a phrase-based SMT system , Pharaoh , , which is based on a log-linear formulation ',Koehn,'num_SMT_2 amod_SMT_Phrase-based nsubj_use_We dep_SMT_use det_system_a amod_system_phrase-based nn_system_SMT dobj_use_system appos_system_Pharaoh appos_system_ nsubjpass_based_which auxpass_based_is rcmod_system_based prep_based_on det__a amod__log-linear nn__formulation pobj_on_ ' +o,2060,' described symmetrized training of a 1-toN log-linear model and a M-to-1 log-linear model ',Fraser,'nsubj_described_ amod_training_symmetrized dobj_described_training prep_training_of det_model_a amod_model_1-toN amod_model_log-linear pobj_of_model cc_model_and det_model_a amod_model_M-to-1 amod_model_log-linear conj_model_model ' +o,2061,'Model parameters are estimated using maximum entropy ',Berger,'nn_parameters_Model nsubjpass_estimated_parameters auxpass_estimated_are xcomp_estimated_using amod_entropy_maximum dobj_using_entropy ' +o,2062,'We use Viterbi training but neighborhood estimation or pegging could also be used ',Och,'nsubj_use_We nn__Viterbi nn__training dobj_use_ cc_use_but nn__neighborhood nn__estimation nsubjpass_used_ cc__or nn__pegging conj__ aux_used_could advmod_used_also auxpass_used_be conj_use_used ' +o,2063,'It also contains tools for tuning these models using minimum error rate training and evaluating the resulting translations using the BLEU score ',Och,'nsubj_contains_It advmod_contains_also dobj_contains_tools prep_contains_for pcomp_for_tuning det_models_these dobj_tuning_models xcomp_tuning_using amod__minimum nn__error nn__rate nn__training dobj_using_ cc_tuning_and conj_tuning_evaluating det_translations_the amod_translations_resulting dobj_evaluating_translations xcomp_evaluating_using det__the amod__BLEU nn__score dobj_using_ ' +o,2064,'2 Problem Setting In the multi-class setting , instances from an input spaceX take labels from a finite setY , Y = K 496 We use a standard approach for generalizing binary classification and assume a feature function f -LRB- x , y -RRB- Rd mapping instances xX and labels yY into a common space ',Collins,'dep_spaceX_2 nn_Setting_Problem dep_In_Setting prep_spaceX_In det_setting_the amod_setting_multi-class pobj_In_setting nsubj_spaceX_instances prep_instances_from det_input_an pobj_from_input dep_spaceX_take dobj_take_labels prep_take_from det_setY_a amod_setY_finite pobj_from_setY advmod_=_Y conj_take_= dobj_=_K num_K_496 nsubj_use_We dep_=_use det__a amod__standard nn__approach dobj_use_ prep_use_for pcomp_for_generalizing amod_classification_binary dobj_generalizing_classification cc_take_and conj_take_assume det_function_a nn_function_feature dobj_assume_function dobj_xX_f dep_f_x appos_x_y nn_instances_Rd nn_instances_mapping nsubj_xX_instances rcmod_function_xX cc_xX_and conj_xX_labels dobj_labels_yY prep_labels_into det_space_a amod_space_common pobj_into_space ' +o,2065,'First , the Wikipedia gazetteer improved the accuracy as expected , ie , it reproduced the result of for Japanese NER ',Kazama,'advmod_improved_First det_gazetteer_the nn_gazetteer_Wikipedia nsubj_improved_gazetteer det_accuracy_the dobj_improved_accuracy prep_improved_as pobj_as_expected appos_expected_ie nsubj_reproduced_it dep_improved_reproduced det_result_the dobj_reproduced_result prep_result_of pobj_of_ prep_reproduced_for amod_NER_Japanese pobj_for_NER ' +o,2066,'At this point , one can imagine estimating a linear matching model in multiple ways , including using conditional likelihood estimation , an averaged perceptron update -LRB- see which matchings are proposed and adjust the weights according to the dierence between the guessed and target structures -RRB- , or in large-margin fashion ',Collins,'prep_imagine_At det_point_this pobj_At_point nsubj_imagine_one aux_imagine_can dep_imagine_estimating det_model_a amod_model_linear amod_model_matching dobj_estimating_model prep_model_in amod_ways_multiple pobj_in_ways prep_estimating_including pcomp_including_using amod_estimation_conditional nn_estimation_likelihood dobj_using_estimation det_update_an amod_update_averaged nn_update_perceptron appos_estimation_update dep_estimation_see dobj_proposed_which nsubjpass_proposed_matchings auxpass_proposed_are ccomp_see_proposed cc_see_and conj_see_adjust det_weights_the dobj_adjust_weights prep_adjust_according dep_according_to det_dierence_the pobj_to_dierence prep_dierence_between det_guessed_the pobj_between_guessed cc_guessed_and nn__target nn__structures conj_guessed_ cc_estimating_or conj_estimating_in amod_fashion_large-margin pobj_in_fashion ' +o,2067,'For example , both papers propose minimum-risk decoding , and McDonald and Satta discuss unsupervised learning and language modeling , while and define hiddenvariable models based on spanning trees ',Smith,'prep_propose_For pobj_For_example det_papers_both nsubj_propose_papers amod_decoding_minimum-risk dobj_propose_decoding cc_propose_and nsubj__McDonald cc_McDonald_and conj_McDonald_Satta conj_propose_ ccomp__discuss amod_learning_unsupervised dobj_discuss_learning cc_learning_and nn_modeling_language conj_learning_modeling mark_define_while nn__ cc__and conj__ nsubj_define_ advcl__define amod_models_hiddenvariable dobj_define_models prep_define_based dep_based_on pcomp_on_spanning dobj_spanning_trees ' +o,2068,'For example , only requires sense number and a few seeds for each sense of an ambiguous word -LRB- hereafter called keyword -RRB- ',Yarowsky,'prep_requires_For pobj_For_example nsubj_requires_ advmod_requires_only nn_number_sense dobj_requires_number cc_number_and det_seeds_a amod_seeds_few conj_number_seeds prep_requires_for det_sense_each pobj_for_sense prep_sense_of det_word_an amod_word_ambiguous pobj_of_word advmod_called_hereafter dep_word_called dep_called_keyword ' +o,2069,'2 Background : Overview of BLEU This section briefly describes the original BLEU 1 , which was designed for English translation evaluation , so English sentences are used as examples in this section ',Papineni,'num_Background_2 nsubj_describes_Overview prep_Overview_of pcomp_of_BLEU det_section_This dobj_BLEU_section advmod_describes_briefly dep_Background_describes det_BLEU_the amod_BLEU_original dobj_describes_BLEU number_1_ num_BLEU_1 nsubjpass_designed_which auxpass_designed_was rcmod_BLEU_designed prep_designed_for amod_evaluation_English nn_evaluation_translation pobj_for_evaluation dep_describes_so amod_sentences_English nsubjpass_used_sentences auxpass_used_are dep_describes_used prep_used_as pobj_as_examples prep_examples_in det_section_this pobj_in_section ' +o,2070,'Therefore , to make the phrase-based SMT system robust against data sparseness for the ranking task , we also make use of the IBM Model 4 in both directions ',Brown,'advmod_make_Therefore aux_make_to parataxis_make_make det_system_the amod_system_phrase-based nn_system_SMT nsubj_robust_system xcomp_make_robust prep_robust_against nn_sparseness_data pobj_against_sparseness prep_sparseness_for det_task_the amod_task_ranking pobj_for_task nsubj_make_we advmod_make_also dobj_make_use prep_use_of det__the nn__IBM nn__Model num__4 pobj_of_ prep__in det_directions_both pobj_in_directions ' +o,2071,'Then the alignments are symmetrized using a refined heuristic as described in ',Och,'advmod_alignments_Then det_alignments_the nsubjpass_symmetrized_alignments auxpass_symmetrized_are xcomp_symmetrized_using det_heuristic_a amod_heuristic_refined dobj_using_heuristic advmod_described_as amod_heuristic_described prep_using_in ' +o,2072,'To measure the coherence of sentences , we use a statistical parser Toolkit to assign each sentence a parsers score that is the related log probability of parsing ',Collins,'aux_measure_To dep_use_measure det_coherence_the dobj_measure_coherence prep_coherence_of pobj_of_sentences nsubj_use_we det__a amod__statistical nn__parser nn__Toolkit nsubj_assign_ aux_assign_to xcomp_use_assign det_sentence_each dobj_assign_sentence det_score_a amod_score_parsers tmod_assign_score nsubj_probability_that cop_probability_is det_probability_the amod_probability_related nn_probability_log dep_score_probability prep_probability_of pobj_of_parsing ' +o,2073,'For comparison , we also implemented a different N-best phrase alignment method , where _ _ _ _ the_light_was_red _ _ _ the_light was_red _ _ the_light was red -LRB- 1 -RRB- -LRB- 2 -RRB- -LRB- 3 -RRB- Figure 4 : N-best phrase alignments phrase pairs are extracted using the standard phrase extraction method described in ',Koehn,'prep_implemented_For pobj_For_comparison nsubj_implemented_we advmod_implemented_also det_method_a amod_method_different amod_method_N-best nn_method_phrase nn_method_alignment dobj_implemented_method advmod_the_light_was_red_where nn____ nn____ nn____ nsubj_the_light_was_red__ rcmod_method_the_light_was_red nn_the_light__ nn_the_light__ nn_the_light__ nn_the_light_the_light nn_the_light_was_red nn_the_light__ nn_the_light__ nsubj_red_the_light cop_red_was ccomp_the_light_was_red_red dep_red_1 appos_method_2 dep_described_3 nsubj_described_Figure num_Figure_4 amod_alignments_N-best nn_alignments_phrase dep_Figure_alignments nn_pairs_phrase nsubjpass_extracted_pairs auxpass_extracted_are rcmod_alignments_extracted xcomp_extracted_using det_method_the amod_method_standard nn_method_phrase nn_method_extraction dobj_using_method dep_implemented_described prt_described_in ' +o,2074,'33 Unknown word features Most of the models presented here use a set of unknown word features basically inherited from , which include using character n-gram prefixes and suffixes -LRB- for n up to 4 -RRB- , and detectors for a few other prominent features of words , such as capitalization , hyphens , and numbers ',Ratnaparkhi,'dep_Unknown_33 amod_features_Unknown nn_features_word nsubj_use_Most prep_Most_of det_models_the pobj_of_models partmod_models_presented advmod_presented_here dep_features_use det_set_a dobj_use_set prep_set_of amod_features_unknown nn_features_word pobj_of_features advmod_inherited_basically partmod_features_inherited prep_inherited_from pobj_from_ nsubj_include_which rcmod__include xcomp_include_using nn_prefixes_character amod_prefixes_n-gram dobj_using_prefixes cc_prefixes_and conj_prefixes_suffixes dep_prefixes_for pobj_for_n advmod_n_up dep_up_to pobj_to_4 cc_prefixes_and conj_prefixes_detectors prep_detectors_for det_features_a amod_features_few amod_features_other amod_features_prominent pobj_for_features prep_features_of pobj_of_words dep_as_such prep_prefixes_as pobj_as_capitalization conj_capitalization_hyphens cc_capitalization_and conj_capitalization_numbers ' +o,2075,'Concept similarity is often measured by vectors of co-occurrence with context words that are typed with dependency information ',Lin,'amod_similarity_Concept nsubjpass_measured_similarity auxpass_measured_is advmod_measured_often prep_measured_by pobj_by_vectors prep_vectors_of pobj_of_co-occurrence prep_measured_with nn_words_context pobj_with_words nsubjpass_typed_that auxpass_typed_are rcmod_words_typed prep_typed_with amod_information_dependency pobj_with_information ' +o,2076,'Metrics based on word alignment between MT outputs and the references ',Banerjee,'nsubj_based_Metrics prep_based_on nn_alignment_word pobj_on_alignment prep_alignment_between nn_outputs_MT pobj_between_outputs cc_outputs_and det_references_the conj_outputs_references ' +o,2077,'This approach to term clustering is closely related to others from the literature 2 Recall that the mutual information between random variables a0 and a1 can be written : a2a4a3a6a5a8a7a10a9a11a13a12a15a14a17a16a19a18a21a20a23a22a25a24a27a26a29a28 a14a17a16a19a18a21a20a23a22a25a24 a14a17a16a19a18a30a24a31a14a17a16a19a22a32a24 -LRB- 1 -RRB- Here , a0 and a1 correspond to term and context clusters , respectively , each event a18 and a22 the observation of some term and contextual term in the corpus ',Brown,'det_approach_This nsubjpass_related_approach prep_approach_to nn_clustering_term pobj_to_clustering auxpass_related_is advmod_related_closely prep_related_to pobj_to_others prep_others_from det_literature_the pobj_from_literature partmod_literature_ number_Recall_2 dobj__Recall complm_written_that det_information_the amod_information_mutual nsubjpass_written_information prep_information_between amod_a0_random nn_a0_variables pobj_between_a0 cc_a0_and conj_a0_a1 aux_written_can auxpass_written_be ccomp__written amod_a14a17a16a19a18a30a24a31a14a17a16a19a22a32a24_a2a4a3a6a5a8a7a10a9a11a13a12a15a14a17a16a19a18a21a20a23a22a25a24a27a26a29a28 amod_a14a17a16a19a18a30a24a31a14a17a16a19a22a32a24_a14a17a16a19a18a21a20a23a22a25a24 dobj__a14a17a16a19a18a30a24a31a14a17a16a19a22a32a24 appos_a14a17a16a19a18a30a24a31a14a17a16a19a22a32a24_1 advmod_correspond_Here nsubj_correspond_a0 cc_a0_and conj_a0_a1 dep_related_correspond prep_correspond_to nn_clusters_term cc_term_and conj_term_context pobj_to_clusters advmod_correspond_respectively det_a18_each nn_a18_event nsubj_observation_a18 cc_a18_and conj_a18_a22 det_observation_the xcomp_correspond_observation prep_observation_of det_term_some pobj_of_term cc_correspond_and conj_correspond_contextual dobj_contextual_term prep_contextual_in det_corpus_the pobj_in_corpus ' +o,2078,' -RRB- , the tagger for grammatical functions works with lexical -LRB- 1 -RRB- Selbst besucht ADV VVPP himself visited hat Peter Sabine VAFIN NE NE has Peter Sabine ` Peter never visited Sabine himself \' l hie ADV never Figure 2 : Example sentence and contextual probability measures PO -LRB- \' -RRB- depending on the category of a mother node -LRB- Q -RRB- ',Brown,'dep_VVPP_ amod_tagger_the dobj__tagger prep_tagger_for amod_works_grammatical nn_works_functions pobj_for_works prep__with pobj_with_lexical appos_lexical_1 nn_ADV_Selbst nn_ADV_besucht nsubj_VVPP_ADV nsubj_visited_himself ccomp_VVPP_visited nn_NE_hat nn_NE_Peter nn_NE_Sabine nn_NE_VAFIN nn_NE_NE nsubj_has_NE ccomp_visited_has nn_Sabine_Peter dobj_has_Sabine nsubj_visited_Peter neg_visited_never rcmod_Sabine_visited dobj_visited_Sabine dep_hie_himself amod_hie_l nsubj_ADV_hie rcmod_Sabine_ADV advmod_Figure_never dobj_ADV_Figure num_Figure_2 amod_sentence_Example dep_Figure_sentence cc_sentence_and amod_measures_contextual nn_measures_probability conj_sentence_measures dep_sentence_PO dep_VVPP_depending prep_depending_on det_category_the pobj_on_category prep_category_of det_node_a nn_node_mother pobj_of_node appos_node_Q ' +o,2079,' , Liang et al ',Tillmann,'appos__Liang cc_Liang_et conj_Liang_al ' +o,2080,'First , we considered single sentences as documents , and tokens as sentences -LRB- we define a token as a sequence of characters delimited by 1In our case , the score we seek to globally maximize by dynamic programming is not only taking into account the length criteria described in but also a cognate-based one similar to ',Brown,'advmod_considered_First nsubj_considered_we amod_sentences_single dobj_considered_sentences prep_considered_as pobj_as_documents cc_considered_and conj_considered_tokens mark_taking_as nsubj_taking_sentences nsubj_define_we rcmod_sentences_define det_token_a dobj_define_token prep_define_as det_sequence_a pobj_as_sequence prep_sequence_of pobj_of_characters partmod_characters_delimited prep_delimited_by nn_case_1In poss_case_our pobj_by_case det_score_the appos_case_score nsubj_seek_we rcmod_score_seek aux_maximize_to advmod_maximize_globally xcomp_seek_maximize prep_maximize_by amod_programming_dynamic pobj_by_programming aux_taking_is neg_taking_not advmod_taking_only advcl_tokens_taking prep_taking_into pobj_into_account det_criteria_the nn_criteria_length dobj_taking_criteria partmod_criteria_described prep_described_in pobj_in_ dep_also_but cc_criteria_also det_one_a amod_one_cognate-based conj_criteria_one amod_one_similar dep_similar_to dep_similar_ ' +o,2081,'Then the initial precision is 1 , citing , actually uses a superficially different score that is , however , a monotone transform of precision , hence equivalent to precision , since it is used only for sorting ',Yarowsky,'advmod_is_Then det_precision_the amod_precision_initial nsubj_is_precision num__1 nsubj_score_ partmod__citing dobj_citing_ advmod_score_actually dep_score_uses det_score_a advmod_different_superficially amod_score_different dep_is_score nsubj_transform_that cop_transform_is advmod_transform_however det_transform_a amod_transform_monotone rcmod_score_transform prep_transform_of pobj_of_precision advmod_score_hence dep_score_equivalent prep_equivalent_to pobj_to_precision mark_used_since nsubjpass_used_it auxpass_used_is advcl_score_used advmod_used_only prep_used_for pobj_for_sorting ' +o,2082,'The use of Profile HMMs for multiple sequence alignment also presents applications to the acquisition of mapping dictionaries and sentence-level paraphrasing ',Barzilay,'det_use_The nsubj_presents_use prep_use_of nn_HMMs_Profile pobj_of_HMMs prep_HMMs_for amod_alignment_multiple nn_alignment_sequence pobj_for_alignment advmod_presents_also dobj_presents_applications prep_presents_to det_acquisition_the pobj_to_acquisition prep_acquisition_of nn__mapping nn__dictionaries pobj_of_ cc__and amod_paraphrasing_sentence-level conj__paraphrasing ' +o,2083,'Our approach is related to those of and Taskar et al ',Collins,'poss_approach_Our nsubjpass_related_approach auxpass_related_is prep_related_to pobj_to_those prep_those_of pobj_of_ cc__and conj__Taskar cc_Taskar_et conj_Taskar_al ' +o,2084,'The pipeline extracts a Hiero-style synchronous context-free grammar , employs suffix-array based rule extraction , and tunes model parameters with minimum error rate training ',Och,'det_extracts_The nn_extracts_pipeline det_grammar_a amod_grammar_Hiero-style amod_grammar_synchronous amod_grammar_context-free nsubj__grammar dep_extracts_ conj__employs dep_based_suffix-array prep_employs_based nn__rule nn__extraction dep_based_ cc__and conj__tunes nn_parameters_model dobj_tunes_parameters prep_tunes_with amod_training_minimum nn_training_error nn_training_rate pobj_with_training ' +o,2085,'Therefore , the Viterbi alignment is comlmted only approximately using the method described in ',Brown,'advmod_comlmted_Therefore det_alignment_the nn_alignment_Viterbi nsubjpass_comlmted_alignment auxpass_comlmted_is quantmod_approximately_only dobj_comlmted_approximately xcomp_comlmted_using det_method_the dobj_using_method partmod_method_described prep_described_in ' +o,2086,'The most important tree-bank transformation in the literature is lexicalization : Each node in a tree is labeled with its head word , the most important word of the constituent under the node -LRB- Magerman , , Charniak -LRB- 1997 -RRB- , Collins -LRB- 1997 -RRB- , Carroll and Rooth -LRB- 1998 -RRB- , etc -RRB- ',Collins,'det_transformation_The advmod_important_most amod_transformation_important amod_transformation_tree-bank nsubj_lexicalization_transformation prep_transformation_in det_literature_the pobj_in_literature cop_lexicalization_is det_node_Each nsubjpass_labeled_node prep_node_in det_tree_a pobj_in_tree auxpass_labeled_is parataxis_lexicalization_labeled prep_labeled_with poss_word_its nn_word_head pobj_with_word det_word_the advmod_important_most amod_word_important appos_word_word prep_word_of det_constituent_the pobj_of_constituent prep_labeled_under det_node_the pobj_under_node dep_node_Magerman num_Magerman_ num_Magerman_ appos_Magerman_Charniak appos_Charniak_1997 appos_Charniak_Collins appos_Collins_1997 appos_Charniak_Carroll cc_Carroll_and conj_Carroll_Rooth appos_Carroll_1998 appos_Magerman_etc ' +o,2087,'Even a length limit of 3 , as proposed by , would result in almost optimal translation quality ',Koehn,'advmod_limit_Even det_limit_a nn_limit_length nsubj_result_limit prep_limit_of pobj_of_3 mark_proposed_as dep_limit_proposed prep_proposed_by pobj_by_ aux_result_would prep_result_in advmod_optimal_almost amod_quality_optimal nn_quality_translation pobj_in_quality ' +o,2088,'Each element of the resulting vector was replaced with its log-likelihood value -LRB- see Definition 10 in Section 23 -RRB- which can be considered as an estimate of how surprising or distinctive a co-occurrence pair is ',Dunning,'det_element_Each nsubjpass_replaced_element prep_element_of det_vector_the amod_vector_resulting pobj_of_vector auxpass_replaced_was prep_replaced_with poss_value_its amod_value_log-likelihood pobj_with_value dep_value_see dobj_see_Definition num_Definition_10 prep_Definition_in pobj_in_Section num_Section_23 nsubjpass_considered_which aux_considered_can auxpass_considered_be rcmod_value_considered prep_considered_as det_estimate_an pobj_as_estimate prep_estimate_of advmod_is_how dep_how_surprising cc_surprising_or conj_surprising_distinctive det_pair_a amod_pair_co-occurrence nsubj_is_pair pcomp_of_is ' +o,2089,' argue that generic sentence fusion is an ill-defined task ','Daume III','nsubj_argue_ complm_task_that amod_fusion_generic nn_fusion_sentence nsubj_task_fusion cop_task_is det_task_an amod_task_ill-defined ccomp_argue_task ' +o,2090,'MI is defined in general as follows : y -RRB- I ix y -RRB- = log2 P -LRB- x -RRB- P -LRB- y -RRB- We can use this definition to derive an estimate of the connectedness between words , in terms of collocations , but also in terms of phrases and grammatical relations ',Hindle,'nsubjpass_defined_MI auxpass_defined_is prep_defined_in pobj_in_general prep_general_as dep_as_follows dep_follows_y num_P_I nn_P_ix nn_P_y nn_P_= nn_P_log2 pobj_as_P dep__x nsubj__P appos_P_y nsubj_use_We aux_use_can rcmod_P_use det_definition_this dobj_use_definition aux_derive_to infmod_definition_derive det_estimate_an dobj_derive_estimate prep_estimate_of det_connectedness_the pobj_of_connectedness prep_connectedness_between pobj_between_words prep_derive_in pobj_in_terms prep_terms_of pobj_of_collocations dep_defined_ cc__but advmod__also conj__in pobj_in_terms prep_terms_of pobj_of_phrases cc_phrases_and amod_relations_grammatical conj_phrases_relations ' +o,2091,'The original training set -LRB- before the addition of the feedback sets -RRB- consisted of a few dozen examples , in comparison to thousands of examples needed in other corpus-based methods ',Yarowsky,'det_set_The amod_set_original nn_set_training nsubj_consisted_set mark_sets_before det_addition_the nsubj_sets_addition prep_addition_of det_feedback_the pobj_of_feedback dep_set_sets prep_consisted_of quantmod_dozen_a quantmod_dozen_few num_examples_dozen pobj_of_examples prep_consisted_in pobj_in_comparison prep_comparison_to pobj_to_thousands prep_thousands_of pobj_of_examples partmod_examples_needed prep_needed_in amod_methods_other amod_methods_corpus-based pobj_in_methods ' +o,2092,'Its size is compatible to ',Turney,'poss_size_Its nsubj_compatible_size cop_compatible_is prep_compatible_to ' +p,2093,'51 Comparison to self-training For completeness , we also compared our results to the self-learning algorithm , which has commonly been referred to as bootstrapping in natural language processing and originally popularized by the work of Yarowsky in word sense disambiguation ',Yarowsky,'num_Comparison_51 prep_Comparison_to pcomp_to_self-training prep_compared_For pobj_For_completeness nsubj_compared_we advmod_compared_also ccomp_self-training_compared poss_results_our dobj_compared_results prep_compared_to det_algorithm_the amod_algorithm_self-learning pobj_to_algorithm nsubjpass_referred_which aux_referred_has advmod_referred_commonly auxpass_referred_been rcmod_algorithm_referred aux_bootstrapping_to advmod_bootstrapping_as xcomp_referred_bootstrapping prep_bootstrapping_in amod_processing_natural nn_processing_language pobj_in_processing cc_referred_and advmod_popularized_originally conj_referred_popularized prep_popularized_by det_work_the pobj_by_work prep_work_of pobj_of_Yarowsky prep_popularized_in nn_disambiguation_word nn_disambiguation_sense pobj_in_disambiguation ' +o,2094,'2 Related Work There have been various efforts to integrate linguistic knowledge into SMT systems , either from the target side , the source side or both sides , just to name a few ',Koehn,'num_Work_2 amod_Work_Related expl_been_There aux_been_have dep_Work_been amod_efforts_various dobj_been_efforts aux_integrate_to infmod_efforts_integrate amod_knowledge_linguistic dobj_integrate_knowledge prep_integrate_into nn_systems_SMT pobj_into_systems dep_into_either conj_into_from det__the nn__target nn__side pobj_from_ det__the nn__source nn__side conj__ cc__or det__both nn__sides conj__ advmod_name_just aux_name_to xcomp_integrate_name det_few_a dobj_name_few ' +o,2095,'Recently , some kinds of learning techniques have been applied to cumulatively acquire exemplars form large corpora ',Yarowsky,'advmod_applied_Recently det_kinds_some nsubjpass_applied_kinds prep_kinds_of pcomp_of_learning dobj_learning_techniques aux_applied_have auxpass_applied_been aux_acquire_to advmod_acquire_cumulatively xcomp_applied_acquire nsubj_form_exemplars ccomp_acquire_form amod_corpora_large dobj_form_corpora ' +o,2096,'The second approach takes triples -LRB- verb , prep , noun2 -RRB- and -LRB- nounl , prep , noun2 -RRB- , like those in Table 10 , as training data for acquiring semantic knowledge and performs PP-attachment disambiguation on quadruples ',Marcus,'det__The amod__second nn__approach nsubj_takes_ dobj_takes_triples nn_noun2_verb dep_noun2_prep appos_triples_noun2 cc_triples_and conj_triples_nounl dep_nounl_prep dep_nounl_noun2 prep_nounl_like pobj_like_those prep_nounl_in pobj_in_Table num_Table_10 prep_triples_as nn_data_training pobj_as_data prep_data_for pcomp_for_acquiring amod_knowledge_semantic dobj_acquiring_knowledge cc_takes_and conj_takes_performs amod_disambiguation_PP-attachment dobj_performs_disambiguation prep_performs_on pobj_on_quadruples ' +p,2097,'There are only a few successful studies , such as for chunking and on constituency parsing ',McClosky,'expl_are_There quantmod_few_only quantmod_few_a num_studies_few amod_studies_successful nsubj_are_studies dep_as_such prep_studies_as pobj_as_ prep__for pcomp_for_chunking cc_chunking_and conj_chunking_ prep_chunking_on nn_parsing_constituency pobj_on_parsing ' +o,2098,'Unlexicalized parsers , on the other hand , achieved accuracies almost equivalent to those of lexicalized parsers ',Matsuzaki,'amod_parsers_Unlexicalized nsubj_achieved_parsers prep_parsers_on det_hand_the amod_hand_other pobj_on_hand nsubj_equivalent_accuracies advmod_equivalent_almost xcomp_achieved_equivalent prep_equivalent_to pobj_to_those prep_those_of amod_parsers_lexicalized pobj_of_parsers ' +o,2099,'22 Themaximumentropytagger The maximum entropy model used in POStagging is described in detail in andthePOCtaggerhereusesthesame probability model ',Ratnaparkhi,'num_Themaximumentropytagger_22 det_model_The amod_model_maximum nn_model_entropy nsubjpass_described_model partmod_model_used prep_used_in pobj_in_POStagging auxpass_described_is dep_Themaximumentropytagger_described prep_described_in pobj_in_detail prep_detail_in amod_model_ nn_model_andthePOCtaggerhereusesthesame nn_model_probability pobj_in_model ' +p,2100,'To reduce the knowledge engineering burden on the user in constructing and porting an IE system , unsupervised learning has been utilized , eg Riloff , Yangarber et al ',Berger,'aux_reduce_To dep_utilized_reduce det_burden_the nn_burden_knowledge nn_burden_engineering dobj_reduce_burden prep_burden_on det_user_the pobj_on_user prep_reduce_in pcomp_in_constructing cc_constructing_and conj_constructing_porting det_system_an nn_system_IE dobj_constructing_system amod_learning_unsupervised nsubjpass_utilized_learning aux_utilized_has auxpass_utilized_been dep_utilized_eg nn__Riloff dobj_utilized_ conj__Yangarber cc__et conj__al ' +o,2101,'It is worth noting that we observed the same relation between subjectivity detection and polarity classification accuracy as described by and Eriksson -LRB- 2006 -RRB- ',Pang,'nsubj_worth_It cop_worth_is xcomp_worth_noting complm_observed_that nsubj_observed_we ccomp_noting_observed det_relation_the amod_relation_same dobj_observed_relation prep_relation_between amod_detection_subjectivity pobj_between_detection cc_detection_and nn_accuracy_polarity nn_accuracy_classification conj_detection_accuracy mark_described_as advcl_worth_described prep_described_by pobj_by_ cc__and conj__Eriksson dep_described_2006 ' +o,2102,' uses a conceptually similar technique for WSD that learns from a small set of seed examples and then increases recall by bootstrapping , evaluated on 12 idiosyncratically polysemous words ',Yarowsky,'advmod_uses_ det_technique_a advmod_similar_conceptually amod_technique_similar dobj_uses_technique prep_technique_for pobj_for_WSD nsubj_learns_that dep_uses_learns prep_learns_from det_set_a amod_set_small pobj_from_set prep_set_of nn_examples_seed pobj_of_examples cc_learns_and advmod_increases_then conj_learns_increases dobj_increases_recall prep_increases_by pobj_by_bootstrapping partmod_bootstrapping_evaluated prep_evaluated_on num_words_12 amod_words_idiosyncratically nn_words_polysemous pobj_on_words ' +o,2103,'2 RelatedWork 21 Sentiment Classification Most previous work on the problem of categorizing opinionated texts has focused on the binary classification of positive and negative sentiment ',Pang,'num_Sentiment_2 nn_Sentiment_RelatedWork num_Sentiment_21 dep_Classification_Sentiment amod_work_Most amod_work_previous dobj_Classification_work prep_work_on det_problem_the pobj_on_problem prep_problem_of pcomp_of_categorizing amod_texts_opinionated nsubj_focused_texts aux_focused_has ccomp_categorizing_focused prep_focused_on det_classification_the amod_classification_binary pobj_on_classification prep_classification_of amod_sentiment_positive cc_positive_and conj_positive_negative pobj_of_sentiment ' +o,2104,'We generate POS tags using the MXPOST tagger for English and Chinese , and Connexor for Spanish ',Ratnaparkhi,'nsubj_generate_We dep_tags_POS nsubj_using_tags dep_generate_using det__the amod__MXPOST nn__tagger dobj_using_ prep__for pobj_for_English cc_English_and conj_English_Chinese cc__and conj__Connexor prep_Connexor_for pobj_for_Spanish ' +o,2105,'Motivation There have been quite a number of recent papers on parallel text : , Chen , Church , Church et al , Dagan et al , Gale and Church , Isabelle , Kay and Rgsenschein , Klavans and Tzoukermann , Kupiec , Matsumoto , Ogden and Gonzales , Shemtov , Simard et al , WarwickArmstrong and Russell , Wu -LRB- to appear -RRB- ',Brown,'expl_been_There aux_been_have rcmod_Motivation_been advmod_number_quite det_number_a dobj_been_number prep_number_of amod_papers_recent pobj_of_papers prep_been_on amod_text_parallel pobj_on_text dep_Motivation_ nn__Chen conj__ nn__Church nn__ conj__Church cc__et conj__al conj__ nn__Dagan cc_Dagan_et conj_Dagan_al conj__ nn__Gale cc_Gale_and conj_Gale_Church conj__ nn__Isabelle conj__ nn__Kay cc_Kay_and conj_Kay_Rgsenschein conj__ nn__Klavans cc_Klavans_and conj_Klavans_Tzoukermann conj__ nn__Kupiec conj__ nn__Matsumoto conj__ nn__Ogden cc_Ogden_and conj_Ogden_Gonzales conj__ nn__Shemtov nn__ conj__Simard cc__et conj__al conj__ conj__WarwickArmstrong cc__and nn_Wu_Russell nn_Wu_ conj__Wu aux_appear_to dep_Motivation_appear ' +o,2106,'In this paper we will describe extensions to tile Hidden-Markov alignment model froln and compare tlmse to Models 1 4 of ',Brown,'prep_describe_In det_paper_this pobj_In_paper nsubj_describe_we aux_describe_will dobj_describe_extensions prep_describe_to amod__tile nn__Hidden-Markov nn__alignment nn__model nn__froln pobj_to_ cc_describe_and conj_describe_compare acomp_compare_tlmse aux_Models_to xcomp_tlmse_Models number_4_1 dobj_Models_4 prep_4_of ' +o,2107,'Unlike Church and Hanks , goes beyond the ` two-word \' limitation and deals with ` collocations of arbitrary length \' ',Smadja,'prep_goes_Unlike nn__Church cc_Church_and conj_Church_Hanks pobj_Unlike_ nsubj_goes_ prep_goes_beyond det_limitation_the amod_limitation_two-word pobj_beyond_limitation cc_limitation_and conj_limitation_deals prep_goes_with pobj_with_collocations prep_collocations_of amod_length_arbitrary pobj_of_length ' +o,2108,'Automatic text summarization approaches have offered reasonably well-performing approximations for identifiying important sentences but , not surprisingly , text -LRB- re -RRB- generation has been a major challange despite some work on sub-sentential modification ',Jing,'nn_approaches_Automatic nn_approaches_text nn_approaches_summarization nsubj_offered_approaches aux_offered_have advmod_well-performing_reasonably amod_approximations_well-performing dobj_offered_approximations prep_offered_for pcomp_for_identifiying amod_sentences_important dobj_identifiying_sentences dep_sentences_ cc__but dep_surprisingly_not advmod_sentences_surprisingly appos_sentences_text appos_text_re nsubj_been_generation aux_been_has rcmod_text_been det_challange_a amod_challange_major dobj_been_challange prep_been_despite det_work_some pobj_despite_work prep_work_on amod_modification_sub-sentential pobj_on_modification ' +o,2109,'Tagging can also be done using maximum entropy modeling -LRB- see Section 24 -RRB- : a maximum entropy tagger , called MXPOST , was developed by -LRB- we will refer to this tagger as MXP below -RRB- ',Ratnaparkhi,'nsubjpass_done_Tagging aux_done_can advmod_done_also auxpass_done_be purpcl_done_using amod_modeling_maximum nn_modeling_entropy dobj_using_modeling dep_modeling_see dobj_see_Section num_Section_24 det_tagger_a amod_tagger_maximum nn_tagger_entropy nsubjpass_developed_tagger partmod_tagger_called dobj_called_MXPOST auxpass_developed_was parataxis_done_developed prep_developed_by pobj_by_ nsubj_refer_we aux_refer_will dep_developed_refer prep_refer_to det_tagger_this pobj_to_tagger prep_tagger_as pobj_as_MXP advmod_refer_below ' +o,2110,'Further , it has been shown that performance of Lins distributional similarity score decreases more significantly than other measures for low frequency nouns ',Weeds,'advmod_shown_Further nsubjpass_shown_it aux_shown_has auxpass_shown_been xcomp_shown_ complm_decreases_that nsubj_decreases_performance prep_performance_of nn_score_Lins amod_score_distributional nn_score_similarity pobj_of_score ccomp__decreases advmod_significantly_more advmod_decreases_significantly prep_decreases_than amod_measures_other pobj_than_measures prep_measures_for amod_nouns_low nn_nouns_frequency pobj_for_nouns ' +p,2111,'Recently so-called reranking techniques , such as maximum entropy models and gradient methods , have been applied to machine translation -LRB- MT -RRB- , and have provided significant improvements ',Och,'amod_techniques_Recently amod_techniques_so-called amod_techniques_reranking nsubjpass_applied_techniques dep_as_such prep_techniques_as amod__maximum nn__entropy nn__models pobj_as_ cc__and nn__gradient nn__methods conj__ aux_applied_have auxpass_applied_been prep_applied_to nn_translation_machine pobj_to_translation abbrev_translation_MT cc_applied_and aux_provided_have conj_applied_provided amod_improvements_significant dobj_provided_improvements ' +o,2112,'Zero derivation pointed out that it is helpful to identify zero-derived noun\\/verb pairs for such tasks as normalization of the semantics of expressions that are only superficially different ',Dolan,'nn__Zero nn__derivation nsubj_pointed_ prt_pointed_out complm_helpful_that nsubj_helpful_it cop_helpful_is ccomp_pointed_helpful aux_identify_to xcomp_helpful_identify amod_pairs_zero-derived nn_pairs_noun\\/verb dobj_identify_pairs prep_identify_for amod_tasks_such pobj_for_tasks prep_identify_as pobj_as_normalization prep_normalization_of det_semantics_the pobj_of_semantics prep_semantics_of pobj_of_expressions nsubj_different_that cop_different_are advmod_different_only advmod_different_superficially rcmod_expressions_different ' +o,2113,'2 Automatic Annotation Schemes Using ROUGE Similarity Measures ROUGE -LRB- Recall-Oriented Understudy for Gisting Evaluation -RRB- is an automatic tool to determine the quality of a summary using a collection of measures ROUGE-N -LRB- N = 1,2,3,4 -RRB- , ROUGE-L , ROUGE-W and ROUGE-S which count the number of overlapping units such as n-gram , word-sequences , and word-pairs between the extract and the abstract summaries ',Lin,'num_Schemes_2 nn_Schemes_Automatic nn_Schemes_Annotation nsubj_tool_Schemes partmod_Schemes_Using nn_ROUGE_ROUGE nn_ROUGE_Similarity nn_ROUGE_Measures dobj_Using_ROUGE nn_Understudy_Recall-Oriented dep_ROUGE_Understudy prep_Understudy_for nn_Evaluation_Gisting pobj_for_Evaluation cop_tool_is det_tool_an amod_tool_automatic aux_determine_to infmod_tool_determine det_quality_the dobj_determine_quality prep_quality_of det_summary_a pobj_of_summary partmod_summary_using det_collection_a dobj_using_collection prep_collection_of pobj_of_measures amod_measures_ROUGE-N nsubj_1,2,3,4_N dep_1,2,3,4_= dep_ROUGE-N_1,2,3,4 amod_measures_ROUGE-L conj_ROUGE-L_ROUGE-W cc_ROUGE-L_and conj_ROUGE-L_ROUGE-S nsubj_count_which rcmod_measures_count det_number_the dobj_count_number prep_number_of amod_units_overlapping pobj_of_units dep_as_such prep_units_as nn_word-pairs_n-gram conj_word-pairs_word-sequences cc_word-pairs_and pobj_as_word-pairs prep_units_between det_extract_the pobj_between_extract cc_extract_and det_summaries_the amod_summaries_abstract conj_extract_summaries ' +o,2114,'a176 Base NP standard data set -LRB- baseNP-S -RRB- This data set was first introduced by , and taken as the standard data set for baseNP identification task2 ',Ramshaw,'num_data_a176 nn_data_Base nn_data_NP amod_data_standard nsubjpass_introduced_data partmod_data_set dep_data_baseNP-S det_data_This dep_set_data partmod_data_set auxpass_introduced_was advmod_introduced_first prep_introduced_by pobj_by_ cc_introduced_and conj_introduced_taken prep_taken_as det_data_the amod_data_standard pobj_as_data partmod_data_set prep_set_for nn_task2_baseNP nn_task2_identification pobj_for_task2 ' +o,2115,'2 The Data Our experiments on joint syntactic and semantic parsing use data that is produced automatically by merging the Penn Treebank -LRB- PTB -RRB- with PropBank -LRB- PRBK -RRB- , as shown in Figure 1 ',Marcus,'dep_experiments_2 det_experiments_The nn_experiments_Data nn_experiments_Our prep_experiments_on amod_syntactic_joint pobj_on_syntactic cc_syntactic_and amod_data_semantic amod_data_parsing nn_data_use conj_syntactic_data nsubjpass_produced_that auxpass_produced_is rcmod_syntactic_produced advmod_produced_automatically prep_produced_by pcomp_by_merging det_Treebank_the nn_Treebank_Penn dobj_merging_Treebank abbrev_experiments_PTB prep_experiments_with nn__PropBank abbrev__PRBK pobj_with_ mark_shown_as dep__shown prep_shown_in pobj_in_Figure num_Figure_1 ' +o,2116,'In all the experiments , our source side language is English , and the Stanford Named Entity Recognizer was used to extract NEs from the source side article ',Finkel,'prep_English_In predet_experiments_all det_experiments_the pobj_In_experiments poss_language_our nn_language_source nn_language_side nsubj_English_language cop_English_is cc_English_and det__the nn__Stanford nn__Named nn__Entity nn__Recognizer nsubjpass_used_ auxpass_used_was conj_English_used aux_extract_to xcomp_used_extract dobj_extract_NEs prep_extract_from det_article_the nn_article_source nn_article_side pobj_from_article ' +p,2117,'Previous work for English has shown that lexicalization leads to a sizable improvement in parsing performance ',Collins,'amod_work_Previous nsubj_shown_work prep_work_for nn__English pobj_for_ aux_shown_has complm_leads_that nsubj_leads_lexicalization ccomp_shown_leads prep_leads_to det_improvement_a amod_improvement_sizable pobj_to_improvement prep_improvement_in amod_performance_parsing pobj_in_performance ' +o,2118,'The Duluth Word Alignment System is a Perl implementation of IBM Model 2 ',Brown,'det_System_The nn_System_Duluth nn_System_Word nn_System_Alignment nsubj_implementation_System cop_implementation_is det_implementation_a nn_implementation_Perl prep_implementation_of nn_Model_IBM pobj_of_Model num_Model_2 ' +o,2119,'Other systems also look at Web product reviews but they do not extract 345 opinions about particular product features ',Pang,'amod__Other nn__systems nsubj_look_ advmod_look_also prep_look_at nn_reviews_Web nn_reviews_product pobj_at_reviews cc_look_but nsubj_extract_they aux_extract_do neg_extract_not conj_look_extract amod_opinions_345 dobj_extract_opinions prep_extract_about amod_features_particular nn_features_product pobj_about_features ' +o,2120,'Minimum error rate training was used to tune the model feature weights ',Och,'nn_training_Minimum nn_training_error nn_training_rate dep_used_training auxpass_used_was prep_used_to pobj_to_tune det_weights_the nn_weights_model nn_weights_feature nsubjpass_used_weights ' +o,2121,'Some studies have been done for acquiring collocation translations using parallel corpora ',Brown,'det_studies_Some nsubjpass_done_studies aux_done_have auxpass_done_been prep_done_for pcomp_for_acquiring nn_translations_collocation dobj_acquiring_translations xcomp_acquiring_using amod__parallel nn__corpora dobj_using_ ' +o,2122,'31 Maximum Entropy This section presents a brief description of ME A more detailed and informative description can be found in Berger 4 , , Manning and Shutze -LRB- 2000 -RRB- to name just a few ',Ratnaparkhi,'num_Entropy_31 nn_Entropy_Maximum det_section_This nsubj_presents_section dep_Entropy_presents det_description_a amod_description_brief dobj_presents_description prep_description_of pobj_of_ME det_description_A advmod_detailed_more amod_description_detailed cc_detailed_and conj_detailed_informative nsubjpass_found_description aux_found_can auxpass_found_be dep_Entropy_found prep_found_in nn__Berger pobj_in_ num__4 num__ appos__Manning cc_Manning_and conj_Manning_Shutze appos_Manning_2000 aux_name_to purpcl_found_name quantmod_few_just quantmod_few_a dobj_name_few ' +o,2123,'PairClass generates probability estimates , whereas uses a cosine measure of similarity ',Turney,'nsubj_generates_PairClass nn_estimates_probability dobj_generates_estimates dep_generates_whereas nsubj_uses_ dep_generates_uses det_measure_a amod_measure_cosine dobj_uses_measure prep_measure_of pobj_of_similarity ' +o,2124,'For every class the weights of the active features are combined and the best scoring class is chosen ',Berger,'det_class_every pobj_For_class det_weights_the nsubjpass_combined_weights prep_weights_of det_features_the amod_features_active pobj_of_features auxpass_combined_are rcmod_class_combined cc_combined_and det_class_the amod_class_best nn_class_scoring nsubjpass_chosen_class auxpass_chosen_is conj_combined_chosen ' +o,2125,'The sentences were processed with the Collins parser to generate automatic parse trees ',Collins,'det_sentences_The nsubjpass_processed_sentences auxpass_processed_were prep_processed_with det__the nn__Collins nn__parser pobj_with_ aux_generate_to xcomp_processed_generate amod_trees_automatic nn_trees_parse dobj_generate_trees ' +o,2126,'7 Independently , quote a performance of 800 words per second for their part-of-speech tagger based on hidden Markov models ',Cutting,'num__7 nn__Independently nsubj_quote_ det_performance_a dobj_quote_performance prep_performance_of num_words_800 pobj_of_words prep_words_per pobj_per_second prep_second_for poss_tagger_their amod_tagger_part-of-speech pobj_for_tagger prep_quote_based dep_based_on amod_models_hidden nn_models_Markov pobj_on_models ' +p,2127,'Tighter integration of semantics into the parsing models , possibly in the form of discriminative reranking models , is a promising way forward in this regard ',McClosky,'amod_integration_Tighter nsubj_way_integration prep_integration_of pobj_of_semantics prep_semantics_into det_models_the amod_models_parsing pobj_into_models rcmod_integration_possibly dep_possibly_in det_form_the pobj_in_form prep_form_of amod__discriminative amod__reranking nn__models pobj_of_ cop_way_is det_way_a amod_way_promising dep_in_forward prep_way_in det_regard_this pobj_in_regard ' +o,2128,'The value of Dist -LRB- D -LRB- T -RRB- -RRB- can be defined in various ways , and they found that using log-likelihood ratio worked best which is represented as follows : 0 # log -RRB- -LRB- # log D K k TD k k i M ii i i M ii i = = , where k i and K i are the frequency of a word w i in D -LRB- W -RRB- and D 0 respectively , and -LCB- w 1 , , w M -RCB- is the set of all words in D 0 As stated in introduction , Dist -LRB- D -LRB- T -RRB- -RRB- is normalized by the baseline function , which is referred as B Dist -LRB- -RRB- here ',Dunning,'det_value_The nsubjpass_defined_value prep_value_of pobj_of_Dist nn_T_D dep_Dist_T aux_defined_can auxpass_defined_be prep_defined_in amod_ways_various pobj_in_ways cc_defined_and nsubj_found_they conj_defined_found complm_worked_that csubj_worked_using amod__log-likelihood nn__ratio dobj_using_ ccomp_found_worked dobj_worked_best nsubjpass_represented_which auxpass_represented_is rcmod_best_represented prep_represented_as pobj_as_follows dep_follows_0 dep_log_# dep_0_log dep_log_# dep_follows_log nn_K_D nsubj_k_K rcmod_log_k prt_k_TD dobj_k_k punct_k_k nn_k_i nn_k_M nn_k_ii nn_k_i nn_k_i nn_k_M nn_k_ii nn_k_i nn_k_= dep_k_= advmod_frequency_where nsubj_frequency_k dep_k_i cc_k_and nn_i_K conj_k_i cop_frequency_are det_frequency_the rcmod_k_frequency prep_frequency_of pobj_of_a nn_a_word nn_a_w dep_a_i prep_a_in pobj_in_D appos_D_W cc_D_and conj_D_D num_D_0 advmod_frequency_respectively cc_k_and dep_k_w num_w_1 amod_M_w appos_w_M cop_set_is det_set_the conj_k_set prep_set_of det_words_all pobj_of_words prep_words_in pobj_in_D num_D_0 mark_normalized_As csubjpass_normalized_stated prep_stated_in pobj_in_introduction appos_introduction_Dist dep_Dist_D appos_Dist_T auxpass_normalized_is dep_k_normalized prep_normalized_by det_function_the amod_function_baseline pobj_by_function nsubjpass_referred_which auxpass_referred_is rcmod_function_referred prep_referred_as nn_-LRB-_B nn_-LRB-_Dist advmod_worked_here ' +o,2129,' , Ponzetto and Strube -RRB- and the exploitation of advanced techniques that involve joint learning -LRB- eg , and Marcu -RRB- and joint inference -LRB- eg , Denis and Baldridge -LRB- 2007 -RRB- -RRB- for coreference resolution and a related extraction task ','Daume III','nn__ conj__Ponzetto cc__and conj__Strube cc__and det_exploitation_the conj__exploitation prep_exploitation_of amod_techniques_advanced pobj_of_techniques nsubj_involve_that rcmod_exploitation_involve amod_learning_joint dobj_involve_learning dep_learning_eg conj_eg_ cc_eg_and nn__Marcu conj_eg_ cc_learning_and amod_inference_joint conj_learning_inference dep_inference_eg conj_eg_Denis cc_eg_and conj_eg_Baldridge appos_inference_2007 prep_involve_for amod_resolution_coreference pobj_for_resolution cc_resolution_and det_task_a amod_task_related nn_task_extraction conj_resolution_task ' +o,2130,'In this paper , we present Phramer , an open-source system that embeds a phrase-based decoder , a minimum error rate training module and various tools related to Machine Translation -LRB- MT -RRB- ',Och,'prep_present_In det_paper_this pobj_In_paper nsubj_present_we dobj_present_Phramer det_system_an amod_system_open-source appos_Phramer_system nsubj_embeds_that rcmod_system_embeds det_decoder_a amod_decoder_phrase-based dobj_embeds_decoder det_module_a amod_module_minimum nn_module_error nn_module_rate nn_module_training nn_module_ conj_decoder_module cc_decoder_and amod_tools_various conj_decoder_tools partmod_tools_related prep_related_to nn_Translation_Machine pobj_to_Translation abbrev_Translation_MT ' +p,2131,'The creation of the Penn English Treebank , a syntactically interpreted corpus , played a crucial role in the advances in natural language parsing technology for English ',Marcus,'det_creation_The nsubj_played_creation prep_creation_of det__the nn__Penn nn__English nn__Treebank pobj_of_ det_corpus_a advmod_interpreted_syntactically amod_corpus_interpreted appos_creation_corpus det_role_a amod_role_crucial dobj_played_role prep_played_in det_advances_the pobj_in_advances prep_advances_in amod_language_natural pobj_in_language xcomp_played_parsing nn__technology dobj_parsing_ prep_parsing_for pobj_for_English ' +o,2132,'For example , the HMM aligner achieves an AER of 207 when using the competitive thresholding heuristic of ',DeNero,'prep_achieves_For pobj_For_example det_aligner_the nn_aligner_HMM nsubj_achieves_aligner det_AER_an dobj_achieves_AER prep_AER_of pobj_of_207 advmod_using_when advcl_achieves_using det_heuristic_the amod_heuristic_competitive nn_heuristic_thresholding dobj_using_heuristic prep_using_of ' +o,2133,'We used GIZA + + package to train IBM translation models ',Och,'nsubj_used_We amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__package dobj_used_ aux_train_to xcomp_used_train nn_models_IBM nn_models_translation dobj_train_models ' +o,2134,'Others , such as , Pang and Vaithyanathan -LRB- 2002 -RRB- , have examined the positive or negative polarity , rather than presence or absence , of affective content in text ',Turney,'nsubj_examined_Others dep_as_such prep_Others_as pobj_as_ conj__Pang cc__and conj__Vaithyanathan appos__2002 aux_examined_have det_polarity_the amod_polarity_positive cc_positive_or conj_positive_negative xcomp_examined_polarity cc_polarity_rather dep_rather_than conj_polarity_presence cc_presence_or conj_presence_absence prep_examined_of amod_content_affective pobj_of_content prep_content_in pobj_in_text ' +o,2135,'We will employ the structural correspondence learning -LRB- SCL -RRB- domain adaption algorithm used in for linking the translated text and the natural text ',Blitzer,'nsubj_employ_We aux_employ_will det_correspondence_the amod_correspondence_structural dobj_employ_correspondence xcomp_employ_learning dep_algorithm_SCL nn_algorithm_domain nn_algorithm_adaption dep_learning_algorithm partmod_algorithm_used prep_used_in pobj_in_ prep_used_for pcomp_for_linking det_text_the amod_text_translated dobj_linking_text cc_text_and det_text_the amod_text_natural conj_text_text ' +o,2136,'53 Systematic Sense Shift contend that there is strong evidence to suggest that a large part of word sense ambiguity is not arbitrary but follows regular patterns ',Ostler,'num__53 nn__Systematic nn__Sense nn__Shift nsubj_contend_ complm_is_that expl_is_there ccomp_contend_is amod_evidence_strong nsubj_is_evidence aux_suggest_to infmod_evidence_suggest complm_arbitrary_that det_part_a amod_part_large nsubj_arbitrary_part prep_part_of nn_ambiguity_word nn_ambiguity_sense pobj_of_ambiguity cop_arbitrary_is neg_arbitrary_not ccomp_suggest_arbitrary cc_arbitrary_but conj_arbitrary_follows amod_patterns_regular dobj_follows_patterns ' +o,2137,'Among all the language modeling approaches , ngram models have been most widely used in speech recognition and other applications ',Brown,'prep_used_Among predet_modeling_all det_modeling_the nn_modeling_language pobj_Among_modeling dep_modeling_approaches nn_models_ngram nsubjpass_used_models aux_used_have auxpass_used_been advmod_widely_most advmod_used_widely prep_used_in nn__speech nn__recognition pobj_in_ cc__and amod_applications_other conj__applications ' +o,2138,'Inter-sentential contexts as in our approach were used as a clue also for subjectivity analysis , which is two-fold classification into subjective and objective sentences ',Pang,'amod_contexts_Inter-sentential nsubjpass_used_contexts dep_in_as prep_contexts_in poss_approach_our pobj_in_approach auxpass_used_were prep_used_as det_clue_a pobj_as_clue dep_for_also prep_used_for amod__subjectivity nn__analysis pobj_for_ nsubj_classification_which cop_classification_is amod_classification_two-fold rcmod__classification prep_classification_into amod_sentences_subjective cc_subjective_and conj_subjective_objective pobj_into_sentences ' +p,2139,'Bleu is fast and easy to run , and it can be used as a target function in parameter optimization training procedures that are commonly used in state-of-the-art statistical MT systems ',Och,'nsubj_fast_Bleu cop_fast_is cc_fast_and conj_fast_easy aux_run_to xcomp_fast_run cc_fast_and nsubjpass_used_it aux_used_can auxpass_used_be conj_fast_used prep_used_as det_function_a nn_function_target pobj_as_function prep_function_in nn_procedures_parameter nn_procedures_optimization nn_procedures_training pobj_in_procedures nsubjpass_used_that auxpass_used_are advmod_used_commonly rcmod_function_used prep_used_in amod_systems_state-of-the-art amod_systems_statistical nn_systems_MT pobj_in_systems ' +o,2140,'We also employ the voted perceptron algorithm and the early update technique as in ',Collins,'nsubj_employ_We advmod_employ_also det__the amod__voted nn__perceptron nn__algorithm dobj_employ_ cc__and det_technique_the amod_technique_early nn_technique_update conj__technique prep_employ_as dep_as_in ' +o,2141,'High quality word alignments can yield more accurate phrase-pairs which improve quality of a phrase-based SMT system ',Och,'nn_alignments_High nn_alignments_quality nn_alignments_word nsubj_yield_alignments aux_yield_can advmod_accurate_more amod_phrase-pairs_accurate dobj_yield_phrase-pairs nsubj_improve_which rcmod_phrase-pairs_improve dobj_improve_quality prep_quality_of det_system_a amod_system_phrase-based nn_system_SMT pobj_of_system ' +o,2142,'Co-occurrence information between neighboring words and words in the same sentence has been used in phrase extraction , phrasal translation , target word selection , domain word translation , sense disambiguation , and even recently for query translation in cross-language IR as well ',Smadja,'amod_information_Co-occurrence nsubjpass_used_information prep_information_between amod_words_neighboring pobj_between_words cc_words_and conj_words_words prep_words_in det_sentence_the amod_sentence_same pobj_in_sentence aux_used_has auxpass_used_been prep_used_in nn__phrase nn__extraction pobj_in_ amod__phrasal nn__translation conj__ nn__target nn__word nn__selection conj__ nn__domain nn__word nn__translation conj__ nn__sense nn__disambiguation conj__ cc__and advmod_recently_even advmod__recently dep_recently_for dep_for_query conj__translation prep_used_in amod_IR_cross-language pobj_in_IR dep_well_as cc_IR_well conj_IR_ ' +o,2143,' , by contrast to the above , used Wikipedia primarily for Named Entity Disambiguation , following the path of Bunescu and Paca ',Cucerzan,'prep__by dep_by_contrast dep_by_to det_Wikipedia_the amod_Wikipedia_above amod_Wikipedia_used pobj_by_Wikipedia advmod__primarily prep__for nn_Disambiguation_Named nn_Disambiguation_Entity pobj_for_Disambiguation prep_Disambiguation_following det_path_the pobj_following_path prep_path_of nn__Bunescu cc_Bunescu_and conj_Bunescu_Paca pobj_of_ ' +o,2144,'Recently there have been some works on using multiple treebanks for domain adaptation of parsers , where these treebanks have the same grammar formalism ',McClosky,'advmod_been_Recently advmod_been_there aux_been_have det_works_some dobj_been_works prep_works_on pcomp_on_using amod_treebanks_multiple dobj_using_treebanks prep_using_for pobj_for_domain nsubj_been_adaptation prep_adaptation_of pobj_of_parsers advmod_have_where det_treebanks_these nsubj_have_treebanks rcmod_adaptation_have det_formalism_the amod_formalism_same nn_formalism_grammar dobj_have_formalism ' +o,2145,'Given a collection of facts , ME chooses a model consistent with all the facts , but otherwise as uniform as possible ',Berger,'prep_chooses_Given det_collection_a dep_Given_collection prep_collection_of pobj_of_facts nsubj_chooses_ME det_model_a dobj_chooses_model amod_model_consistent prep_consistent_with predet_facts_all det_facts_the pobj_with_facts cc_consistent_but advmod_uniform_otherwise advmod_uniform_as conj_consistent_uniform mark_possible_as dep_model_possible ' +o,2146,'The earliest work in this direction are those of , , , , and ',Lin,'det_work_The amod_work_earliest nsubj_those_work prep_work_in det_direction_this pobj_in_direction cop_those_are prep_those_of pobj_of_ conj__ conj__ conj__ conj__ cc__and conj__ ' +o,2147,'Compared with their string-based counterparts , treebased systems offer some attractive features : they are much faster in decoding -LRB- linear time vs cubic time , see -RRB- , do not require a binary-branching grammar as in string-based models , and can have separate grammars for parsing and translation , say , a context-free grammar for the former and a tree substitution grammar for the latter ',Zhang,'prep_offer_Compared dep_Compared_with poss_counterparts_their amod_counterparts_string-based pobj_with_counterparts amod_systems_treebased nsubj_offer_systems det_features_some amod_features_attractive dobj_offer_features nsubj_faster_they cop_faster_are advmod_faster_much parataxis_offer_faster prep_faster_in pcomp_in_decoding amod_time_linear dep_decoding_time cc_time_vs amod_time_cubic conj_time_time dep_time_see dobj_see_ aux_require_do neg_require_not conj_faster_require det_grammar_a amod_grammar_binary-branching dobj_require_grammar prep_require_as dep_as_in amod__string-based nn__models pobj_in_ cc_faster_and aux_have_can conj_faster_have amod_grammars_separate dobj_have_grammars prep_grammars_for pobj_for_parsing cc_parsing_and conj_parsing_translation dep_have_say det_grammar_a amod_grammar_context-free dobj_have_grammar prep_grammar_for det_former_the pobj_for_former cc_grammar_and det_grammar_a nn_grammar_tree nn_grammar_substitution conj_grammar_grammar prep_grammar_for det__the amod__latter pobj_for_ ' +o,2148,'The studies presented by Goldwater and Griffiths and differed in the number of states that they used ',Johnson,'det_studies_The nsubj_presented_studies prep_presented_by nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths pobj_by_ cc__and nn__ conj__ partmod__differed prep_differed_in det_number_the pobj_in_number prep_number_of pobj_of_states complm_used_that nsubj_used_they ccomp_differed_used ' +o,2149,'BLEU score In order to measure the extent to which whole chunks of text from the prompt are reproduced in the student essays , we used the BLEU score , known from studies of machine translation ',Papineni,'nsubj_score_BLEU mark_measure_In dep_measure_order aux_measure_to dep_used_measure det_extent_the dobj_measure_extent rel_reproduced_to pobj_to_which amod_chunks_whole nsubj_reproduced_chunks prep_chunks_of pobj_of_text prep_text_from det_prompt_the pobj_from_prompt aux_reproduced_are rcmod_extent_reproduced prep_reproduced_in det_essays_the nn_essays_student pobj_in_essays nsubj_used_we ccomp_score_used det_score_the amod_score_BLEU dobj_used_score partmod_score_known prep_known_from pobj_from_studies prep_studies_of nn_translation_machine pobj_of_translation ' +n,2150,'1 Introduction Recent works in statistical machine translation -LRB- SMT -RRB- shows how phrase-based modeling significantly outperform the historical word-based modeling ',Brown,'num_works_1 nn_works_Introduction amod_works_Recent nsubj_shows_works prep_works_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT advmod_outperform_how amod__phrase-based nn__modeling nsubj_outperform_ advmod_outperform_significantly ccomp_shows_outperform det_modeling_the amod_modeling_historical amod_modeling_word-based dobj_outperform_modeling ' +o,2151,'09595 09590 09611 09085 09134 09152 Table 8 : Comparison of F1 results of our baseline model with Nakagawa and Uchimoto and on CTB 30 ',Zhang,'num_Table_09595 num_Table_09590 num_Table_09611 num_Table_09085 num_Table_09134 nn_Table_09152 dep_Table_8 dep_Table_Comparison prep_Comparison_of nn_results_F1 pobj_of_results prep_results_of poss_model_our amod_model_baseline pobj_of_model prep_model_with nn__Nakagawa cc_Nakagawa_and conj_Nakagawa_Uchimoto pobj_with_ cc_Comparison_and conj_Comparison_ prep__on pobj_on_CTB num_CTB_30 ' +o,2152,' -RRB- , the tagger for grammatical functions works with lexical -LRB- 1 -RRB- Selbst besucht ADV VVPP himself visited hat Peter Sabine VAFIN NE NE has Peter Sabine ` Peter never visited Sabine himself \' l hie ADV never Figure 2 : Example sentence and contextual probability measures PO -LRB- \' -RRB- depending on the category of a mother node -LRB- Q -RRB- ',Cutting,'dep_VVPP_ amod_tagger_the dobj__tagger prep_tagger_for amod_works_grammatical nn_works_functions pobj_for_works prep__with pobj_with_lexical appos_lexical_1 nn_ADV_Selbst nn_ADV_besucht nsubj_VVPP_ADV nsubj_visited_himself ccomp_VVPP_visited nn_NE_hat nn_NE_Peter nn_NE_Sabine nn_NE_VAFIN nn_NE_NE nsubj_has_NE ccomp_visited_has nn_Sabine_Peter dobj_has_Sabine nsubj_visited_Peter neg_visited_never rcmod_Sabine_visited dobj_visited_Sabine dep_hie_himself amod_hie_l nsubj_ADV_hie rcmod_Sabine_ADV advmod_Figure_never dobj_ADV_Figure num_Figure_2 amod_sentence_Example dep_Figure_sentence cc_sentence_and amod_measures_contextual nn_measures_probability conj_sentence_measures dep_sentence_PO dep_VVPP_depending prep_depending_on det_category_the pobj_on_category prep_category_of det_node_a nn_node_mother pobj_of_node appos_node_Q ' +p,2153,'For our experiments , we chose GIZA + + and the RA approach the best known alignment combination technique as our initial aligners1 42 TBL Templates Our templates consider consecutive words -LRB- of size 1 , 2 or 3 -RRB- in both languages ',Koehn,'prep_chose_For poss_experiments_our pobj_For_experiments nsubj_chose_we nsubj_+_GIZA xcomp_chose_+ cc_+_+ conj_+_ cc_chose_and det_approach_the nn_approach_RA nsubj__approach conj_chose_ det_technique_the dep_known_best amod_technique_known nn_technique_alignment nn_technique_combination dobj__technique prep_technique_as poss_Templates_our amod_Templates_initial nn_Templates_aligners1 num_Templates_42 nn_Templates_TBL pobj_as_Templates poss_templates_Our nsubj_consider_templates dep_chose_consider amod_words_consecutive dobj_consider_words prep_1_of pobj_of_size dep_words_1 appos_1_2 cc_2_or conj_2_3 prep_words_in det_languages_both pobj_in_languages ' +o,2154,'This leads to 49 methods that use semi-supervised techniques on a treebank-infered grammar backbone , such as ',Matsuzaki,'nsubj_leads_This prep_leads_to num_methods_49 pobj_to_methods nsubj_use_that rcmod_methods_use amod_techniques_semi-supervised dobj_use_techniques prep_use_on det_backbone_a amod_backbone_treebank-infered nn_backbone_grammar pobj_on_backbone advmod_as_such advmod_use_as ' +o,2155,'Smadja proposed a statistical model by measuring the spread of the distribution of cooccurring pairs of words with higher strength ',Smadja,'nn__Smadja nsubj_proposed_ det_model_a amod_model_statistical dobj_proposed_model prep_proposed_by pcomp_by_measuring det_spread_the dobj_measuring_spread prep_spread_of det_distribution_the pobj_of_distribution prep_distribution_of amod_pairs_cooccurring pobj_of_pairs prep_pairs_of pobj_of_words prep_words_with amod_strength_higher pobj_with_strength ' +o,2156,'Existing automatic evaluation measures such as BLEU and ROUGE -LRB- Lin 2The collections are available from http://wwwcsail ',Papineni,'amod_measures_Existing amod_measures_automatic nn_measures_evaluation dep_as_such prep_measures_as nn__BLEU pobj_as_ cc__and nn_Lin_ROUGE conj__Lin amod_collections_2The nsubj_available_collections cop_available_are dep_measures_available prep_available_from pobj_from_http://wwwcsail ' +o,2157,'To generate the n-best lists , a phrase based SMT was used ',Koehn,'aux_generate_To dep_based_generate det_lists_the amod_lists_n-best dobj_generate_lists det_phrase_a nsubj_based_phrase nn__SMT nsubjpass_used_ auxpass_used_was ccomp_based_used ' +o,2158,'EnglishChinese and EnglishSpanish ',Wellington,'amod__EnglishChinese cc__and amod__EnglishSpanish conj__ ' +o,2159,'CIT -RRB- ',Rapp,'nn_-RRB-_CIT ' +o,2160,'A structured perceptron learns weights for our transliteration features , which are drawn from two broad classes : indicator and hybrid generative features ',Collins,'det__A amod__structured nn__perceptron dep__learns dobj_learns_weights prep_learns_for poss_features_our nn_features_transliteration pobj_for_features nsubjpass_drawn_which auxpass_drawn_are rcmod_features_drawn prep_drawn_from num_classes_two amod_classes_broad pobj_from_classes dep__indicator cc_indicator_and amod_features_hybrid nn_features_generative conj_indicator_features ' +p,2161,'It has been known for some years that good performance can be realized with partial tagging and a hidden Markov model ',Cutting,'nsubjpass_known_It aux_known_has auxpass_known_been prep_known_for det_years_some pobj_for_years complm_realized_that amod_performance_good nsubjpass_realized_performance aux_realized_can auxpass_realized_be ccomp_known_realized prep_realized_with amod_tagging_partial pobj_with_tagging cc_tagging_and det_model_a amod_model_hidden nn_model_Markov conj_tagging_model ' +o,2162,'Four alternatives are proposed in these special issues : -LRB- 1 -RRB- Brent , -LRB- 2 -RRB- Briscoe and Carroll -LRB- this issue -RRB- , -LRB- 3 -RRB- Hindle and Rooth -LRB- this issue -RRB- , and -LRB- 4 -RRB- Weischedel et al ',Brown,'num_alternatives_Four nsubjpass_proposed_alternatives auxpass_proposed_are prep_proposed_in det_issues_these amod_issues_special pobj_in_issues dep__1 nsubj__Brent parataxis_proposed_ dep_Hindle_2 nsubj_Hindle_Briscoe cc_Briscoe_and conj_Briscoe_Carroll det_issue_this appos_Carroll_issue appos_Carroll_3 ccomp_proposed_Hindle cc_Rooth_and dobj_Hindle_Rooth det_issue_this appos_Rooth_issue cc_proposed_and dep_al_4 nn_et_Weischedel nsubj_al_et conj_proposed_al ' +o,2163,'Building upon the large body of research to improve tagging performance for various languages using various models -LRB- eg , -RRB- and the recent work on PCFG grammars with latent annotations , we will investigate the use of fine-grained latent annotations for Chinese POS tagging ',Matsuzaki,'dep_investigate_Building prep_Building_upon det_body_the amod_body_large pobj_upon_body prep_body_of pobj_of_research aux_improve_to xcomp_Building_improve amod_performance_tagging dobj_improve_performance prep_performance_for amod_languages_various pobj_for_languages partmod_languages_using amod_models_various dobj_using_models appos_models_eg dep_eg_ cc_models_and det_work_the amod_work_recent conj_models_work prep_using_on pcomp_on_PCFG dobj_PCFG_grammars prep_PCFG_with amod__latent nn__annotations pobj_with_ nsubj_investigate_we aux_investigate_will det_use_the dobj_investigate_use prep_use_of amod_annotations_fine-grained amod_annotations_latent pobj_of_annotations prep_investigate_for amod_POS_Chinese pobj_for_POS amod_POS_tagging ' +o,2164,'The parser has been trained , developed and tested on a large collection of syntactically analyzed sentences , the Penn Treebank ',Marcus,'det_parser_The nsubjpass_trained_parser aux_trained_has auxpass_trained_been conj_trained_developed cc_trained_and conj_trained_tested prep_tested_on det_collection_a amod_collection_large pobj_on_collection prep_collection_of amod_sentences_syntactically amod_sentences_analyzed pobj_of_sentences det_Treebank_the nn_Treebank_Penn appos_sentences_Treebank ' +o,2165,' discuss three approaches : hand-crafted rules ; grammatical inference of subsequential transducers ; and log-linear classifiers with bigram and trigram features used as taggers ',Ratnaparkhi,'advmod_discuss_ num_approaches_three dobj_discuss_approaches amod_rules_hand-crafted dep_approaches_rules amod_inference_grammatical dep_approaches_inference prep_inference_of amod_transducers_subsequential pobj_of_transducers cc_approaches_and amod_classifiers_log-linear conj_approaches_classifiers prep_discuss_with nn_features_bigram cc_bigram_and conj_bigram_trigram pobj_with_features partmod_features_used prep_used_as pobj_as_taggers ' +o,2166,'Both taggers used the Penn Treebank tagset and were trained on the Wall Street Journal corpus ',Marcus,'det_taggers_Both nsubjpass_used_taggers det_tagset_the nn_tagset_Penn nn_tagset_Treebank dobj_used_tagset cc_used_and auxpass_trained_were conj_used_trained prep_trained_on det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_on_corpus ' +o,2167,' , and et al ',Rosti,'cc__and conj__ nn_al_et dobj__al ' +o,2168,'Nevertheless , EM sometimes fails to find good parameter values2 The reason is that EM tries to assign roughly the same number of word tokens to each of the hidden states ',Johnson,'advmod_fails_Nevertheless nsubj_fails_EM advmod_fails_sometimes aux_find_to xcomp_fails_find amod_values2_good nn_values2_parameter dobj_find_values2 det_reason_The nsubj_is_reason dep_fails_is complm_tries_that nsubj_tries_EM ccomp_is_tries aux_assign_to xcomp_tries_assign advmod_number_roughly det_number_the amod_number_same dobj_assign_number prep_number_of nn_tokens_word pobj_of_tokens prep_assign_to pobj_to_each prep_each_of det_states_the amod_states_hidden pobj_of_states ' +o,2169,'21 Global linear models We follow the framework outlined in ',Collins,'num_models_21 amod_models_Global amod_models_linear nsubj_follow_We dep_models_follow det_framework_the nsubj_outlined_framework ccomp_follow_outlined prt_outlined_in ' +o,2170,'When updating model parameters , we employ a memorizationvariant of a local updating strategy in which parameters are optimized toward a set of good translations found in the k-best list across iterations ',Liang,'advmod_employ_When dep_employ_updating nn_parameters_model dobj_updating_parameters nsubj_employ_we det_memorizationvariant_a dobj_employ_memorizationvariant prep_memorizationvariant_of det__a amod__local amod__updating nn__strategy pobj_of_ rel_optimized_in pobj_in_which nsubjpass_optimized_parameters auxpass_optimized_are rcmod__optimized prep_optimized_toward det_set_a pobj_toward_set prep_set_of amod_translations_good pobj_of_translations partmod_translations_found prep_found_in det_list_the amod_list_k-best pobj_in_list prep_list_across pobj_across_iterations ' +o,2171,'We tested several measures , such as ROUGE and the cosine distance ',Lin,'nsubj_tested_We amod_measures_several dobj_tested_measures dep_as_such prep_measures_as nn__ROUGE pobj_as_ cc__and det_distance_the amod_distance_cosine conj__distance ' +o,2172,'Following extraction , O-CRF applies the RESOLVER algorithm to find relation synonyms , the various ways in which a relation is expressed in text ',Yates,'prep_applies_Following pobj_Following_extraction appos_extraction_O-CRF det__the amod__RESOLVER nn__algorithm dobj_applies_ aux_find_to xcomp_applies_find nn_synonyms_relation nsubj_applies_synonyms det_ways_the amod_ways_various appos_synonyms_ways rel_expressed_in pobj_in_which det_relation_a nsubjpass_expressed_relation auxpass_expressed_is rcmod_ways_expressed prep_expressed_in pobj_in_text ' +o,2173,'Computational approaches to prosodic modeling of DAs have aimed to automatically extract various prosodic parameters -- such as duration , pitch , and energy patterns -- from the speech signal ',Berger,'amod_approaches_Computational nsubj_aimed_approaches prep_approaches_to amod_modeling_prosodic pobj_to_modeling prep_modeling_of pobj_of_DAs aux_aimed_have aux_extract_to advmod_extract_automatically xcomp_aimed_extract amod_parameters_various amod_parameters_prosodic dobj_extract_parameters dep_as_such dep_parameters_as pobj_as_duration conj_duration_pitch cc_duration_and nn_patterns_energy conj_duration_patterns prep_parameters_from det__the nn__speech nn__signal pobj_from_ ' +o,2174,'1 Introduction A recent theme in parsing research has been the application of statistical methods to linguistically motivated grammars , for example LFG , HPSG , TAG and CCG ',Cahill,'num_theme_1 nn_theme_Introduction nn_theme_A amod_theme_recent nsubj_been_theme prep_theme_in amod_research_parsing pobj_in_research aux_been_has det_application_the dobj_been_application prep_application_of amod_methods_statistical pobj_of_methods prep_methods_to advmod_motivated_linguistically amod_grammars_motivated pobj_to_grammars prep_been_for nn__example nn__LFG pobj_for_ nn__HPSG conj__ nn__TAG conj__ cc__and nn__CCG conj__ ' +o,2175,'Furthermore , they extended WSD to phrase sense disambiguation -LRB- PSD -RRB- ',Carpuat,'advmod_extended_Furthermore nsubj_extended_they dobj_extended_WSD prep_extended_to nn_disambiguation_phrase nn_disambiguation_sense pobj_to_disambiguation abbrev_disambiguation_PSD ' +o,2176,'As in fstructures are generated from the -LRB- now altered -RRB- treebank and from this data , along with the treebank trees , the PCFG-based grammar , which is used for training the generation model , is extracted ',Cahill,'mark_generated_As prep_generated_in amod_fstructures_ pobj_in_fstructures auxpass_generated_are advcl_extracted_generated prep_generated_from amod_treebank_the dep_altered_now dep_the_altered pobj_from_treebank cc_from_and conj_from_from det_data_this pobj_from_data advmod_extracted_along dep_along_with det_trees_the nn_trees_treebank pobj_with_trees det_grammar_the amod_grammar_PCFG-based nsubjpass_extracted_grammar nsubjpass_used_which auxpass_used_is rcmod_grammar_used prep_used_for pcomp_for_training det_model_the nn_model_generation dobj_training_model auxpass_extracted_is ' +o,2177,'We also test an MI model inspired by Erk : MISIM -LRB- n , v -RRB- = log summationdisplay nSIMS -LRB- n -RRB- Sim -LRB- n , n -RRB- Pr -LRB- v , n -RRB- Pr -LRB- v -RRB- Pr -LRB- n -RRB- We gather similar words using , mining similar verbs from a comparable-sized parsed corpus , and collecting similar nouns from a broader 10 GB corpus of English text4 We also use Keller and Lapata s approach to obtaining web-counts ',Lin,'nsubj_test_We advmod_test_also det_model_an nn_model_MI dobj_test_model partmod_model_inspired prep_inspired_by nn__Erk pobj_by_ nsubj_=_MISIM appos_MISIM_n dep_n_v rcmod_model_= nn_nSIMS_log nn_nSIMS_summationdisplay dobj_=_nSIMS appos_nSIMS_n nn_Pr_Sim appos_Pr_n dep_n_n prep_nSIMS_Pr appos_Pr_v dep_v_n nn_Pr_Pr appos_Pr_v dep_Pr_Pr appos_Pr_n nsubj_gather_We rcmod_Pr_gather amod_words_similar dobj_gather_words dep_gather_using dobj_using_ conj_using_mining amod_verbs_similar dobj_mining_verbs prep_mining_from det_corpus_a amod_corpus_comparable-sized amod_corpus_parsed pobj_from_corpus cc_using_and conj_using_collecting amod_nouns_similar dobj_collecting_nouns prep_=_from det_corpus_a amod_corpus_broader dep_GB_10 amod_corpus_GB pobj_from_corpus prep_corpus_of amod_text4_English pobj_of_text4 nsubj_use_We advmod_use_also dep_test_use nn__Keller cc_Keller_and conj_Keller_Lapata nsubj_approach_ dep_approach_s ccomp_use_approach prep_approach_to pcomp_to_obtaining dobj_obtaining_web-counts ' +o,2178,'The search also uses a Tag Dictionary constructed from training data , described in , that reduces the number of actions explored by the tagging model ',Ratnaparkhi,'det_search_The nsubj_uses_search advmod_uses_also det_Dictionary_a nn_Dictionary_Tag dobj_uses_Dictionary partmod_Dictionary_constructed prep_constructed_from nn_data_training pobj_from_data partmod_data_described prep_described_in pobj_in_ nsubj_reduces_that rcmod_data_reduces det_number_the nsubj_explored_number prep_number_of pobj_of_actions ccomp_reduces_explored prep_explored_by det_model_the amod_model_tagging pobj_by_model ' +o,2179,'~ gtPdl= |&.allm~WI.Lqlf IDW,t~lIO, r I~\'1~~ ~ II, Mlmulm, IP, il~,,lllb, l~ ~ I I I I I I I I I 0 200 400 600 800 1000 1200 1400 1600 1800 Article# 2000 Figure 1: Distribution of Tags for the word \'about\' vs. Article# Training Size(wrds)I Test571190 Size(wrds) I Baseline44478 97.04\% Specialized 197.13\% Table 10: Performance of Baseline ~ Specialized Model When Tested on Consistent Subset of Development Set 139 POS Tag 35 30 25 2O 15 10 5 0 1 I o. Oho m I I I B ~ m M I I I 2 3 4 Annotator Figure 2: Distribution of Tags for the word \'about\' vs. Annotator (Weischedel et al. , 1993) provide the results from a battery of \'tri-tag\' Markov Model experiments, in which the probability P(W,T) of observing a word sequence W = {wl,w2,,wn} together with a tag sequence T = {tl,t2,,tn} is given by: P(TIW)P(W) p(tl)p(t21tl) H P(tilti-lti-2) p(wilti i=3 Furthermore, p(wilti) for unknown words is computed by the following heuristic, which uses a set of 35 pre-determined endings: p(wilti) p(unknownwordlti ) x p(capitalfeature[ti) x p(endings, hypenationlti ) This approximation works as well as the MaxEnt model, giving 85\% unknown word accuracy(Weischedel et al. , 1993) on the Wall St. Journal, but cannot be generalized to handle more diverse information sources.',Marcus,'' +o,2180,'Phrase pairs are extracted up to a fixed maximum length , since very long phrases rarely have a tangible impact during translation ',Koehn,'amod_pairs_Phrase nsubjpass_extracted_pairs auxpass_extracted_are prt_extracted_up prep_extracted_to det_length_a amod_length_fixed amod_length_maximum pobj_to_length mark_have_since advmod_long_very amod_phrases_long nsubj_have_phrases advmod_have_rarely dep_length_have det_impact_a amod_impact_tangible dobj_have_impact prep_have_during pobj_during_translation ' +o,2181,'The resulting model has an exponential form with free parameters a102 a91 a24a94a93 a8 a87 a24 a10a11a10a11a10 a24a46a95 The parameter values which maximize the likelihood for a given training corpus can be computed with the socalled GIS algorithm -LRB- general iterative scaling -RRB- or its improved version IIS ',Berger,'det_model_The amod_model_resulting nsubj_has_model det_form_an amod_form_exponential dobj_has_form prep_form_with amod_a24a46a95_free nn_a24a46a95_parameters nn_a24a46a95_a102 nn_a24a46a95_a91 nn_a24a46a95_a24a94a93 nn_a24a46a95_a8 nn_a24a46a95_a87 nn_a24a46a95_a24 amod_a24a46a95_a10a11a10a11a10 pobj_with_a24a46a95 det_values_The nn_values_parameter nsubjpass_computed_values nsubj_maximize_which rcmod_values_maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_for det_corpus_a amod_corpus_given nn_corpus_training pobj_for_corpus aux_computed_can auxpass_computed_be dep_has_computed prep_computed_with det_algorithm_the amod_algorithm_socalled nn_algorithm_GIS pobj_with_algorithm amod_scaling_general nn_scaling_iterative appos_algorithm_scaling cc_algorithm_or poss_IIS_its amod_IIS_improved nn_IIS_version conj_algorithm_IIS ' +o,2182,'0 005 01 015 02 025 03 035 04 45 50 55 60 65 70 75 80 85Correlation Coefficient with Human Judgement -LRB- R -RRB- Human-Likeness Classifier Accuracy -LRB- \% -RRB- Figure 1 : This scatter plot compares classifiers accuracy with their corresponding metrics correlations with human assessments been previously observed by ',Liu,'num_Coefficient_0 num_Coefficient_005 num_Coefficient_01 num_Coefficient_015 num_Coefficient_02 num_Coefficient_025 num_Coefficient_03 num_Coefficient_035 num_Coefficient_04 num_Coefficient_45 num_Coefficient_50 num_Coefficient_55 num_Coefficient_60 num_Coefficient_65 num_Coefficient_70 num_Coefficient_75 num_Coefficient_80 nn_Coefficient_85Correlation nsubj_correlations_Coefficient prep_Coefficient_with nn_Judgement_Human nn_Figure_Judgement appos_Figure_R nn_Figure_Human-Likeness nn_Figure_Classifier nn_Figure_Accuracy appos_Figure_\% pobj_with_Figure dep_Figure_1 det_plot_This nn_plot_scatter nsubj_compares_plot dep_Figure_compares amod_accuracy_classifiers dobj_compares_accuracy prep_compares_with poss_metrics_their amod_metrics_corresponding pobj_with_metrics prep_correlations_with amod_assessments_human pobj_with_assessments auxpass_observed_been advmod_observed_previously partmod_assessments_observed prep_observed_by ' +o,2183,'Two error rates : the sentence error rate -LRB- SER -RRB- and the word error rate -LRB- WER -RRB- that we seek to minimize , and BLEU , that we seek to maximize ',Papineni,'num_rates_Two nn_rates_error det_rate_the nn_rate_sentence nn_rate_error dep_rates_rate abbrev_rate_SER cc_rate_and det_rate_the nn_rate_word nn_rate_error conj_rate_rate abbrev_rate_WER complm_seek_that nsubj_seek_we dep_rates_seek aux_minimize_to xcomp_seek_minimize cc_minimize_and advmod__BLEU conj_minimize_ dep_seek_that nsubj_seek_we ccomp__seek aux_maximize_to xcomp_seek_maximize ' +o,2184,'We report precision , recall and balanced F-measure ',Och,'nsubj_report_We nn_recall_precision nn_F-measure_recall cc_recall_and conj_recall_balanced dobj_report_F-measure ' +o,2185,'51 Pharaoh The baseline system we used for comparison was Pharaoh , a freely available decoder for phrase-based translation models : p -LRB- e f -RRB- = p -LRB- f e -RRB- pLM -LRB- e -RRB- LM pD -LRB- e , f -RRB- D length -LRB- e -RRB- W -LRB- e -RRB- -LRB- 10 -RRB- We ran GIZA + + on the training corpus in both directions using its default setting , and then applied the refinement rule diagand described in to obtain a single many-to-many word alignment for each sentence pair ',Och,'num_Pharaoh_51 det_system_The amod_system_baseline nsubj__system nsubj_used_we rcmod_system_used prep_used_for pobj_for_comparison cop__was nn__Pharaoh dep_Pharaoh_ det_decoder_a advmod_available_freely amod_decoder_available appos__decoder prep_decoder_for amod_models_phrase-based nn_models_translation pobj_for_models dep_decoder_p dep_f_e dep_-LRB-_f rcmod_p_= amod_W_p dep_-LRB-_f dep_f_e amod_W_pLM dep_-LRB-_e nn_W_LM amod_W_pD dep_W_e appos_e_f nn_W_D nn_W_length dep_-LRB-_e dobj_=_W dep_-LRB-_e appos_W_10 nsubj_ran_We rcmod_W_ran nsubj_+_GIZA xcomp_ran_+ cc_+_+ conj_+_ prep_=_on det_corpus_the nn_corpus_training pobj_on_corpus prep_corpus_in det_directions_both pobj_in_directions dep_Pharaoh_using poss_setting_its nn_setting_default dobj_using_setting cc_using_and advmod_using_then conj_using_applied det_diagand_the amod_diagand_refinement nn_diagand_rule dobj_applied_diagand partmod_diagand_described prep_described_in pobj_in_ aux_obtain_to xcomp_applied_obtain det_alignment_a amod_alignment_single amod_alignment_many-to-many nn_alignment_word dobj_obtain_alignment prep_alignment_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,2186,'The mapping of answer terms to question terms is modeled using Black et als simplest model , called IBM Model 1 ',Brown,'det_mapping_The nsubjpass_modeled_mapping prep_mapping_of nn_terms_answer pobj_of_terms prep_terms_to nn_terms_question pobj_to_terms auxpass_modeled_is ccomp_called_modeled xcomp_modeled_using nn_model_Black cc_Black_et conj_Black_als nn_model_ nn_model_simplest dobj_using_model nn_Model_IBM dobj_called_Model nsubj_called_1 ' +o,2187,'Recent research shows that using different clusters for predicted and conditional words can lead to cluster models that are superior to classical cluster models , which use the same clusters for both words ',Brown,'amod__Recent nn__research nsubj_shows_ complm_lead_that csubj_lead_using amod_clusters_different dobj_using_clusters prep_using_for amod_words_predicted cc_predicted_and conj_predicted_conditional pobj_for_words aux_lead_can ccomp_shows_lead prep_lead_to nn_models_cluster pobj_to_models nsubj_superior_that cop_superior_are rcmod_models_superior prep_superior_to amod_models_classical nn_models_cluster pobj_to_models nsubj_use_which rcmod_models_use det_clusters_the amod_clusters_same dobj_use_clusters prep_clusters_for det_words_both pobj_for_words ' +o,2188,'1 Introduction Recent trends in machine translation illustrate that highly accurate word and phrase translations can be learned automatically given enough parallel training data ',Koehn,'num_trends_1 nn_trends_Introduction amod_trends_Recent nsubj_illustrate_trends prep_trends_in nn_translation_machine pobj_in_translation complm_learned_that advmod_accurate_highly amod_word_accurate nsubjpass_learned_word cc_word_and nn_translations_phrase conj_word_translations aux_learned_can auxpass_learned_be ccomp_illustrate_learned dep_given_automatically prep_learned_given amod_data_enough amod_data_parallel nn_data_training dep_given_data ' +o,2189,'In the SUMMAC experiments , the Kappa score for interannotator agreement was reported to be 038 ',Carletta,'prep_reported_In det_experiments_the nn_experiments_SUMMAC pobj_In_experiments det__the nn__Kappa nn__score nsubjpass_reported_ prep__for nn_agreement_interannotator pobj_for_agreement auxpass_reported_was aux__to cop__be number__038 xcomp_reported_ ' +p,2190,'21 Lexicalized parse trees The first successful work on syntactic disambiguation was based on lexicalized probabilistic context-free grammar -LRB- LPCFG -RRB- ',Collins,'num_trees_21 nn_trees_Lexicalized amod_trees_parse det_work_The amod_work_first amod_work_successful nsubjpass_based_work prep_work_on amod_disambiguation_syntactic pobj_on_disambiguation auxpass_based_was dep_trees_based prep_based_on amod_grammar_lexicalized dep_lexicalized_probabilistic amod_grammar_context-free pobj_on_grammar abbrev_grammar_LPCFG ' +o,2191,'It has been shown by Shapiro and Stephens and -LRB- 1997 , Sec ',Wu,'nsubjpass_shown_It aux_shown_has auxpass_shown_been prep_shown_by pobj_by_Shapiro cc_Shapiro_and nn__Stephens conj_Shapiro_ cc__and nn_Sec_ nn_Sec_1997 conj__Sec ' +o,2192,'Alternatively , one can train them with respect to the final translation quality measured by an error criterion ',Och,'advmod_train_Alternatively nsubj_train_one aux_train_can dobj_train_them prep_train_with pobj_with_respect prep_train_to det_quality_the amod_quality_final nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_by det_criterion_an nn_criterion_error pobj_by_criterion ' +o,2193,'First , we need to determine whether or not the positive effect of SVD feature selection is preserved in more complex feature spaces such as syntactic feature spaces as those used in ',Snow,'advmod_need_First nsubj_need_we aux_determine_to xcomp_need_determine complm_preserved_whether cc_preserved_or conj_preserved_not det_effect_the amod_effect_positive nsubjpass_preserved_effect prep_effect_of nn_selection_SVD nn_selection_feature pobj_of_selection auxpass_preserved_is ccomp_determine_preserved prep_preserved_in advmod_complex_more amod_spaces_complex nn_spaces_feature pobj_in_spaces dep_as_such prep_spaces_as amod_spaces_syntactic nn_spaces_feature pobj_as_spaces prep_preserved_as pobj_as_those partmod_those_used prep_used_in ' +o,2194,'Och showed thatsystemperformanceisbestwhenparametersare optimizedusingthesameobjectivefunctionthatwill be used for evaluation ; BLEU remains common for both purposes and is often retained for parameter optimization even when alternative evaluation measures are used , eg , ',Banerjee,'nsubj_showed_Och amod_optimizedusingthesameobjectivefunctionthatwill_thatsystemperformanceisbestwhenparametersare nsubjpass_used_optimizedusingthesameobjectivefunctionthatwill auxpass_used_be ccomp_showed_used prep_used_for pobj_for_evaluation nn__BLEU nsubjpass_common_ cop_common_remains parataxis_showed_common prep_common_for det_purposes_both pobj_for_purposes cc_common_and auxpass_retained_is advmod_retained_often conj_common_retained prep_retained_for nn_optimization_parameter pobj_for_optimization advmod_used_even advmod_used_when amod_measures_alternative nn_measures_evaluation nsubjpass_used_measures auxpass_used_are ccomp_retained_used dep_used_eg ' +o,2195,'2 Phrasal Inversion Transduction Grammar We use a phrasal extension of Inversion Transduction Grammar as the generative framework ',Wu,'num_Grammar_2 nn_Grammar_Phrasal nn_Grammar_Inversion nn_Grammar_Transduction nsubj_use_We dep_Grammar_use det_extension_a amod_extension_phrasal dobj_use_extension prep_extension_of nn__Inversion nn__Transduction nn__Grammar pobj_of_ prep__as det_framework_the amod_framework_generative pobj_as_framework ' +o,2196,'6 Training Similar to most state-of-the-art phrase-based SMT systems , we use the SRI toolkit for language model training and Giza + + toolkit for word alignment ',Och,'dep_use_Training acomp_Training_Similar prep_Similar_to advmod_systems_most amod_systems_state-of-the-art nn_systems_phrase-based nn_systems_SMT pobj_to_systems nsubj_use_we dep_6_use det__the nn__SRI nn__toolkit dobj_use_ prep__for nn_training_language nn_training_model pobj_for_training cc__and amod__Giza cc_Giza_+ conj_Giza_+ nn__toolkit conj__ prep__for nn_alignment_word pobj_for_alignment ' +p,2197,'32 Statistical Learning Model 321 Nave Bayes Learning Nave Bayes learning has been widely used in natural language processing with good results such as statistical syntactic parsing , hidden language understanding ',Collins,'num_learning_32 nn_learning_Statistical nn_learning_Learning nn_learning_Model nn_learning_321 nn_learning_Nave nn_learning_Bayes nn_learning_Learning nn_learning_Nave nn_learning_Bayes nsubjpass_used_learning aux_used_has auxpass_used_been advmod_used_widely prep_used_in amod_processing_natural nn_processing_language pobj_in_processing prep_processing_with amod_results_good pobj_with_results dep_as_such prep_results_as amod__statistical amod__syntactic nn__parsing pobj_as_ amod__hidden nn__language nn__understanding appos__ ' +o,2198,'We chose to train maximum entropy models ',Berger,'nsubj_chose_We aux_train_to xcomp_chose_train amod_models_maximum nn_models_entropy dobj_train_models ' +o,2199,'Because of this , and Zens and Ney -LRB- 2003 -RRB- introduced a normal form ITG which avoids this over-counting ',Wu,'prep_introduced_Because dep_Because_of pobj_of_this nsubj_introduced_ cc__and conj__Zens cc__and conj__Ney appos_Ney_2003 det_ITG_a amod_ITG_normal nn_ITG_form dobj_introduced_ITG nsubj_avoids_which rcmod_ITG_avoids det_over-counting_this dobj_avoids_over-counting ' +o,2200,'The second approximation proposed in takes into consideration the fact that , after each decision is made , all the preceding latent variables should have their means i updated ',Titov,'det_approximation_The amod_approximation_second nsubj_proposed_approximation dep_takes_proposed prep_proposed_in pobj_in_ prep_takes_into pobj_into_consideration det_fact_the dobj_takes_fact dobj_have_that mark_made_after det_decision_each nsubjpass_made_decision auxpass_made_is advcl_have_made predet_variables_all det_variables_the amod_variables_preceding amod_variables_latent nsubj_have_variables aux_have_should rcmod_fact_have poss_means_their dobj_have_means nn_updated_i nsubj_takes_updated ' +o,2201,'It is also possible to train statistical models using unlabeled data with the expectation maximization algorithm ',Cutting,'nsubj_possible_It cop_possible_is advmod_possible_also aux_train_to xcomp_possible_train amod_models_statistical dobj_train_models xcomp_train_using amod_data_unlabeled dobj_using_data prep_using_with det_algorithm_the nn_algorithm_expectation nn_algorithm_maximization pobj_with_algorithm ' +o,2202,'44 Related Work implemented an MEMM model for supertagging which is analogous to the POS tagging model of ',Ratnaparkhi,'num__44 amod__Related nn__Work partmod__implemented det_model_an nn_model_MEMM dobj_implemented_model prep_implemented_for pcomp_for_supertagging nsubj_analogous_which cop_analogous_is ccomp_supertagging_analogous prep_analogous_to det_model_the dep_model_POS amod_model_tagging pobj_to_model dep__of ' +o,2203,'24 Maximum Entropy Classifier Maximum Entropy Models seek to maximize the conditional probability of classes , given certain observations -LRB- features -RRB- ',Berger,'number_Maximum_24 num__Maximum nn__Entropy nn__Classifier nn__Maximum nn__Entropy nn__Models nsubj_seek_ aux_maximize_to xcomp_seek_maximize det_probability_the amod_probability_conditional dobj_maximize_probability prep_probability_of pobj_of_classes amod_observations_given amod_observations_certain appos_classes_observations appos_observations_features ' +n,2204,'Our system outperforms competing approaches , including the standard machine translation alignment models and the state-of-the-art Cut and Paste summary alignment technique ',Brown,'poss_system_Our nsubj_outperforms_system amod_approaches_competing dobj_outperforms_approaches prep_approaches_including det_translation_the amod_translation_standard nn_translation_machine pobj_including_translation amod__alignment nn__models dep_translation_ cc_translation_and det_Cut_the amod_Cut_state-of-the-art conj_translation_Cut cc_Cut_and nn__Paste nn__summary nn__alignment nn__technique conj_Cut_ ' +o,2205,'3 Maximum Entropy Taggers The taggers are based on Maximum Entropy tagging methods , and can all be trained on new annotated data , using either GIS or BFGS training code ',Ratnaparkhi,'num_Taggers_3 amod_Taggers_Maximum nn_Taggers_Entropy det_taggers_The nsubjpass_based_taggers auxpass_based_are dep_Taggers_based prep_based_on nn_Entropy_Maximum pobj_on_Entropy amod_methods_tagging nsubjpass__methods dep_Taggers_ cc__and aux_trained_can advmod_trained_all auxpass_trained_be conj__trained prep_trained_on amod_data_new amod_data_annotated pobj_on_data conj__using preconj_GIS_either dobj_using_GIS cc_GIS_or nn_code_BFGS amod_code_training conj_GIS_code ' +p,2206,'1 Introduction IBM Model 1 is a wordalignment model that is widely used in working with parallel bilingual corpora ',Brown,'num__1 nn__Introduction nn__IBM nn__Model num__1 nsubj_model_ cop_model_is det_model_a amod_model_wordalignment nsubjpass_used_that auxpass_used_is advmod_used_widely rcmod_model_used prep_used_in pcomp_in_working prep_working_with amod_corpora_parallel amod_corpora_bilingual pobj_with_corpora ' +o,2207,'The fact that the error rate more than doubles when the seeds in experiments are reduced from a sense \'s best collocations to just one word per sense suggests that the error rate would increase further if no seeds were provided ',Yarowsky,'det_fact_The nsubj_suggests_fact complm_doubles_that det_rate_the nn_rate_error nsubj_doubles_rate advmod_doubles_more dep_more_than dep_fact_doubles advmod_reduced_when det_seeds_the nsubjpass_reduced_seeds prep_seeds_in amod_experiments_ pobj_in_experiments auxpass_reduced_are advcl_doubles_reduced prep_reduced_from det_sense_a poss_collocations_sense possessive_sense_\'s amod_collocations_best pobj_from_collocations prep_reduced_to advmod_word_just num_word_one pobj_to_word prep_word_per pobj_per_sense complm_increase_that det_rate_the nn_rate_error nsubj_increase_rate aux_increase_would ccomp_suggests_increase advmod_increase_further mark_provided_if det_seeds_no nsubjpass_provided_seeds auxpass_provided_were dep_further_provided ' +o,2208,'Any linguistic annotation required during the extraction process , therefore , is produced through automatic means , and it is only for reasons of accessibility and comparability with other research that we choose to work over the Wall Street Journal section of the Penn Treebank ',Marcus,'det_annotation_Any amod_annotation_linguistic nsubjpass_produced_annotation partmod_annotation_required prep_required_during det_process_the nn_process_extraction pobj_during_process advmod_produced_therefore auxpass_produced_is prep_produced_through amod_means_automatic pobj_through_means cc_produced_and nsubj_is_it conj_produced_is advmod_is_only prep_is_for pobj_for_reasons prep_reasons_of pobj_of_accessibility cc_accessibility_and conj_accessibility_comparability prep_accessibility_with amod_research_other pobj_with_research complm_choose_that nsubj_choose_we ccomp_is_choose aux_work_to xcomp_choose_work prt_work_over det_section_the nn_section_Wall nn_section_Street nn_section_Journal dobj_work_section prep_section_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,2209,'-LRB- General grammars with infinite numbers of nonterminals were studied by -RRB- ',Liang,'nsubj_grammars_General mark_studied_with amod_numbers_infinite nsubjpass_studied_numbers prep_numbers_of pobj_of_nonterminals auxpass_studied_were advcl_grammars_studied prep_studied_by pobj_by_ ' +o,2210,'Such tasks will require an extension of the current framework of beyond evidence from the direct cooccurrence of target word pairs ',Turney,'amod_tasks_Such nsubj_require_tasks aux_require_will det_extension_an dobj_require_extension prep_extension_of det_framework_the amod_framework_current pobj_of_framework prep_framework_of pobj_of_ prep_require_beyond pobj_beyond_evidence prep_require_from det_cooccurrence_the amod_cooccurrence_direct pobj_from_cooccurrence prep_cooccurrence_of nn_pairs_target nn_pairs_word pobj_of_pairs ' +o,2211,'51 Baseline System We trained Moses on all Spanish-English Europarl sentences up to length 20 -LRB- 177k sentences -RRB- using GIZA + + Model 4 word alignments and the growdiag-final-and combination heuristic , which performed better than any alternative combination heuristic13 The baseline estimates -LRB- Heuristic -RRB- come fromextractingphrasesuptolength7fromtheword alignment ',Och,'num_System_51 nn_System_Baseline nsubj_trained_We dep_System_trained dobj_trained_Moses prep_Moses_on det_sentences_all nn_sentences_Spanish-English nn_sentences_Europarl pobj_on_sentences prt_trained_up prep_trained_to pobj_to_length nsubj_using_20 num_sentences_177k appos_20_sentences dep_System_using nn_alignments_GIZA nn_alignments_+ nn_alignments_+ nn_alignments_Model num_alignments_4 nn_alignments_word dobj_using_alignments cc_alignments_and det_combination_the amod_combination_growdiag-final-and conj_alignments_combination amod__heuristic tmod_using_ nsubj_performed_which dep__performed advmod_performed_better dep_better_than det_heuristic13_any amod_heuristic13_alternative nn_heuristic13_combination pobj_than_heuristic13 det_estimates_The nn_estimates_baseline dobj_performed_estimates dep_performed_Heuristic dep_performed_come amod_alignment_fromextractingphrasesuptolength7fromtheword dobj_come_alignment ' +o,2212,'A Head Percolation Table has previously been used in several statistical parsers to find heads of phrases ',Collins,'det_Table_A nn_Table_Head nn_Table_Percolation nsubjpass_used_Table aux_used_has advmod_used_previously auxpass_used_been prep_used_in amod_parsers_several amod_parsers_statistical pobj_in_parsers partmod_parsers_ aux_find_to xcomp__find dobj_find_heads prep_heads_of pobj_of_phrases ' +n,2213,'For example , we would like to know that if a -LRB- JJ , JJ -RRB- 7We also tried using word clusters instead of POS but found that POS was more helpful ',Brown,'prep_like_For pobj_For_example nsubj_like_we aux_like_would aux_know_to xcomp_like_know complm_tried_that dep_tried_if det_7We_a appos_7We_JJ dep_JJ_JJ nsubj_tried_7We advmod_tried_also ccomp_know_tried xcomp_tried_using nn__word nn__clusters dobj_using_ dep_of_instead prep__of pobj_of_POS cc_tried_but conj_tried_found complm_helpful_that nsubj_helpful_POS cop_helpful_was advmod_helpful_more ccomp_found_helpful ' +o,2214,'Numerous approaches have been explored for exploiting situations where some amount of annotated data is available and a much larger amount of data exists unannotated , eg Marialdo \'s HMM part-of-speech tagger training , Charniak \'s parser retraining experiment , and Nigam et al \'s -LRB- 1998 -RRB- topic classifier learned in part from unlabelled documents ',Yarowsky,'amod_approaches_Numerous nsubj_explored_approaches aux_explored_have dep_explored_been prep_explored_for pcomp_for_exploiting dobj_exploiting_situations advmod_available_where det_amount_some nsubj_available_amount prep_amount_of amod_data_annotated pobj_of_data cop_available_is advcl_explored_available cc_available_and det_amount_a advmod_larger_much amod_amount_larger nsubj_exists_amount prep_amount_of pobj_of_data conj_available_exists dobj_exists_unannotated nn_Marialdo_eg poss__Marialdo possessive_Marialdo_\'s nn__HMM amod__part-of-speech nn__tagger nn__training conj_unannotated_ poss_experiment_Charniak possessive_Charniak_\'s nn_experiment_parser amod_experiment_retraining conj_unannotated_experiment partmod_experiment_ conj_unannotated_ cc_unannotated_and poss_classifier_Nigam cc_Nigam_et conj_Nigam_al possessive_Nigam_\'s appos_classifier_1998 nn_classifier_topic conj_unannotated_classifier partmod_classifier_learned prep_learned_in pobj_in_part prep_learned_from amod_documents_unlabelled pobj_from_documents ' +o,2215,'2 Inversion transduction grammars Inversion transduction grammars (ITGs) (Wu, 1997) are a notational variant of binary syntax-directed translation schemas (Aho and Ullman, 1972) and are usually presented with a normal form: A [BC] A BC A e|f A e| A |f where A,B,C N and e,f T. The first production rule, intuitively, says that the subtree [[]B[]C]A in the source language translates into 62 a subtree [[]B[]C]A, whereas the second production rule inverts the order in the target language, i.e. [[]C[]B]A. The universal recognition problem of ITGs can be solved in time O(n6|G|) by a CYKstyle parsing algorithm with two charts.',Wu,'' +o,2216,'Tuning -LRB- learning the values discussed in section 41 -RRB- was done using minimum error rate training ',Och,'nsubjpass_done_Tuning dep_Tuning_learning det_values_the dobj_learning_values partmod_values_discussed prep_discussed_in pobj_in_section num_section_41 auxpass_done_was xcomp_done_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +o,2217,'In NLP community , it has been shown that having more data results in better performance ',Brants,'prep_shown_In nn_community_NLP pobj_In_community nsubjpass_shown_it aux_shown_has auxpass_shown_been complm_having_that ccomp_shown_having amod_results_more nn_results_data dobj_having_results prep_results_in amod_performance_better pobj_in_performance ' +o,2218,' use a syntaxbased distance in an HMM word alignment model to favor syntax-friendly alignments ',DeNero,'nsubj_use_ det_distance_a amod_distance_syntaxbased nsubj_favor_distance prep_distance_in det_model_an nn_model_HMM nn_model_word nn_model_alignment pobj_in_model aux_favor_to xcomp_use_favor amod_alignments_syntax-friendly dobj_favor_alignments ' +n,2219,'This is because their training data , the Penn Treebank , does not fully annotate NP structure ',Marcus,'nsubj_is_This mark_annotate_because poss_data_their nn_data_training nsubj_annotate_data det__the nn__Penn nn__Treebank appos_data_ aux_annotate_does neg_annotate_not advmod_annotate_fully advcl_is_annotate nn_structure_NP dobj_annotate_structure ' +o,2220,' , these models have non-uniform linguistically motivated structure , at present coded by hand ',Dunning,'advmod_have_ det_models_these nsubj_have_models amod_structure_non-uniform advmod_motivated_linguistically amod_structure_motivated dobj_have_structure prep_have_at pobj_at_present partmod_present_coded prep_coded_by pobj_by_hand ' +o,2221,'Additional evidence for this distinction is given in Pustejovsky and Anick and Briscoe et al ',Hobbs,'amod_evidence_Additional nsubjpass_given_evidence prep_evidence_for det_distinction_this pobj_for_distinction auxpass_given_is prep_given_in nn__Pustejovsky cc_Pustejovsky_and conj_Pustejovsky_Anick pobj_in_ cc__and conj__Briscoe cc_Briscoe_et conj_Briscoe_al ' +p,2222,'Such methods have also been a key driver of progress in statistical machine translation , which depends heavily on unsupervised word alignments ',Brown,'amod_methods_Such nsubj_been_methods aux_been_have advmod_been_also det_driver_a amod_driver_key dobj_been_driver prep_driver_of pobj_of_progress prep_progress_in amod_translation_statistical nn_translation_machine pobj_in_translation nsubj_depends_which rcmod_translation_depends advmod_depends_heavily prep_depends_on amod_alignments_unsupervised nn_alignments_word pobj_on_alignments ' +p,2223,' improves the F1 score from 882 \% to 897 \% , while Charniak and Johnson -LRB- 2005 -RRB- improve from 903 \% to 914 \% ',Collins,'nsubj_improves_ det_score_the nn_score_F1 dobj_improves_score prep_improves_from number_\%_882 dep_\%_\% dep_\%_to number_\%_897 pobj_from_\% mark_improve_while nsubj_improve_Charniak cc_Charniak_and conj_Charniak_Johnson appos_Charniak_2005 advcl_improves_improve prep_improve_from number_\%_903 dep_\%_\% dep_\%_to number_\%_914 pobj_from_\% ' +o,2224,'22 Perceptron algorithm Our discriminative n-gram model training approach uses the perceptron algorithm , as presented in , which follows the general approach presented in ',Collins,'num_algorithm_22 amod_algorithm_Perceptron poss_approach_Our amod_approach_discriminative amod_approach_n-gram nn_approach_model nn_approach_training nsubj_uses_approach dep_algorithm_uses det_algorithm_the nn_algorithm_perceptron dobj_uses_algorithm advmod_presented_as partmod_algorithm_presented prep_presented_in pobj_in_ nsubj_follows_which rcmod_algorithm_follows det_approach_the amod_approach_general dobj_follows_approach partmod_approach_presented prep_presented_in ' +o,2225,'As expected , Malt and MST have very similar accuracy for short sentences but Malt degrades more rapidly with increasing sentence length because of error propagation ',McDonald,'mark_expected_As advcl_have_expected nsubj_have_Malt cc_Malt_and conj_Malt_MST advmod_similar_very amod_accuracy_similar dobj_have_accuracy prep_accuracy_for amod_sentences_short pobj_for_sentences cc_have_but nsubj_degrades_Malt conj_have_degrades advmod_rapidly_more advmod_degrades_rapidly prep_degrades_with amod_length_increasing nn_length_sentence pobj_with_length dep_of_because prep_length_of nn_propagation_error pobj_of_propagation ' +o,2226,' explores whether lexical reordering or the phrase discontiguity inherent in hierarchical rules explains improvements over phrase-based systems ',Lopez,'advmod_explores_ dep_explains_whether amod_reordering_lexical nsubj_explains_reordering cc_reordering_or det_phrase_the conj_reordering_phrase advmod_inherent_discontiguity amod_phrase_inherent prep_phrase_in amod_rules_hierarchical pobj_in_rules dep_explores_explains dobj_explains_improvements prep_improvements_over amod_systems_phrase-based pobj_over_systems ' +p,2227,'In , anotherstate-of-the-artWSDengine -LRB- acombination of naive Bayes , maximum entropy , boosting and Kernel PCA models -RRB- is used to dynamically determine the score of a phrase pair under consideration and , thus , let the phrase selection adapt to the context of the sentence ',Carpuat,'prep_used_In pobj_In_ nsubjpass_used_anotherstate-of-the-artWSDengine dep_anotherstate-of-the-artWSDengine_acombination prep_acombination_of amod_models_naive tmod_naive_Bayes amod_entropy_maximum dep_Bayes_entropy dep_Bayes_boosting dep_Bayes_and nn_PCA_Kernel dep_Bayes_PCA pobj_of_models auxpass_used_is aux_determine_to advmod_determine_dynamically xcomp_used_determine det_score_the dobj_determine_score prep_score_of det_pair_a nn_pair_phrase pobj_of_pair prep_pair_under pobj_under_consideration cc_determine_and dep_determine_thus conj_determine_let det_selection_the nn_selection_phrase nsubj_adapt_selection ccomp_let_adapt prep_adapt_to det_context_the pobj_to_context prep_context_of det_sentence_the pobj_of_sentence ' +o,2228,'Clearly the present research task is quite considerably harder than the parsing and tagging tasks undertaken in , which would seem to be the closest work to ours , and any comparison between this work and ours must be approached with extreme caution ',Marcus,'advmod_harder_Clearly det_task_the amod_task_present nn_task_research nsubj_harder_task cop_harder_is advmod_harder_quite advmod_harder_considerably prep_harder_than det_tasks_the nn_tasks_parsing cc_parsing_and conj_parsing_tagging pobj_than_tasks partmod_tasks_undertaken prep_undertaken_in pobj_in_ nsubj_seem_which aux_seem_would rcmod__seem aux_work_to cop_work_be det_work_the amod_work_closest xcomp_seem_work prep_work_to pobj_to_ours cc_harder_and det_comparison_any nsubjpass_approached_comparison prep_comparison_between det_work_this pobj_between_work cc_work_and conj_work_ours aux_approached_must auxpass_approached_be conj_harder_approached prep_approached_with amod_caution_extreme pobj_with_caution ' +o,2229,'3 A Categorization of Block Styles In , multi-word cepts -LRB- which are realized in our block concept -RRB- are discussed and the authors state that when a target sequence is sufficiently different from a word by word translation , only then should the target sequence should be promoted to a cept ',Brown,'dep_Styles_3 det_Categorization_A nsubj_Styles_Categorization prep_Categorization_of pobj_of_Block prep_discussed_In pobj_In_ amod_cepts_multi-word nsubjpass_discussed_cepts nsubjpass_realized_which auxpass_realized_are dep_cepts_realized prep_realized_in poss_concept_our nn_concept_block pobj_in_concept auxpass_discussed_are dep_Styles_discussed cc_discussed_and det_authors_the nsubj_state_authors conj_discussed_state complm_should_that advmod_different_when det_sequence_a nn_sequence_target nsubj_different_sequence cop_different_is advmod_different_sufficiently dep_should_different prep_different_from det_word_a pobj_from_word prep_word_by nn_translation_word pobj_by_translation advmod_then_only advmod_should_then ccomp_state_should det_sequence_the nn_sequence_target nsubjpass_promoted_sequence aux_promoted_should auxpass_promoted_be dep_Styles_promoted prep_promoted_to det_cept_a pobj_to_cept ' +o,2230,'5http : \\/ \\/ wwwstatmtorg\\/wmt08 185 the BLEU score , and tested on test2008 ',Papineni,'dep_5http_\\/ amod_wwwstatmtorg\\/wmt08_\\/ dobj_\\/_wwwstatmtorg\\/wmt08 xcomp_\\/_185 det__the amod__BLEU nn__score dobj_185_ cc_\\/_and conj_\\/_tested prep_tested_on pobj_on_test2008 ' +n,2231,'Although this Wikipedia gazetteer is much smaller than the English version used by that has over 2,000,000 entries , it is the largest gazetteer that can be freely used for Japanese NER ',Kazama,'mark_smaller_Although det_gazetteer_this nn_gazetteer_Wikipedia nsubj_smaller_gazetteer cop_smaller_is advmod_smaller_much advcl_gazetteer_smaller prep_smaller_than det_version_the amod_version_English pobj_than_version partmod_version_used prep_used_by pobj_by_ nsubj_has_that rcmod__has quantmod_2,000,000_over num_entries_2,000,000 dobj_has_entries nsubj_gazetteer_it cop_gazetteer_is det_gazetteer_the amod_gazetteer_largest nsubjpass_used_that aux_used_can auxpass_used_be advmod_used_freely rcmod_gazetteer_used prep_used_for amod_NER_Japanese pobj_for_NER ' +o,2232,'6 Experiment 61 Setup The experiments we report were done on the Penn Treebank WSJ Corpus ',Marcus,'num_Setup_6 nn_Setup_Experiment num_Setup_61 det_experiments_The nsubjpass_done_experiments nsubj_report_we rcmod_experiments_report auxpass_done_were dep_Setup_done prep_done_on det_Corpus_the nn_Corpus_Penn nn_Corpus_Treebank nn_Corpus_WSJ pobj_on_Corpus ' +o,2233,'INTRODUCTION Word associations have been studied for some time in the fields of psycholinguistics -LRB- by testing human subjects on words -RRB- , linguistics -LRB- where meaning is often based on how words co-occur with each other -RRB- , and more recently , by researchers in natural language processing using statistical measures to identify sets of associated words for use in various natural language processing tasks ',Church,'nn_associations_INTRODUCTION nn_associations_Word nsubjpass_studied_associations aux_studied_have auxpass_studied_been prep_studied_for det_time_some pobj_for_time prep_time_in det_fields_the pobj_in_fields prep_fields_of pobj_of_psycholinguistics dep_psycholinguistics_by pcomp_by_testing amod_subjects_human dobj_testing_subjects prep_testing_on pobj_on_words appos_psycholinguistics_linguistics advmod_based_where nsubjpass_based_meaning auxpass_based_is advmod_based_often dep_linguistics_based prep_based_on advmod_co-occur_how nsubj_co-occur_words pcomp_on_co-occur prep_co-occur_with det_other_each pobj_with_other cc_studied_and advmod_recently_more advmod_studied_recently conj_studied_by nsubj_using_researchers prep_researchers_in amod__natural nn__language nn__processing pobj_in_ pcomp_by_using amod_measures_statistical dobj_using_measures aux_identify_to xcomp_using_identify dobj_identify_sets prep_sets_of amod_words_associated pobj_of_words prep_words_for pobj_for_use prep_use_in amod_tasks_various amod_tasks_natural nn_tasks_language nn_tasks_processing pobj_in_tasks ' +o,2234,'In a next step , chunk information was added by a rule-based language-independent chunker that contains distituency rules , which implies that chunk boundaries are added between two PoS codes that can not occur in the same constituent ',Macken,'prep_added_In det_step_a amod_step_next pobj_In_step nn_information_chunk nsubjpass_added_information auxpass_added_was prep_added_by det__a amod__rule-based amod__language-independent nn__chunker pobj_by_ nsubj_contains_that rcmod__contains amod_rules_distituency dobj_contains_rules nsubj_implies_which rcmod_rules_implies complm_added_that nn_boundaries_chunk nsubjpass_added_boundaries auxpass_added_are ccomp_implies_added prep_added_between num_codes_two nn_codes_PoS pobj_between_codes nsubj_occur_that aux_occur_can neg_occur_not rcmod_codes_occur prep_occur_in det_constituent_the amod_constituent_same pobj_in_constituent ' +o,2235,'A few exceptions are the hierarchical -LRB- possibly syntaxbased -RRB- transduction models and the string transduction models ',Wu,'det_exceptions_A amod_exceptions_few nsubj__exceptions cop__are det__the amod__hierarchical dep_syntaxbased_possibly dep_hierarchical_syntaxbased nn__transduction nn__models cc__and det_string_the conj__string nn_models_transduction nsubj__models dep__ ' +o,2236,'Yet , the very nature of these alignments , as defined in the IBM modeling approach , lead to descriptions of the correspondences between sourcelanguage -LRB- SL -RRB- and target-language -LRB- TL -RRB- words of a translation that are often unsatisfactory , at least from a human perspective ',Brown,'cc_lead_Yet det_nature_the amod_nature_very nsubj_lead_nature prep_nature_of det_alignments_these pobj_of_alignments mark_defined_as dep_nature_defined prep_defined_in det__the nn__IBM nn__modeling nn__approach pobj_in_ prep_lead_to pobj_to_descriptions prep_descriptions_of det_correspondences_the pobj_of_correspondences prep_correspondences_between amod_words_sourcelanguage dep_sourcelanguage_SL cc_sourcelanguage_and conj_sourcelanguage_target-language abbrev_words_TL pobj_between_words prep_descriptions_of det_translation_a pobj_of_translation nsubj_are_that rcmod_translation_are advmod_are_often advmod_are_unsatisfactory dep_from_at dep_at_least prep_are_from det_perspective_a amod_perspective_human pobj_from_perspective ' +o,2237,'Previous attempts have used , for instance , the similarities between case frames , anchor words , and a web-based method ',Barzilay,'amod_attempts_Previous nsubj_used_attempts aux_used_have prep__for pobj_for_instance det_similarities_the nsubj__similarities prep_similarities_between nn_frames_case pobj_between_frames ccomp_used_ amod_words_anchor nsubj__words ccomp_used_ cc_used_and det_method_a amod_method_web-based nsubj__method conj_used_ ' +o,2238,'In order to calculate a global score or probability for a transition sequence , two systems used a Markov chain approach ',Dredze,'mark_calculate_In dep_calculate_order aux_calculate_to dep_used_calculate det_score_a amod_score_global dobj_calculate_score cc_score_or conj_score_probability prep_score_for det_sequence_a nn_sequence_transition pobj_for_sequence num_systems_two nsubj_used_systems det__a nn__Markov nn__chain nn__approach dobj_used_ ' +o,2239,'We measured stability -LRB- the degree to which the same annotator will produce an annotation after 6 weeks -RRB- and reproducibility -LRB- the degree to which two unrelated annotators will produce the same annotation -RRB- , using the Kappa coefficient K , which controls agreement P -LRB- A -RRB- for chance agreement P -LRB- E -RRB- : K = PA -RRB- - P -LRB- E -RRB- 1-P -LRB- Z -RRB- Kappa is 0 for if agreement is only as would be expected by chance annotation following the same distribution as the observed distribution , and 1 for perfect agreement ',Carletta,'nsubj_measured_We dobj_measured_stability det_degree_the dep_stability_degree rel_produce_to pobj_to_which det_annotator_the amod_annotator_same nsubj_produce_annotator aux_produce_will rcmod_degree_produce det_annotation_an dobj_produce_annotation prep_produce_after num_weeks_6 pobj_after_weeks cc_stability_and conj_stability_reproducibility det_degree_the dep_reproducibility_degree rel_produce_to pobj_to_which num_annotators_two amod_annotators_unrelated nsubj_produce_annotators aux_produce_will rcmod_degree_produce det_annotation_the amod_annotation_same dobj_produce_annotation xcomp_measured_using det_coefficient_the nn_coefficient_Kappa dobj_using_coefficient nn__K dep_coefficient_ nsubj_controls_which rcmod_coefficient_controls nn_P_agreement dobj_controls_P appos_P_A prep_controls_for nn_P_chance nn_P_agreement pobj_for_P appos_P_E nn_PA_K nn_PA_= dep_P_PA dep_coefficient_P appos_P_E dep_coefficient_1-P appos_1-P_Z nsubj_0_Kappa cop_0_is rcmod_1-P_0 prep_0_for mark_is_if nsubj_is_agreement pcomp_for_is advmod_expected_only dep_expected_as aux_expected_would auxpass_expected_be ccomp_is_expected prep_expected_by nn_annotation_chance pobj_by_annotation prep_annotation_following det_distribution_the amod_distribution_same pobj_following_distribution prep_distribution_as det_distribution_the amod_distribution_observed pobj_as_distribution cc_coefficient_and conj_coefficient_1 prep_1_for amod_agreement_perfect pobj_for_agreement ' +o,2240,'2.2.1 The evaluator The evaluator is a function p(t[t\', s) which assigns to each target-text unit t an estimate of its probability given a source text s and the tokens t\' which precede t in the current translation of s. Our approach to modeling this distribution is based to a large extent on that of the IBM group (Brown et al. , 1993), but it diflhrs in one significant aspect: whereas the IBM model involves a \'noisy channel\' decomposition, we use a linear combination of separate predictions from a language model p(t[t\') and a translation model p(t[s).',Brown,'' +o,2241,'First , a non-anaphoric NP classifier identifies definite noun phrases that are existential , using both syntactic rules and our learned existential NP recognizer , and removes them from the resolution process ',Bean,'advmod_identifies_First det_classifier_a amod_classifier_non-anaphoric nn_classifier_NP nsubj_identifies_classifier amod_phrases_definite nn_phrases_noun dobj_identifies_phrases nsubj_existential_that cop_existential_are rcmod_phrases_existential xcomp_identifies_using det_rules_both amod_rules_syntactic dobj_using_rules cc_rules_and poss__our amod__learned amod__existential nn__NP nn__recognizer conj_rules_ cc_identifies_and conj_identifies_removes dobj_removes_them prep_removes_from det_process_the nn_process_resolution pobj_from_process ' +o,2242,'The algorithm is slightly different from other online training algorithms in that we keep and update oracle translations , which is a set of good translations reachable by a decoder according to a metric , ie BLEU ',Tillmann,'det_algorithm_The nsubj_different_algorithm cop_different_is advmod_different_slightly prep_different_from amod_algorithms_other amod_algorithms_online nn_algorithms_training pobj_from_algorithms partmod_algorithms_ prep__in complm_keep_that nsubj_keep_we pcomp_in_keep cc_keep_and conj_keep_update amod_translations_oracle dobj_keep_translations nsubj_set_which cop_set_is det_set_a rcmod_algorithms_set prep_set_of amod_translations_good pobj_of_translations partmod_translations_reachable prep_reachable_by det_decoder_a pobj_by_decoder prep_reachable_according dep_according_to det__a amod__metric amod__ie nn__BLEU pobj_to_ ' +o,2243,'Inter-annotator agreement was determined for six pairs of two annotators each , resulting in kappa values -LRB- -RRB- ranging from 062 to 082 for the whole database -LRB- Carlson et al ',Carletta,'nn_agreement_Inter-annotator nsubjpass_determined_agreement auxpass_determined_was prep_determined_for num_pairs_six pobj_for_pairs prep_pairs_of num_annotators_two pobj_of_annotators dep_annotators_each partmod_annotators_resulting prep_resulting_in nn_values_kappa pobj_in_values appos_values_ partmod_values_ranging prep_ranging_from number_082_062 dep_082_to pobj_from_082 prep_082_for det_al_the amod_al_whole nn_al_database nn_al_Carlson nn_al_et pobj_for_al ' +o,2244,'Discriminative, context-specific training seems to yield a better set of similar predicates, e.g. the highest-ranked contexts for DSPcooc on the verb join,3 lead 1.42, rejoin 1.39, form 1.34, belong to 1.31, found 1.31, quit 1.29, guide 1.19, induct 1.19, launch (subj) 1.18, work at 1.14 give a better SIMS(join) for Equation (1) than the top similarities returned by (Lin, 1998a): participate 0.164, lead 0.150, return to 0.148, say 0.143, rejoin 0.142, sign 0.142, meet 0.142, include 0.141, leave 0.140, work 0.137 Other features are also weighted intuitively.',Lin,'' +o,2245,'Then , some manual and automatic symbol splitting methods are presented , which get comparable performance with lexicalized parsers ',Matsuzaki,'advmod_presented_Then det_manual_some nsubjpass_presented_manual cc_manual_and amod_methods_automatic nn_methods_symbol nn_methods_splitting conj_manual_methods auxpass_presented_are nsubj_get_which ccomp_presented_get amod_performance_comparable dobj_get_performance prep_get_with amod_parsers_lexicalized pobj_with_parsers ' +o,2246,' , Johnson -- that conditioning the probabilities of structures on the context within which they appear , for example on the lexical head of a constituent , on the label of its parent nonterrninal , or , ideally , on both and many other things besides , leads to a much better parsing model and results in higher parsing accuracies ',Collins,'advmod_leads_ nn__Johnson nsubj_leads_ nsubj_conditioning_that rcmod__conditioning det_probabilities_the dobj_conditioning_probabilities prep_probabilities_of pobj_of_structures prep_structures_on det_context_the pobj_on_context prep_conditioning_within dep_appear_which nsubj_appear_they pcomp_within_appear prep_appear_for pobj_for_example prep_example_on det_head_the amod_head_lexical pobj_on_head prep_head_of det__a amod__constituent pobj_of_ conj_for_on det_label_the pobj_on_label prep_label_of poss__its amod__parent amod__nonterrninal pobj_of_ cc_for_or dep_for_ideally conj_for_on pobj_on_both cc_both_and amod_things_many amod_things_other conj_both_things prep_appear_besides prep_leads_to det_model_a advmod_better_much amod_model_better nn_model_parsing pobj_to_model cc_model_and conj_model_results prep_results_in amod_accuracies_higher nn_accuracies_parsing pobj_in_accuracies ' +o,2247,'The normalization is visualized as a translation problem where messages in the SMS language are to be translated to normal English using a similar phrase-based statistical MT method ',Koehn,'det_normalization_The nsubjpass_visualized_normalization auxpass_visualized_is prep_visualized_as det_problem_a nn_problem_translation pobj_as_problem advmod_are_where nsubj_are_messages prep_messages_in det_language_the nn_language_SMS pobj_in_language advcl_visualized_are aux_translated_to auxpass_translated_be xcomp_are_translated prep_translated_to amod_English_normal pobj_to_English purpcl_translated_using det_method_a amod_method_similar dep_similar_phrase-based amod_method_statistical nn_method_MT dobj_using_method ' +o,2248,'However there has recently been much work drawing connections between the two methods ; in this section we review this work ',Collins,'advmod_been_However expl_been_there aux_been_has advmod_been_recently amod_connections_much nn_connections_work nn_connections_drawing dobj_been_connections prep_connections_between det__the num__two nn__methods pobj_between_ prep_review_in det_section_this pobj_in_section nsubj_review_we parataxis_been_review det_work_this dobj_review_work ' +o,2249,'To circumvent these computational limitations , various pruning techniques are usually needed , eg , ',Huang,'aux_circumvent_To dep_needed_circumvent det_limitations_these amod_limitations_computational dobj_circumvent_limitations amod_techniques_various nn_techniques_pruning nsubjpass_needed_techniques auxpass_needed_are advmod_needed_usually dep_needed_eg ' +o,2250,'In addition to raw inter-tagger agreement , the kappa statistic , which removes from the agreement rate the amount of agreement that is expected by chance , was also determined ',Carletta,'prep_determined_In pobj_In_addition prep_addition_to amod_agreement_raw amod_agreement_inter-tagger pobj_to_agreement det_statistic_the nn_statistic_kappa nsubjpass_determined_statistic nsubj_removes_which rcmod_statistic_removes prep_removes_from det_rate_the nn_rate_agreement pobj_from_rate det_amount_the dobj_removes_amount prep_amount_of pobj_of_agreement nsubjpass_expected_that auxpass_expected_is rcmod_amount_expected prep_expected_by nn__chance pobj_by_ auxpass_determined_was advmod_determined_also ' +o,2251,'of the works of (Kuplec, Pedersen, and Chen, 1995) and (Brandow, Mltze, .and Ran, 1995), and advances summarmatlon technology by applynag corpus-based statistical NLP teehmques, robust information extraction, and readily avaalable on-hne resources Our prehxmnary experiments with combining different summarization features have been reported, and our current effort to learn to combine these features to produce the best summaries has been described The features derived by these robust NLP techmques were also utihzed m presentmg multiple summary.vtews to the user m a novel way References Advanced Research Projects Agency 1995 Proceed:rigs of S:zth Message Understanding Conference (MUC-6) Morgan Kanfmann Pubhshers Brandow, Ron, Karl Mltze, and Lisa Ran 1995 Automatic condensation of electromc pubhcatlous by sentence selection Information Processing and Management, 31, forthcoming .Bull, Eric 1993 A Comps-based Approach to Language Learning Ph D thesm, Umverslty of Pennsylvania Church, Kenneth and Patrick Hanks 1990 Word Aesoclatlon Norrns, Mutual Information, and Lexicography Computational Lmgmstscs, 16(1) Church, Kenneth W 1995 One term or two 9 In Proceedings of the 17th Annual International SIGIR Conference on Research and Development In Informatzon Retrzeral, pages 310-318 Edmundson, H P 1969 New methods m automatic abstracting Journal of the ACM, 16(2) 264-228 Fum, Dando, Glovanm Gmda, and Carlo Tasso 1985 Evalutatmg Importance A step towards text surnmarlzatlon In I3CAI85, pages 840-844IJCAi, AAAI Hahn, Udo 1990 Topic parsing Accounting for text macro structures m full-text analysm In format:on Processing and Management, 26(1)135170 Harman, Donna 1991 How effective is suttixang ~ Journal of the Amerlcan Sot:cry for Informatwn Sc:ence, 42(1) 7-15 Harman, Donna 1996 Overview of the fifth text retrieval conference (tree-5) In TREC-5 Conference Proceedings Jmg, Y and B Croft 1994 An Assoc:atwn Thesaurns for Informatzon Retrseval Umass Techmcal Report 94-I7 Center for Intelligent Information Retrieval, University of Massachusetts Johnson, F C, C D Prate, W J Black, and A P Neal 1993.',Church,'' +o,2252,'The dataset is available only in English and has been widely used in previous semantic relatedness evaluations -LRB- eg , -RRB- ',Hughes,'det_dataset_The nsubjpass_available_dataset cop_available_is dep_in_only prep_available_in pobj_in_English cc_available_and aux_used_has auxpass_used_been advmod_used_widely conj_available_used prep_used_in amod_evaluations_previous amod_evaluations_semantic amod_evaluations_relatedness pobj_in_evaluations appos_evaluations_eg dep_eg_ ' +o,2253,'For practical reasons , the maximum size of a token was set at three for Chinese , andfor forKorean2 Minimum error rate training was run on each system afterwardsand BLEU score was calculated on the test sets ',Papineni,'prep_set_For amod_reasons_practical pobj_For_reasons det_size_the amod_size_maximum nsubjpass_set_size prep_size_of det_token_a pobj_of_token auxpass_set_was prep_set_at pobj_at_three prep_set_for amod_rate_Chinese amod_rate_andfor amod_rate_forKorean2 nn_rate_Minimum nn_rate_error pobj_for_rate nn__training nsubjpass_run_ auxpass_run_was dep_set_run prep_run_on det_system_each pobj_on_system partmod_system_afterwardsand amod__BLEU nn__score nsubjpass_calculated_ auxpass_calculated_was ccomp_afterwardsand_calculated prep_calculated_on det_sets_the nn_sets_test pobj_on_sets ' +o,2254,'Assuming that the parameters P -LRB- etk fsk -RRB- are known , the most likely alignment is computed by a simple dynamic-programming algorithm1 Instead of using an Expectation-Maximization algorithm to estimate these parameters , as commonly done when performing word alignment , we directly compute these parameters by relying on the information contained within the chunks ',Och,'dep_compute_Assuming complm_known_that det_P_the nn_P_parameters nsubjpass_known_P nn_fsk_etk appos_P_fsk auxpass_known_are ccomp_Assuming_known det_alignment_the advmod_likely_most amod_alignment_likely nsubjpass_computed_alignment auxpass_computed_is parataxis_compute_computed prep_computed_by det_algorithm1_a amod_algorithm1_simple amod_algorithm1_dynamic-programming pobj_by_algorithm1 dep_of_Instead prep_algorithm1_of pcomp_of_using det_algorithm_an amod_algorithm_Expectation-Maximization dobj_using_algorithm aux_estimate_to xcomp_using_estimate det_parameters_these dobj_estimate_parameters advmod_commonly_as advmod_computed_commonly dep_computed_done advmod_performing_when advcl_done_performing nn__word nn__alignment dobj_performing_ nsubj_compute_we advmod_compute_directly det_parameters_these dobj_compute_parameters prep_compute_by pcomp_by_relying prep_relying_on det_information_the pobj_on_information partmod_information_contained prep_contained_within det_chunks_the pobj_within_chunks ' +o,2255,'287 System Train + base Test + base 1 Baseline 8789 8789 2 Contrastive 8870 082 8845 056 -LRB- 5 trials\\/fold -RRB- 3 Contrastive 8882 093 8855 066 -LRB- greedy selection -RRB- Table 1 : Average F1 of 7-way cross-validation To generate the alignments , we used Model 4 , as implemented in GIZA + + ',Marcus,'nn_Train_System nsubj_+_Train ccomp_287_+ amod_base_base nn_base_Test amod_base_+ iobj_+_base num_Baseline_1 dobj_+_Baseline number_8789_8789 num_Baseline_8789 num_Contrastive_2 dobj_+_Contrastive number_082_8870 num_Contrastive_082 number_056_8845 dobj_+_056 num_trials\\/fold_5 appos_056_trials\\/fold num_Contrastive_3 dobj_+_Contrastive number_093_8882 num_Contrastive_093 number_066_8855 dobj_+_066 amod_selection_greedy appos_066_selection prep_+_Table pobj_Table_1 amod_F1_Average dobj_+_F1 prep_F1_of amod_cross-validation_7-way pobj_of_cross-validation aux_generate_To dep_287_generate det_alignments_the dobj_generate_alignments nsubj_used_we parataxis_287_used nn__Model num__4 dobj_used_ mark_implemented_as dep_287_implemented prep_implemented_in pobj_in_GIZA amod_GIZA_+ cc_+_+ conj_+_ ' +o,2256,'While the need for annotation by multiple raters has been well established in NLP tasks , most previous work in error detection has surprisingly relied on only one rater to either create an annotated corpus of learner errors , or to check the systems output ',Carletta,'mark_established_While det_need_the nsubjpass_established_need prep_need_for pobj_for_annotation prep_annotation_by amod_raters_multiple pobj_by_raters aux_established_has auxpass_established_been advmod_established_well advcl_relied_established prep_established_in nn__NLP nn__tasks pobj_in_ amod_work_most amod_work_previous nsubj_relied_work prep_work_in nn_detection_error pobj_in_detection aux_relied_has advmod_relied_surprisingly prep_relied_on advmod_rater_only num_rater_one pobj_on_rater aux_create_to dep_create_either ccomp_relied_create det_corpus_an amod_corpus_annotated dobj_create_corpus prep_corpus_of nn_errors_learner pobj_of_errors cc_create_or aux_check_to conj_create_check det_output_the nn_output_systems dobj_check_output ' +o,2257,'The model scaling factors M1 are optimized with respect to the BLEU score as described in ',Och,'det_M1_The nn_M1_model nn_M1_scaling nn_M1_factors nsubjpass_optimized_M1 auxpass_optimized_are prep_optimized_with pobj_with_respect prep_optimized_to det_score_the amod_score_BLEU pobj_to_score mark_described_as advcl_optimized_described prt_described_in ' +o,2258,'We implemented an N-gram indexer\\/estimator using MPI inspired by the MapReduce implementation of N-gram language model indexing\\/estimation pipeline ',Brants,'nsubj_implemented_We det_indexer\\/estimator_an amod_indexer\\/estimator_N-gram dobj_implemented_indexer\\/estimator xcomp_implemented_using dobj_using_MPI partmod_MPI_inspired prep_inspired_by det_implementation_the nn_implementation_MapReduce pobj_by_implementation prep_implementation_of amod_pipeline_N-gram nn_pipeline_language nn_pipeline_model nn_pipeline_indexing\\/estimation pobj_of_pipeline ' +p,2259,'The creation of the Penn English Treebank , a syntactically interpreted corpus , played a crucial role in the advances in natural language parsing technology for English ',Collins,'det_creation_The nsubj_played_creation prep_creation_of det__the nn__Penn nn__English nn__Treebank pobj_of_ det_corpus_a advmod_interpreted_syntactically amod_corpus_interpreted appos_creation_corpus det_role_a amod_role_crucial dobj_played_role prep_played_in det_advances_the pobj_in_advances prep_advances_in amod_language_natural pobj_in_language xcomp_played_parsing nn__technology dobj_parsing_ prep_parsing_for pobj_for_English ' +o,2260,'The idea of bidirectional parsing is related to the bidirectional sequential classification method described in ',Shen,'det_idea_The nsubjpass_related_idea prep_idea_of amod_parsing_bidirectional pobj_of_parsing auxpass_related_is prep_related_to det_method_the amod_method_bidirectional amod_method_sequential nn_method_classification pobj_to_method partmod_method_described prep_described_in ' +o,2261,'There bas recently been work in the detection of semantically related nouns via , for example , shared argument structures , and shared dictionary definition context ',Hindle,'expl_been_There aux_been_bas advmod_been_recently dobj_been_work prep_work_in det_detection_the pobj_in_detection prep_detection_of advmod_related_semantically amod_nouns_related pobj_of_nouns prep_nouns_via dep_via_for pobj_for_example amod__shared nn__argument nn__structures pobj_via_ cc_nouns_and amod_definition_shared nn_definition_dictionary conj_nouns_definition prep_definition_context dep_context_ ' +o,2262,'It is believed that improvement can be achieved by training the generative model based on a discriminative optimization criteria in which the training procedure is designed to maximize the conditional probability of the parses given the sentences in the training corpus ',Klein,'nsubjpass_believed_It auxpass_believed_is complm_achieved_that nsubjpass_achieved_improvement aux_achieved_can auxpass_achieved_be ccomp_believed_achieved prep_achieved_by pcomp_by_training det_model_the amod_model_generative dobj_training_model prep_training_based dep_based_on det__a amod__discriminative amod__optimization nn__criteria pobj_on_ rel_designed_in pobj_in_which det_procedure_the nn_procedure_training nsubjpass_designed_procedure auxpass_designed_is rcmod__designed aux_maximize_to xcomp_designed_maximize det_probability_the amod_probability_conditional dobj_maximize_probability prep_probability_of det_parses_the pobj_of_parses partmod_parses_given det_sentences_the dobj_given_sentences prep_given_in det_corpus_the nn_corpus_training pobj_in_corpus ' +o,2263,'In this paper , we used CTB 50 as our main corpus , defined the training , development and test sets according to , and designed our experiments to explore the impact of the training corpus size on our approach ',Jiang,'prep_used_In det_paper_this pobj_In_paper nsubj_used_we dobj_used_CTB num_CTB_50 dep_CTB_ prep__as poss_corpus_our amod_corpus_main pobj_as_corpus conj__defined det_sets_the nn_sets_training conj_training_development cc_training_and conj_training_test dobj_defined_sets prep_defined_according dep_according_to pobj_to_ cc__and conj__designed poss_experiments_our dobj_designed_experiments aux_explore_to xcomp_designed_explore det_impact_the dobj_explore_impact prep_impact_of det_size_the nn_size_training nn_size_corpus pobj_of_size prep_explore_on poss_approach_our pobj_on_approach ' +o,2264,'Expectation Evaluation is the soul of parameter estimation , ',Brown,'nn_Evaluation_Expectation nsubj_soul_Evaluation cop_soul_is det_soul_the prep_soul_of nn__parameter nn__estimation nn__ pobj_of_ ' +o,2265,'The samplers that Goldwater and Griffiths and describe are pointwise collapsed Gibbs samplers ',Johnson,'det_samplers_The nsubj_pointwise_samplers complm_describe_that nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths nsubj_describe_ cc__and nn__ conj__ dep_samplers_describe cop_pointwise_are dep_collapsed_pointwise nn_samplers_Gibbs nsubj_collapsed_samplers ' +o,2266,'Recently , methods from nonparametric Bayesian statistics have been gaining popularity as a way to approach unsupervised learning for a variety of tasks , including language modeling , word and morpheme segmentation , parsing , and machine translation ',Liang,'advmod_gaining_Recently nsubj_gaining_methods prep_methods_from amod_statistics_nonparametric nn_statistics_Bayesian pobj_from_statistics aux_gaining_have aux_gaining_been dobj_gaining_popularity prep_gaining_as det_way_a pobj_as_way aux_approach_to infmod_way_approach acomp_approach_unsupervised xcomp_approach_learning prep_learning_for det_variety_a pobj_for_variety prep_variety_of pobj_of_tasks prep_approach_including nn_modeling_language pobj_including_modeling nn_segmentation_word cc_word_and conj_word_morpheme conj_modeling_segmentation conj_modeling_parsing cc_modeling_and nn_translation_machine conj_modeling_translation ' +o,2267,'The bigram translation probability t2 -LRB- f f , e -RRB- specifies the likelihood that target word f is to follow f in a phrase generated by source word e 170 21 Properties of the Model and Prior Work The formulation of the WtoP alignment model was motivated by both the HMM word alignment model and IBM Model-4 with the goal of building on the strengths of each ',Berger,'det_bigram_The nn_t2_translation nn_t2_probability nsubj_likelihood_t2 nn_f_f appos_t2_f dep_f_e dep_likelihood_specifies det_likelihood_the ccomp_bigram_likelihood complm_is_that nn_f_target nn_f_word nsubj_is_f ccomp_likelihood_is aux_follow_to xcomp_is_follow dobj_follow_f prep_f_in det_phrase_a pobj_in_phrase partmod_phrase_generated prep_generated_by nn_word_source pobj_by_word dep_Properties_e number_21_170 num_Properties_21 dep_bigram_Properties prep_Properties_of det_Work_the nn_Work_Model cc_Model_and conj_Model_Prior pobj_of_Work det_formulation_The nsubjpass_motivated_formulation prep_formulation_of det_model_the nn_model_WtoP nn_model_alignment pobj_of_model auxpass_motivated_was rcmod_Work_motivated prep_motivated_by preconj__both det__the nn__HMM nn__word nn__alignment nn__model pobj_by_ cc__and nn_Model-4_IBM conj__Model-4 prep_motivated_with det_goal_the pobj_with_goal prep_goal_of pcomp_of_building prep_building_on det_strengths_the pobj_on_strengths prep_strengths_of pobj_of_each ' +p,2268,'1 Introduction As with many other statistical natural language processing tasks , statistical machine translation produces high quality results when ample training data is available ',Brown,'num_Introduction_1 dep_As_Introduction dep_with_As prep_produces_with amod_tasks_many amod_tasks_other amod_tasks_statistical amod_tasks_natural nn_tasks_language amod_tasks_processing pobj_with_tasks amod__statistical nn__machine nn__translation nsubj_produces_ amod_results_high nn_results_quality dobj_produces_results advmod_available_when amod_data_ample nn_data_training nsubj_available_data cop_available_is dep_results_available ' +o,2269,'The automatic alignments were extracted by appending the manually aligned sentences on to the respective Europarl v3 corpora and aligning them using GIZA + + and the growfinal-diag algorithm ',Koehn,'det_alignments_The amod_alignments_automatic nsubjpass_extracted_alignments auxpass_extracted_were prep_extracted_by pcomp_by_appending det_manually_the dobj_appending_manually partmod_manually_aligned dobj_aligned_sentences prep_aligned_on dep_on_to det_corpora_the amod_corpora_respective nn_corpora_Europarl nn_corpora_v3 pobj_to_corpora cc_appending_and conj_appending_aligning nsubj_using_them dep_aligning_using nn__GIZA nn__+ nn__+ dobj_using_ cc__and det_algorithm_the amod_algorithm_growfinal-diag conj__algorithm ' +o,2270,'This scoring function has been successfully applied to resolve ambiguity problems in an English-to-Chinese machine translation system -LRB- BehaviorTran -RRB- and a spoken language processing system ',Brown,'det_function_This amod_function_scoring nsubjpass_applied_function aux_applied_has auxpass_applied_been advmod_applied_successfully aux_resolve_to xcomp_applied_resolve amod_problems_ambiguity dobj_resolve_problems prep_problems_in det_system_an amod_system_English-to-Chinese nn_system_machine nn_system_translation pobj_in_system appos_problems_BehaviorTran dep_problems_ cc__and det__a amod__spoken nn__language amod__processing nn__system conj__ ' +o,2271,'This paper proposes a method for building a bilingual lexicon through a pivot language by using phrase-based statistical machine translation -LRB- SMT -RRB- ',Koehn,'det_paper_This nsubj_proposes_paper det_method_a dobj_proposes_method prep_proposes_for pcomp_for_building det_lexicon_a amod_lexicon_bilingual dobj_building_lexicon prep_building_through det_language_a amod_language_pivot pobj_through_language prep_building_by pcomp_by_using amod_translation_phrase-based amod_translation_statistical nn_translation_machine dobj_using_translation abbrev_translation_SMT ' +p,2272,'With non-local features , we can not use efcient procedures such as forward-backward procedures and the Viterbi algorithm that are required in training CRFs and perceptrons ',Collins,'prep_use_With amod_features_non-local pobj_With_features nsubj_use_we aux_use_can neg_use_not amod_procedures_efcient dobj_use_procedures dep_as_such prep_procedures_as amod_procedures_forward-backward pobj_as_procedures cc_procedures_and det_algorithm_the nn_algorithm_Viterbi conj_procedures_algorithm nsubjpass_required_that auxpass_required_are rcmod_procedures_required prep_required_in pcomp_in_training nn__CRFs dobj_training_ cc__and conj__perceptrons ' +o,2273,'Networks (Toutanova et al., 2003) 97.24 SVM (Gimenez and M`arquez, 2003) 97.05 ME based a bidirectional inference (Tsuruoka and Tsujii, 2005) 97.15 Guided learning for bidirectional sequence classification (Shen et al., 2007) 97.33 AdaBoost.SDF with candidate features (=2,=1,=100, W-dist) 97.32 AdaBoost.SDF with candidate features (=2,=10,=10, F-dist) 97.32 SVM with candidate features (C=0.1, d=2) 97.32 Text Chunking F=1 Regularized Winnow + full parser output (Zhang et al., 2001) 94.17 SVM-voting (Kudo and Matsumoto, 2001) 93.91 ASO + unlabeled data (Ando and Zhang, 2005) 94.39 CRF+Reranking(Kudo et al., 2005) 94.12 ME based a bidirectional inference (Tsuruoka and Tsujii, 2005) 93.70 LaSo (Approximate Large Margin Update) (Daume III and Marcu, 2005) 94.4 HySOL (Suzuki et al., 2007) 94.36 AdaBoost.SDF with candidate featuers (=2,=1,=, W-dist) 94.32 AdaBoost.SDF with candidate featuers (=2,=10,=10,W-dist) 94.30 SVM with candidate features (C=1, d=2) 94.31 One of the reasons that boosting-based classifiers realize faster classification speed is sparseness of rules.',Shen,'' +o,2274,'The second uses the decoder to search for the highest-B translation , which Arun and Koehn -LRB- 2007 -RRB- call max-B updating ',Tillmann,'det_second_The nsubj_uses_second det_decoder_the dobj_uses_decoder aux_search_to infmod_decoder_search prep_search_for det__the amod__highest-B nn__translation pobj_for_ dobj_call_which nsubj_call_Arun cc_Arun_and conj_Arun_Koehn appos_Arun_2007 rcmod__call dobj_call_max-B partmod_max-B_updating ' +o,2275,' show that this model is a member of an exponential family with one parameter for each constraint , specifically a model of the form 1 ~ I ~ -LRB- x , ~ -RRB- p -LRB- yl -RRB- = E \' in which z -LRB- x -RRB- = eZ , Y The parameters A1 , , An are Lagrange multipliers that impose the constraints corresponding to the chosen features fl , - , fnThe term Z -LRB- x -RRB- normalizes the probabilities by summing over all possible outcomes y et al ',Berger,'advmod_show_ complm_member_that det_model_this nsubj_member_model cop_member_is det_member_a ccomp_show_member prep_member_of det_family_an amod_family_exponential pobj_of_family prep_family_with num_parameter_one pobj_with_parameter prep_parameter_for det_constraint_each pobj_for_constraint advmod_constraint_specifically det_model_a dep_specifically_model dep_specifically_of det_p_the nn_p_form num_p_1 advmod_~_~ dep_~_I amod_p_~ dep_~_x appos_x_~ pobj_of_p appos_p_yl nn_E_= dep_constraint_E rel_normalizes_in pobj_in_which dep_multipliers_z dep_eZ_x amod_eZ_= nsubj_multipliers_eZ appos_eZ_Y det_A1_The nn_A1_parameters dep_Y_A1 nsubj_multipliers_An cop_multipliers_are amod_multipliers_Lagrange dep_normalizes_multipliers nsubj_impose_that rcmod_multipliers_impose det_constraints_the dobj_impose_constraints amod_constraints_corresponding prep_constraints_to det_fl_the amod_fl_chosen nn_fl_features pobj_to_fl amod_Z_fnThe nn_Z_term nsubj_normalizes_Z appos_Z_x rcmod_member_normalizes det_probabilities_the dobj_normalizes_probabilities prep_normalizes_by pcomp_by_summing prep_summing_over det_y_all amod_y_possible nn_y_outcomes pobj_over_y nn_al_ nn_al_et nsubj_show_al ' +o,2276,'While this is certainly a daunting task , it is possible that for annotation studies that do not require expert annotators and extensive annotator training , the newly available access to a large pool of inexpensive annotators , such as the Amazon Mechanical Turk scheme ,4 or embedding the task in an online game played by volunteers could provide some solutions ',Poesio,'mark_task_While nsubj_task_this cop_task_is advmod_task_certainly det_task_a amod_task_daunting advcl_possible_task nsubj_possible_it cop_possible_is complm_provide_that prep_provide_for nn_studies_annotation pobj_for_studies nsubj_require_that aux_require_do neg_require_not rcmod_studies_require nn_annotators_expert dobj_require_annotators cc_annotators_and amod_training_extensive nn_training_annotator conj_annotators_training det_access_the advmod_available_newly amod_access_available appos_annotators_access prep_access_to det_pool_a amod_pool_large pobj_to_pool prep_pool_of amod_annotators_inexpensive pobj_of_annotators dep_as_such prep_annotators_as det_scheme_the nn_scheme_Amazon nn_scheme_Mechanical nn_scheme_Turk pobj_as_scheme dep_scheme_ dobj__,4 cc__or conj__embedding det_task_the dobj_embedding_task prep_embedding_in det_game_an amod_game_online pobj_in_game partmod_game_played prep_played_by pobj_by_volunteers nsubj_provide_ aux_provide_could ccomp_possible_provide det_solutions_some dobj_provide_solutions ' +o,2277,'Translation scores are reported using caseinsensitive BLEU with a single reference translation ',Papineni,'nn_scores_Translation nsubjpass_reported_scores auxpass_reported_are xcomp_reported_using amod__caseinsensitive amod__BLEU dobj_using_ prep_using_with det_translation_a amod_translation_single nn_translation_reference pobj_with_translation ' +o,2278,'We used GIZA + + to align approximately 751,000 sentences from the German-English portion of the Europarl corpus , in both the German-to-English and English-to-German directions ',Och,'nsubj_used_We nsubj_align_GIZA dep_align_+ cc_+_+ conj_+_ aux_align_to xcomp_used_align quantmod_751,000_approximately num_sentences_751,000 dobj_align_sentences prep_align_from det_portion_the amod_portion_German-English pobj_from_portion prep_portion_of det__the nn__Europarl nn__corpus pobj_of_ prep_align_in preconj_German-to-English_both dep_German-to-English_the pobj_in_German-to-English cc_German-to-English_and amod_directions_English-to-German conj_German-to-English_directions ' +o,2279,'A Greek model was trained on 440,082 aligned sentences of Europarl v3 , tuned with Minimum Error Training ',Och,'det_model_A amod_model_Greek nsubjpass_trained_model auxpass_trained_was prep_trained_on num_sentences_440,082 amod_sentences_aligned pobj_on_sentences prep_sentences_of pobj_of_Europarl num_Europarl_v3 num_Europarl_tuned prep_trained_with nn_Training_Minimum nn_Training_Error pobj_with_Training ' +n,2280,'As the tagger of can not tag a word lattice , we can not back off to this tagging ',Ratnaparkhi,'mark_tag_As det_tagger_the nsubj_tag_tagger prep_tagger_of pobj_of_ aux_tag_can neg_tag_not advcl_back_tag det_lattice_a nn_lattice_word dobj_tag_lattice nsubj_back_we aux_back_can neg_back_not prt_back_off prep_back_to det_tagging_this pobj_to_tagging ' +p,2281,'1 Introduction In recent years , phrase-based systems for statistical machine translation have delivered state-of-the-art performance on standard translation tasks ',Koehn,'dep_delivered_1 dep_In_Introduction prep_delivered_In amod_years_recent pobj_In_years amod_systems_phrase-based nsubj_delivered_systems prep_systems_for amod__statistical nn__machine nn__translation pobj_for_ aux_delivered_have amod_performance_state-of-the-art dobj_delivered_performance prep_delivered_on amod_tasks_standard nn_tasks_translation pobj_on_tasks ' +o,2282,'Therefore , defined the translation candidate with the minimum word-error rate as pseudo reference translation ',Och,'advmod_defined_Therefore nsubj_defined_ det_candidate_the nn_candidate_translation dobj_defined_candidate prep_defined_with det_rate_the amod_rate_minimum amod_rate_word-error pobj_with_rate prep_defined_as amod_translation_pseudo nn_translation_reference pobj_as_translation ' +o,2283,'In the field of eomputationa1 linguistics , mutual information , 2 , or a likelihood ratio test -LRB- Dunning , 199a -RRB- are suggested ',Church,'prep_suggested_In det_field_the pobj_In_field prep_field_of amod_linguistics_eomputationa1 pobj_of_linguistics amod__mutual nn__information nsubjpass_suggested_ num__2 appos__ cc__or det_test_a nn_test_likelihood nn_test_ratio conj__test appos_test_Dunning dep_Dunning_199a auxpass_suggested_are ' +o,2284,'Previous authors have used numerous HMM-based models and other types of networks including maximum entropy models , conditional Markov models , conditional random elds -LRB- CRF -RRB- , and cyclic dependency networks ',Klein,'amod_authors_Previous nsubj_used_authors aux_used_have amod_models_numerous amod_models_HMM-based dobj_used_models amod_types_ cc__and conj__other nsubj__types prep_types_of pobj_of_networks prep_networks_including amod_models_maximum nn_models_entropy pobj_including_models rcmod_models_ amod_models_conditional nn_models_Markov dep__models ccomp_used_ amod_elds_conditional amod_elds_random nsubj__elds abbrev_elds_CRF ccomp_used_ cc_used_and amod_networks_cyclic nn_networks_dependency nsubj__networks conj_used_ ' +o,2285,'On the other hand , other authors -LRB- eg , -RRB- do use the expression phrase-based models ',Och,'prep_do_On det_hand_the amod_hand_other pobj_On_hand amod_authors_other nsubj_do_authors appos_authors_eg dep_eg_ dep_use_do det_expression_the dobj_use_expression amod_models_phrase-based nsubj_use_models ' +o,2286,'4 Methodology 41 Data In order to be able to compare our results with the results obtained by other researchers , we worked with the same data sets already used by for NP and SV detection ',Ramshaw,'num_Data_4 nn_Data_Methodology num_Data_41 mark_able_In dep_able_order aux_able_to cop_able_be dep_worked_able aux_compare_to xcomp_able_compare poss_results_our dobj_compare_results prep_compare_with det_results_the pobj_with_results partmod_results_obtained prep_obtained_by amod_researchers_other pobj_by_researchers nsubj_worked_we dep_Data_worked prep_worked_with det_sets_the amod_sets_same nn_sets_data pobj_with_sets advmod_used_already partmod_sets_used prep_used_by amod_detection_ prep__for pobj_for_NP cc_NP_and conj_NP_SV pobj_by_detection ' +o,2287,'Alternatively , one can view -LRB- 2 -RRB- as inducing an alignment between terms in the h to the terms in the t , somewhat similar to alignment models in statistical MT ',Brown,'advmod_view_Alternatively nsubj_view_one aux_view_can dobj_view_2 prep_view_as pcomp_as_inducing det_alignment_an dobj_inducing_alignment prep_alignment_between pobj_between_terms prep_terms_in det_h_the pobj_in_h prep_h_to det_terms_the pobj_to_terms prep_terms_in det_t_the pobj_in_t advmod_similar_somewhat amod_alignment_similar aux_alignment_to infmod_alignment_alignment dobj_alignment_models prep_alignment_in amod_MT_statistical pobj_in_MT ' +p,2288,'1 Introduction There is a pressing need for a consensus on a taskoriented level of semantic representation that can enable the development of powerful new semantic analyzers in the same way that the Penn Treebank enabled the development of statistical syntactic parsers ',Marcus,'num_Introduction_1 expl_is_There dep_Introduction_is det_need_a amod_need_pressing nsubj_is_need prep_need_for det_consensus_a pobj_for_consensus prep_consensus_on det_level_a amod_level_taskoriented pobj_on_level prep_level_of amod_representation_semantic pobj_of_representation nsubj_enable_that aux_enable_can rcmod_consensus_enable det_development_the dobj_enable_development prep_development_of amod_analyzers_powerful amod_analyzers_new amod_analyzers_semantic pobj_of_analyzers prep_analyzers_in det_way_the amod_way_same pobj_in_way complm_enabled_that det__the nn__Penn nn__Treebank nsubj_enabled_ ccomp_enable_enabled det_development_the dobj_enabled_development prep_development_of amod__statistical amod__syntactic nn__parsers pobj_of_ ' +o,2289,'4 Semantic Class Induction from Wikipedia Wikipedia has recently been used as a knowledge source for various language processing tasks , including taxonomy construction , coreference resolution , and English NER -LRB- eg , Bunescu and Pasca , , Kazama and Torisawa -LRB- 2007 -RRB- , Watanabe et al ',Cucerzan,'num_Induction_4 nn_Induction_Semantic nn_Induction_Class dep_used_Induction prep_Induction_from nn_Wikipedia_Wikipedia pobj_from_Wikipedia aux_used_has advmod_used_recently auxpass_used_been prep_used_as det_source_a nn_source_knowledge pobj_as_source prep_source_for amod_tasks_various nn_tasks_language nn_tasks_processing pobj_for_tasks prep_tasks_including amod__taxonomy nn__construction pobj_including_ amod__coreference nn__resolution appos__ cc__and nn_NER_English conj__NER dep_NER_eg dep_eg_Bunescu cc__and nn__Pasca conj__ conj__ conj__Kazama cc__and conj__Torisawa appos__2007 nn_al_Watanabe nn_al_et nsubjpass_used_al ' +o,2290,'Such a lexicon can be used , eg , to classify individual sentences or phrases as subjective or not , and as bearing positive or negative sentiments ',Pang,'predet_lexicon_Such det_lexicon_a nsubjpass_used_lexicon aux_used_can auxpass_used_be dep_used_eg aux_classify_to dep_used_classify amod_sentences_individual dobj_classify_sentences cc_sentences_or conj_sentences_phrases prep_classify_as pobj_as_subjective cc_subjective_or conj_subjective_not cc_as_and conj_as_as pcomp_as_bearing amod_sentiments_positive cc_positive_or conj_positive_negative dobj_bearing_sentiments ' +o,2291,'accuracy Training data 66 \% unsupervised Pang & Lee -LRB- 2004 -RRB- 8715 \% supervised Aue & Gamon -LRB- 2005 -RRB- 914 \% supervised SO 7395 \% unsupervised SM+SO to increase seed words , then SO 7485 \% weakly supervised Table 7 : Classification accuracy on the movie review domain achieves 66 \% accuracy on the movie review domain using the PMI-IR algorithm to gather association scores from the web ',Turney,'nn_data_accuracy nn_data_Training nsubj_unsupervised_data partmod_data_ num_\%_66 dobj__\% dobj_unsupervised_Pang cc_Pang_& conj_Pang_Lee appos_Pang_2004 num_\%_8715 measure_supervised_\% amod_Pang_supervised nsubj_supervised_Aue cc_Aue_& conj_Aue_Gamon appos_Aue_2005 dep_\%_914 amod_Aue_\% dep_unsupervised_supervised nn_SM+SO_SO num_\%_7395 measure_unsupervised_\% amod_SM+SO_unsupervised dobj_supervised_SM+SO aux_increase_to xcomp_supervised_increase nn_words_seed dobj_increase_words advmod_supervised_then number_7485_SO num_\%_7485 dep_Table_\% advmod_supervised_weakly amod_\%_supervised ccomp_supervised_Table num_Table_7 amod_accuracy_Classification nsubj_achieves_accuracy prep_accuracy_on det__the nn__movie nn__review nn__domain pobj_on_ parataxis_supervised_achieves dep_\%_66 amod_accuracy_\% dobj_achieves_accuracy prep_achieves_on det_domain_the nn_domain_movie nn_domain_review pobj_on_domain partmod_domain_using det_algorithm_the nn_algorithm_PMI-IR dobj_using_algorithm aux_gather_to xcomp_using_gather nn_scores_association dobj_gather_scores prep_gather_from det_web_the pobj_from_web ' +o,2292,'In some recent grammar induction and MT work it has been shown that even a small amount of knowledge about a language , in the form of grammar fragments , treelets or prototypes , can go a long way in helping with the induction of a grammar from raw text or with alignment of parallel corpora ',Haghighi,'prep_shown_In det_induction_some amod_induction_recent nn_induction_grammar pobj_In_induction cc_induction_and nn__MT nn__work conj_induction_ nsubjpass_shown_it aux_shown_has auxpass_shown_been complm_go_that advmod_amount_even det_amount_a amod_amount_small nsubj_go_amount prep_amount_of pobj_of_knowledge prep_knowledge_about det_language_a pobj_about_language prep_amount_in det_form_the pobj_in_form prep_form_of amod_fragments_grammar pobj_of_fragments conj_fragments_treelets cc_fragments_or conj_fragments_prototypes aux_go_can ccomp_shown_go det_way_a amod_way_long dobj_go_way prep_go_in pcomp_in_helping prep_helping_with det_induction_the pobj_with_induction prep_induction_of det_grammar_a pobj_of_grammar prep_helping_from amod_text_raw pobj_from_text cc_from_or conj_from_with pobj_with_alignment prep_alignment_of amod_corpora_parallel pobj_of_corpora ' +o,2293,'The results have demonstrated the existence of priming effects in corpus data : they occur for specific syntactic constructions , consistent with the experimental literature , but also generalize to syntactic rules across the board , which repeated more often than expected by chance ',Dubey,'det_results_The nsubj_demonstrated_results aux_demonstrated_have det_existence_the dobj_demonstrated_existence prep_existence_of amod_effects_priming pobj_of_effects prep_effects_in nn_data_corpus pobj_in_data nsubj_occur_they parataxis_demonstrated_occur prep_occur_for amod_constructions_specific amod_constructions_syntactic pobj_for_constructions dobj_occur_ dep__consistent prep_consistent_with det_literature_the amod_literature_experimental pobj_with_literature cc_consistent_but advmod_consistent_also dep_consistent_generalize prep_generalize_to amod_rules_syntactic pobj_to_rules prep_occur_across det_board_the pobj_across_board nsubj_repeated_which rcmod_board_repeated advmod_often_more advmod_repeated_often mark_expected_than dep_often_expected prep_expected_by pobj_by_chance ' +p,2294,'Corpus-based or example-based MT and statistical MT systems provide the easiest customizability , since users have only to supply a collection of source and target sentence pairs -LRB- a bilingual corpus -RRB- ',Brown,'amod__Corpus-based cc_Corpus-based_or conj_Corpus-based_example-based nn__MT nsubj_provide_ cc__and amod_systems_statistical nn_systems_MT nn_systems_ conj__systems det_easiest_the dobj_provide_easiest advmod_provide_customizability mark_have_since nsubj_have_users advcl_provide_have advmod_have_only aux_supply_to xcomp_have_supply det_collection_a dobj_supply_collection prep_collection_of nn_pairs_source cc_source_and conj_source_target nn_pairs_sentence pobj_of_pairs det_corpus_a amod_corpus_bilingual appos_pairs_corpus ' +o,2295,'Treebanks have been used within the field of natural language processing as a source of training data for statistical part og speech taggers and for statistical parsers ',Marcus,'nsubjpass_used_Treebanks aux_used_have auxpass_used_been prep_used_within det_field_the pobj_within_field prep_field_of amod_processing_natural nn_processing_language pobj_of_processing prep_used_as det_source_a pobj_as_source prep_source_of nn_data_training pobj_of_data prep_data_for amod__statistical nn__part nn__og nn__speech nn__taggers pobj_for_ cc_as_and conj_as_for amod_parsers_statistical pobj_for_parsers ' +o,2296,'The next two methods are heuristic -LRB- H -RRB- in and grow-diagonal -LRB- GD -RRB- proposed in ',Och,'det_methods_The amod_methods_next num_methods_two nsubj_proposed_methods cop_heuristic_are rcmod_methods_heuristic dep_heuristic_H prep_heuristic_in pobj_in_ cc__and conj__grow-diagonal abbrev_methods_GD prep_proposed_in pobj_in_ ' +o,2297,'number of words in target string These statistics are combined into a log-linear model whose parameters are adjusted by minimum error rate training ',Och,'prep_number_of pobj_of_words prep_words_in nn_string_target pobj_in_string det_statistics_These nsubjpass_combined_statistics auxpass_combined_are rcmod_string_combined prep_combined_into det_model_a amod_model_log-linear pobj_into_model poss_parameters_whose nsubjpass_adjusted_parameters auxpass_adjusted_are rcmod_model_adjusted prep_adjusted_by amod_training_minimum nn_training_error nn_training_rate pobj_by_training ' +o,2298,'It has also obtained competitive scores on general GR evaluation corpora ',Cahill,'nsubj_obtained_It aux_obtained_has advmod_obtained_also amod_scores_competitive dobj_obtained_scores prep_obtained_on amod_corpora_general nn_corpora_GR nn_corpora_evaluation pobj_on_corpora ' +o,2299,'The following four metrics were used speci cally in this study : BLEU : A weighted geometric mean of the n-gram matches between test and reference sentences multiplied by a brevity penalty that penalizes short translation sentences ',Papineni,'det_metrics_The amod_metrics_following num_metrics_four nsubjpass_used_metrics auxpass_used_were dobj_used_speci advmod_speci_cally prep_used_in det_study_this pobj_in_study nn__BLEU dep_matches_ det_mean_A amod_mean_weighted amod_mean_geometric nsubj_matches_mean prep_mean_of det_n-gram_the pobj_of_n-gram dep_used_matches prep_matches_between nn_sentences_test cc_test_and conj_test_reference pobj_between_sentences partmod_sentences_multiplied prep_multiplied_by det_penalty_a amod_penalty_brevity pobj_by_penalty nsubj_penalizes_that rcmod_penalty_penalizes amod_sentences_short nn_sentences_translation dobj_penalizes_sentences ' +o,2300,'In addition , we use the measure from Resnik , which is computed using an intrinsic information content measure relying on the hierarchical structure of the category tree ',Pedersen,'prep_use_In pobj_In_addition nsubj_use_we det_measure_the dobj_use_measure prep_use_from nn__Resnik pobj_from_ nsubjpass_computed_which auxpass_computed_is rcmod__computed xcomp_computed_using det_measure_an amod_measure_intrinsic nn_measure_information nn_measure_content dobj_using_measure partmod_measure_relying prep_relying_on det_structure_the amod_structure_hierarchical pobj_on_structure prep_structure_of det__the nn__category nn__tree pobj_of_ ' +o,2301,' -RRB- , training on a corpus of one type and then applying the tagger to a corpus of a different type usually results in a tagger with low accuracy ',Marcus,'dep_on_training dep_results_on det_corpus_a pobj_on_corpus prep_corpus_of num_type_one pobj_of_type cc_of_and conj_of_then conj_of_applying det_tagger_the pobj_applying_tagger prep_tagger_to det_corpus_a pobj_to_corpus prep_corpus_of det_type_a amod_type_different pobj_of_type dep_of_usually dobj__results prep_results_in det_tagger_a pobj_in_tagger prep__with amod__low nn__accuracy pobj_with_ ' +o,2302,'Rules have the form X e , f , where e and f are phrases containing terminal symbols -LRB- words -RRB- and possibly co-indexed instances of the nonterminal symbol X2 Associated with each rule is a set of translation model features , i -LRB- f , e -RRB- ; for example , one intuitively natural feature of a rule is the phrase translation -LRB- log - -RRB- probability -LRB- f , e -RRB- = log p -LRB- e f -RRB- , directly analogous to the corresponding feature in non-hierarchical phrase-based models like Pharaoh ',Koehn,'nsubj_have_Rules det_X_the nn_X_form dobj_have_X dep_f_e nsubj_=_f advmod_phrases_where nsubj_phrases_e cc_e_and conj_e_f aux_phrases_are dep_f_phrases xcomp_phrases_containing amod_symbols_terminal dobj_containing_symbols appos_symbols_words cc_f_and advmod_f_possibly amod_instances_co-indexed dep_f_instances prep_instances_of det_symbol_the amod_symbol_nonterminal pobj_of_symbol nn_Associated_X2 nsubj_set_Associated prep_Associated_with det_rule_each pobj_with_rule cop_set_is det_set_a rcmod_instances_set prep_set_of nn_features_translation nn_features_model pobj_of_features appos_features_i dep_i_f appos_f_e prep_probability_for pobj_for_example num_feature_one amod_feature_intuitively amod_feature_natural nsubj_probability_feature prep_feature_of det_rule_a pobj_of_rule cop_probability_is det_probability_the nn_probability_phrase nn_probability_translation dep_-LRB-_log parataxis_set_probability dep_probability_f appos_f_e rcmod_X_= nn_p_log dobj_=_p dep_f_e dep_-LRB-_f advmod_analogous_directly amod_p_analogous prep_=_to det_feature_the amod_feature_corresponding pobj_to_feature prep_feature_in amod_models_non-hierarchical amod_models_phrase-based pobj_in_models prep_models_like pobj_like_Pharaoh ' +o,2303,'This paper continues a line of research on online discriminative training , extending that of Watanabe et al ',Liang,'det_paper_This nsubj_continues_paper det_line_a dobj_continues_line prep_line_of pobj_of_research prep_research_on amod__online amod__discriminative nn__training pobj_on_ xcomp_continues_extending dobj_extending_that prep_that_of pobj_of_Watanabe cc_Watanabe_et conj_Watanabe_al ' +o,2304,'We used the preprocessed data to train the phrase-based translation model by using GIZA + + and the Pharaoh tool kit ',Och,'nsubj_used_We det_data_the amod_data_preprocessed dobj_used_data aux_train_to xcomp_used_train det_model_the amod_model_phrase-based nn_model_translation dobj_train_model prep_train_by pcomp_by_using dep_+_GIZA dep_using_+ cc_+_+ conj_+_ cc_+_and det_kit_the nn_kit_Pharaoh nn_kit_tool dep__kit conj_+_ ' +o,2305,'We can find some other machine-learning approaches that use more sophisticated LMs, such as Decision Trees (Mhrquez and Rodrfguez, 1998)(Magerman, 1996), memory-based approaclms to learn special decision trees (Daelemans et al. , 1996), maximmn entropy approaches that combine statistical information from different sources (Ratnaparkhi, 1996), finite state autonmt2 inferred using Grammatical Inference (Pla and Prieto, 1998), etc. The comparison among different al)t)roaches is dif ficult due to the nmltiple factors that can be eonsid614 ered: tile languagK, tile mmfl)er and tyt)e of the tags, the size of tilt vocabulary, thK ambiguity, the diiticulty of the test ski, Kte.',Ratnaparkhi,'' +o,2306,'The principle of our approach is more similar to ',Yarowsky,'det_principle_The nsubj_similar_principle prep_principle_of poss_approach_our pobj_of_approach cop_similar_is advmod_similar_more xcomp_similar_to ' +o,2307,'We adopt the similarity score proposed by as the distributional similarity score and use 50 nearest neighbors in line with McCarthy et al For the random baseline we select one word sense at random for each word token and average the precision over 100 trials ',Lin,'nsubj_adopt_We det_score_the nn_score_similarity nsubj_proposed_score ccomp_adopt_proposed prep_proposed_by pobj_by_ mark_score_as det_similarity_the amod_similarity_distributional nsubj_score_similarity advcl_proposed_score cc_score_and conj_score_use dobj_score_50 prep_50_nearest pobj_nearest_neighbors prep_neighbors_in pobj_in_line prep_score_with pobj_with_McCarthy cc_McCarthy_et conj_McCarthy_al prep_select_For det_baseline_the amod_baseline_random pobj_For_baseline nsubj_select_we dep_adopt_select num_sense_one nn_sense_word dobj_select_sense prep_select_at pobj_at_random prep_random_for det_word_each pobj_for_word amod_random_token cc_random_and amod_precision_average det_precision_the conj_random_precision prep_precision_over num_trials_100 pobj_over_trials ' +o,2308,'3 Implementation 31 Feature Structure To implement the twin model , we adopt the log linear or maximum entropy -LRB- MaxEnt -RRB- model for its flexibility of combining diverse sources of information ',Berger,'num_Implementation_3 num_Implementation_31 nn_Structure_Feature nsubj_implement_Structure aux_implement_To infmod_Implementation_implement det_model_the amod_model_twin dobj_implement_model nsubj_adopt_we dep_implement_adopt det__the dep_linear_log nn__linear cc_linear_or amod_entropy_maximum conj_linear_entropy appos_linear_MaxEnt nn__model dobj_adopt_ prep__for poss_flexibility_its pobj_for_flexibility prep_Implementation_of pcomp_of_combining amod_sources_diverse dobj_combining_sources prep_sources_of pobj_of_information ' +o,2309,'In non-terminals in a standard PCFG model are augmented with latent variables ',Matsuzaki,'prep_augmented_In amod_non-terminals_ pobj_In_non-terminals prep_non-terminals_in det_PCFG_a amod_PCFG_standard pobj_in_PCFG nsubj_augmented_model aux_augmented_are prep_augmented_with amod_variables_latent pobj_with_variables ' +o,2310,'The tag propagation\\/elimination scheme is adopted from ',Yarowsky,'det_scheme_The nn_scheme_tag nn_scheme_propagation\\/elimination nsubjpass_adopted_scheme auxpass_adopted_is prep_adopted_from ' +o,2311,'The implementation includes path-length , information-content and text-overlap measures , as described in Strube & Ponzetto -LRB- 2006 -RRB- ',Pedersen,'det_implementation_The nsubj_includes_implementation amod__path-length dobj_includes_ amod__information-content conj__ cc__and amod_measures_text-overlap amod_measures_ conj__measures mark_described_as advcl_includes_described prep_described_in pobj_in_Strube cc_Strube_& conj_Strube_Ponzetto dep_described_2006 ' +o,2312,'But it makes obvious that were tackling a problem different from given the fact that their baseline was at 59 \% guessing noun attachment -LRB- rather than 67 \% in the Hindle and Rooth experiments -RRB- 3 Of course , the baseline is not a direct indicator of the difficulty of the disambiguation task ',Marcus,'cc_obvious_But nsubj_obvious_it dep_obvious_makes complm_tackling_that nsubj_tackling_ aux_tackling_were ccomp_obvious_tackling det_problem_a nsubj_different_problem xcomp_tackling_different prep_different_from pobj_from_ dep_fact_given det_fact_the partmod__fact complm_guessing_that poss_baseline_their nsubj_guessing_baseline aux_guessing_was advmod_guessing_at num_\%_59 dep_at_\% ccomp_fact_guessing nn_attachment_noun dobj_guessing_attachment dep_\%_rather dep_rather_than num_\%_67 dep_attachment_\% prep_\%_in det_experiments_the nn_experiments_Hindle cc_Hindle_and conj_Hindle_Rooth pobj_in_experiments dep_attachment_3 prep_indicator_Of pobj_Of_course det_baseline_the nsubj_indicator_baseline cop_indicator_is neg_indicator_not det_indicator_a amod_indicator_direct dep_obvious_indicator prep_indicator_of det_difficulty_the pobj_of_difficulty prep_difficulty_of det_task_the nn_task_disambiguation pobj_of_task ' +o,2313,'For a detailed introduction to IBM translation models , please see ',Brown,'prep_please_For det_introduction_a amod_introduction_detailed pobj_For_introduction prep_introduction_to nn_models_IBM nn_models_translation pobj_to_models dep_please_see ' +o,2314,'By using only the bidirectional word alignment links , one can implement a very robust such filter , as the bidirectional links are generally reliable , even though they have low recall for overall translational correspondences ',Koehn,'prep_filter_By pcomp_By_using advmod_word_only det_word_the amod_word_bidirectional measure_alignment_word amod_links_alignment dobj_using_links nsubj_filter_one aux_filter_can dep_filter_implement det_filter_a advmod_robust_very amod_filter_robust amod_filter_such mark_reliable_as det_links_the amod_links_bidirectional nsubj_reliable_links cop_reliable_are advmod_reliable_generally advcl_filter_reliable advmod_have_even dep_have_though nsubj_have_they ccomp_reliable_have amod_recall_low dobj_have_recall prep_recall_for amod_correspondences_overall amod_correspondences_translational pobj_for_correspondences ' +p,2315,'Automated evaluation metrics that rate system behavior based on automatically computable properties have been developed in a number of other fields : widely used measures include BLEU for machine translation and ROUGE for summarisation , for example ',Papineni,'amod_metrics_Automated nn_metrics_evaluation nsubj_include_metrics complm_developed_that nn_behavior_rate nn_behavior_system nsubjpass_developed_behavior prep_behavior_based dep_based_on dep_on_automatically amod_properties_computable pobj_on_properties aux_developed_have auxpass_developed_been dep_metrics_developed prep_developed_in det_number_a pobj_in_number prep_number_of amod_fields_other pobj_of_fields advmod_used_widely amod_measures_used dep_number_measures nn__BLEU dobj_include_ prep__for nn_translation_machine pobj_for_translation cc__and nn__ROUGE conj__ prep__for pobj_for_summarisation prep_include_for pobj_for_example ' +o,2316,'Agglomerative clustering iteratively merges the most similar clusters into bigger clusters , which need to be labeled ',Brown,'advmod_clustering_Agglomerative amod_iteratively_ dobj_clustering_iteratively dep_clustering_merges det_clusters_the advmod_clusters_most amod_clusters_similar nsubj_clustering_clusters prep_clusters_into amod_clusters_bigger pobj_into_clusters nsubj_need_which rcmod_clusters_need aux_labeled_to auxpass_labeled_be xcomp_need_labeled ' +o,2317,'After parsing the corpus , we artificially introduced verb form errors into these sentences , and observed the resulting disturbances to the parse trees ',Collins,'prep_introduced_After pcomp_After_parsing det__the nn__corpus dobj_parsing_ nsubj_introduced_we advmod_introduced_artificially amod_errors_verb nn_errors_form dobj_introduced_errors prep_introduced_into det_sentences_these pobj_into_sentences cc_introduced_and conj_introduced_observed det_disturbances_the amod_disturbances_resulting dobj_observed_disturbances prep_observed_to det_trees_the amod_trees_parse pobj_to_trees ' +o,2318,'3 Statistical Word Alignment According to the IBM models , the statistical word alignment model can be generally represented as in equation -LRB- 1 -RRB- ',Brown,'num_Alignment_3 nn_Alignment_Statistical nn_Alignment_Word prep_represented_According dep_According_to det__the nn__IBM nn__models pobj_to_ det_model_the amod_model_statistical nn_model_word nn_model_alignment nsubjpass_represented_model aux_represented_can auxpass_represented_be advmod_represented_generally dep_Alignment_represented prep_represented_as dep_as_in pobj_in_equation appos_equation_1 ' +p,2319,'Specifically , in the task of word alignment , heuristic approaches such as the Dice coefficient consistently underperform their re-estimated counterparts , such as the IBM word alignment models ',Brown,'advmod_coefficient_Specifically prep_coefficient_in det_task_the pobj_in_task prep_task_of nn_alignment_word pobj_of_alignment amod_approaches_heuristic nsubj_coefficient_approaches dep_as_such prep_approaches_as det_Dice_the pobj_as_Dice advmod_coefficient_consistently dep_coefficient_underperform poss_counterparts_their amod_counterparts_re-estimated dobj_underperform_counterparts dep_as_such prep_counterparts_as det_models_the nn_models_IBM nn_models_word nn_models_alignment pobj_as_models ' +p,2320,'More recent work has achieved state-of-the-art results with Maxi101 mum entropy conditional Markov models -LRB- MaxEnt CMMs , or MEMMs for short -RRB- ',Ratnaparkhi,'advmod_achieved_More amod_work_recent dep_More_work aux_achieved_has amod_results_state-of-the-art dobj_achieved_results prep_achieved_with nn_mum_Maxi101 pobj_with_mum amod_models_entropy amod_models_conditional nn_models_Markov nsubj_achieved_models dep_models_MaxEnt dep_MaxEnt_CMMs cc_MaxEnt_or conj_MaxEnt_MEMMs prep_MEMMs_for pobj_for_short ' +o,2321,'1 Introduction In a classical statistical machine translation , a foreign language sentence f J1 = f1 , f2 , fJ is translated into another language , ie English , eI1 = e1 , e2 , , eI by seeking a maximum likely solution of : eI1 = argmax eI1 Pr -LRB- eI1 f J1 -RRB- -LRB- 1 -RRB- = argmax eI1 Pr -LRB- f J1 eI1 -RRB- Pr -LRB- eI1 -RRB- -LRB- 2 -RRB- The source channel approach in Equation 2 independently decomposes translation knowledge into a translation model and a language model , respectively ',Brown,'dep_translated_1 nsubjpass_translated_Introduction prep_f_In det_translation_a amod_translation_classical amod_translation_statistical nn_translation_machine pobj_In_translation det_sentence_a amod_sentence_foreign nn_sentence_language nsubj_f_sentence rcmod_Introduction_f nsubj_fJ_J1 dep_fJ_= amod_fJ_f1 amod_fJ_f2 ccomp_f_fJ auxpass_translated_is prep_translated_into det_language_another pobj_into_language nn_English_ie appos_language_English amod_language_eI1 tmod_eI1_= num_=_e1 num_=_e2 partmod_language_eI prep_eI_by pcomp_by_seeking det_solution_a amod_solution_maximum amod_solution_likely dobj_seeking_solution prep_solution_of nsubj_argmax_eI1 dep_argmax_= dep_Pr_argmax amod_Pr_eI1 pobj_of_Pr amod_J1_eI1 amod_J1_f appos_Pr_J1 appos_solution_1 dep_solution_= nn_eI1_argmax iobj_=_eI1 nn_Pr_Pr dep_-LRB-_f nn_eI1_J1 dep_f_eI1 dobj_=_Pr appos_Pr_eI1 dep_decomposes_2 det_approach_The nn_approach_source nn_approach_channel nsubj_decomposes_approach prep_approach_in pobj_in_Equation num_Equation_2 advmod_decomposes_independently dep_translated_decomposes nn_knowledge_translation dobj_decomposes_knowledge prep_decomposes_into det_model_a nn_model_translation pobj_into_model cc_model_and det_model_a nn_model_language conj_model_model dep_into_respectively ' +o,2322,'This is an unsuitable measure for inferring reliability , and it was the use of this measure that prompted to recommend chance-corrected measures ',Carletta,'nsubj_measure_This cop_measure_is det_measure_an amod_measure_unsuitable prep_measure_for pcomp_for_inferring advmod_inferring_reliability cc_measure_and nsubj_use_it cop_use_was det_use_the conj_measure_use prep_use_of det_measure_this pobj_of_measure nsubj_prompted_that rcmod_measure_prompted acomp_prompted_ aux_recommend_to xcomp__recommend amod_measures_chance-corrected dobj_recommend_measures ' +o,2323,'Once the set of features functions are selected , algorithm such as improved iterative scaling or sequential conditional generalized iterative scaling can be used to find the optimal parameter values of fkg and fig ',Berger,'mark_selected_Once det_set_the nsubjpass_selected_set prep_set_of nn_functions_features pobj_of_functions auxpass_selected_are advcl_used_selected nsubjpass_used_algorithm dep_as_such prep_algorithm_as amod__improved amod__iterative amod__scaling pobj_as_ cc__or amod__sequential dep_sequential_conditional amod__generalized nn__iterative amod__scaling conj__ aux_used_can auxpass_used_be aux_find_to purpcl_used_find det_values_the amod_values_optimal nn_values_parameter dobj_find_values prep_values_of pobj_of_fkg cc_fkg_and conj_fkg_fig ' +o,2324,'In the similaritybased approaches , rather than a class , each word is modelled by its own set of similar words derived from statistical data collected from corpora ',Dagan,'prep_modelled_In det__the amod__similaritybased nn__approaches pobj_In_ cc__rather dep_rather_than det_class_a conj__class det_word_each nsubjpass_modelled_word auxpass_modelled_is prep_modelled_by poss_set_its amod_set_own pobj_by_set prep_set_of amod_words_similar pobj_of_words partmod_words_derived prep_derived_from amod_data_statistical pobj_from_data partmod_data_collected prep_collected_from pobj_from_corpora ' +o,2325,'Estimation of the parameters has been described elsewhere ',Brown,'nsubjpass_described_Estimation prep_Estimation_of det_parameters_the pobj_of_parameters aux_described_has auxpass_described_been advmod_described_elsewhere ' +o,2326,'According to one account the majority of errors arise because of the statistical filtering process , which is reported to be particularly unreliable for low frequency SCFs ',Dunning,'prep__According dep_According_to num_account_one pobj_to_account det_majority_the nsubj__majority prep_majority_of pobj_of_errors rcmod_errors_arise dep_of_because prep_arise_of det_process_the amod_process_statistical amod_process_filtering pobj_of_process nsubjpass_reported_which auxpass_reported_is rcmod_process_reported aux_unreliable_to cop_unreliable_be advmod_unreliable_particularly xcomp_reported_unreliable prep_unreliable_for amod__low nn__frequency nn__SCFs pobj_for_ ' +o,2327,'Sentiment summarization has been well studied in the past decade ',Turney,'nn_summarization_Sentiment nsubjpass_studied_summarization aux_studied_has auxpass_studied_been advmod_studied_well prep_studied_in det_decade_the amod_decade_past pobj_in_decade ' +o,2328,'This is also true for reranking and discriminative training , where the k-best list of candidates serves as an approximation of the full set ',Och,'nsubj_true_This cop_true_is advmod_true_also prep_true_for amod_training_reranking cc_reranking_and conj_reranking_discriminative pobj_for_training advmod_serves_where det_list_the amod_list_k-best nsubj_serves_list prep_list_of pobj_of_candidates rcmod_training_serves prep_serves_as det_approximation_an pobj_as_approximation prep_approximation_of det_set_the amod_set_full pobj_of_set ' +o,2329,'The tree-based reranker includes the features described in as well as features based on non-projective edge attributes explored in ',Hall,'det_reranker_The amod_reranker_tree-based nsubj_includes_reranker det_features_the nsubj_explored_features partmod_features_described prep_described_in pobj_in_ dep_well_as cc__well dep_well_as conj__features partmod__based prep_based_on amod_attributes_non-projective nn_attributes_edge pobj_on_attributes ccomp_includes_explored prep_explored_in pobj_in_ ' +o,2330,'The chunking classification was made by based on the parsing information in the WSJ corpus ',Ramshaw,'det_classification_The amod_classification_chunking nsubjpass_made_classification auxpass_made_was prep_made_by pobj_by_ prep_made_based dep_based_on det_information_the amod_information_parsing pobj_on_information prep_information_in det_corpus_the nn_corpus_WSJ pobj_in_corpus ' +o,2331,'A similar approach was taken in where an unknown word was guessed given the probabilities for an unknown word to be of a particular POS , its capitalization feature and its ending ',Marcus,'det_approach_A amod_approach_similar nsubjpass_taken_approach auxpass_taken_was prep_taken_in pcomp_in_ advmod_given_where det_word_an amod_word_unknown nsubjpass_given_word auxpass_given_was advmod_given_guessed advcl__given det_probabilities_the dobj_given_probabilities prep_given_for det_word_an amod_word_unknown pobj_for_word aux_be_to xcomp_given_be prep_be_of det_POS_a amod_POS_particular pobj_of_POS poss_feature_its nn_feature_capitalization conj_POS_feature cc_POS_and poss_ending_its conj_POS_ending ' +o,2332,'In retrospect , however , there are perhaps even greater similarities to that of ',Matsuzaki,'prep_are_In pobj_In_retrospect advmod_are_however expl_are_there advmod_are_perhaps advmod_greater_even amod_similarities_greater nsubj_are_similarities prep_similarities_to pobj_to_that prep_that_of ' +o,2333,'Decoding weights are optimized using Ochs algorithm to set weights for the four components of the loglinear model : language model , phrase translation model , distortion model , and word-length feature ',Och,'amod_weights_Decoding auxpass_optimized_are dep_weights_optimized xcomp_optimized_using nn__Ochs nn__algorithm nsubj_set_ aux_set_to xcomp_using_set dobj_set_weights prep_set_for det_components_the num_components_four pobj_for_components prep_components_of det_model_the nn_model_loglinear pobj_of_model nn_model_language dep_weights_model nn_model_phrase nn_model_translation conj_model_model nn_model_distortion conj_model_model cc_model_and amod_feature_word-length conj_model_feature ' +o,2334,'We use the GIZA + + implementation of IBM Model 4 coupled with the phrase extraction heuristics of Koehn et al ',Brown,'nsubj_use_We det_implementation_the amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dobj_use_implementation prep_implementation_of nn__IBM nn__Model num__4 pobj_of_ partmod__coupled prep_coupled_with det_heuristics_the nn_heuristics_phrase nn_heuristics_extraction pobj_with_heuristics prep_heuristics_of pobj_of_Koehn cc_Koehn_et conj_Koehn_al ' +o,2335,'While simple statistical alignment models like IBM-1 and the symmetric alignment approach by Hiemstra -LRB- 1996 -RRB- treat sentences as unstructured bags of words , the more sophisticated IBM-models by Brown et al ',Brown,'mark_treat_While amod_models_simple amod_models_statistical nn_models_alignment nsubj_treat_models prep_models_like nn__IBM-1 pobj_like_ cc_models_and det_approach_the amod_approach_symmetric nn_approach_alignment conj_models_approach prep_approach_by pobj_by_Hiemstra appos_models_1996 dobj_treat_sentences prep_sentences_as amod_bags_unstructured pobj_as_bags prep_bags_of pobj_of_words det_IBM-models_the advmod_sophisticated_more amod_IBM-models_sophisticated appos_words_IBM-models prep_treat_by pobj_by_Brown cc_Brown_et conj_Brown_al ' +p,2336,'51 The AUGMENT technique for Domain Adaptation The AUGMENT technique introduced by is a simple yet very effective approach to performing domain adaptation ','Daume III','det_technique_The amod_technique_AUGMENT nsubj_approach_technique prep_technique_for nn_Adaptation_Domain pobj_for_Adaptation det_technique_The nn_technique_AUGMENT nsubj_introduced_technique rcmod_Adaptation_introduced prep_introduced_by pobj_by_ cop_approach_is det_approach_a amod_approach_simple cc_simple_yet advmod_effective_very conj_simple_effective dep_51_approach prep_approach_to pcomp_to_performing nn_adaptation_domain dobj_performing_adaptation ' +o,2337,'Two block sets are derived for each of the training sets using a phrase-pair selection algorithm similar to ',Koehn,'num_sets_Two nn_sets_block nsubjpass_derived_sets auxpass_derived_are prep_derived_for pobj_for_each prep_each_of det_sets_the nn_sets_training pobj_of_sets xcomp_derived_using det_algorithm_a amod_algorithm_phrase-pair nn_algorithm_selection nsubj_similar_algorithm xcomp_using_similar prep_similar_to ' +o,2338,'The different approaches vary largely according to the methods used and the number of SCFS being extracted ',Dunning,'det__The amod__different nn__approaches nsubj_vary_ advmod_vary_largely prep_vary_according dep_according_to det_methods_the pobj_to_methods partmod_methods_used cc_methods_and det_number_the conj_methods_number prep_number_of pobj_of_SCFS auxpass_extracted_being partmod_SCFS_extracted ' +o,2339,' and -LRB- and see below for discussions -RRB- , so in this paper we focus on the less studied , but equally important problem of annotationstyle adaptation ','Daume III','cc__and conj__ dep_see_and dep__see advmod_see_below prep_see_for pobj_for_discussions mark_focus_so prep_focus_in det_paper_this pobj_in_paper nsubj_focus_we advcl__focus prep_focus_on det_problem_the advmod_studied_less amod_problem_studied cc_studied_but advmod_important_equally conj_studied_important pobj_on_problem prep_problem_of amod_adaptation_annotationstyle pobj_of_adaptation ' +o,2340,'Syntactic context information is used to compute term similarities , based on which similar words to a particular word can directly be returned ',Hindle,'amod_information_Syntactic nn_information_context nsubjpass_used_information auxpass_used_is advmod_used_ aux_compute_to xcomp_used_compute nn_similarities_term dobj_compute_similarities prep_used_based dep_based_on dobj_returned_which amod_words_similar nsubjpass_returned_words prep_words_to det_word_a amod_word_particular pobj_to_word aux_returned_can advmod_returned_directly auxpass_returned_be pcomp_on_returned ' +o,2341,'To obtain their corresponding weights , we adapted the minimum-error-rate training algorithm to train the outside-layer model ',Och,'aux_obtain_To dep_adapted_obtain poss_weights_their amod_weights_corresponding dobj_obtain_weights nsubj_adapted_we det_algorithm_the amod_algorithm_minimum-error-rate nn_algorithm_training nsubj__algorithm ccomp_adapted_ aux_train_to xcomp__train det_model_the amod_model_outside-layer dobj_train_model ' +o,2342,'For a full description of the algorithm , see ',Collins,'prep_see_For det_description_a amod_description_full pobj_For_description prep_description_of det_algorithm_the pobj_of_algorithm ' +o,2343,'Following , we adopt the view that the syntactic structure of sentences paraphrasing some sentence s should be inspired by the structure of s Because dependency syntax is still only a crude approximation to semantic structure , we augment the model with a lexical semantics component , based on WordNet , that models how words are probabilistically altered in generating a paraphrase ',Smith,'prep_adopt_Following pobj_Following_ nsubj_adopt_we det_view_the dobj_adopt_view complm_inspired_that det_structure_the amod_structure_syntactic nsubjpass_inspired_structure prep_structure_of pobj_of_sentences partmod_sentences_paraphrasing det_s_some nn_s_sentence dobj_paraphrasing_s aux_inspired_should auxpass_inspired_be ccomp_adopt_inspired prep_inspired_by det_structure_the pobj_by_structure prep_structure_of pobj_of_s mark_approximation_Because amod_syntax_dependency nsubj_approximation_syntax cop_approximation_is advmod_approximation_still advmod_approximation_only det_approximation_a amod_approximation_crude advcl_model_approximation prep_approximation_to amod_structure_semantic pobj_to_structure nsubj_model_we dep_model_augment det_model_the dep_adopt_model prep_model_with det_component_a amod_component_lexical nn_component_semantics pobj_with_component partmod_component_based prep_based_on nn__WordNet pobj_on_ det_models_that appos__models advmod_altered_how nsubjpass_altered_words auxpass_altered_are advmod_altered_probabilistically dep_model_altered prep_altered_in pcomp_in_generating det_paraphrase_a dobj_generating_paraphrase ' +o,2344,'We consider the outputs of the top 3 allwords WSD systems that participated in Senseval-3 : Gambl , SenseLearner , and KOC University ',Pedersen,'nsubj_consider_We det_outputs_the dobj_consider_outputs prep_outputs_of det_allwords_the amod_allwords_top num_allwords_3 pobj_of_allwords nn_systems_WSD dep_consider_systems nsubj_participated_that rcmod_systems_participated prep_participated_in pobj_in_Senseval-3 nn__Gambl dobj_consider_ nn__SenseLearner conj__ cc__and nn__KOC nn__University conj__ ' +o,2345,'33 System evaluation Since both the system translations and the reference translations are available for the tuning 43 set , we first compare each output to the reference translation using BLEU and METEOR and a combined scoring scheme provided by the ULC toolkit ',Banerjee,'num_evaluation_33 nn_evaluation_System mark_available_Since preconj_translations_both det_translations_the nn_translations_system nsubj_available_translations cc_translations_and det_translations_the nn_translations_reference conj_translations_translations cop_available_are advcl_compare_available prep_available_for det_set_the amod_set_tuning num_set_43 pobj_for_set nsubj_compare_we advmod_compare_first dep_evaluation_compare det_output_each dobj_compare_output prep_compare_to det_translation_the nn_translation_reference pobj_to_translation partmod_translation_using nn__BLEU dobj_using_ cc__and nn__METEOR conj__ cc__and det_scheme_a amod_scheme_combined nn_scheme_scoring conj__scheme partmod_scheme_provided prep_provided_by det__the nn__ULC nn__toolkit pobj_by_ ' +o,2346,'Specifically , we will consider a system which was developed for the ACE -LRB- Automatic Content Extraction -RRB- task 3 and includes the following stages : name structure parsing , coreference , semantic relation extraction and event extraction ',Ji,'advmod_consider_Specifically nsubj_consider_we aux_consider_will det_system_a dobj_consider_system nsubjpass_developed_which auxpass_developed_was rcmod_system_developed prep_developed_for det_task_the nn_task_ACE nn_Extraction_Automatic nn_Extraction_Content appos_task_Extraction pobj_for_task dep_task_3 cc_consider_and conj_consider_includes det_stages_the amod_stages_following dobj_includes_stages nn_parsing_name nn_parsing_structure dep_stages_parsing conj_parsing_coreference amod_extraction_semantic nn_extraction_relation conj_parsing_extraction cc_parsing_and nn_extraction_event conj_parsing_extraction ' +o,2347,'The preprocessed training data was filtered for length and aligned using the GIZA + + implementation of IBM Model 4 in both directions and symmetrized using the grow-diag-final-and heuristic ',Och,'det_data_The amod_data_preprocessed nn_data_training nsubjpass_filtered_data auxpass_filtered_was prep_filtered_for pobj_for_length cc_filtered_and conj_filtered_aligned xcomp_aligned_using det_implementation_the amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dobj_using_implementation prep_implementation_of nn_Model_IBM pobj_of_Model num__4 dobj_filtered_ prep_filtered_in det_directions_both pobj_in_directions cc_filtered_and conj_filtered_symmetrized xcomp_symmetrized_using det_heuristic_the amod_heuristic_grow-diag-final-and dobj_using_heuristic ' +o,2348,'This gives the translation model more information about the structure of the source language , and further constrains the reorderings to match not just a possible bracketing as in , but the specific bracketing of the parse tree provided ',Wu,'nsubj_gives_This det_model_the nn_model_translation iobj_gives_model amod_information_more dobj_gives_information prep_information_about det_structure_the pobj_about_structure prep_structure_of det_language_the nn_language_source pobj_of_language cc_gives_and advmod_constrains_further conj_gives_constrains det_reorderings_the nsubj_match_reorderings aux_match_to xcomp_constrains_match neg_just_not dobj_match_just det_as_a amod_as_possible amod_as_bracketing dep_constrains_as prep_as_in pobj_in_ cc_constrains_but det_bracketing_the amod_bracketing_specific nsubj_provided_bracketing prep_bracketing_of det_tree_the amod_tree_parse pobj_of_tree conj_constrains_provided ' +o,2349,'To support a more rigorous analysis , however , wc have followed of using the K coettMcnt as a measure of coder agreement ',Carletta,'aux_support_To dep_followed_support det_analysis_a amod_analysis_more amod_analysis_rigorous dobj_support_analysis advmod_followed_however nsubj_followed_wc aux_followed_have dobj_followed_ prep__of pcomp_of_using det__the nn__K nn__coettMcnt dobj_using_ prep_using_as det_measure_a pobj_as_measure prep_measure_of nn_agreement_coder pobj_of_agreement ' +o,2350,'Co-selection measures include precision and recall of co-selected sentences , relative utility , and Kappa ',Carletta,'amod_measures_Co-selection nsubj_include_measures dobj_include_precision cc_precision_and conj_precision_recall prep_precision_of amod_sentences_co-selected pobj_of_sentences amod__relative nn__utility conj_sentences_ cc_sentences_and conj_sentences_Kappa ' +o,2351,'Goodman and Johnson both suggest this strategy ',Collins,'nn__Goodman nsubj__ cc__and conj__Johnson nsubj_suggest_both ccomp__suggest det_strategy_this dobj_suggest_strategy ' +o,2352,'61 Hiero Results Using the MT 2002 test set , we ran the minimumerror rate training -LRB- MERT -RRB- with the decoder to tune the weights for each feature ',Och,'num_Results_61 nn_Results_Hiero dep_ran_Using det_test_the nn_test_MT num_test_2002 dobj_Using_test partmod_test_set nsubj_ran_we dep_Results_ran det_training_the nn_training_minimumerror nn_training_rate nsubj__training abbrev_training_MERT ccomp_ran_ prep__with det_decoder_the pobj_with_decoder aux_tune_to infmod_decoder_tune det_weights_the dobj_tune_weights prep_tune_for det_feature_each pobj_for_feature ' +o,2353,'An alternative representation for baseNPs has been put forward by ',Ramshaw,'det_representation_An amod_representation_alternative nsubjpass_put_representation prep_representation_for pobj_for_baseNPs aux_put_has auxpass_put_been advmod_put_forward prep_put_by ' +o,2354,'We compare those algorithms to generalized iterative scaling -LRB- GIS -RRB- , non-preconditioned CG , and voted perceptron training ',Collins,'nsubj_compare_We det_algorithms_those dobj_compare_algorithms prep_compare_to amod__generalized dep_generalized_iterative nn__scaling abbrev__GIS pobj_to_ amod_CG_non-preconditioned appos__CG cc_compare_and conj_compare_voted nn_training_perceptron dobj_voted_training ' +o,2355,'The other main difference is the apparently nonlocal nature of the problem , which motivates our choice of a Maximum Entropy -LRB- ME -RRB- model for the tagging task ',Berger,'det_difference_The amod_difference_other amod_difference_main nsubj_nature_difference cop_nature_is det_nature_the advmod_nonlocal_apparently amod_nature_nonlocal prep_nature_of det_problem_the pobj_of_problem nsubj_motivates_which rcmod_problem_motivates poss_choice_our dobj_motivates_choice prep_choice_of det_model_a amod_model_Maximum nn_model_Entropy abbrev_model_ME pobj_of_model prep_model_for det_task_the amod_task_tagging pobj_for_task ' +o,2356,'1 Introduction Statistical machine translation has seen many improvements in recent years , most notably the transition from wordto phrase-based models ',Brown,'num__1 nn__Introduction nn__Statistical nn__machine nn__translation nsubj_seen_ aux_seen_has amod_improvements_many dobj_seen_improvements prep_seen_in amod_years_recent pobj_in_years advmod_notably_most advmod_years_notably det_transition_the dep_years_transition prep_transition_from amod__wordto amod__phrase-based nn__models pobj_from_ ' +o,2357,' To reduce the inference time , following , we collapsed the 45 different POS labels contained in the original data ',Ramshaw,'dep_collapsed_ aux_reduce_To xcomp__reduce det_time_the amod_time_inference dobj_reduce_time prep_time_following pobj_following_ nsubj_collapsed_we det_labels_the num_labels_45 amod_labels_different dep_labels_POS nsubj_contained_labels ccomp_collapsed_contained prep_contained_in det_data_the amod_data_original pobj_in_data ' +o,2358,'For example , in phrase-based SMT systems , distortion model is used , in which reordering probabilities depend on relative positions of target side phrases between adjacent blocks ',Koehn,'prep_used_For pobj_For_example prep_used_in amod__phrase-based nn__SMT nn__systems pobj_in_ nn_model_distortion nsubjpass_used_model auxpass_used_is prep_used_in dep_depend_which nn_probabilities_reordering nsubj_depend_probabilities pcomp_in_depend prep_depend_on amod_positions_relative pobj_on_positions prep_positions_of nn_phrases_target nn_phrases_side pobj_of_phrases prep_phrases_between amod_blocks_adjacent pobj_between_blocks ' +p,2359,'Thus , as a powerful sequence tagging model , CRF became the dominant method in the Bakeoff 2006 ',Levow,'advmod_method_Thus prep_method_as det_sequence_a amod_sequence_powerful pobj_as_sequence partmod_sequence_tagging dobj_tagging_model nsubj_method_CRF cop_method_became det_method_the amod_method_dominant prep_method_in det_Bakeoff_the pobj_in_Bakeoff num_Bakeoff_2006 ' +p,2360,' improves over by suggesting a simpler approach ',Dasgupta,'nsubj_improves_ prep_improves_over pcomp_over_ prep__by pcomp_by_suggesting det_approach_a amod_approach_simpler dobj_suggesting_approach ' +p,2361,'Nowadays , most of the state-of-the-art SMT systems are based on bilingual phrases ',Och,'nsubjpass_based_Nowadays nsubjpass_based_most prep_most_of det_systems_the amod_systems_state-of-the-art nn_systems_SMT pobj_of_systems auxpass_based_are prep_based_on amod_phrases_bilingual pobj_on_phrases ' +o,2362,'The word alignment models implemented in GIZA + + , the so-called IBM and HMM alignment models are typical implementation of the EM algorithm ',Brown,'det_models_The nn_models_word nn_models_alignment nsubj_implementation_models partmod_models_implemented prep_implemented_in nn_+_GIZA nn_+_+ pobj_in_+ det__the amod__so-called nn__IBM conj_+_ cc_+_and nn__HMM amod__alignment nn__models conj_+_ cop_implementation_are amod_implementation_typical prep_implementation_of det__the nn__EM nn__algorithm pobj_of_ ' +o,2363,'c2009 Association for Computational Linguistics Improving Mid-Range Reordering using Templates of Factors Hieu Hoang School of Informatics University of Edinburgh hhoang @ smsedacuk Philipp Koehn School of Informatics University of Edinburgh pkoehn @ infedacuk Abstract We extend the factored translation model to allow translations of longer phrases composed of factors such as POS and morphological tags to act as templates for the selection and reordering of surface phrase translation ',Koehn,'nsubj_smsedacuk_Association prep_Association_for nn_Reordering_Computational nn_Reordering_Linguistics nn_Reordering_Improving nn_Reordering_Mid-Range pobj_for_Reordering partmod_Association_using dobj_using_Templates prep_Templates_of nn_School_Factors nn_School_Hieu nn_School_Hoang pobj_of_School prep_School_of nn_University_Informatics pobj_of_University prep_University_of nn_hhoang_Edinburgh pobj_of_hhoang dep_Templates_@ rcmod_c2009_smsedacuk nn_School_Philipp nn_School_Koehn dobj_smsedacuk_School prep_School_of nn_University_Informatics pobj_of_University prep_University_of nn_pkoehn_Edinburgh pobj_of_pkoehn dep_Abstract_@ nn_Abstract_infedacuk dep_c2009_Abstract nsubj_extend_We rcmod_Abstract_extend det_model_the amod_model_factored nn_model_translation nsubj__model ccomp_extend_ aux_allow_to xcomp__allow dobj_allow_translations prep_Abstract_of amod_phrases_longer pobj_of_phrases partmod_phrases_composed prep_composed_of pobj_of_factors dep_as_such prep_factors_as pobj_as_POS cc_POS_and amod_tags_morphological conj_POS_tags aux_act_to xcomp_composed_act prep_act_as pobj_as_templates prep_templates_for det_selection_the pobj_for_selection cc_selection_and conj_selection_reordering prep_selection_of nn_translation_surface nn_translation_phrase pobj_of_translation ' +p,2364,'Global information is known to be useful in other NLP tasks , especially in the named entity recognition task , and several studies successfully used global features ',Finkel,'amod_information_Global nsubjpass_known_information auxpass_known_is aux_useful_to cop_useful_be xcomp_known_useful prep_useful_in amod_tasks_other nn_tasks_NLP pobj_in_tasks dep_in_especially prep_useful_in det_task_the amod_task_named nn_task_entity nn_task_recognition pobj_in_task cc_known_and amod_studies_several nsubj_used_studies advmod_used_successfully conj_known_used amod_features_global dobj_used_features ' +p,2365,'All state-of-the-art wide-coverage parsers relax this assumption in some way , for instance by -LRB- i -RRB- changing the parser in step -LRB- 3 -RRB- , such that the application of rules is conditioned on other steps in the derivation process , or by -LRB- ii -RRB- enriching the nonterminal labels in step -LRB- 1 -RRB- with context-information , along with suitable backtransforms in step -LRB- 4 -RRB- ',Collins,'det_parsers_All amod_parsers_state-of-the-art amod_parsers_wide-coverage nsubj_relax_parsers det_assumption_this dobj_relax_assumption prep_assumption_in det_way_some pobj_in_way prep_relax_for pobj_for_instance prep_instance_by nsubj_changing_i pcomp_by_changing det_parser_the dobj_changing_parser prep_parser_in pobj_in_step dep_such_3 dep_step_such complm_conditioned_that det_application_the nsubjpass_conditioned_application prep_application_of pobj_of_rules auxpass_conditioned_is dep_such_conditioned prep_conditioned_on amod_steps_other pobj_on_steps prep_steps_in det__the nn__derivation nn__process pobj_in_ cc_by_or conj_by_by dep_labels_ii advmod_labels_enriching det_labels_the amod_labels_nonterminal pobj_by_labels prep_labels_in pobj_in_step appos_step_1 prep_labels_with amod__context-information pobj_with_ advmod_labels_along dep_along_with amod_backtransforms_suitable pobj_with_backtransforms prep_backtransforms_in pobj_in_step appos_labels_4 ' +o,2366,' s Inversion Transduction Grammar , as well as tree-transformation models of translation such as Yamada and Knight , Galley et al ',Wu,'nsubj_Grammar_ dep_Grammar_s nn_Grammar_Inversion nn_Grammar_Transduction dep_well_as cc_Grammar_well dep_well_as amod_models_tree-transformation conj_Grammar_models prep_models_of pobj_of_translation dep_as_such prep_translation_as pobj_as_Yamada cc_Yamada_and nn__Knight conj_Yamada_ conj__Galley cc__et conj__al ' +o,2367,'Finally , recent work has explored learning to map sentences to lambda-calculus meaning representations ',Zettlemoyer,'advmod_explored_Finally amod_work_recent nsubj_explored_work aux_explored_has xcomp_explored_learning prep_learning_to nn_sentences_map pobj_to_sentences prep_learning_to amod_representations_lambda-calculus nn_representations_meaning pobj_to_representations ' +o,2368,'However , we do not rely on linguistic resources or on search engines to determine the semantic orientation , but rather rely on econometrics for this task ',Turney,'advmod_rely_However nsubj_rely_we aux_rely_do neg_rely_not prep_rely_on amod__linguistic nn__resources pobj_on_ cc_on_or conj_on_on nn_engines_search pobj_on_engines partmod_engines_ aux_determine_to xcomp__determine det_orientation_the amod_orientation_semantic dobj_determine_orientation dep_rather_but cc_determine_rather conj_determine_rely prep_rely_on pobj_on_econometrics prep_econometrics_for det_task_this pobj_for_task ' +o,2369,'We perform term disambiguation on each document using an entity extractor ',Cucerzan,'nsubj_perform_We nn_disambiguation_term dobj_perform_disambiguation prep_perform_on det_document_each pobj_on_document partmod_document_using det_extractor_an nn_extractor_entity dobj_using_extractor ' +o,2370,'In recent years , reranking techniques have been successfully applied to the so-called history-based models , especially to parsing ',Marcus,'prep_applied_In amod_years_recent pobj_In_years amod_techniques_reranking nsubjpass_applied_techniques aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_to det__the amod__so-called amod__history-based nn__models pobj_to_ dep_to_especially dep_to_to amod__parsing pobj_to_ ' +n,2371,'Point-wise mutual information -LRB- PMI -RRB- is commonly used for computing the association of two terms , which is defined as : nullnullnull null null , null null nullnullnull nullnullnullnull , nullnull nullnull null null null nullnullnullnullnull However , we argue that PMI is not a suitable measure for our purpose ',Turney,'amod_information_Point-wise amod_information_mutual nsubjpass_used_information abbrev_information_PMI auxpass_used_is advmod_used_commonly prep_used_for pcomp_for_computing det_association_the dobj_computing_association prep_association_of num_terms_two pobj_of_terms amod_terms_ nsubjpass_defined_which auxpass_defined_is rcmod_terms_defined prep_defined_as amod_nullnullnullnullnull_nullnullnull dep_nullnullnull_null amod_nullnullnullnullnull_null amod_nullnullnullnullnull_null dep_null_null amod_nullnullnullnullnull_nullnullnull amod_nullnullnullnullnull_nullnullnullnull amod_nullnullnullnullnull_nullnull dep_nullnull_nullnull amod_nullnullnullnullnull_null amod_nullnullnullnullnull_null nn_nullnullnullnullnull_null dep_association_nullnullnullnullnull advmod_argue_However nsubj_argue_we dep_used_argue complm_measure_that nsubj_measure_PMI cop_measure_is neg_measure_not det_measure_a amod_measure_suitable ccomp_argue_measure prep_measure_for poss_purpose_our pobj_for_purpose ' +o,2372,'We measured inter-annotator agreement with the Kappa statistic using the 1,391 items that two annotators scored in common ',Carletta,'nsubj_measured_We amod_agreement_inter-annotator dobj_measured_agreement prep_measured_with det__the nn__Kappa nn__statistic pobj_with_ xcomp_measured_using det_items_the num_items_1,391 dobj_using_items complm_scored_that num_annotators_two nsubj_scored_annotators ccomp_using_scored prep_scored_in pobj_in_common ' +o,2373,'This setup provides an elegant solution to the fairly complex task of integrating multiple MT results that may differ in word order using only standard software modules , in particular GIZA + + for the identification of building blocks and Moses for the recombination , but the authors were not able to observe improvements in 1see http://wwwstatmtorg/moses/ terms of BLEU score ',Och,'det_setup_This nsubj_provides_setup det_solution_an amod_solution_elegant dobj_provides_solution prep_provides_to det_task_the advmod_complex_fairly amod_task_complex pobj_to_task prep_task_of pcomp_of_integrating amod_results_multiple nn_results_MT dobj_integrating_results nsubj_differ_that aux_differ_may rcmod_results_differ prep_differ_in nn_order_word pobj_in_order partmod_order_using advmod_modules_only amod_modules_standard nn_modules_software dobj_using_modules prep_provides_in amod__particular nn__GIZA nn__+ nn__+ pobj_in_ prep__for det_identification_the pobj_for_identification prep_identification_of pcomp_of_building dobj_building_blocks cc_blocks_and conj_blocks_Moses prep_building_for det_recombination_the pobj_for_recombination cc_provides_but det_authors_the nsubj_able_authors cop_able_were neg_able_not conj_provides_able aux_observe_to xcomp_able_observe dobj_observe_improvements prep_observe_in amod_terms_1see amod_terms_http://wwwstatmtorg/moses/ pobj_in_terms prep_terms_of amod_score_BLEU pobj_of_score ' +o,2374,'Our baseline uses Giza + + alignments symmetrized with the grow-diag-final-and heuristic ',Koehn,'poss_baseline_Our nsubj_uses_baseline amod_alignments_Giza cc_Giza_+ conj_Giza_+ dobj_uses_alignments advmod_symmetrized_ amod_alignments_symmetrized prep_uses_with det_heuristic_the amod_heuristic_grow-diag-final-and pobj_with_heuristic ' +o,2375,'2 Machine Translation using Inversion Transduction Grammar The Inversion Transduction Grammar -LRB- ITG -RRB- of is a type of context-free grammar -LRB- CFG -RRB- for generating two languages synchronously ',Wu,'num_Translation_2 nn_Translation_Machine dep_Translation_using nn_Grammar_Inversion nn_Grammar_Transduction dobj_using_Grammar det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction nsubj_type_Grammar abbrev_Grammar_ITG prep_Grammar_of pobj_of_ cop_type_is det_type_a rcmod_Grammar_type prep_type_of amod_grammar_context-free pobj_of_grammar abbrev_Translation_CFG prep_Translation_for pcomp_for_generating num_languages_two dobj_generating_languages advmod_generating_synchronously ' +p,2376,'Recent work includes improved model variants and applications such as web data extraction , scientific citation extraction , and word alignment ',Jiao,'amod_work_Recent nsubj_includes_work amod__improved nn__model amod__variants dobj_includes_ cc__and conj__applications dep_as_such prep__as nn__web nn__data nn__extraction pobj_as_ amod__scientific nn__citation nn__extraction conj__ cc__and nn__word nn__alignment conj__ ' +o,2377,'All of the features of the ATR\\/Lancaster Treebank that are described below represent a radical departure from extant large-scale treebanks ',Marcus,'nsubj_represent_All prep_All_of det_features_the pobj_of_features prep_features_of det_Treebank_the nn_Treebank_ATR\\/Lancaster pobj_of_Treebank nsubjpass_described_that auxpass_described_are rcmod_features_described advmod_described_below det_departure_a amod_departure_radical dobj_represent_departure prep_represent_from amod_treebanks_extant amod_treebanks_large-scale nn_treebanks_ pobj_from_treebanks ' +o,2378,'C0 , C , q 1 , q xq xq1 xq1 xq xr xr +1 Table 6 : Lexicalized Features for Joint Models aging of the weights suggested by ',Collins,'dep_6_C0 appos_C0_C appos_C0_q num_q_1 nn_xr_q nn_xr_xq nn_xr_xq1 nn_xr_xq1 nn_xr_xq appos_C0_xr dep_6_xr num_xr_+1 number_6_Table nn_Features_Lexicalized dep_6_Features prep_Features_for nn_aging_Joint nn_aging_Models pobj_for_aging prep_6_of det_weights_the pobj_of_weights dep_6_suggested prep_suggested_by ' +o,2379,'31 Experiments The model described in section 2 has been tested on the Brown corpus , tagged with the 45 tags of the Penn treebank tagset , which constitute the initial tagset T0 ',Marcus,'num_Experiments_31 det_model_The nsubjpass_tested_model partmod_model_described prep_described_in pobj_in_section num_section_2 aux_tested_has auxpass_tested_been dep_Experiments_tested prep_tested_on det__the nn__Brown nn__corpus pobj_on_ partmod__tagged prep_tagged_with det_tags_the num_tags_45 pobj_with_tags prep_tags_of det__the nn__Penn nn__treebank nn__tagset pobj_of_ nsubj_constitute_which rcmod__constitute det_T0_the amod_T0_initial nn_T0_tagset dobj_constitute_T0 ' +o,2380,'In order to avoid this problem we implemented a simple bootstrapping procedure in which a seed data set of 100 instances of each of the eight categories was hand tagged and used to generate a decision list classifier using the C45 algorithm with the word frequency and topic signature features described below ',Dunning,'mark_avoid_In dep_avoid_order aux_avoid_to det_problem_this dobj_avoid_problem nsubj_implemented_we rcmod_problem_implemented det_procedure_a amod_procedure_simple nn_procedure_bootstrapping dobj_implemented_procedure rel_tagged_in pobj_in_which det_set_a nn_set_seed nn_set_data nsubjpass_tagged_set prep_set_of num_instances_100 pobj_of_instances prep_instances_of pobj_of_each prep_each_of det_categories_the num_categories_eight pobj_of_categories auxpass_tagged_was advmod_tagged_hand rcmod_procedure_tagged cc_tagged_and conj_tagged_used aux_generate_to xcomp_tagged_generate det_classifier_a nn_classifier_decision nn_classifier_list dobj_generate_classifier xcomp_generate_using det__the nn__C45 nn__algorithm dobj_using_ prep_using_with det_frequency_the nn_frequency_word pobj_with_frequency cc_tagged_and nn_features_topic nn_features_signature nsubj_described_features conj_tagged_described advmod_described_below ' +o,2381,'A null Assuming that one SMS word is mapped exactly to one English word in the channel model under an alignment , we need to consider only two types of probabilities : the alignment probabilities denoted by Pm and the lexicon mapping probabilities denoted by ',Brown,'det_null_A nsubj_Assuming_null complm_mapped_that num_word_one nn_word_SMS nsubjpass_mapped_word auxpass_mapped_is ccomp_Assuming_mapped advmod_mapped_exactly prep_mapped_to num_word_one amod_word_English pobj_to_word prep_word_in det_model_the nn_model_channel pobj_in_model prep_mapped_under det_alignment_an pobj_under_alignment nsubj_need_we conj_Assuming_need aux_consider_to xcomp_need_consider advmod_types_only num_types_two dobj_consider_types prep_types_of pobj_of_probabilities det_probabilities_the amod_probabilities_alignment dep_types_probabilities partmod_probabilities_denoted prep_denoted_by pobj_by_Pm cc_Assuming_and det_probabilities_the nn_probabilities_lexicon nn_probabilities_mapping nsubj_denoted_probabilities conj_Assuming_denoted prep_denoted_by ' +o,2382,'Computational linguists have demonstrated that a words meaning is captured to some extent by the distribution of words and phrases with which it commonly co-occurs ',Church,'amod_linguists_Computational nsubj_demonstrated_linguists aux_demonstrated_have complm_captured_that det_meaning_a nn_meaning_words nsubjpass_captured_meaning auxpass_captured_is ccomp_demonstrated_captured prep_captured_to det_extent_some pobj_to_extent prep_captured_by det_distribution_the pobj_by_distribution prep_distribution_of pobj_of_words cc_words_and conj_words_phrases rel_co-occurs_with pobj_with_which nsubj_co-occurs_it advmod_co-occurs_commonly rcmod_distribution_co-occurs ' +p,2383,'2 Related Work To model the syntactic transformation process , researchers in these fieldsespecially in machine translationhave developed powerful grammatical formalisms and statistical models for representing and learning these tree-to-tree relations ',Smith,'num_Work_2 amod_Work_Related nsubj_developed_Work aux_model_To infmod_Work_model det_process_the amod_process_syntactic nn_process_transformation dobj_model_process appos_process_researchers prep_researchers_in det_fieldsespecially_these pobj_in_fieldsespecially prep_fieldsespecially_in nn_translationhave_machine pobj_in_translationhave amod_formalisms_powerful amod_formalisms_grammatical dobj_developed_formalisms cc_formalisms_and amod_models_statistical conj_formalisms_models prep_developed_for pcomp_for_representing cc_representing_and conj_representing_learning det_relations_these amod_relations_tree-to-tree dobj_representing_relations ' +o,2384,'In our experiments using BLEU as the metric , the interpolated synthetic model achieves a relative improvement of 117 \% over the best RBMT system that is used to produce the synthetic bilingual corpora ',Papineni,'prep_achieves_In poss_experiments_our nsubj_using_experiments pcomp_In_using nn__BLEU dobj_using_ prep__as det_metric_the pobj_as_metric det_model_the amod_model_interpolated amod_model_synthetic nsubj_achieves_model det_improvement_a amod_improvement_relative dobj_achieves_improvement prep_improvement_of num_\%_117 pobj_of_\% prep_\%_over det_system_the amod_system_best nn_system_RBMT pobj_over_system nsubjpass_used_that auxpass_used_is rcmod_system_used aux_produce_to xcomp_used_produce det_corpora_the amod_corpora_synthetic amod_corpora_bilingual dobj_produce_corpora ' +o,2385,'We suggest two ways to do it : a version of -LRB- -LRB- obbs et al \'s Generation as Abduction ; and the Interactive Defaults strategy introduced by aoshi et al ',Hobbs,'nsubj_suggest_We num_ways_two dobj_suggest_ways aux_do_to infmod_ways_do dobj_do_it det_version_a dep_ways_version prep_version_of nn_al_obbs nn_al_et nsubj_Generation_al cop_Generation_\'s nn_Generation_ pcomp_of_Generation prep_Generation_as pobj_as_Abduction cc_ways_and det_strategy_the amod_strategy_Interactive nn_strategy_Defaults conj_ways_strategy partmod_strategy_introduced prep_introduced_by nn__aoshi cc_aoshi_et conj_aoshi_al pobj_by_ ' +o,2386,'of Linguistics University of Potsdam kuhn @ linguni-potsdamde Abstract The empirical adequacy of synchronous context-free grammars of rank two -LRB- 2-SCFGs -RRB- , used in syntaxbased machine translation systems such as , Zhang et al ',Wu,'nn_University_Linguistics pobj_of_University prep_University_of nn_kuhn_Potsdam pobj_of_kuhn dep_Abstract_@ amod_Abstract_linguni-potsdamde dep_adequacy_Abstract det_adequacy_The amod_adequacy_empirical dep_University_adequacy prep_adequacy_of amod_grammars_synchronous amod_grammars_context-free pobj_of_grammars prep_grammars_of amod_two_rank pobj_of_two appos_University_2-SCFGs amod_University_ partmod_University_used prep_used_in amod_systems_syntaxbased nn_systems_machine nn_systems_translation pobj_in_systems dep_as_such prep_systems_as pobj_as_ conj__Zhang cc__et conj__al ' +o,2387,'42 Word alignment We have used IBM models proposed by Brown for word aligning the parallel corpus ',Marcus,'num_alignment_42 nn_alignment_Word nsubj_used_We aux_used_have dep_alignment_used nn_models_IBM nsubj_proposed_models dep_alignment_proposed prep_proposed_by nn__Brown pobj_by_ prep__for pobj_for_word xcomp_proposed_aligning det_corpus_the amod_corpus_parallel dobj_aligning_corpus ' +o,2388,'In Experiment 1 , we applied three standard parsing models from the literature to Negra : an unlexicalized PCFG model -LRB- the baseline -RRB- , Carroll and Rooths head-lexicalized model , and model based on head-head dependencies ',Collins,'prep_applied_In pobj_In_Experiment num_Experiment_1 nsubj_applied_we num_models_three amod_models_standard nn_models_parsing dobj_applied_models prep_applied_from det_literature_the pobj_from_literature prep_literature_to pobj_to_Negra det_model_an amod_model_unlexicalized nn_model_PCFG dep_literature_model det_baseline_the appos_model_baseline conj_literature_Carroll cc_Carroll_and nn_model_Rooths nn_model_ amod_model_head-lexicalized conj_Carroll_model cc_literature_and amod_model_ conj_literature_model partmod_model_based prep_based_on amod_dependencies_head-head pobj_on_dependencies ' +o,2389,'The IBM models benefit from a one-tomany constraint , where each target word has ex105 the tax causes unrest l \' impt cause le malaise Figure 1 : A cohesion constraint violation ',Brown,'det_models_The nn_models_IBM nsubj__models dobj__benefit prep__from det_constraint_a amod_constraint_one-tomany pobj_from_constraint advmod_has_where det_word_each nn_word_target nsubj_has_word advcl__has dobj_has_ex105 det_tax_the nsubj_causes_tax rcmod_ex105_causes nn_l_unrest poss_impt_l nsubj_cause_impt ccomp_causes_cause nn_Figure_le nn_Figure_malaise dobj_cause_Figure dep_Figure_1 det_violation_A nn_violation_cohesion nn_violation_constraint dep_Figure_violation ' +o,2390,'The first step is to label each node as either a head , complement , or adjunct based on the approaches of Magerman and ',Collins,'det_step_The amod_step_first nsubj_is_step aux_label_to xcomp_is_label det_node_each dobj_label_node prep_label_as advmod_head_either det_head_a pobj_as_head conj_label_complement cc_label_or conj_label_adjunct prep_adjunct_based dep_based_on det_approaches_the pobj_on_approaches prep_approaches_of nn_and_Magerman nn_and_ pobj_of_and ' +o,2391,' have build a chunker by applying transformation-based learning to sections of the Penn Treebank ',Ramshaw,'nsubj_build_ aux_build_have det_chunker_a dobj_build_chunker prep_build_by pcomp_by_applying amod_learning_transformation-based dobj_applying_learning prep_applying_to pobj_to_sections prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,2392,'Unfortunately , determining the optimal segmentation is challenging , typically requiring extensive experimentation ',Koehn,'advmod_challenging_Unfortunately advmod_segmentation_determining det_segmentation_the amod_segmentation_optimal nsubj_challenging_segmentation cop_challenging_is advmod_requiring_typically xcomp_challenging_requiring amod_experimentation_extensive dobj_requiring_experimentation ' +o,2393,'Unlike a full blown machine translation task , annotators and systems will not be required to translate the whole context but just the target word ',Carpuat,'prep_required_Unlike det__a amod__full amod__blown nn__machine nn__translation nn__task pobj_Unlike_ nsubjpass_required_annotators cc_annotators_and conj_annotators_systems aux_required_will neg_required_not auxpass_required_be aux_translate_to purpcl_required_translate det_context_the amod_context_whole dobj_translate_context dep_just_but cc_context_just det_word_the nn_word_target conj_context_word ' +p,2394,'1 Introduction The availability of large amounts of so-called parallel texts has motivated the application of statistical techniques to the problem of machine translation starting with the seminal work at IBM in the early 90s ',Brown,'num_Introduction_1 det_availability_The nsubj_motivated_availability prep_availability_of amod_amounts_large pobj_of_amounts prep_amounts_of amod_texts_so-called amod_texts_parallel pobj_of_texts aux_motivated_has dep_Introduction_motivated det_application_the dobj_motivated_application prep_application_of amod_techniques_statistical pobj_of_techniques prep_motivated_to det_problem_the pobj_to_problem prep_problem_of nn_translation_machine pobj_of_translation prep_motivated_starting dep_starting_with det_work_the amod_work_seminal pobj_with_work prep_work_at pobj_at_IBM prep_IBM_in det_90s_the amod_90s_early pobj_in_90s ' +p,2395,'This combination of the perceptron algorithm with beam-search is similar to that described by 5 The perceptron algorithm is a convenient choice because it converges quickly usually taking only a few iterations over the training set ',Collins,'det_combination_This nsubj_similar_combination prep_combination_of det_algorithm_the nn_algorithm_perceptron pobj_of_algorithm prep_algorithm_with pobj_with_beam-search cop_similar_is prep_similar_to pobj_to_that partmod_that_described prep_described_by pobj_by_ num__5 det_algorithm_The nn_algorithm_perceptron nsubj_choice_algorithm cop_choice_is det_choice_a amod_choice_convenient dep_similar_choice mark_converges_because nsubj_converges_it advcl_choice_converges advmod_converges_quickly advmod_converges_usually xcomp_converges_taking quantmod_few_only quantmod_few_a num_iterations_few dobj_taking_iterations prep_iterations_over det_set_the nn_set_training pobj_over_set ' +p,2396,'In their seminal work , demonstrated that supervised learning signi cantly outperformed a competing body of work where hand-crafted dictionaries are used to assign sentiment labels based on relative frequencies of positive and negative terms ',Pang,'prep_demonstrated_In poss_work_their amod_work_seminal pobj_In_work nsubj_demonstrated_ complm_outperformed_that amod_signi_supervised nn_signi_learning nsubj_outperformed_signi advmod_outperformed_cantly ccomp_demonstrated_outperformed det_body_a amod_body_competing dobj_outperformed_body prep_body_of pobj_of_work advmod_used_where amod_dictionaries_hand-crafted nsubjpass_used_dictionaries auxpass_used_are rcmod_body_used aux_assign_to xcomp_used_assign nn_labels_sentiment dobj_assign_labels prep_assign_based dep_based_on amod_frequencies_relative pobj_on_frequencies prep_frequencies_of amod_terms_positive cc_positive_and conj_positive_negative pobj_of_terms ' +o,2397,'aoifecahill @ imsuni-stuttgartde and van Genabith , which do not rely on handcrafted grammars and thus can easily be ported to new languages ',Cahill,'dep_aoifecahill_@ amod__imsuni-stuttgartde cc_imsuni-stuttgartde_and conj_imsuni-stuttgartde_van nn__Genabith nsubj_aoifecahill_ nsubj_rely_which aux_rely_do neg_rely_not rcmod__rely dep_rely_on amod_grammars_handcrafted pobj_on_grammars cc_on_and conj_on_thus aux_ported_can advmod_ported_easily auxpass_ported_be dep_on_ported prep_ported_to amod_languages_new pobj_to_languages ' +o,2398,'We adopted log-likelihood ratio , which gave the best pertbrmance among crude non-iterative methods in our test experiments 6 ',Dunning,'nsubj_adopted_We amod__log-likelihood nn__ratio dobj_adopted_ nsubj_gave_which rcmod__gave det_pertbrmance_the amod_pertbrmance_best dobj_gave_pertbrmance prep_pertbrmance_among amod_methods_crude amod_methods_non-iterative pobj_among_methods prep_gave_in poss_experiments_our nn_experiments_test pobj_in_experiments tmod_gave_6 ' +o,2399,'Translation quality is reported using case-insensitive BLEU ',Papineni,'nn_quality_Translation nsubjpass_reported_quality auxpass_reported_is xcomp_reported_using amod_BLEU_case-insensitive dobj_using_BLEU ' +o,2400,'These heuristics define a phrase pair to consist of a source and target ngrams of a word-aligned source-target sentence pair such that if one end of an alignment is in the one ngram , the other end is in the other ngram -LRB- and there is at least one such alignment -RRB- ',Koehn,'det_heuristics_These nsubj_define_heuristics det_pair_a nn_pair_phrase nsubj_consist_pair aux_consist_to xcomp_define_consist prep_consist_of det_ngrams_a nn_ngrams_source cc_source_and conj_source_target pobj_of_ngrams prep_ngrams_of det_pair_a amod_pair_word-aligned amod_pair_source-target nn_pair_sentence pobj_of_pair prep_consist_such complm_is_that mark_is_if num_end_one nsubj_is_end prep_end_of det_alignment_an pobj_of_alignment advcl_is_is prep_is_in det_ngram_the num_ngram_one pobj_in_ngram det_end_the amod_end_other nsubj_is_end ccomp_consist_is prep_is_in det_ngram_the amod_ngram_other pobj_in_ngram dep_is_and expl_is_there dep_ngram_is quantmod_one_at dep_at_least num_alignment_one amod_alignment_such nsubj_is_alignment ' +o,2401,'Zens and Ney compute the viterbi alignments for German-English and French-English sentences pairs using IBM Model 5 , and then measure how many of the resulting alignments fall within the hard constraints of both and Berger et al ',Wu,'nsubj_compute_Zens cc_Zens_and nn__Ney conj_Zens_ det_alignments_the nn_alignments_viterbi dobj_compute_alignments prep_compute_for amod_sentences_German-English cc_German-English_and conj_German-English_French-English pobj_for_sentences nsubj_using_pairs dep_compute_using nn_Model_IBM dobj_using_Model num_Model_5 cc_using_and advmod_using_then nsubj_fall_measure advmod_many_how dep_measure_many prep_measure_of det_alignments_the amod_alignments_resulting pobj_of_alignments dep_using_fall prep_fall_within det_constraints_the amod_constraints_hard pobj_within_constraints prep_constraints_of det__both pobj_of_ cc__and conj__Berger cc_Berger_et conj_Berger_al ' +o,2402,'There are many choices for modeling co-occurrence data ',Marcus,'expl_are_There amod_choices_many nsubj_are_choices prep_choices_for amod__modeling amod__co-occurrence nn__data pobj_for_ ' +o,2403,'ProAlign models P -LRB- A E , F -RRB- directly , using a different decomposition of terms than the model used by IBM ',Brown,'nn_P_models dobj_ProAlign_P det_E_A dep_P_E appos_E_F advmod_ProAlign_directly xcomp_ProAlign_using det_decomposition_a amod_decomposition_different dobj_using_decomposition prep_decomposition_of pobj_of_terms prep_using_than det_model_the pobj_than_model partmod_model_used prep_used_by pobj_by_IBM ' +o,2404,'This test set was tagged using MXPOST which was itself trained on Switchboard ',Ratnaparkhi,'det_set_This nn_set_test nsubjpass_tagged_set auxpass_tagged_was xcomp_tagged_using nn__MXPOST dobj_using_ nsubjpass_trained_which auxpass_trained_was dobj_trained_itself rcmod__trained prep_trained_on pobj_on_Switchboard ' +o,2405,'Experiments are presented in table 1 , using BLEU and METEOR5 , and we also show the length ratio -LRB- ratio of hypothesized tokens to reference tokens -RRB- ',Banerjee,'nsubjpass_presented_Experiments auxpass_presented_are prep_presented_in pobj_in_table num_table_1 xcomp_presented_using amod__BLEU dobj_using_ cc__and nn__METEOR5 conj__ cc_presented_and nsubj_show_we advmod_show_also conj_presented_show det_ratio_the nn_ratio_length dobj_show_ratio dep_ratio_ratio prep_ratio_of amod_tokens_hypothesized pobj_of_tokens prep_tokens_to nn_tokens_reference pobj_to_tokens ' +o,2406,'We use MXPOST tagger for POS tagging , Charniak parser for extracting syntactic relations , SVMlight1 for SVM classifier and David Bleis version of LDA2 for LDA training and inference ',Ratnaparkhi,'nsubj_use_We amod__MXPOST nn__tagger dobj_use_ prep_use_for dep_tagging_POS pobj_for_tagging nn__Charniak nn__parser dobj_use_ prep__for pcomp_for_extracting amod_relations_syntactic dobj_extracting_relations conj_relations_SVMlight1 prep_SVMlight1_for nn_classifier_SVM pobj_for_classifier cc_relations_and nn_version_David nn_version_Bleis conj_relations_version prep_version_of pobj_of_LDA2 prep_LDA2_for nn_training_LDA pobj_for_training cc_training_and conj_training_inference ' +o,2407,'CIT -RRB- ',Koo,'nn_-RRB-_CIT ' +o,2408,'Domain adaptation deals with these feature distribution changes ',Blitzer,'amod_deals_Domain nn_deals_adaptation prep_deals_with det_feature_these pobj_with_feature nn_changes_distribution dep_deals_changes ' +o,2409,' work on classiflcation of reviews is perhaps the closest to ours2 He applied a speciflc unsupervised learning technique based on the mutual information between document phrases and the words excellent \' and poor \' , where the mutual information is computed using statistics gathered by a search engine ',Turney,'amod_work_ nsubj_closest_work prep_work_on pobj_on_classiflcation prep_classiflcation_of pobj_of_reviews cop_closest_is advmod_closest_perhaps det_closest_the prep_closest_to pobj_to_ours2 nsubj_technique_He dep_technique_applied det_technique_a amod_technique_speciflc amod_technique_unsupervised nn_technique_learning dep_closest_technique prep_technique_based dep_based_on det_information_the amod_information_mutual pobj_on_information prep_information_between nn_phrases_document pobj_between_phrases cc_information_and det_excellent_the nn_excellent_words conj_information_excellent cc_excellent_and conj_excellent_poor advmod_computed_where det_information_the amod_information_mutual nsubjpass_computed_information auxpass_computed_is advcl_technique_computed xcomp_computed_using dobj_using_statistics partmod_statistics_gathered prep_gathered_by det_engine_a nn_engine_search pobj_by_engine ' +o,2410,'These distributions are modeled using a maximum entropy formulation , using training data which consists of human judgments of question answer pairs ',Berger,'det_distributions_These nsubjpass_modeled_distributions auxpass_modeled_are xcomp_modeled_using det__a amod__maximum amod__entropy nn__formulation dobj_using_ xcomp_modeled_using nn_data_training dobj_using_data nsubj_consists_which rcmod_data_consists prep_consists_of amod_judgments_human pobj_of_judgments prep_judgments_of nn_pairs_question nn_pairs_answer pobj_of_pairs ' +o,2411,'SMT has evolved from the original word-based approach into phrase-based approaches and syntax-based approaches ',Wu,'nsubj_evolved_SMT aux_evolved_has prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep_evolved_into amod__phrase-based nn__approaches pobj_into_ cc__and amod_approaches_syntax-based conj__approaches ' +o,2412,'Given a wordq , its set of featuresFq and feature weightswq -LRB- f -RRB- for f Fq , a common symmetric similarity measure is Lin similarity : Lin -LRB- u , v -RRB- = summationtext fFuFv -LRB- wu -LRB- f -RRB- + wv -LRB- f -RRB- -RRB- summationtext fFu wu -LRB- f -RRB- + summationtext fFv wv -LRB- f -RRB- where the weight of each feature is the pointwise mutual information -LRB- pmi -RRB- between the word and the feature : wq -LRB- f -RRB- = log -LRB- Pr -LRB- f q -RRB- Pr -LRB- f -RRB- -RRB- ',Lin,'prep__Given det_wordq_a dep_Given_wordq poss_set_its appos_wordq_set prep_set_of nn_weightswq_featuresFq cc_featuresFq_and conj_featuresFq_feature pobj_of_weightswq dep_-LRB-_f prep_weightswq_for nn_Fq_f pobj_for_Fq det_measure_a amod_measure_common amod_measure_symmetric nn_measure_similarity nsubj__measure cop__is nn__Lin nn__similarity dep__Lin appos_Lin_u dep_u_v rcmod_Lin_= amod_fFu_summationtext amod_fFu_fFuFv dep_fFu_wu dep_wv_f amod_wv_+ dobj_wu_wv dep_-LRB-_f nn_fFu_summationtext dobj_=_fFu nsubj_wq_wu dep_-LRB-_f prep_wu_+ amod_wv_summationtext nn_wv_fFv pobj_+_wv dep_-LRB-_f advmod_information_where det_weight_the nsubj_information_weight prep_weight_of det_feature_each pobj_of_feature cop_information_is det_information_the amod_information_pointwise amod_information_mutual rcmod_wv_information appos_information_pmi prep_information_between det_word_the pobj_between_word cc_word_and det_feature_the conj_word_feature dep__wq dep_-LRB-_f dep_log_= ccomp_wq_log parataxis_wq_Pr dep_-LRB-_f dep_f_q dobj_Pr_Pr dep_-LRB-_f ' +o,2413,'This is seen in that each time we check for the nearest intersection to the current 1-best for some n-best list l, we Algorithm 1 Och (2003)s line search method to find the global minimum in the loss, lscript, when starting at the point w and searching along the direction d using the candidate translations given in the collection of n-best lists L. Input: L, w, d, lscript I {} for l L do for e l do m{e} e.features d b{e} e.features w end for bestn argmaxel m{e}{b{e} breaks ties} loop bestn+1 = argminel max parenleftBig 0, b{bestn}b{e}m{e}m{bestn} parenrightBig intercept max parenleftBig 0, b{bestn}b{bestn+1}m{bestn+1}m{bestn} parenrightBig if intercept > 0 then add(I, intercept) else break end if end loop end for add(I, max(I)+2epsilon1) ibest = argminiI evallscript(L,w+(iepsilon1)d) return w+(ibest epsilon1)d must calculate its intersection with all other candidate translations that have yet to be selected as the 1-best.',Och,'' +o,2414,'We generated for each phrase pair in the translation table 5 features : phrase translation probability -LRB- both directions -RRB- , lexical weighting -LRB- both directions -RRB- and phrase penalty -LRB- constant value -RRB- ',Koehn,'nsubj_generated_We prep_generated_for det_pair_each nn_pair_phrase pobj_for_pair prep_generated_in det_table_the nn_table_translation pobj_in_table num_features_5 dobj_generated_features nn_probability_phrase nn_probability_translation dep_features_probability det_directions_both dep_probability_directions amod__lexical nn__weighting conj_probability_ det_directions_both dep__directions cc_probability_and nn_penalty_phrase conj_probability_penalty amod_value_constant appos_penalty_value ' +o,2415,'Formally , transformational rules ri presented in are equivalent to 1-state xRs transducers mapping a given pattern -LRB- subtree to match in pi -RRB- to a right hand side string ',Galley,'advmod_equivalent_Formally amod_ri_transformational nn_ri_rules nsubj_equivalent_ri partmod_ri_presented prep_presented_in pobj_in_ cop_equivalent_are prep_equivalent_to amod_transducers_1-state nn_transducers_xRs pobj_to_transducers partmod_transducers_mapping det_pattern_a amod_pattern_given dobj_mapping_pattern dep_pattern_subtree aux_match_to dep_subtree_match prep_match_in pobj_in_pi prep_mapping_to det_string_a amod_string_right nn_string_hand nn_string_side pobj_to_string ' +o,2416,'Mincuts have been used 4As of this writing , WordNet is available for more than 40 world languages -LRB- http://wwwglobalwordnetorg -RRB- Figure 2 : Semi-supervised classification using mincuts in semi-supervised learning for various tasks , including document level sentiment analysis ',Pang,'nsubjpass_used_Mincuts aux_used_have auxpass_used_been dobj_used_4As prep_4As_of det_writing_this pobj_of_writing appos_4As_WordNet cop_available_is rcmod_WordNet_available prep_available_for dep_than_more quantmod_40_than num_languages_40 nn_languages_world pobj_for_languages appos_languages_http://wwwglobalwordnetorg dep_4As_Figure num_Figure_2 amod_classification_Semi-supervised dep_Figure_classification partmod_classification_using dobj_using_mincuts prep_using_in amod_learning_semi-supervised pobj_in_learning prep_learning_for amod_tasks_various pobj_for_tasks prep_4As_including nn_analysis_document nn_analysis_level nn_analysis_sentiment pobj_including_analysis ' +n,2417,'Moreover , the parameters of the model must be estimated using averaged perceptron training , which can be unstable ',Collins,'advmod_estimated_Moreover det_parameters_the nsubjpass_estimated_parameters prep_parameters_of det_model_the pobj_of_model aux_estimated_must auxpass_estimated_be purpcl_estimated_using amod__averaged nn__perceptron nn__training dobj_using_ nsubj_unstable_which aux_unstable_can cop_unstable_be rcmod__unstable ' +o,2418,'The original intention of assignment 2 was that students then use this maxent classifier as a building block of a maxent part-of-speech tagger like that of ',Ratnaparkhi,'det_intention_The amod_intention_original nsubj_was_intention prep_intention_of pobj_of_assignment num_assignment_2 complm_use_that nsubj_use_students advmod_use_then ccomp_was_use det_classifier_this amod_classifier_maxent dobj_use_classifier prep_use_as det_block_a nn_block_building pobj_as_block prep_block_of det_tagger_a amod_tagger_maxent amod_tagger_part-of-speech pobj_of_tagger prep_tagger_like pobj_like_that prep_that_of ' +o,2419,'Finally, we show in Section 7.3 that our SCL PoS 124 (a) 100 500 1k 5k 40k75 80 85 90 Results for 561 MEDLINE Test Sentences Number of WSJ Training Sentences Accuracy supervised semiASO SCL (b) Accuracy on 561-sentence test set Words Model All Unknown Ratnaparkhi (1996) 87.2 65.2 supervised 87.9 68.4 semi-ASO 88.4 70.9 SCL 88.9 72.0 (c) Statistical Significance (McNemars) for all words Null Hypothesis p-value semi-ASO vs. super 0.0015 SCL vs. super 2.1 1012 SCL vs. semi-ASO 0.0003 Figure 5: PoS tagging results with no target labeled training data (a) 50 100 200 500 86 88 90 92 94 96 Number of MEDLINE Training Sentences Accuracy Results for 561 MEDLINE Test Sentences 40kSCL 40ksuper 1kSCL 1ksuper nosource (b) 500 target domain training sentences Model Testing Accuracy nosource 94.5 1k-super 94.5 1k-SCL 95.0 40k-super 95.6 40k-SCL 96.1 (c) McNemars Test (500 training sentences) Null Hypothesis p-value 1k-super vs. nosource 0.732 1k-SCL vs. 1k-super 0.0003 40k-super vs. nosource 1.9 1012 40k-SCL vs. 40k-super 6.5 107 Figure 6: PoS tagging results with no target labeled training data tagger improves the performance of a dependency parser on the target domain.',Ratnaparkhi,'' +o,2420,' , et al ',Rosti,'appos__ cc__et conj__al ' +o,2421,'Many researchers , have observed consistent gains by using more flexible matching criteria ',Banerjee,'amod_researchers_Many nsubj_observed_researchers partmod_researchers_ aux_observed_have amod_gains_consistent dobj_observed_gains prep_observed_by pcomp_by_using advmod_flexible_more amod_criteria_flexible amod_criteria_matching dobj_using_criteria ' +o,2422,'Our approach is based on earlier work on LFG semantic form extraction and recent progress in automatically annotating the Penn-II treebank with LFG f-structures ',Cahill,'poss_approach_Our nsubjpass_based_approach auxpass_based_is prep_based_on amod_work_earlier pobj_on_work prep_work_on nn__LFG amod__semantic nn__form nn__extraction pobj_on_ cc__and amod_progress_recent conj__progress prep_based_in advmod_annotating_automatically pcomp_in_annotating det_treebank_the amod_treebank_Penn-II dobj_annotating_treebank prep_annotating_with nn_f-structures_LFG pobj_with_f-structures ' +o,2423,'Following , the input to the NP chunker consists of the words in a sentence annotated automatically with part-of-speech -LRB- POS -RRB- tags ',Ramshaw,'nn__Following nsubj_consists_ det_input_the nsubj_consists_input prep_input_to det_chunker_the nn_chunker_NP pobj_to_chunker prep_consists_of det_words_the pobj_of_words prep_words_in det_sentence_a pobj_in_sentence partmod_sentence_annotated advmod_annotated_automatically prep_annotated_with amod_tags_part-of-speech appos_tags_POS pobj_with_tags ' +o,2424,'3 The Framework 31 The Algorithm Our transductive learning algorithm , Algorithm 1 , is inspired by the Yarowsky algorithm ',Abney,'det_algorithm_The nn_algorithm_Framework num_algorithm_31 nn_algorithm_The nn_algorithm_Algorithm nn_algorithm_Our nn_algorithm_transductive nn_algorithm_learning nsubjpass_inspired_algorithm appos_algorithm_Algorithm num_Algorithm_1 auxpass_inspired_is dep_3_inspired prep_inspired_by det_algorithm_the nn_algorithm_Yarowsky pobj_by_algorithm ' +o,2425,'The features we used are as follows : word posterior probability ; 3 , 4-gram target language model ; word length penalty ; Null word length penalty ; Also , we use MERT to tune the weights of confusion network ',Och,'det_features_The nsubj_are_features nsubj_used_we rcmod_features_used prep_are_as pobj_as_follows nn__word nn__posterior nn__probability dep_follows_ dep_follows_3 amod_model_4-gram nn_model_target nn_model_language appos_3_model nn_penalty_word nn_penalty_length dep_follows_penalty nn_penalty_Null nn_penalty_word nn_penalty_length dep_follows_penalty advmod_use_Also nsubj_use_we parataxis_are_use nn__MERT nsubj_tune_ aux_tune_to xcomp_use_tune det_weights_the dobj_tune_weights prep_weights_of nn_network_confusion pobj_of_network ' +o,2426,'The score combination weights are trained by a minimum error rate training procedure similar to ',Och,'det_weights_The nn_weights_score nn_weights_combination nsubjpass_trained_weights auxpass_trained_are prep_trained_by det_procedure_a amod_procedure_minimum nn_procedure_error nn_procedure_rate nn_procedure_training pobj_by_procedure amod_procedure_similar prep_similar_to ' +o,2427,'We use the likelihood ratio for a binomial distribution , which tests the hypothesis whether the term occurs independently in texts of biographical nature given a large corpus of biographical and non-biographical texts ',Dunning,'nsubj_use_We det_ratio_the nn_ratio_likelihood dobj_use_ratio prep_ratio_for det__a amod__binomial nn__distribution pobj_for_ nsubj_tests_which rcmod__tests det_hypothesis_the dobj_tests_hypothesis complm_occurs_whether det_term_the nsubj_occurs_term ccomp_tests_occurs advmod_occurs_independently prep_occurs_in pobj_in_texts prep_texts_of amod_nature_biographical pobj_of_nature partmod_nature_given det_corpus_a amod_corpus_large dobj_given_corpus prep_corpus_of amod_texts_biographical cc_biographical_and conj_biographical_non-biographical pobj_of_texts ' +o,2428,'41 The test environment For our experiments , we used a manually corrected version of the Air Travel Information System -LRB- ATIS -RRB- spoken language corpus annotated in the Pennsylvania Treebank ',Marcus,'tmod_used_41 det_environment_The nn_environment_test dep_41_environment prep_used_For poss_experiments_our pobj_For_experiments nsubj_used_we det_version_a amod_version_manually amod_version_corrected dobj_used_version prep_version_of det_System_the nn_System_Air nn_System_Travel nn_System_Information pobj_of_System abbrev_System_ATIS partmod_System_spoken nn__language nn__corpus dobj_spoken_ partmod__annotated prep_annotated_in det_Treebank_the nn_Treebank_Pennsylvania pobj_in_Treebank ' +o,2429,'Following , 765 Feature Sets Templates Error \% A Ratnaparkhis 305 B A + -LRB- t0 , t1 -RRB- , -LRB- t0 , t1 , t1 -RRB- , -LRB- t0 , t1 , t2 -RRB- 292 C B + -LRB- t0 , t2 -RRB- , -LRB- t0 , t2 -RRB- , -LRB- t0 , t2 , w0 -RRB- , -LRB- t0 , t1 , w0 -RRB- , -LRB- t0 , t1 , w0 -RRB- , -LRB- t0 , t2 , w0 -RRB- , -LRB- t0 , t2 , t1 , w0 -RRB- , -LRB- t0 , t1 , t1 , w0 -RRB- , -LRB- t0 , t1 , t2 , w0 -RRB- 284 D C + -LRB- t0 , w1 , w0 -RRB- , -LRB- t0 , w1 , w0 -RRB- 278 E D + -LRB- t0 , X = prefix or suffix of w0 -RRB- ,4 -LRB- X 9 272 Table 2 : Experiments on the development data with beam width of 3 we cut the PTB into the training , development and test sets as shown in Table 1 ',Berger,'prep_Templates_Following pobj_Following_ nn_Sets_765 nn_Sets_Feature appos__Sets num_\%_Error dobj_Templates_\% det_B_A nn_B_Ratnaparkhis num_B_305 nsubj_Templates_B det_+_A dep_B_+ appos_+_t0 dep_t0_t1 nn_t1_t0 dep_t1_t1 appos_+_t1 nn_t2_t0 dep_t2_t1 appos_+_t2 nn_D_292 nn_D_C nn_D_B nn_D_+ appos_D_t0 dep_t0_t2 appos_D_t0 dep_t0_t2 nn_w0_t0 dep_w0_t2 appos_D_w0 nn_w0_t0 dep_w0_t1 appos_D_w0 nn_w0_t0 dep_w0_t1 appos_D_w0 nn_w0_t0 dep_w0_t2 appos_D_w0 appos_D_t0 nn_w0_t2 dep_w0_t1 dep_t0_w0 appos_D_t0 nn_w0_t1 dep_w0_t1 dep_t0_w0 appos_D_t0 nn_w0_t1 dep_w0_t2 dep_t0_w0 nn_D_284 measure_E_D nn_+_C dep_D_+ nn_w0_t0 dep_w0_w1 appos_D_w0 nn_w0_t0 dep_w0_w1 appos_D_w0 num_D_278 amod_D_E nsubj_+_D rcmod_+_+ dep_=_t0 nsubj_=_X dep_Experiments_= dobj_=_prefix cc_prefix_or conj_prefix_suffix prep_prefix_of pobj_of_w0 nsubj_Experiments_,4 partmod_,4_X number_272_9 num_Table_272 dobj_X_Table num_Table_2 dep_cut_Experiments prep_Experiments_on det_data_the nn_data_development pobj_on_data prep_Experiments_with nn_width_beam pobj_with_width prep_width_of pobj_of_3 nsubj_cut_we ccomp_+_cut det_PTB_the dobj_cut_PTB prep_cut_into det_sets_the nn_sets_training conj_training_development cc_training_and conj_training_test pobj_into_sets mark_shown_as advcl_cut_shown prep_shown_in pobj_in_Table num_Table_1 ' +o,2430,'We estimated the probabilities P -LRB- c I Pi -RRB- and P -LRB- c -RRB- similarly to Resnik by using relative frequencies from the BNC , together with WordNet as a source of taxonomic semantic class information ',Dunning,'nsubj_estimated_We det_P_the amod_P_probabilities dobj_estimated_P dep_Pi_c nsubj_Pi_I dep_P_Pi cc_P_and amod_information_P dep_-LRB-_c amod_information_similarly prep_similarly_to nn__Resnik pobj_to_ prep__by pcomp_by_using amod_frequencies_relative dobj_using_frequencies prep_using_from det_BNC_the pobj_from_BNC advmod_BNC_together prep_using_with nn__WordNet pobj_with_ prep__as det_source_a pobj_as_source prep_source_of amod_class_taxonomic amod_class_semantic pobj_of_class conj_P_information ' +o,2431,'Thus , GCNF is a more restrictive normal form than those used by and Melamed -LRB- 2003 -RRB- ',Wu,'advmod_form_Thus nsubj_form_GCNF cop_form_is det_form_a advmod_restrictive_more amod_form_restrictive amod_form_normal prep_form_than pobj_than_those partmod_those_used prep_used_by pobj_by_ cc__and conj__Melamed dep_used_2003 ' +o,2432,'The CDR is assigned with access to clinical and cognitive test information , independent of performance on the battery of neuropsychological tests used for this research study , and has been shown to have high expert inter-annotator reliability ',Marcus,'det__The nn__CDR nsubjpass_assigned_ auxpass_assigned_is prep_assigned_with pobj_with_access aux_clinical_to xcomp_assigned_clinical cc_clinical_and conj_clinical_cognitive nn_information_test dobj_clinical_information conj_clinical_independent prep_independent_of pobj_of_performance prep_assigned_on det_battery_the pobj_on_battery prep_battery_of amod_tests_neuropsychological pobj_of_tests partmod_tests_used prep_used_for det_study_this nn_study_research pobj_for_study cc_assigned_and aux_shown_has auxpass_shown_been conj_assigned_shown aux_have_to xcomp_shown_have amod_inter-annotator_high nn_inter-annotator_expert dobj_have_inter-annotator advmod__reliability amod_inter-annotator_ ' +o,2433,'We evaluate its performance on the standard Penn English Treebank -LRB- PTB -RRB- dependency parsing task , ie , train on sections 02-21 and test on section 23 with automatically assigned POS tags -LRB- at 972 \% accuracy -RRB- using a tagger similar to , and using the headrules of Yamada and Matsumoto -LRB- 2003 -RRB- for conversion into dependency trees ',Collins,'nsubj_evaluate_We poss_performance_its dobj_evaluate_performance prep_evaluate_on det_dependency_the amod_dependency_standard nn_dependency_Penn nn_dependency_English nn_dependency_Treebank abbrev_dependency_PTB pobj_on_dependency partmod_dependency_parsing dobj_parsing_task conj_evaluate_ie conj_evaluate_train prep_train_on pobj_on_sections num_sections_02-21 cc_evaluate_and conj_evaluate_test prep_test_on pobj_on_section num_section_23 prep_test_with advmod_tags_automatically nn_tags_assigned dep_tags_POS pobj_with_tags advmod_accuracy_at num_\%_972 dep_at_\% appos_tags_accuracy ccomp_test_using det_tagger_a dobj_using_tagger amod_tagger_similar dep_similar_to dep_similar_ cc_using_and conj_using_using det_headrules_the dobj_using_headrules prep_headrules_of pobj_of_Yamada cc_Yamada_and conj_Yamada_Matsumoto dep_test_2003 prep_test_for pobj_for_conversion prep_conversion_into amod_trees_dependency pobj_into_trees ' +o,2434,'The outcomes of CW resemble those of MinCut : Dense regions in the graph are grouped into one cluster while sparsely connected regions are separated ',Dunning,'det_outcomes_The nsubj_those_outcomes prep_outcomes_of pobj_of_CW cop_those_resemble prep_those_of nn__MinCut pobj_of_ amod_regions_Dense nsubjpass_grouped_regions prep_regions_in det_graph_the pobj_in_graph auxpass_grouped_are parataxis_those_grouped prep_grouped_into num_cluster_one pobj_into_cluster mark_separated_while advmod_connected_sparsely amod_regions_connected nsubj_separated_regions cop_separated_are advcl_grouped_separated ' +o,2435,'Bilingual alignments have so far shown that they can play multiple roles in a wide range of linguistic applications , such as computer assisted translation , terminology lexicography , and cross-language information retrieval -LRB- Nie et al , \\* This research was funded by the Canadian Department of Foreign Affairs and International Trade -LRB- http://~dfait-maecigcca/ -RRB- , via the Agence de la francophonie -LRB- http : \\/ \\/ ~ ',Brown,'amod_alignments_Bilingual nsubj_shown_alignments aux_shown_have advmod_far_so advmod_shown_far complm_play_that nsubj_play_they aux_play_can ccomp_shown_play amod_roles_multiple dobj_play_roles prep_play_in det_range_a amod_range_wide pobj_in_range prep_range_of amod_applications_linguistic pobj_of_applications dep_as_such prep_range_as pobj_as_computer partmod_computer_assisted nn__translation dobj_assisted_ advmod__terminology amod__ nn__lexicography conj__ cc__and amod_retrieval_cross-language nn_retrieval_information conj__retrieval nn_al_Nie nn_al_et nsubjpass_funded_al dep_funded_\\* det_research_This nsubjpass_funded_research auxpass_funded_was dep_shown_funded prep_funded_by det_~_the nn_Department_Canadian nn_~_Department prep_Department_of nn_Affairs_Foreign pobj_of_Affairs cc_Affairs_and nn_Trade_International conj_Affairs_Trade appos_Department_http://~dfait-maecigcca/ prep_Department_via det_francophonie_the nn_francophonie_Agence nn_francophonie_de nn_francophonie_la pobj_via_francophonie dep_francophonie_http amod_~_\\/ nn_~_\\/ pobj_by_~ ' +n,2436,'This method has the advantage that it is not limited to the model scaling factors as the method described in ',Och,'det_method_This nsubj_has_method det_advantage_the dobj_has_advantage complm_limited_that nsubj_limited_it cop_limited_is neg_limited_not dep_advantage_limited prep_limited_to det_factors_the amod_factors_model nn_factors_scaling pobj_to_factors mark_described_as det_method_the nsubj_described_method advcl_limited_described prt_described_in ' +p,2437,'7Another related measure is \'s likelihood ratio tests for binomial and multinomial distributions , which are claimed to be effective even with very much smaller volumes of text than is necessary for other tests based on assumed normal distributions ',Dunning,'amod_measure_7Another amod_measure_related nsubj_tests_measure cop_tests_is poss_tests_ possessive__\'s nn_tests_likelihood nn_tests_ratio prep_tests_for amod_distributions_binomial cc_binomial_and conj_binomial_multinomial pobj_for_distributions nsubjpass_claimed_which auxpass_claimed_are rcmod_distributions_claimed aux_effective_to cop_effective_be xcomp_claimed_effective dep_with_even prep_effective_with advmod_smaller_very advmod_smaller_much amod_volumes_smaller pobj_with_volumes prep_volumes_of pobj_of_text mark_necessary_than cop_necessary_is advcl_effective_necessary prep_necessary_for amod_tests_other pobj_for_tests partmod_tests_based prep_based_on amod_distributions_assumed amod_distributions_normal pobj_on_distributions ' +o,2438,'To identify these , we use a word-aligned corpus annotated with parse trees generated by statistical syntactic parsers ',Collins,'aux_identify_To dep_use_identify dobj_identify_these nsubj_use_we det_corpus_a amod_corpus_word-aligned dobj_use_corpus partmod_corpus_annotated prep_annotated_with amod_trees_parse pobj_with_trees partmod_trees_generated prep_generated_by amod_parsers_statistical amod_parsers_syntactic pobj_by_parsers ' +o,2439,'Therefore , we also carried out evaluations using the NIST , METEOR , WER , PER and TER machine translation evaluation techniques ',Banerjee,'advmod_carried_Therefore nsubj_carried_we advmod_carried_also prt_carried_out dobj_carried_evaluations xcomp_carried_using det__the amod__NIST dobj_using_ nn__METEOR conj__ nn__WER conj__ nn__PER conj__ cc__and nn_techniques_TER nn_techniques_ nn_techniques_machine nn_techniques_translation nn_techniques_evaluation conj__techniques ' +o,2440,'Similar techniques are used in for socalled direct translation models instead of those proposed in ',Berger,'amod_techniques_Similar nsubjpass_used_techniques auxpass_used_are prep_used_in pobj_in_ prep_used_for amod_models_socalled amod_models_direct nn_models_translation pobj_for_models dep_of_instead prep_models_of pobj_of_those partmod_those_proposed prep_proposed_in pobj_in_ ' +o,2441,'In this case it is possible to perform the correct selection if we used only statistics about the cooccurrences of ` corruption \' with either ` investigator \' or ` researcher \' , without looking for any syntactic relation -LRB- as in -RRB- ',Church,'prep_possible_In det_case_this pobj_In_case nsubj_possible_it cop_possible_is aux_perform_to xcomp_possible_perform det_selection_the amod_selection_correct dobj_perform_selection mark_used_if nsubj_used_we advcl_perform_used amod_statistics_only dobj_used_statistics prep_used_about det_cooccurrences_the pobj_about_cooccurrences prep_cooccurrences_of pobj_of_corruption prep_used_with preconj_investigator_either pobj_with_investigator cc_investigator_or conj_investigator_researcher prep_used_without pcomp_without_looking prep_looking_for det_relation_any amod_relation_syntactic pobj_for_relation dep_in_as dep_relation_in pobj_in_ ' +o,2442,'In and -LRB- Pereira et al , ! 993 -RRB- , clusters of similar words are evaluated by how well they are able to recover data items that are removed from the input corpus one at a time ',Dagan,'prep_evaluated_In pobj_In_ cc__and appos__Pereira cc_Pereira_et conj_Pereira_al num_!_993 conj__clusters prep_clusters_of amod_words_similar pobj_of_words auxpass_evaluated_are prep_evaluated_by advmod_able_how advmod_how_well nsubj_able_they cop_able_are pcomp_by_able aux_recover_to xcomp_able_recover nn_items_data dobj_recover_items nsubjpass_removed_that auxpass_removed_are rcmod_items_removed prep_removed_from det_corpus_the nn_corpus_input pobj_from_corpus nsubjpass_evaluated_one prep_one_at det_time_a pobj_at_time ' +p,2443,'The ROUGE suite of metrics are n-gram overlap based metrics that have been shown to highly correlate with human evaluations on content responsiveness ',Lin,'det_suite_The nn_suite_ROUGE nn_suite_ nsubj_n-gram_suite prep_suite_of pobj_of_metrics cop_n-gram_are ccomp_n-gram_overlap prep_overlap_based dep_based_metrics nsubjpass_shown_that aux_shown_have auxpass_shown_been rcmod_metrics_shown aux_correlate_to advmod_correlate_highly xcomp_shown_correlate prep_correlate_with amod_evaluations_human pobj_with_evaluations prep_shown_on nn_responsiveness_content pobj_on_responsiveness ' +o,2444,'7 Related Work Much work on sentiment analysis classifies documents by their overall sentiment , for example determining whether a review is positive or negative -LRB- eg , -RRB- ',Pang,'num_work_7 nn_work_Related nn_work_Work amod_work_Much nsubj_classifies_work prep_work_on nn_analysis_sentiment pobj_on_analysis dobj_classifies_documents prep_classifies_by poss_sentiment_their amod_sentiment_overall pobj_by_sentiment prep_classifies_for pobj_for_example partmod_example_determining complm_positive_whether det_review_a nsubj_positive_review cop_positive_is ccomp_determining_positive cc_positive_or conj_positive_negative dep_positive_eg dep_eg_ ' +o,2445,'5http : \\/ \\/ nlpcsberkeleyedu\\/Mainhtml # Parsing 47 Figure 3 : Predicate argument structure timized automatically by assigning latent variables to each nonterminal node and estimating the parameters of the latent variables by the EM algorithm ',Matsuzaki,'nn_\\/_\\/ nsubj_nlpcsberkeleyedu\\/Mainhtml_\\/ dep_5http_nlpcsberkeleyedu\\/Mainhtml dep_47_# number_47_Parsing num_Figure_47 dobj_nlpcsberkeleyedu\\/Mainhtml_Figure num_Figure_3 nn_structure_Predicate nn_structure_argument nsubj_timized_structure parataxis_nlpcsberkeleyedu\\/Mainhtml_timized advmod_timized_automatically prep_timized_by pcomp_by_assigning amod_variables_latent dobj_assigning_variables prep_assigning_to det_node_each amod_node_nonterminal pobj_to_node cc_assigning_and conj_assigning_estimating det_parameters_the dobj_estimating_parameters prep_parameters_of det_variables_the amod_variables_latent pobj_of_variables prep_estimating_by det_algorithm_the nn_algorithm_EM pobj_by_algorithm ' +o,2446,' , it is much higher than the 26 \% unknown word rate in the test set for English POS tagging experiments ',Ratnaparkhi,'advmod_higher_ nsubj_higher_it cop_higher_is advmod_higher_much prep_higher_than det_rate_the dep_\%_26 amod_rate_\% amod_rate_unknown nn_rate_word pobj_than_rate prep_rate_in det_test_the pobj_in_test partmod_test_set prep_set_for nn_POS_ nn_POS_English pobj_for_POS partmod_POS_tagging dobj_tagging_experiments ' +o,2447,'Our appoach is based on Maximum Entropy -LRB- MaxEnt henceforth -RRB- technique ',Berger,'poss_appoach_Our nsubjpass_based_appoach auxpass_based_is prep_based_on nn_technique_Maximum nn_technique_Entropy nn_henceforth_MaxEnt appos_technique_henceforth pobj_on_technique ' +o,2448,'Our approach is based on earlier work on LFG semantic form extraction and recent progress in automatically annotating the Penn-II and Penn-III Treebanks with LFG f-structures ',Cahill,'poss_approach_Our nsubjpass_based_approach auxpass_based_is prep_based_on amod_work_earlier pobj_on_work prep_work_on nn__LFG amod__semantic nn__form nn__extraction pobj_on_ cc__and amod_progress_recent conj__progress prep_based_in advmod_annotating_automatically pcomp_in_annotating det_Treebanks_the nn_Treebanks_Penn-II cc_Penn-II_and conj_Penn-II_Penn-III dobj_annotating_Treebanks prep_annotating_with nn_f-structures_LFG pobj_with_f-structures ' +o,2449,'The parse trees on the English side of the bitexts were generated using a parser implementing the Collins parsing models ',Collins,'det_trees_The amod_trees_parse nsubjpass_generated_trees prep_trees_on det_side_the amod_side_English pobj_on_side prep_side_of det_bitexts_the pobj_of_bitexts auxpass_generated_were xcomp_generated_using det__a nn__parser dobj_using_ xcomp_using_implementing det_models_the nn_models_Collins nn_models_parsing dobj_implementing_models ' +p,2450,' improve their F-score by 3 \% by including a Wikipedia-based feature in their machine learner ',Kazama,'nsubj_improve_ poss_F-score_their dobj_improve_F-score prep_improve_by num_\%_3 pobj_by_\% prep_improve_by pcomp_by_including det_feature_a amod_feature_Wikipedia-based dobj_including_feature prep_including_in poss_learner_their nn_learner_machine pobj_in_learner ' +o,2451,'In this paper , we make a direct comparison of a syntactically unsupervised alignment model , based on , with a syntactically supervised model , based on Yamada and Knight -LRB- 2001 -RRB- ',Wu,'prep_make_In det_paper_this pobj_In_paper nsubj_make_we det_comparison_a amod_comparison_direct dobj_make_comparison prep_comparison_of det_model_a advmod_unsupervised_syntactically amod_model_unsupervised nn_model_alignment pobj_of_model partmod_model_based prep_based_on pobj_on_ prep_based_with det_model_a advmod_supervised_syntactically amod_model_supervised pobj_with_model partmod_model_based prep_based_on pobj_on_Yamada cc_Yamada_and conj_Yamada_Knight dep_based_2001 ' +o,2452,'31 Data and Experimental Setup The data set by consists of 2000 movie reviews -LRB- 1000-pos , 1000-neg -RRB- from the IMDb review archive ',Pang,'num_Data_31 nsubj_set_Data cc_Data_and nn_Setup_Experimental conj_Data_Setup det_data_The dep_Setup_data prep_set_by nn_consists_ pobj_by_consists prep_consists_of num_reviews_2000 nn_reviews_movie pobj_of_reviews dep_reviews_1000-pos dep_1000-pos_1000-neg prep_set_from det_archive_the nn_archive_IMDb nn_archive_review pobj_from_archive ' +p,2453,'Support Vector Machines -LRB- SVMs -RRB- and Maximum Entropy -LRB- ME -RRB- method are powerful learning methods that satisfy such requirements , and are applied successfully to other NLP tasks ',Ratnaparkhi,'nn__Support nn__Vector nn__Machines abbrev__SVMs nsubj_powerful_ cc__and nn_Entropy_Maximum nn__Entropy abbrev__ME nn__method conj__ cop_powerful_are dep_powerful_learning dobj_learning_methods nsubj_satisfy_that rcmod_methods_satisfy amod_requirements_such dobj_satisfy_requirements cc_learning_and auxpass_applied_are conj_learning_applied advmod_applied_successfully prep_applied_to amod_tasks_other nn_tasks_NLP pobj_to_tasks ' +o,2454,'We measure translation performance by the BLEU score and Translation Error Rate -LRB- TER -RRB- with one reference for each hypothesis ',Papineni,'nsubj_measure_We nn_performance_translation dobj_measure_performance prep_measure_by det__the amod__BLEU nn__score pobj_by_ cc_measure_and conj_measure_Translation nn__Error nn__Rate abbrev__TER dobj_Translation_ prep__with num_reference_one pobj_with_reference prep_reference_for det_hypothesis_each pobj_for_hypothesis ' +o,2455,' predict the results of an election by analyzing forums discussing the elections ',Kim,'nsubj_predict_ det_results_the dobj_predict_results prep_results_of det_election_an pobj_of_election prep_predict_by pcomp_by_analyzing dobj_analyzing_forums xcomp_analyzing_discussing det_elections_the dobj_discussing_elections ' +p,2456,'Among these techniques , SCL -LRB- Structural Correspondence Learning -RRB- is regarded as a promising method to tackle transfer-learning problem ',Blitzer,'prep_regarded_Among det_techniques_these pobj_Among_techniques appos_techniques_SCL nn_Learning_Structural nn_Learning_Correspondence appos_SCL_Learning nsubjpass_regarded_ auxpass_regarded_is prep_regarded_as det_method_a amod_method_promising pobj_as_method aux_tackle_to xcomp_regarded_tackle amod_problem_transfer-learning dobj_tackle_problem ' +o,2457,'5 Analysis Over the last few years , several automatic metrics for machine translation evaluation have been introduced , largely to reduce the human cost of iterative system evaluation during the development cycle ',Och,'num_Analysis_5 prep_introduced_Over det_years_the amod_years_last amod_years_few pobj_Over_years amod_metrics_several amod_metrics_automatic nsubjpass_introduced_metrics prep_metrics_for nn_evaluation_machine nn_evaluation_translation pobj_for_evaluation aux_introduced_have auxpass_introduced_been rcmod_Analysis_introduced advmod_introduced_largely aux_reduce_to infmod_Analysis_reduce det_cost_the amod_cost_human dobj_reduce_cost prep_cost_of amod_evaluation_iterative nn_evaluation_system pobj_of_evaluation prep_reduce_during det_cycle_the nn_cycle_development pobj_during_cycle ' +o,2458,'Conditional probability , the log-likelihood ratio , and Resnik \'s selectional association measure were also significantly correlated with plausibility ratings ',Dunning,'amod_probability_Conditional nsubjpass_correlated_probability det_ratio_the amod_ratio_log-likelihood appos_probability_ratio cc_probability_and poss_measure_Resnik possessive_Resnik_\'s amod_measure_ amod_measure_selectional nn_measure_association conj_probability_measure auxpass_correlated_were advmod_correlated_also advmod_correlated_significantly prep_correlated_with amod_ratings_plausibility pobj_with_ratings ' +o,2459,'1153 While much research has explored how to reconcile pairwise decisions to form coherent clusters , we simply take the transitive closure of our pairwise decision -LRB- as in Ng and Cardie and Bengston and Roth -RRB- which can and does cause system errors ',Haghighi,'dep_explored_1153 dep_explored_While amod__much nn__research nsubj_explored_ aux_explored_has advcl_take_explored advmod_reconcile_how aux_reconcile_to ccomp_explored_reconcile amod_decisions_pairwise dobj_reconcile_decisions aux_form_to xcomp_reconcile_form amod_clusters_coherent dobj_form_clusters nsubj_take_we advmod_take_simply det_closure_the amod_closure_transitive dobj_take_closure prep_closure_of poss_decision_our amod_decision_pairwise pobj_of_decision dep_in_as dep_decision_in nn__Ng cc_Ng_and conj_Ng_Cardie pobj_in_ cc__and nn__Bengston cc_Bengston_and conj_Bengston_Roth conj__ nsubj_can_which ccomp_decision_can cc_take_and conj_take_does nn_errors_cause nn_errors_system dobj_does_errors ' +o,2460,'The problem itself has started to get attention only recently ','Daume III','det_problem_The nsubj_started_itself aux_started_has dep_problem_started aux_get_to xcomp_started_get dobj_get_attention advmod_recently_only advmod_get_recently ' +o,2461,'In this form , the distinction between our two models is sometimes referred to as joint versus conditional \' rather than generative versus discriminative \' ',Klein,'prep_referred_In det_form_this pobj_In_form det_distinction_the nsubjpass_referred_distinction prep_distinction_between poss_models_our num_models_two pobj_between_models auxpass_referred_is advmod_referred_sometimes prep_referred_to advmod_joint_as pobj_to_joint prep_joint_versus poss__conditional pobj_versus_ cc_joint_rather dep_rather_than conj_joint_generative prep_generative_versus poss__discriminative pobj_versus_ ' +o,2462,' demonstrates a discriminatively trained system for machine translation that has the following characteristics : 1 -RRB- requires a varying update strategy -LRB- local vs bold -RRB- depending on whether the reference sentence is reachable or not , 2 -RRB- uses sentence level BLEU as a criterion for selecting which output to update towards , and 3 -RRB- only trains on limited length -LRB- 5-15 words -RRB- sentences ',Liang,'advmod_demonstrates_ det_system_a advmod_trained_discriminatively amod_system_trained nsubj_requires_system prep_system_for nn_translation_machine pobj_for_translation nsubj_has_that rcmod_translation_has det_characteristics_the amod_characteristics_following dobj_has_characteristics nsubj_requires_1 ccomp_demonstrates_requires det_strategy_a amod_strategy_varying nn_strategy_update dobj_requires_strategy dep_strategy_local cc_local_vs conj_local_bold partmod_strategy_depending prt_depending_on complm_reachable_whether det_sentence_the nn_sentence_reference nsubj_reachable_sentence cop_reachable_is ccomp_depending_reachable cc_reachable_or conj_reachable_not dep_BLEU_2 amod_BLEU_uses nn_BLEU_sentence nn_BLEU_level conj_strategy_BLEU prep_BLEU_as det_criterion_a pobj_as_criterion prep_criterion_for pcomp_for_selecting dobj_update_which nsubj_update_output aux_update_to ccomp_selecting_update prep_update_towards cc_strategy_and dep_trains_3 amod_trains_only conj_strategy_trains prep_trains_on amod_length_limited pobj_on_length num_words_5-15 appos_length_words nsubj_demonstrates_sentences ' +o,2463,'Walker et al -LRB- forthcoming -RRB- and Boguraev and Briscoe -RRB- ',Hobbs,'cc_Walker_et conj_Walker_al dep_Walker_forthcoming cc_Walker_and conj_Walker_Boguraev cc_Boguraev_and nn__Briscoe conj_Boguraev_ ' +o,2464,'Introduction Verb subcategorizafion probabilities play an important role in both computational linguistic applications and psycholinguisfic models of language processing ',Collins,'nn_probabilities_Introduction nn_probabilities_Verb nn_probabilities_subcategorizafion nsubj_play_probabilities det_role_an amod_role_important dobj_play_role prep_role_in det__both amod__computational amod__linguistic nn__applications pobj_in_ cc_role_and amod_models_psycholinguisfic conj_role_models prep_models_of nn__language nn__processing pobj_of_ ' +o,2465,'Standard data sets for machine learning approaches to this task were put forward by ',Ramshaw,'nn_data_Standard nsubj_sets_data mark_put_for nsubjpass_put_machine partmod_machine_learning dobj_learning_approaches prep_learning_to det_task_this pobj_to_task auxpass_put_were advcl_sets_put advmod_put_forward prep_put_by ' +o,2466,'Its distribution is asymptotic to a 2 distribution and can hence be used as a test statistic ',Dunning,'poss_distribution_Its nsubjpass_asymptotic_distribution cop_asymptotic_is prep_asymptotic_to det_distribution_a num_distribution_2 pobj_to_distribution cc_asymptotic_and aux_used_can advmod_used_hence auxpass_used_be conj_asymptotic_used prep_used_as det_statistic_a nn_statistic_test pobj_as_statistic ' +o,2467,'Ultinmtely , however , it seems that a more complex ai -RRB- t -RRB- roach incorporating back-off and smoothing is necessary ill order to achieve the parsing accuracy achieved by Charniak and ',Collins,'advmod_seems_Ultinmtely advmod_seems_however nsubj_seems_it complm_order_that det_roach_a amod_roach_more nsubj_ai_complex dep_more_ai dep_ai_t nsubj_order_roach partmod_roach_incorporating acomp_incorporating_back-off cc_back-off_and conj_back-off_smoothing cop_order_is amod_order_necessary dep_necessary_ill ccomp_seems_order aux_achieve_to infmod_order_achieve det_accuracy_the amod_accuracy_parsing dobj_achieve_accuracy partmod_accuracy_achieved prep_achieved_by nn__Charniak pobj_by_ cc__and nn__ conj__ ' +o,2468,'Instead researchers condition parsing decisions on many other features , such as parent phrase-marker , and , famously , the lexical-head of the phrase -LRB- and others -RRB- ',Collins,'nn_condition_researchers dep_Instead_condition rcmod_condition_parsing dobj_parsing_decisions prep_decisions_on amod_features_many amod_features_other pobj_on_features dep_as_such prep_decisions_as nn_phrase-marker_parent pobj_as_phrase-marker cc_parsing_and conj_parsing_famously det__the amod__lexical-head prep_lexical-head_of det_phrase_the pobj_of_phrase dobj_famously_ cc__and conj__others ' +o,2469,'Use of probability estimates is not a serious limitation of this approach because in practice candidates are normally provided by some probabilistic model and its probability estimates are used as additional features in the reranker ',Koo,'nsubj_limitation_Use prep_Use_of nn_estimates_probability pobj_of_estimates cop_limitation_is neg_limitation_not det_limitation_a amod_limitation_serious prep_limitation_of det_approach_this pobj_of_approach mark_provided_because prep_provided_in pobj_in_practice nsubjpass_provided_candidates auxpass_provided_are advmod_provided_normally advcl_limitation_provided prep_provided_by det_model_some amod_model_probabilistic pobj_by_model cc_provided_and poss_estimates_its nn_estimates_probability nsubjpass_used_estimates auxpass_used_are conj_provided_used prep_used_as amod_features_additional pobj_as_features prep_features_in det_reranker_the pobj_in_reranker ' +o,2470,'Related Works Generally speaking , approaches to MWE extraction proposed so far can be divided into three categories : a -RRB- statistical approaches based on frequency and co-occurrence affinity , b -RRB- knowledgebased or symbolic approaches using parsers , lexicons and language filters , and c -RRB- hybrid approaches combining different methods ',Wu,'nn_Works_Related nsubjpass_divided_Works advmod_speaking_Generally partmod_Works_speaking appos_Works_approaches prep_Works_to nn_extraction_MWE pobj_to_extraction partmod_extraction_proposed advmod_far_so advmod_proposed_far aux_divided_can auxpass_divided_be prep_divided_into num_categories_three pobj_into_categories dep_approaches_a amod_approaches_statistical dep_categories_approaches prep_approaches_based dep_based_on pobj_on_frequency cc_frequency_and amod_affinity_co-occurrence conj_frequency_affinity dep_filters_b amod_approaches_knowledgebased cc_knowledgebased_or conj_knowledgebased_symbolic nsubj_filters_approaches partmod_approaches_using dobj_using_parsers conj_parsers_lexicons cc_parsers_and conj_parsers_language ccomp_divided_filters cc_divided_and dep_approaches_c nsubj_approaches_hybrid conj_divided_approaches xcomp_approaches_combining amod_methods_different dobj_combining_methods ' +o,2471,'In statistical computational linguistics , maximum conditional likelihood estimators have mostly been used with general exponential or maximum entropy models because standard maximum likelihood estimation is usually computationally intractable ',Berger,'prep_used_In amod_linguistics_statistical amod_linguistics_computational pobj_In_linguistics amod_estimators_maximum amod_estimators_conditional nn_estimators_likelihood nsubjpass_used_estimators aux_used_have advmod_used_mostly auxpass_used_been prep_used_with pobj_with_general dep_general_exponential cc_general_or amod_models_maximum nn_models_entropy conj_general_models mark_intractable_because amod_estimation_standard nn_estimation_maximum nn_estimation_likelihood nsubj_intractable_estimation cop_intractable_is advmod_intractable_usually advmod_intractable_computationally advcl_used_intractable ' +o,2472,' observed that sense division in MRD is frequently too free for the purpose of WSD ',Dolan,'nsubj_observed_ complm_free_that nn_division_sense nsubj_free_division prep_division_in pobj_in_MRD cop_free_is advmod_free_frequently advmod_free_too ccomp_observed_free prep_free_for det_purpose_the pobj_for_purpose prep_purpose_of pobj_of_WSD ' +o,2473,'Some researchers classify terms by similarities based on their distributional syntactic patterns ',Hindle,'det_researchers_Some nsubj__researchers amod_terms_classify dobj__terms prep__by pobj_by_similarities prep__based dep_based_on poss_patterns_their amod_patterns_distributional amod_patterns_syntactic pobj_on_patterns ' +n,2474,'They reported that their method is superior to BLEU in terms of the correlation between human assessment and automatic evaluation ',Papineni,'nsubj_reported_They complm_superior_that poss_method_their nsubj_superior_method cop_superior_is ccomp_reported_superior prep_superior_to nn__BLEU pobj_to_ prep__in pobj_in_terms prep_terms_of det_correlation_the pobj_of_correlation prep_correlation_between amod_assessment_human pobj_between_assessment cc_assessment_and amod_evaluation_automatic conj_assessment_evaluation ' +o,2475,'Discriminative parsing has been investigated before , such as in Johnson , Clark and Curran , Henderson , , Turian et al ',Koo,'amod_parsing_Discriminative nsubjpass_investigated_parsing aux_investigated_has auxpass_investigated_been prep_investigated_before dep_as_such dep_before_as dep_as_in nn__Johnson nn__ conj__Clark cc__and conj__Curran pobj_in_ nn__Henderson pobj_before_ appos__ conj__Turian cc__et conj__al ' +o,2476,'For example , shows how to train a log-linear translation model not by maximizing the likelihood of training data , but maximizing the BLEU score -LRB- among other metrics -RRB- of the model on 53 the data ',Och,'prep_shows_For pobj_For_example nsubj_shows_ advmod_train_how aux_train_to ccomp_shows_train det_model_a amod_model_log-linear nn_model_translation dobj_train_model dep_by_not prep_train_by pcomp_by_maximizing det_likelihood_the dobj_maximizing_likelihood prep_likelihood_of nn_data_training pobj_of_data cc_maximizing_but conj_maximizing_maximizing det_score_the amod_score_BLEU nsubj_data_score dep_score_among amod_metrics_other pobj_among_metrics prep_score_of det_model_the pobj_of_model prep_model_on pobj_on_53 det_data_the xcomp_maximizing_data ' +p,2477,'Several general-purpose off-the-shelf -LRB- OTS -RRB- parsers have become widely available ',Collins,'amod_off-the-shelf_Several amod_off-the-shelf_general-purpose nn_parsers_off-the-shelf abbrev_parsers_OTS nsubj_become_parsers aux_become_have advmod_available_widely dep_become_available ' +o,2478,'3 Related work Word collocation Various collocation metrics have been proposed , including mean and variance , the t-test , the chi-square test , pointwise mutual information -LRB- MI -RRB- , and binomial loglikelihood ratio test -LRB- BLRT -RRB- ',Dunning,'num_work_3 amod_work_Related nsubj_Word_work amod_metrics_collocation amod_metrics_Various nn_metrics_collocation nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been ccomp_Word_proposed prep_proposed_including nn__mean cc_mean_and conj_mean_variance pobj_including_ det__the amod__t-test conj__ det_test_the amod_test_chi-square conj__test amod__pointwise amod__mutual nn__information abbrev__MI appos_test_ cc__and amod_test_binomial amod_test_loglikelihood nn_test_ratio conj__test abbrev_test_BLRT ' +o,2479,'This may stem from the differences between the two models \' feature templates , thresholds , and approximations of the expected values for the features , as discussed in the beginning of the section , or may just reflect differences in the choice of training and test sets -LRB- which are not precisely specified in -RRB- ',Ratnaparkhi,'nsubj_stem_This aux_stem_may prep_stem_from det_differences_the pobj_from_differences prep_differences_between det_models_the num_models_two poss_templates_models nn_templates_feature pobj_between_templates conj_differences_thresholds cc_differences_and conj_differences_approximations prep_approximations_of det_values_the amod_values_expected pobj_of_values dep_values_for det_features_the pobj_for_features mark_discussed_as dep_values_discussed prep_discussed_in det_beginning_the pobj_in_beginning prep_beginning_of det_section_the pobj_of_section cc_discussed_or aux_reflect_may advmod_reflect_just conj_discussed_reflect dobj_reflect_differences prep_reflect_in det_choice_the pobj_in_choice prep_choice_of pobj_of_training cc_values_and dep_sets_test conj_values_sets nsubjpass_specified_which auxpass_specified_are neg_specified_not advmod_specified_precisely dep_values_specified prep_specified_in pobj_in_ ' +o,2480,'1 Introduction Most -LRB- if not all -RRB- statistical machine translation systems employ a word-based alignment model , which treats words in a sentence as independent entities and ignores the structural relationship among them ',Brown,'num_Most_1 nn_Most_Introduction dep_all_if dep_all_not dep_-LRB-_all amod_systems_statistical nn_systems_machine nn_systems_translation nsubj_employ_systems dep_Most_employ det__a amod__word-based amod__alignment nn__model dobj_employ_ nsubj_treats_which rcmod__treats dobj_treats_words prep_treats_in det_sentence_a pobj_in_sentence prep_sentence_as amod_entities_independent pobj_as_entities cc_treats_and conj_treats_ignores det_relationship_the amod_relationship_structural dobj_ignores_relationship prep_ignores_among pobj_among_them ' +o,2481,'len : median length of sequences of co-specifying referring expressions with Cohen \'s n ',Carletta,'amod_length_median dep_len_length prep_length_of pobj_of_sequences prep_sequences_of amod_expressions_co-specifying amod_expressions_referring pobj_of_expressions prep_expressions_with poss_n_Cohen possessive_Cohen_\'s pobj_with_n ' +o,2482,'While close attention has been paid to multi-document summarization technologies , the inherent properties of humanwritten multi-document summaries have not yet been quantified ',Goldstein,'mark_paid_While amod_attention_close nsubjpass_paid_attention aux_paid_has auxpass_paid_been advcl_quantified_paid prep_paid_to amod__multi-document nn__summarization nn__technologies pobj_to_ det_properties_the amod_properties_inherent nsubjpass_quantified_properties prep_properties_of amod_summaries_humanwritten amod_summaries_multi-document pobj_of_summaries aux_quantified_have neg_quantified_not advmod_quantified_yet auxpass_quantified_been ' +o,2483,'Similar ideas were explored in ',He,'amod_ideas_Similar nsubjpass_explored_ideas auxpass_explored_were prep_explored_in ' +o,2484,'1 Introduction In the first SMT systems , word alignment was introduced as a hidden variable of the translation model ',Brown,'num_Introduction_1 prep_introduced_In det__the amod__first nn__SMT nn__systems pobj_In_ nn_alignment_word nsubjpass_introduced_alignment auxpass_introduced_was dep_Introduction_introduced prep_introduced_as det_variable_a amod_variable_hidden pobj_as_variable prep_variable_of det_model_the nn_model_translation pobj_of_model ' +o,2485,'SMT has evolved from the original word-based approach into phrase-based approaches and syntax-based approaches ',Och,'nsubj_evolved_SMT aux_evolved_has prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep_evolved_into amod__phrase-based nn__approaches pobj_into_ cc__and amod__syntax-based nn__approaches conj__ ' +o,2486,' cites the convention from the domain of content analysis indicating that 67 K K -LRB- 8 indicates marginal agreement , while K -RRB- 8 is an indication of good agreement ',Carletta,'nsubj_cites_ det_convention_the dobj_cites_convention prep_cites_from det_domain_the nsubj_indicating_domain prep_domain_of nn_analysis_content pobj_of_analysis pcomp_from_indicating complm_indication_that nsubj_indication_67 nn_K_K dep_67_K nsubj_indicates_8 dep_67_indicates amod_agreement_marginal dobj_indicates_agreement prep_indicates_while pobj_while_K dep_67_8 cop_indication_is det_indication_an ccomp_indicating_indication prep_indication_of amod_agreement_good pobj_of_agreement ' +n,2487,'While the idea of exploiting multiple news reports for paraphrase acquisition is not new , previous efforts have been restricted to at most two news sources ',Barzilay,'mark_new_While det_idea_the nsubj_new_idea prep_idea_of pcomp_of_exploiting amod_reports_multiple nn_reports_news dobj_exploiting_reports prep_exploiting_for amod_acquisition_paraphrase pobj_for_acquisition cop_new_is neg_new_not advcl_restricted_new amod__previous nn__efforts nsubjpass_restricted_ aux_restricted_have auxpass_restricted_been prep_restricted_to quantmod_two_at dep_at_most num_sources_two nn_sources_news pobj_to_sources ' +o,2488,'The synchronous grammar rules are extracted from word aligned sentence pairs where the target sentence is annotated with a syntactic parse ',Galley,'det_rules_The amod_rules_synchronous nn_rules_grammar nsubjpass_extracted_rules auxpass_extracted_are prep_extracted_from pobj_from_word dep_pairs_aligned nn_pairs_sentence partmod_word_pairs advmod_annotated_where det_sentence_the nn_sentence_target nsubjpass_annotated_sentence auxpass_annotated_is advcl_pairs_annotated prep_annotated_with det_parse_a amod_parse_syntactic pobj_with_parse ' +o,2489,'Since , numerous works have used patterns for discovery and identification of instances of semantic relationships -LRB- eg , -RRB- ',Snow,'prep_used_Since pobj_Since_ amod_works_numerous nsubj_used_works aux_used_have dobj_used_patterns prep_patterns_for pobj_for_discovery cc_discovery_and conj_discovery_identification prep_patterns_of pobj_of_instances prep_instances_of amod_relationships_semantic pobj_of_relationships appos_relationships_eg dep_eg_ ' +o,2490,'31 Selecting Coreference Systems A learning-based coreference system can be defined by four elements : the learning algorithm used to train the coreference classifier , the method of creating training instances for the learner , the feature set 2Examples of such scoring functions include the DempsterShafer rule -LRB- see Kehler and -RRB- and its variants -LRB- see Harabagiu et al ',Bean,'num_Systems_31 nn_Systems_Selecting nn_Systems_Coreference nsubjpass_defined_Systems det_system_A amod_system_learning-based nn_system_coreference dep_Systems_system aux_defined_can auxpass_defined_be prep_defined_by num_elements_four pobj_by_elements det_algorithm_the nn_algorithm_learning nsubj_used_algorithm parataxis_defined_used aux_train_to xcomp_used_train det_classifier_the amod_classifier_coreference dobj_train_classifier det_method_the nsubj_include_method prep_method_of pcomp_of_creating nn_instances_training dobj_creating_instances prep_creating_for det_learner_the pobj_for_learner det_2Examples_the nn_2Examples_feature nn_2Examples_set appos_learner_2Examples prep_2Examples_of amod_functions_such amod_functions_scoring pobj_of_functions conj_defined_include det_rule_the nn_rule_DempsterShafer dobj_include_rule dep_rule_see nn__Kehler dobj_see_ cc__and conj__ cc_defined_and nsubj_variants_its conj_defined_variants xcomp_variants_see advmod_see_Harabagiu nn_al_et dobj_see_al ' +o,2491,'However , after several advances in tasks such as automatic tagging of text with high level semantics such as parts-of-speech , named-entities , sentence-parsing , etc , there is increasing hope that one could leverage this information into IR techniques ',Ratnaparkhi,'advmod_is_However prep_is_after amod_advances_several pobj_after_advances prep_advances_in pobj_in_tasks dep_as_such prep_tasks_as amod_tagging_automatic pobj_as_tagging prep_tagging_of pobj_of_text prep_text_with amod_semantics_high nn_semantics_level pobj_with_semantics dep_as_such prep_semantics_as amod__parts-of-speech pobj_as_ amod__named-entities appos__ amod__sentence-parsing appos__ dep__etc expl_is_there dep_is_increasing dobj_increasing_hope complm_leverage_that nsubj_leverage_one aux_leverage_could ccomp_increasing_leverage det_information_this dobj_leverage_information prep_leverage_into nn_techniques_IR pobj_into_techniques ' +o,2492,'Instead , we follow a simplified form of previous work on biography creation , where a classifier is trained to distinguish biographical text ',Biadsy,'advmod_follow_Instead nsubj_follow_we det_form_a amod_form_simplified dobj_follow_form prep_form_of amod_work_previous pobj_of_work prep_form_on nn_creation_biography pobj_on_creation advmod_trained_where det_classifier_a nsubj_trained_classifier cop_trained_is rcmod_creation_trained aux_distinguish_to xcomp_trained_distinguish amod_text_biographical dobj_distinguish_text ' +n,2493,'Furthermore , we provide a 638 \% error reduction compared to IBM Model 4 ',Brown,'advmod_provide_Furthermore nsubj_provide_we det_reduction_a num_reduction_638 nn_reduction_\% nn_reduction_error dobj_provide_reduction prep_reduction_compared dep_compared_to nn_Model_IBM pobj_to_Model num_Model_4 ' +o,2494,' , mention about substrings of collocations ',Smadja,'dep__mention prep__about pobj_about_substrings prep__of pobj_of_collocations ' +o,2495,'In , a maximum entropy tagger is presented ',Ratnaparkhi,'prep_presented_In pobj_In_ det_tagger_a amod_tagger_maximum nn_tagger_entropy nsubjpass_presented_tagger auxpass_presented_is ' +o,2496,'Identifying subjectivity helps separate opinions from fact , which may be useful in question answering , summarization , etc Sentiment detection is the task of determining positive or negative sentiment of words , phrases and sentences , or documents ',Pang,'advmod_Identifying_subjectivity dep_Identifying_helps amod_opinions_separate dobj_helps_opinions prep_helps_from pobj_from_fact nsubj_useful_which aux_useful_may cop_useful_be rcmod_fact_useful prep_useful_in nn_answering_question pobj_in_answering nsubj_Identifying_summarization nn_detection_etc nn_detection_Sentiment appos_summarization_detection cop_task_is det_task_the rcmod_detection_task prep_task_of pcomp_of_determining amod_sentiment_positive cc_positive_or conj_positive_negative dobj_determining_sentiment prep_sentiment_of nn__words pobj_of_ conj__phrases cc_phrases_and nn__sentences conj_phrases_ cc__or conj__documents ' +o,2497,'CIT -RRB- ',Smith,'nn_-RRB-_CIT ' +o,2498,'We selected four binary NLP datasets for evaluation : 20 Newsgroups1 and Reuters -LRB- used by Tong and Koller -RRB- and sentiment classification and spam ',Blitzer,'nsubj_selected_We num_datasets_four amod_datasets_binary nn_datasets_NLP dobj_selected_datasets prep_datasets_for pobj_for_evaluation num_Newsgroups1_20 dep_datasets_Newsgroups1 cc_Newsgroups1_and nn__Reuters conj_Newsgroups1_ dep_Newsgroups1_used prep_used_by pobj_by_Tong cc_Tong_and conj_Tong_Koller cc_Newsgroups1_and nn__sentiment nn__classification conj_Newsgroups1_ cc__and nn__spam conj__ ' +p,2499,'The best example of such an approach is , who proposes a method that automatically identifies collocations that are indicative of the sense of a word , and uses those to iteratively label more examples ',Yarowsky,'det_example_The amod_example_best nsubj__example prep_example_of predet_approach_such det_approach_an pobj_of_approach cop__is dep_proposes_ nsubj_proposes_who det_method_a dobj_proposes_method nsubj_identifies_that advmod_identifies_automatically rcmod_method_identifies dobj_identifies_collocations nsubj_indicative_that cop_indicative_are rcmod_collocations_indicative prep_indicative_of det_sense_the pobj_of_sense prep_sense_of det_word_a pobj_of_word cc_indicative_and conj_indicative_uses nsubj_label_those aux_label_to advmod_label_iteratively xcomp_uses_label amod_examples_more dobj_label_examples ' +p,2500,'1 Introduction There has been a great deal of progress in statistical parsing in the past decade ',Collins,'num_Introduction_1 expl_been_There aux_been_has dep_Introduction_been det_deal_a amod_deal_great dobj_been_deal prep_deal_of pobj_of_progress prep_been_in amod_parsing_statistical pobj_in_parsing prep_parsing_in det_decade_the amod_decade_past pobj_in_decade ' +o,2501,'We set the feature weights by optimizing the Bleu score directly using minimum error rate training on the development set ',Och,'nsubj_set_We det_weights_the nn_weights_feature dobj_set_weights prep_set_by pcomp_by_optimizing det_score_the nn_score_Bleu dobj_optimizing_score advmod_using_directly xcomp_optimizing_using amod__minimum nn__error nn__rate nn__training dobj_using_ prep_using_on det_set_the nn_set_development pobj_on_set ' +o,2502,'Such measures as mutual information , latent semantic analysis , log-likelihood ratio have been proposed to evaluate word semantic similarity based on the co-occurrence information on a large corpus ',Dunning,'amod_measures_Such nsubjpass_proposed_measures prep_measures_as amod__mutual nn__information pobj_as_ amod__latent amod__semantic nn__analysis appos__ amod__log-likelihood nn__ratio appos__ aux_proposed_have auxpass_proposed_been aux_evaluate_to xcomp_proposed_evaluate nn_similarity_word amod_similarity_semantic dobj_evaluate_similarity prep_similarity_based dep_based_on det_information_the amod_information_co-occurrence pobj_on_information prep_information_on det_corpus_a amod_corpus_large pobj_on_corpus ' +o,2503,'33 Methods We parsed the English side of each bilingual bitext and both sides of each English\\/English bitext using an off-the-shelf syntactic parser , which was trained on sections 02-21 of the Penn English Treebank ',Marcus,'num_Methods_33 nsubj_parsed_We dep_Methods_parsed det_side_the amod_side_English dobj_parsed_side prep_side_of det_bitext_each amod_bitext_bilingual pobj_of_bitext cc_side_and det_sides_both conj_side_sides prep_sides_of det_bitext_each nn_bitext_English\\/English pobj_of_bitext xcomp_parsed_using det__an amod__off-the-shelf amod__syntactic nn__parser dobj_using_ nsubjpass_trained_which auxpass_trained_was rcmod__trained prep_trained_on pobj_on_sections num_sections_02-21 prep_sections_of det_Treebank_the nn_Treebank_Penn nn_Treebank_English pobj_of_Treebank ' +p,2504,'An especially well-founded framework for doing this is maximum entropy ',Berger,'det_framework_An advmod_well-founded_especially amod_framework_well-founded nsubj_entropy_framework prep_framework_for pcomp_for_doing dobj_doing_this cop_entropy_is amod_entropy_maximum ' +o,2505,'To implement this method , we rst use the Stanford Named Entity Recognizer4 toidentifythesetofpersonandorganisation entities , E , from each article in the corpus ',Finkel,'aux_implement_To dep_rst_implement det_method_this dobj_implement_method nsubj_rst_we xcomp_rst_use det_entities_the nn_entities_Stanford nn_entities_Named nn_entities_Entity nn_entities_Recognizer4 nn_entities_ nn_entities_toidentifythesetofpersonandorganisation dobj_use_entities appos_entities_E prep_entities_from det_article_each pobj_from_article prep_article_in det_corpus_the pobj_in_corpus ' +o,2506,'Then , by using evaluations similar to those described in and by , we show that the best distance-based measures correlate better overall with human association scores than do the best window based configurations -LRB- see Section 4 -RRB- , and that they also serve as better predictors of the strongest human associations -LRB- see Section 5 -RRB- ',Rapp,'advmod_show_Then prep_show_by pcomp_by_using nsubj_similar_evaluations xcomp_using_similar prep_similar_to pobj_to_those partmod_those_described prep_described_in pobj_in_ cc_by_and conj_by_by pobj_by_ nsubj_show_we complm_correlate_that det_measures_the amod_measures_best amod_measures_distance-based nsubj_correlate_measures ccomp_show_correlate advmod_correlate_better advmod_better_overall prep_correlate_with amod_scores_human nn_scores_association pobj_with_scores mark_do_than dep_correlate_do det_window_the amod_window_best dobj_do_window partmod_window_based dobj_based_configurations parataxis_do_see dobj_see_Section num_Section_4 cc_do_and dep_serve_that nsubj_serve_they advmod_serve_also conj_do_serve prep_serve_as amod_predictors_better pobj_as_predictors prep_predictors_of det_associations_the amod_associations_strongest amod_associations_human pobj_of_associations dep_associations_see dobj_see_Section num_Section_5 ' +o,2507,'My guess is that the features used in eg , the Collins or Charniak parsers are probably close to optimal for English Penn Treebank parsing , but that other features might improve parsing of other languages or even other English genres ',Marcus,'poss_guess_My nsubj_is_guess complm_used_that det_features_the nsubj_used_features ccomp_is_used prep_used_in pobj_in_eg det__the nn__Collins nsubj_close_ cc__or nn_parsers_Charniak nn_parsers_ conj__parsers cop_close_are advmod_close_probably dep_used_close aux_optimal_to xcomp_close_optimal prep_optimal_for nn__English nn__Penn nn__Treebank amod__parsing pobj_for_ cc_close_but dep_improve_that amod_features_other nsubj_improve_features aux_improve_might conj_close_improve dobj_improve_parsing prep_parsing_of amod_languages_other pobj_of_languages cc_languages_or advmod_genres_even amod_genres_other amod_genres_English conj_languages_genres ' +o,2508,'Typically , a small set of seed polar phrases are prepared , and new polar phrases are detected based on the strength of co-occurrence with the seeds ',Turney,'advmod_prepared_Typically det_set_a amod_set_small nsubj_prepared_set prep_set_of nn_phrases_seed amod_phrases_polar pobj_of_phrases cop_prepared_are cc_prepared_and amod_phrases_new amod_phrases_polar nsubjpass_detected_phrases auxpass_detected_are conj_prepared_detected prep_detected_based dep_based_on det_strength_the pobj_on_strength prep_strength_of pobj_of_co-occurrence prep_co-occurrence_with det_seeds_the pobj_with_seeds ' +o,2509,'For example , aspects of a digital camera could include picture quality , battery life , size , color , value , etc Finding such aspects is a challenging research problem that has been addressed in a number of ways ',Titov,'prep_include_For pobj_For_example nsubj_include_aspects prep_aspects_of det_camera_a amod_camera_digital pobj_of_camera aux_include_could nn_quality_picture dep_value_quality nn_life_battery appos_quality_life appos_life_size dep_value_color dobj_include_value dep_value_etc xcomp_include_Finding amod_aspects_such nsubj_problem_aspects cop_problem_is det_problem_a amod_problem_challenging nn_problem_research ccomp_Finding_problem nsubjpass_addressed_that aux_addressed_has auxpass_addressed_been rcmod_problem_addressed prep_addressed_in det_number_a pobj_in_number prep_number_of pobj_of_ways ' +o,2510,'The approach made use of a maximum entropy model formulated from frequency information for various combinations of the observed features ',Berger,'det_approach_The nsubj_made_approach dobj_made_use prep_use_of det__a amod__maximum amod__entropy nn__model pobj_of_ partmod__formulated prep_formulated_from nn_information_frequency pobj_from_information prep_formulated_for amod_combinations_various pobj_for_combinations prep_combinations_of det_features_the amod_features_observed pobj_of_features ' +o,2511,'Instead , we opt to utilize the Stanford NER tagger over the sentences in a document and annotate each NP with the NER label assigned to that mention head ',Finkel,'advmod_opt_Instead nsubj_opt_we aux_utilize_to xcomp_opt_utilize det__the nn__Stanford nn__NER nn__tagger dobj_utilize_ prep__over det_sentences_the pobj_over_sentences prep_sentences_in det_document_a pobj_in_document cc_utilize_and conj_utilize_annotate det_NP_each dobj_annotate_NP prep_annotate_with det_label_the nn_label_NER pobj_with_label partmod_label_assigned prep_assigned_to det_head_that nn_head_mention pobj_to_head ' +o,2512,'The simple model 1 for the translation of a SL sentence d = dldt in a TL sentence e = el em assumes that every TL word is generated independently as a mixture of the SL words : m l P -LRB- e -LRB- d -RRB- , , ~ H ~ t -LRB- ej -LRB- di -RRB- -LRB- 2 -RRB- j = l i = O In the equation above t -LRB- ej -LRB- di -RRB- stands for the probability that ej is generated by di ',Brown,'det_model_The amod_model_simple dep_dldt_model num__1 dep_model_ prep__for det_translation_the pobj_for_translation prep_translation_of det_sentence_a nn_sentence_SL pobj_of_sentence dep_dldt_d amod_dldt_= dep_assumes_dldt prep_dldt_in det_sentence_a nn_sentence_TL pobj_in_sentence dep_assumes_e amod_em_= nn_em_el nsubj_assumes_em complm_generated_that det_word_every nn_word_TL nsubjpass_generated_word auxpass_generated_is ccomp_assumes_generated advmod_generated_independently prep_generated_as det_mixture_a pobj_as_mixture prep_mixture_of det_di_the nn_words_SL nn_di_words nn_P_m nn_P_l dep_words_P dep_P_e appos_e_d nn_t_~ nn_t_H nn_t_~ appos_P_t nn_di_ej pobj_of_di dep_O_2 nsubj_O_j dep_j_= dobj_=_l nn_l_i dep_l_= dep_assumes_O prep_stands_In det_equation_the pobj_In_equation prep_equation_above pobj_above_t nn_di_ej nsubj_stands_di ccomp_O_stands prep_stands_for det_probability_the pobj_for_probability complm_generated_that nsubjpass_generated_ej auxpass_generated_is ccomp_stands_generated prep_generated_by pobj_by_di ' +o,2513,'Moreover , as mentions , some of the benefits of Model 2 are already captured by inclusion of the distance measure ',Collins,'advmod_captured_Moreover mark_mentions_as nsubj_mentions_ dep_captured_mentions nsubjpass_captured_some prep_some_of det_benefits_the pobj_of_benefits prep_benefits_of pobj_of_Model num_Model_2 auxpass_captured_are advmod_captured_already prep_captured_by pobj_by_inclusion prep_inclusion_of det_measure_the nn_measure_distance pobj_of_measure ' +o,2514,'The translation models they presented in various papers between 1988 and 1993 are commonly referred to as IBM models 15 , based on the numbering in Brown , Della Pietra , Della Pietra , and Mercer ',Brown,'det_models_The nn_models_translation nsubj_presented_they dep_models_presented prep_presented_in amod_papers_various pobj_in_papers prep_papers_between pobj_between_1988 cc_1988_and conj_1988_1993 nsubjpass_referred_ auxpass_referred_are advmod_referred_commonly dep_models_referred xcomp_referred_to prep_referred_as nn_models_IBM pobj_as_models dep_models_15 prep_referred_based dep_based_on det_numbering_the pobj_on_numbering prep_numbering_in pobj_in_Brown nn_Pietra_Della conj_Brown_Pietra nn_Pietra_Della conj_Brown_Pietra cc_Brown_and nn__Mercer conj_Brown_ ' +o,2515,'To tune all lambda weights above , we perform minimum error rate training on the development set described in Section 7 ',Och,'aux_tune_To dep_perform_tune det_weights_all nn_weights_lambda dobj_tune_weights advmod_tune_above nsubj_perform_we amod__minimum nn__error nn__rate nn__training dobj_perform_ prep_perform_on det_set_the nn_set_development pobj_on_set partmod_set_described prep_described_in pobj_in_Section num_Section_7 ' +o,2516,' ` Ontological Promiscuity \' , Proceedings of the 23rd Annual Meeting of the Association for Computational Linguistics , Chicago , Illinois , pp ',Hobbs,'nn_Promiscuity_ nn_Promiscuity_Ontological nsubj_pp_Promiscuity nsubj_pp_Proceedings prep_Proceedings_of det_Meeting_the amod_Meeting_23rd amod_Meeting_Annual pobj_of_Meeting prep_Meeting_of det_Association_the pobj_of_Association prep_Association_for nn_Illinois_Computational nn_Illinois_Linguistics appos_Illinois_Chicago pobj_for_Illinois ' +o,2517,'Examples of monolingual parallel corpora that have been used are multiple translations of classical French novels into English , and data created for machine translation evaluation methods such as Bleu which use multiple reference translations ',Papineni,'nsubj_translations_Examples prep_Examples_of amod_corpora_monolingual amod_corpora_parallel pobj_of_corpora nsubjpass_used_that aux_used_have auxpass_used_been rcmod_corpora_used cop_translations_are amod_translations_multiple prep_translations_of amod_novels_classical amod_novels_French pobj_of_novels prep_translations_into pobj_into_English cc_translations_and nsubj_created_data conj_translations_created prep_created_for nn_methods_machine nn_methods_translation nn_methods_evaluation pobj_for_methods dep_as_such prep_methods_as nn__Bleu pobj_as_ nsubj_use_which rcmod__use amod_translations_multiple nn_translations_reference dobj_use_translations ' +o,2518,'22 Motivation from previous work 221 Parsing In recent years , the success of statistical parsing techniques can be attributed to several factors , such as the increasing size of computing machinery to accommodate larger models , the availability of resources such as the Penn Treebank and the success of machine learning techniques for lowerlevel NLP problems , such as part-of-speech tagging , and PPattachment ',Marcus,'num_Motivation_22 nsubjpass_attributed_Motivation prep_Motivation_from amod_Parsing_previous nn_Parsing_work nn_Parsing_221 pobj_from_Parsing prep_attributed_In amod_years_recent pobj_In_years det_success_the nsubjpass_attributed_success prep_success_of amod_techniques_statistical amod_techniques_parsing pobj_of_techniques aux_attributed_can auxpass_attributed_be prep_attributed_to amod_factors_several pobj_to_factors dep_as_such prep_factors_as det_size_the amod_size_increasing pobj_as_size prep_size_of nn_machinery_computing pobj_of_machinery aux_accommodate_to infmod_size_accommodate amod_models_larger dobj_accommodate_models det_availability_the conj_models_availability prep_availability_of pobj_of_resources dep_as_such prep_resources_as det__the nn__Penn nn__Treebank pobj_as_ cc_models_and det_success_the conj_models_success prep_success_of pobj_of_machine partmod_machine_learning dobj_learning_techniques prep_techniques_for amod_problems_lowerlevel nn_problems_NLP pobj_for_problems dep_as_such prep_learning_as amod__part-of-speech amod__tagging pobj_as_ cc_models_and nn__PPattachment conj_models_ ' +o,2519,' -RRB- and view the POS tags and word identities as two separate sources of information ',Brown,'nn_view_and dobj__view det_tags_the dep_tags_POS dep_as_tags cc_tags_and nn_identities_word conj_tags_identities prep__as num_sources_two amod_sources_separate pobj_as_sources prep_sources_of pobj_of_information ' +o,2520,'In all of the cited approaches , the Penn Wall Street Journal Treebank is used , the availability of whichobviates the standard eort required for treebank traininghandannotating large corpora of specic domains of specic languages with specic parse types ',Marcus,'prep_used_In pobj_In_all prep_all_of det_approaches_the amod_approaches_cited pobj_of_approaches det__the nn__Penn nn__Wall nn__Street nn__Journal nn__Treebank nsubjpass_used_ auxpass_used_is det_availability_the nsubj_eort_availability prep_availability_of pobj_of_whichobviates det_eort_the amod_eort_standard xcomp_used_eort partmod_eort_required prep_required_for pobj_for_treebank xcomp_required_traininghandannotating amod_corpora_large dobj_traininghandannotating_corpora prep_corpora_of amod_domains_specic pobj_of_domains prep_domains_of amod_languages_specic pobj_of_languages prep_traininghandannotating_with amod_types_specic nn_types_parse pobj_with_types ' +o,2521,'Many corpus based methods have been proposed to deal with the sense disambiguation problem when given de nition for each possible sense of a target word or a tagged corpus with the instances of each possible sense , eg , supervised sense disambiguation , and semi-supervised sense disambiguation ',Yarowsky,'amod_corpus_Many dep_corpus_based dep_based_methods aux_proposed_have auxpass_proposed_been rcmod_methods_proposed aux_deal_to xcomp_proposed_deal prep_deal_with det_problem_the nn_problem_sense nn_problem_disambiguation pobj_with_problem advmod_given_when advcl_deal_given iobj_given_de dobj_given_nition prep_nition_for det_sense_each amod_sense_possible pobj_for_sense prep_sense_of det_word_a nn_word_target pobj_of_word cc_given_or det_corpus_a amod_corpus_tagged conj_given_corpus prep_corpus_with det_instances_the pobj_with_instances prep_instances_of det_sense_each amod_sense_possible pobj_of_sense conj_sense_eg amod__supervised nn__sense nn__disambiguation conj_sense_ cc_sense_and amod_disambiguation_semi-supervised nn_disambiguation_sense conj_sense_disambiguation ' +o,2522,'Additionally , our approach makes it possible to do inference in just about twice the inference time with a single sequential CRF ; in contrast , approaches like Gibbs Sampling that model the dependencies directly can increase inference time by a factor of 30 ',Finkel,'advmod_makes_Additionally poss_approach_our nsubj_makes_approach nsubj_possible_it xcomp_makes_possible aux_do_to xcomp_possible_do dobj_do_inference prep_do_in quantmod_twice_just quantmod_twice_about pobj_in_twice det_time_the amod_time_inference tmod_in_time prep_do_with det_CRF_a amod_CRF_single amod_CRF_sequential pobj_with_CRF prep_approaches_in pobj_in_contrast parataxis_makes_approaches prep_approaches_like nn_Sampling_Gibbs pobj_like_Sampling complm_increase_that nsubj_increase_model det_dependencies_the dep_directly_dependencies advmod_increase_directly aux_increase_can ccomp_approaches_increase amod_time_inference dobj_increase_time prep_increase_by det_factor_a pobj_by_factor prep_factor_of pobj_of_30 ' +o,2523,'Construct a parse chart with a CKY parser simultaneously constrained on the foreign string and English tree , similar to the bilingual parsing of 1 ',Wu,'det_chart_a amod_chart_parse dobj_Construct_chart prep_Construct_with det_parser_a nn_parser_CKY pobj_with_parser advmod_constrained_simultaneously partmod_parser_constrained prep_constrained_on det_string_the amod_string_foreign pobj_on_string cc_string_and conj_string_English nsubj_Construct_tree amod_tree_similar prep_similar_to det_parsing_the amod_parsing_bilingual pobj_to_parsing prep_parsing_of pobj_of_ num__1 ' +p,2524,'State-of-art systems for doing word alignment use generative models like GIZA + + ',Brown,'amod_systems_State-of-art nsubj_use_systems prep_systems_for pcomp_for_doing nn_alignment_word dobj_doing_alignment amod_models_generative dobj_use_models prep_use_like nn_+_GIZA nn_+_+ pobj_like_+ ' +o,2525,' used a different update style based on a convex loss function : = L -LRB- e , e ; et -RRB- max parenleftBig 0 , 1 parenleftBig si -LRB- f t , e -RRB- si -LRB- f t , e -RRB- parenrightBigparenrightBig 768 Table 1 : Experimental results obtained by varying normalized tokens used with surface form ',Tillmann,'advmod_used_ det_style_a amod_style_different nn_style_update dobj_used_style prep_used_based dep_based_on det_function_a amod_function_convex nn_function_loss pobj_on_function nn_max_= nn_max_L appos_max_e nn_et_e dep_e_et dep_function_max partmod_max_parenleftBig dobj_parenleftBig_0 num_si_1 amod_si_parenleftBig appos_max_si amod_t_f appos_si_t dep_t_e nsubj_used_si amod_t_f appos_si_t dep_t_e partmod_si_parenrightBigparenrightBig num_Table_768 dobj_parenrightBigparenrightBig_Table num_Table_1 amod_results_Experimental dep_Table_results partmod_results_obtained prep_obtained_by amod_tokens_varying amod_tokens_normalized pobj_by_tokens partmod_tokens_used prep_used_with nn_form_surface pobj_with_form ' +o,2526,'This research has focused mostly on the development of statistical parsers trained on large annotated corpora , in particular the Penn Treebank WSJ corpus ',Marcus,'det_research_This nsubj_focused_research aux_focused_has advmod_focused_mostly prep_focused_on det_development_the pobj_on_development prep_development_of amod_parsers_statistical pobj_of_parsers partmod_parsers_trained prep_trained_on amod_corpora_large amod_corpora_annotated pobj_on_corpora prep_focused_in pcomp_in_particular det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_WSJ dobj_particular_corpus ' +o,2527,'It also contains tools for tuning these models using minimum error rate training and evaluating the resulting translations using the BLEU score ',Papineni,'nsubj_contains_It advmod_contains_also dobj_contains_tools prep_contains_for pcomp_for_tuning det_models_these dobj_tuning_models xcomp_tuning_using amod__minimum nn__error nn__rate nn__training dobj_using_ cc_tuning_and conj_tuning_evaluating det_translations_the amod_translations_resulting dobj_evaluating_translations xcomp_evaluating_using det_score_the amod_score_BLEU dobj_using_score ' +o,2528,'Binarizing the grammars further increases the size of these sets , due to the introduction of virtual nonterminals ',Zhang,'det__the amod__grammars dobj_Binarizing_ advmod_Binarizing_further dep_Binarizing_increases det_size_the nsubj_Binarizing_size prep_size_of det_sets_these pobj_of_sets amod_sets_due prep_due_to det_introduction_the pobj_to_introduction prep_introduction_of amod_nonterminals_virtual pobj_of_nonterminals ' +o,2529,'We argue that linguistic knowledge could not only improve results but is essential when extracting collocations from certain languages : this knowledge provides other applications -LRB- or a lexicon user , respectively -RRB- with a ne-grained description of how the extracted collocations are to be used in context ',Smadja,'nsubj_argue_We complm_improve_that amod_knowledge_linguistic nsubj_improve_knowledge aux_improve_could neg_improve_not advmod_improve_only ccomp_argue_improve dobj_improve_results advmod_improve_ cc_improve_but cop_essential_is conj_improve_essential advmod_extracting_when ccomp_essential_extracting dobj_extracting_collocations prep_extracting_from amod_languages_certain pobj_from_languages det_knowledge_this nsubj_provides_knowledge parataxis_argue_provides amod_applications_other dobj_provides_applications dep_-LRB-_or det_user_a nn_user_lexicon dep_respectively_user dep_-LRB-_respectively prep_provides_with det_description_a amod_description_ne-grained pobj_with_description prep_description_of advmod_are_how det_collocations_the amod_collocations_extracted nsubj_are_collocations pcomp_of_are aux_used_to auxpass_used_be xcomp_are_used prep_used_in pobj_in_context ' +o,2530,'From this aligned training corpus , we extract the phrase pairs according to the heuristics in ',Koehn,'prep_extract_From det_corpus_this amod_corpus_aligned nn_corpus_training pobj_From_corpus nsubj_extract_we det_pairs_the nn_pairs_phrase dobj_extract_pairs prep_extract_according dep_according_to det_heuristics_the pobj_to_heuristics prep_heuristics_in ' +o,2531,'31 Conditional Random Field for Alignment Our conditional random field -LRB- CRF -RRB- for alignment has a graphical model structure that resembles that of IBM Model 1 ',Brown,'num_Field_31 nn_Field_Conditional nn_Field_Random nsubj_has_Field prep_Field_for pobj_for_Alignment poss_field_Our amod_field_conditional amod_field_random dep_Field_field abbrev_Field_CRF prep_Field_for pobj_for_alignment det_structure_a amod_structure_graphical nn_structure_model dobj_has_structure nsubj_that_that cop_that_resembles rcmod_structure_that prep_that_of nn_Model_IBM pobj_of_Model num_Model_1 ' +p,2532,'5 Comparison with Previous Top Systems and Related Work In POS tagging , the previous best performance was reported by as summarized in Table 7 ',Shen,'num_Comparison_5 nsubj_summarized_Comparison prep_Comparison_with amod_Systems_Previous nn_Systems_Top pobj_with_Systems cc_Systems_and nn_Work_Related conj_Systems_Work prep_reported_In dep_tagging_POS pobj_In_tagging det_performance_the amod_performance_previous amod_performance_best nsubjpass_reported_performance auxpass_reported_was rcmod_Comparison_reported prep_reported_by pobj_by_ advmod__as prep_summarized_in pobj_in_Table num_Table_7 ' +o,2533,'As discussed in (Och, 2003), the direct translation model represents the probability of target sentence English e = e1eI being the translation for a source sentence French f = f1 fJ through an exponential, or log-linear model p(e|f) = exp( summationtextm k=1 k hk(e,f))summationtext eprimeE exp( summationtextm k=1 k hk(eprime,f)) (1) where e is a single candidate translation for f from the set of all English translations E, is the parameter vector for the model, and each hk is a feature function of e and f. In practice, we restrict E to the set Gen(f) which is a set of highly likely translations discovered by a decoder (Vogel et al. , 2003).',Och,'' +o,2534,'Evaluation Metrics We evaluated the generated translations using three different evaluation metrics : BLEU score , mWER -LRB- multi-reference word error rate -RRB- , and mPER -LRB- multi-reference positionindependent word error rate -RRB- ',Papineni,'advmod_Metrics_Evaluation nsubj_evaluated_We ccomp_Metrics_evaluated det_translations_the amod_translations_generated nsubj_Metrics_translations partmod_translations_using num_metrics_three amod_metrics_different nn_metrics_evaluation dobj_using_metrics amod__BLEU nn__score dep_translations_ appos__mWER amod_rate_multi-reference nn_rate_word nn_rate_error appos_mWER_rate cc_translations_and amod__mPER nn_rate_multi-reference amod_rate_positionindependent nn_rate_word nn_rate_error appos__rate conj_translations_ ' +o,2535,'The study is conducted on both a simple Air Travel Information System -LRB- ATIS -RRB- corpus and the more complex Wall Street Journal -LRB- WSJ -RRB- corpus ',Marcus,'det_study_The nsubjpass_conducted_study auxpass_conducted_is prep_conducted_on det__both det_System_a amod_System_simple nn_System_Air nn_System_Travel nn_System_Information nn__System abbrev_System_ATIS nn__corpus pobj_on_ cc__and det_corpus_the advmod_complex_more amod_corpus_complex nn_corpus_Wall nn_corpus_Street nn_corpus_Journal abbrev_corpus_WSJ conj__corpus ' +o,2536,'This sparse information , however , can be propagated across all data based on distributional similarity ',Haghighi,'det_information_This amod_information_sparse nsubjpass_propagated_information advmod_information_however aux_propagated_can auxpass_propagated_be prep_propagated_across det_data_all pobj_across_data prep_propagated_based dep_based_on amod_similarity_distributional pobj_on_similarity ' +o,2537,'41 Overview In this work , factored models are experimented with three factors : the surface form , the lemma and the part of speech -LRB- POS -RRB- ',Koehn,'num_Overview_41 dep_Overview_In det_work_this pobj_In_work amod_models_factored appos_work_models nsubjpass_experimented_ auxpass_experimented_are rcmod_models_experimented prep_experimented_with num_factors_three pobj_with_factors det_form_the nn_form_surface dep_models_form det_lemma_the appos_form_lemma cc_models_and det_part_the conj_models_part prep_part_of pobj_of_speech appos_speech_POS ' +o,2538,'ROUGE-L This measure evaluates summaries by longest common subsequence -LRB- LCS -RRB- defined by Equation 4 ',Lin,'nn__ROUGE-L det_measure_This nsubj_evaluates_measure dep__evaluates dobj_evaluates_summaries prep_evaluates_by amod_subsequence_longest amod_subsequence_common pobj_by_subsequence abbrev_subsequence_LCS partmod_subsequence_defined prep_defined_by pobj_by_Equation num_Equation_4 ' +o,2539,'We also report state-of-the-art results for Hebrew full mor1Another notable work , though within a slightly different framework , is the prototype-driven method proposed by , in which the dictionary is replaced with a very small seed of prototypical examples ',Haghighi,'nsubj_report_We advmod_report_also amod_results_state-of-the-art dobj_report_results prep_results_for pobj_for_Hebrew amod_work_full amod_work_mor1Another amod_work_notable nsubj_method_work amod_work_though prep_though_within det_framework_a advmod_different_slightly amod_framework_different pobj_within_framework cop_method_is det_method_the amod_method_prototype-driven dep_report_method partmod_method_proposed prep_proposed_by pobj_by_ rel_replaced_in pobj_in_which det_dictionary_the nsubjpass_replaced_dictionary auxpass_replaced_is rcmod__replaced prep_replaced_with det_seed_a advmod_small_very amod_seed_small pobj_with_seed prep_seed_of amod_examples_prototypical pobj_of_examples ' +o,2540,'24 Factor Model Decomposition Factored translation models extend the phrase-based model by integrating word level factors into the decoding process ',Koehn,'num_models_24 nn_models_Factor nn_models_Model nn_models_Decomposition nn_models_Factored nn_models_translation nsubj__models xcomp__extend det_model_the amod_model_phrase-based dobj_extend_model prep_extend_by pcomp_by_integrating nn_factors_word nn_factors_level dobj_integrating_factors prep_integrating_into det_process_the amod_process_decoding pobj_into_process ' +o,2541,'Given that semantically similar words can be identified automatically on the basis of distributional properties and linguistic cues , identifying the semantic orientation of words would allow a system to further refine the retrieved semantic similarity relationships , extracting antonyms ',Brown,'complm_identified_that amod_words_semantically amod_words_similar nsubjpass_identified_words aux_identified_can auxpass_identified_be ccomp_Given_identified advmod_identified_automatically prep_identified_on det_basis_the pobj_on_basis prep_basis_of amod_properties_distributional pobj_of_properties cc_properties_and amod__linguistic nn__cues conj_properties_ xcomp_identified_identifying det_orientation_the amod_orientation_semantic dobj_identifying_orientation prep_orientation_of pobj_of_words aux_allow_would dep_Given_allow det_system_a dobj_allow_system aux_refine_to advmod_refine_further infmod_system_refine det_similarity_the amod_similarity_retrieved amod_similarity_semantic dobj_refine_similarity nsubj_Given_relationships partmod_relationships_extracting dobj_extracting_antonyms ' +o,2542,'The sequential classi cation approach can handle many correlated features , as demonstrated in work on maximum-entropy and a variety of other linear classi ers , including winnow , AdaBoost , and support-vector machines ',Ratnaparkhi,'det_approach_The amod_approach_sequential nn_approach_classi nn_approach_cation nsubj_handle_approach aux_handle_can advmod_correlated_many amod_features_correlated dobj_handle_features mark_demonstrated_as dep_features_demonstrated prep_demonstrated_in pobj_in_work prep_demonstrated_on amod__maximum-entropy pobj_on_ cc__and det_variety_a conj__variety prep_variety_of amod_ers_other amod_ers_linear nn_ers_classi pobj_of_ers prep_demonstrated_including nn__winnow pobj_including_ nn__AdaBoost conj__ cc__and amod__support-vector nn__machines conj__ ' +o,2543,'lscript1-regularized log-linear models -LRB- lscript1-LLMs -RRB- , on the other hand , provide sparse solutions , in which weights of irrelevant features are exactly zero , by assumingaLaplacianpriorontheweight ',Gao,'amod_models_lscript1-regularized amod_models_log-linear nsubj_provide_models appos_models_lscript1-LLMs prep_models_on det_hand_the amod_hand_other pobj_on_hand amod_solutions_sparse dobj_provide_solutions rel_zero_in pobj_in_which nsubj_zero_weights prep_weights_of amod_features_irrelevant pobj_of_features cop_zero_are advmod_zero_exactly rcmod_solutions_zero prep_zero_by pobj_by_assumingaLaplacianpriorontheweight ' +p,2544,' work is perhaps one of the most notable examples of unsupervised polarity classification ',Turney,'amod_work_ nsubj_one_work cop_one_is advmod_one_perhaps prep_one_of det_examples_the advmod_notable_most amod_examples_notable pobj_of_examples prep_examples_of amod_classification_unsupervised amod_classification_polarity pobj_of_classification ' +o,2545,'Equation -LRB- 2 -RRB- is rewritten as : -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- 2211 21 ce colecolcolcolcol rrpcepcep crpcepcepcep = = -LRB- 3 -RRB- It is equal to a word translation model if we take the relation type in the collocations as an element like a word , which is similar to Model 1 in ',Brown,'dep_colecolcolcolcol_Equation dobj_Equation_2 rcmod_2_is prep_Equation_rewritten dep_rewritten_as nn_ce_2211 num_ce_21 nsubj_colecolcolcolcol_ce nn_crpcepcepcep_rrpcepcep nsubj_=_crpcepcepcep dep_=_= ccomp_colecolcolcolcol_= dep_equal_3 nsubj_equal_It cop_equal_is dep_colecolcolcolcol_equal prep_equal_to det_model_a nn_model_word nn_model_translation pobj_to_model mark_take_if nsubj_take_we advcl_equal_take det_type_the nn_type_relation dobj_take_type prep_type_in det_collocations_the pobj_in_collocations prep_take_as det_element_an pobj_as_element prep_element_like det_word_a pobj_like_word nsubj_similar_which cop_similar_is rcmod_word_similar prep_similar_to pobj_to_Model num_Model_1 prep_similar_in ' +o,2546,'One heuristic approach is to adapt the self-training algorithm to our model ',Yarowsky,'num_approach_One amod_approach_heuristic nsubj_is_approach aux_adapt_to xcomp_is_adapt det__the amod__self-training nn__algorithm dobj_adapt_ prep_adapt_to poss_model_our pobj_to_model ' +o,2547,'This is referred to as an IOB representation ',Ramshaw,'nsubjpass_referred_This auxpass_referred_is xcomp_referred_to prep_referred_as det_representation_an nn_representation_IOB pobj_as_representation ' +n,2548,'With all but two formats IBI-IG achieves better FZ = l rates than the best published result in ',Ramshaw,'prep_achieves_With num_formats_all cc_all_but conj_all_two pobj_With_formats nsubj_achieves_IBI-IG amod_FZ_better nsubj_rates_FZ dep_rates_= amod_rates_l ccomp_achieves_rates prep_achieves_than det_result_the amod_result_best amod_result_published pobj_than_result prep_result_in ' +o,2549,'We participated in the multilingual track of the CoNLL 2007 shared task , and evaluated the system on data sets of 10 languages ',Marcus,'nsubj_participated_We prep_participated_in det_track_the amod_track_multilingual pobj_in_track prep_track_of det__the nn__CoNLL num__2007 amod__shared nn__task pobj_of_ cc_participated_and conj_participated_evaluated det_system_the dobj_evaluated_system prep_evaluated_on nn_sets_data pobj_on_sets prep_sets_of num_languages_10 pobj_of_languages ' +p,2550,'Far from full syntactic complexity , we suggest to go back to the simpler alignment methods first described by ',Brown,'advmod_suggest_Far dep_Far_from amod_complexity_full amod_complexity_syntactic pobj_from_complexity nsubj_suggest_we aux_go_to xcomp_suggest_go prt_go_back prep_go_to det_methods_the amod_methods_simpler amod_methods_alignment pobj_to_methods advmod_described_first partmod_methods_described prep_described_by ' +o,2551,'By associating natural language with concepts as they are entered into a knowledge A Model Of Semantic Analysis All of the following discussion is based on a model of semantic analysis similar to that proposed in ',Hobbs,'prep_entered_By pcomp_By_associating amod_language_natural dobj_associating_language prep_associating_with pobj_with_concepts prep_concepts_as pobj_as_they auxpass_entered_are prep_entered_into det_knowledge_a pobj_into_knowledge det_Model_A nsubjpass_entered_Model prep_based_Of nn_Analysis_Semantic pobj_Of_Analysis nsubjpass_based_All prep_All_of det_discussion_the amod_discussion_following pobj_of_discussion auxpass_based_is rcmod_Model_based prep_based_on det_model_a pobj_on_model prep_Model_of amod_analysis_semantic pobj_of_analysis amod_analysis_similar prep_similar_to pobj_to_that partmod_that_proposed prep_proposed_in ' +o,2552,'Table 3 compares precision , recall , and F scores for our system with CoNLL-2001 results training on sections 15-18 of the Penn Treebank and testing on section 21 ',Marcus,'nsubj_compares_Table num_Table_3 dobj_compares_precision conj_precision_recall cc_precision_and nn_scores_F conj_precision_scores prep_scores_for poss_system_our pobj_for_system prep_compares_with amod_results_CoNLL-2001 pobj_with_results dep_compares_training prep_training_on pobj_on_sections num_sections_15-18 prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank cc_training_and conj_training_testing prep_testing_on pobj_on_section num_section_21 ' +o,2553,'Comparatively , propose to use the N-gram Overlap metric to capture similarities between sentences and automatically create paraphrase corpora ',Barzilay,'advmod_propose_Comparatively nsubj_propose_ aux_use_to xcomp_propose_use det_metric_the nn_metric_N-gram nn_metric_Overlap nsubj_capture_metric aux_capture_to xcomp_use_capture dobj_capture_similarities prep_similarities_between pobj_between_sentences cc_capture_and advmod_create_automatically conj_capture_create amod_corpora_paraphrase dobj_create_corpora ' +p,2554,'The latent-annotation model is one of the most effective un-lexicalized models ',Matsuzaki,'det__The amod__latent-annotation nn__model nsubj_one_ cop_one_is prep_one_of det_models_the advmod_effective_most amod_models_effective amod_models_un-lexicalized pobj_of_models ' +o,2555,'We used the MXPOST tagger for POS annotation ',Ratnaparkhi,'nsubj_used_We det__the amod__MXPOST nn__tagger dobj_used_ prep_used_for dep_annotation_POS pobj_for_annotation ' +o,2556,'Techniques for weakening the independence assumptions made by the IBM models 1 and 2 have been proposed in recent work ',Brown,'nsubj_made_Techniques prep_Techniques_for pcomp_for_weakening det_assumptions_the nn_assumptions_independence dobj_weakening_assumptions prep_made_by det_models_the nn_models_IBM pobj_by_models nsubjpass_proposed_1 cc_1_and conj_1_2 aux_proposed_have auxpass_proposed_been dep_made_proposed prep_proposed_in amod_work_recent pobj_in_work ' +o,2557,'A first family of libraries was based on a word alignment A , produced using the Refined method described in -LRB- combination of two IBM-Viterbi alignments -RRB- : we call these the A libraries ',Och,'det_family_A amod_family_first nsubjpass_based_family prep_family_of pobj_of_libraries auxpass_based_was prep_based_on det_A_a nn_A_word nn_A_alignment pobj_on_A partmod_A_produced xcomp_produced_using det_method_the nn_method_Refined dobj_using_method partmod_method_described prep_described_in pobj_in_ dep__combination prep_combination_of num_alignments_two amod_alignments_IBM-Viterbi pobj_of_alignments nsubj_call_we parataxis_based_call nsubj_libraries_these det_libraries_the dep_libraries_A xcomp_call_libraries ' +p,2558,'Online votedperceptrons have been reported to work well in a number of NLP tasks ',Collins,'amod_votedperceptrons_Online nsubjpass_reported_votedperceptrons aux_reported_have auxpass_reported_been aux_work_to xcomp_reported_work advmod_work_well prep_work_in det_number_a pobj_in_number prep_number_of nn_tasks_NLP pobj_of_tasks ' +o,2559,'D , Noun classification from predicate argument structures , in ',Hindle,'nn_classification_D nn_classification_ appos_classification_Noun prep_classification_from amod_structures_predicate nn_structures_argument pobj_from_structures prep_classification_in pobj_in_ ' +o,2560,'Unlike stochastic approaches to part-of-speech tagging , up to now the knowledge found in finite-state taggers has been handcrafted and was not automatically acquired ',Cutting,'prep_handcrafted_Unlike amod_approaches_stochastic pobj_Unlike_approaches prep_approaches_to amod__part-of-speech nn__tagging pobj_to_ advmod_knowledge_up dep_up_to pobj_to_now det_knowledge_the nsubjpass_handcrafted_knowledge partmod_knowledge_found prep_found_in amod_taggers_finite-state pobj_in_taggers aux_handcrafted_has auxpass_handcrafted_been cc_handcrafted_and auxpass_acquired_was neg_acquired_not advmod_acquired_automatically conj_handcrafted_acquired ' +o,2561,'Even if the idea of using Wikipedia links for disambiguation is not novel , it is applied for the first time to FrameNet lexical units , considering a frame as a sense definition ',Cucerzan,'advmod__Even dep__if det_idea_the nsubj__idea prep_idea_of pcomp_of_using nn_links_Wikipedia dobj_using_links prep_using_for pobj_for_disambiguation cop__is neg__not amod__novel advcl_applied_ nsubjpass_applied_it auxpass_applied_is prep_applied_for det_time_the amod_time_first pobj_for_time aux_FrameNet_to xcomp_applied_FrameNet amod_units_lexical dobj_FrameNet_units xcomp_FrameNet_considering det_frame_a dobj_considering_frame prep_considering_as det_definition_a nn_definition_sense pobj_as_definition ' +o,2562,'A ~ value of 08 or greater indicates a high level of reliability among raters , with values between 067 and 08 indicating only moderate agreement ',Carletta,'det_value_A amod_value_~ nsubj_indicates_value prep_value_of pobj_of_08 cc_08_or conj_08_greater det_level_a amod_level_high dobj_indicates_level prep_level_of pobj_of_reliability prep_reliability_among pobj_among_raters prep_indicates_with nsubj_indicating_values prep_values_between pobj_between_067 cc_067_and conj_067_08 pcomp_with_indicating advmod_agreement_only amod_agreement_moderate dobj_indicating_agreement ' +o,2563,'To accommodate multiple overlapping features on observations , some other approaches view the sequence labeling problem as a sequence of classification problems , including support vector machines -LRB- SVMs -RRB- and a variety of other classifiers ',Ratnaparkhi,'aux_accommodate_To dep_view_accommodate amod_features_multiple amod_features_overlapping dobj_accommodate_features prep_accommodate_on pobj_on_observations det_approaches_some amod_approaches_other nsubj_view_approaches det_problem_the nn_problem_sequence nn_problem_labeling dobj_view_problem prep_view_as det_sequence_a pobj_as_sequence prep_sequence_of nn_problems_classification pobj_of_problems prep_view_including nn_machines_support nn_machines_vector pobj_including_machines abbrev_machines_SVMs dep_machines_ cc_machines_and det_variety_a conj_machines_variety prep_variety_of amod_classifiers_other pobj_of_classifiers ' +o,2564,'Our POS tagger is essentially the maximum entropy tagger by retrained on the CTB-I data ',Ratnaparkhi,'poss_tagger_Our dep_tagger_POS nsubj_tagger_tagger cop_tagger_is advmod_tagger_essentially det_tagger_the amod_tagger_maximum nn_tagger_entropy prep_tagger_by pobj_by_ partmod__retrained prep_retrained_on det_data_the nn_data_CTB-I pobj_on_data ' +p,2565,'Because it is not feasible here to have humans judge the quality of many sets of translated data , we rely on an array of well known automatic evaluation measures to estimate translation quality : BLEU is the geometric mean of the n-gram precisions in the output with respect to a set of reference translations ',Papineni,'mark_feasible_Because nsubj_feasible_it cop_feasible_is neg_feasible_not advcl_rely_feasible advmod_feasible_here aux_have_to xcomp_feasible_have nsubj_judge_humans ccomp_have_judge det_quality_the dobj_judge_quality prep_quality_of amod_sets_many pobj_of_sets prep_sets_of amod_data_translated pobj_of_data nsubj_rely_we prep_rely_on det_array_an pobj_on_array prep_array_of advmod_known_well amod_measures_known amod_measures_automatic nn_measures_evaluation pobj_of_measures aux_estimate_to xcomp_rely_estimate nn_quality_translation dobj_estimate_quality nn__BLEU nsubj_mean_ cop_mean_is det_mean_the amod_mean_geometric parataxis_rely_mean prep_mean_of det_precisions_the amod_precisions_n-gram pobj_of_precisions prep_precisions_in det_output_the pobj_in_output prep_output_with pobj_with_respect prep_respect_to det_set_a pobj_to_set prep_set_of nn_translations_reference pobj_of_translations ' +o,2566,'Most of this work has so far focused either on post-processing to recover non-local dependencies from context-free parse trees , or on incorporating nonlocal dependency information in nonterminal categories in constituency representations or in the categories used to label arcs in dependency representations ',Cahill,'nsubj_focused_Most prep_Most_of det_work_this pobj_of_work aux_focused_has advmod_far_so advmod_focused_far preconj_on_either prep_focused_on amod__post-processing aux_recover_to xcomp_post-processing_recover amod_dependencies_non-local dobj_recover_dependencies prep_recover_from amod_trees_context-free nn_trees_parse pobj_from_trees pobj_on_ cc_on_or conj_on_on pcomp_on_incorporating amod_information_nonlocal amod_information_dependency dobj_incorporating_information prep_incorporating_in amod_categories_nonterminal pobj_in_categories prep_categories_in nn__constituency nn__representations pobj_in_ cc_on_or conj_on_in det_categories_the pobj_in_categories partmod_categories_used aux_label_to xcomp_used_label dobj_label_arcs prep_label_in nn__dependency nn__representations pobj_in_ ' +o,2567,'These tables were computed from a small fragment of the Canadian Hansards that has been used in a number of other studies : Church and Simard et al ',Brown,'det_tables_These auxpass_computed_were dep_tables_computed prep_computed_from det_fragment_a amod_fragment_small pobj_from_fragment prep_fragment_of det_Hansards_the amod_Hansards_Canadian pobj_of_Hansards nsubjpass_used_that aux_used_has auxpass_used_been rcmod_Hansards_used prep_used_in det_number_a pobj_in_number prep_number_of amod_studies_other pobj_of_studies nn__Church dep_tables_ cc__and nn__Simard cc_Simard_et conj_Simard_al conj__ ' +p,2568,'22 Unsupervised Parameter Estimation We can perform maximum likelihood estimation of the parameters of this model in a similar fashion to that of Model 4 , described thoroughly in ',Och,'num_Estimation_22 nn_Estimation_Unsupervised nn_Estimation_Parameter nsubj_described_Estimation nsubj_perform_We aux_perform_can rcmod_Estimation_perform amod_estimation_maximum nn_estimation_likelihood dobj_perform_estimation prep_estimation_of det_parameters_the pobj_of_parameters prep_parameters_of det_model_this pobj_of_model prep_model_in det_fashion_a amod_fashion_similar pobj_in_fashion prep_perform_to pobj_to_that prep_that_of nn__Model num__4 pobj_of_ advmod_described_thoroughly prep_described_in ' +o,2569,'We use 3500 sentences from CoNLL as the NER data and section 20-23 of the WSJ as the POS\\/chunk data -LRB- 8936 sentences -RRB- ',Ramshaw,'nsubj_use_We num_sentences_3500 dobj_use_sentences prep_use_from nn__CoNLL pobj_from_ prep__as det_data_the nn_data_NER pobj_as_data cc_data_and nn_20-23_section conj_data_20-23 prep_data_of det__the nn__WSJ pobj_of_ prep__as det_data_the nn_data_POS\\/chunk pobj_as_data num_sentences_8936 appos_data_sentences ' +o,2570,'Training Data Our source for syntactically annotated training data was the Penn Treebank ',Marcus,'nn_Data_Training poss_source_Our nsubj_Treebank_source prep_source_for advmod_annotated_syntactically amod_data_annotated nn_data_training pobj_for_data cop_Treebank_was det_Treebank_the nn_Treebank_Penn dep_Data_Treebank ' +o,2571,' describes a method -LRB- Latent Relational Analysis -RRB- that extracts subsequence patterns for noun pairs from a large corpus , using query expansion to increase the recall of the search and feature selection and dimensionality reduction to reduce the complexity of the feature space ',Turney,'advmod_describes_ det_method_a nsubj_describes_method nn_Analysis_Latent nn_Analysis_Relational appos_method_Analysis dep_subsequence_that nsubj_subsequence_extracts dep_method_subsequence dobj_subsequence_patterns prep_subsequence_for nn_pairs_noun pobj_for_pairs prep_subsequence_from det_corpus_a amod_corpus_large pobj_from_corpus xcomp_subsequence_using amod_expansion_query dobj_using_expansion aux_increase_to xcomp_using_increase det_recall_the dobj_increase_recall prep_recall_of det_selection_the nn_selection_search cc_search_and conj_search_feature pobj_of_selection cc_selection_and amod_reduction_dimensionality conj_selection_reduction aux_reduce_to xcomp_increase_reduce det_complexity_the dobj_reduce_complexity prep_complexity_of det_space_the nn_space_feature pobj_of_space ' +o,2572,'General purpose text annotations , such as part-of-speech tags and noun-phrase bracketing , are costly to obtain but have wide applicability and have been used successfully to develop statistical NLP systems ',Marcus,'nn_annotations_General nn_annotations_purpose nn_annotations_text nsubj_costly_annotations dep_as_such prep_annotations_as amod_tags_part-of-speech pobj_as_tags cc_tags_and amod_bracketing_noun-phrase conj_tags_bracketing cop_costly_are aux_obtain_to xcomp_costly_obtain cc_have_but ccomp_obtain_have amod_applicability_wide dobj_have_applicability cc_have_and aux_used_have auxpass_used_been conj_have_used advmod_used_successfully aux_develop_to xcomp_used_develop amod_systems_statistical nn_systems_NLP dobj_develop_systems advmod_develop_ ' +o,2573,'To compare different clustering algorithms , results with the standard method of -LRB- SRILMs ngram-class -RRB- are also reported ',Brown,'aux_compare_To dep_reported_compare amod_algorithms_different amod_algorithms_clustering dobj_compare_algorithms nsubjpass_reported_results prep_results_with det_method_the amod_method_standard pobj_with_method prep_method_of pobj_of_ nn_ngram-class_SRILMs abbrev__ngram-class auxpass_reported_are advmod_reported_also ' +o,2574,'In this vein , the CoNLL 2008 shared task sets the challenge of learning jointly both syntactic dependencies -LRB- extracted from the Penn Treebank -RRB- and semantic dependencies -LRB- extracted both from PropBank c2008 ',Marcus,'prep_sets_In det_vein_this pobj_In_vein det_task_the nn_task_CoNLL num_task_2008 amod_task_shared nsubj_sets_task det_challenge_the dobj_sets_challenge prep_challenge_of pcomp_of_learning advmod_learning_jointly det_dependencies_both amod_dependencies_syntactic dobj_learning_dependencies dep_dependencies_extracted prep_extracted_from det__the nn__Penn nn__Treebank pobj_from_ cc_dependencies_and amod_dependencies_semantic conj_dependencies_dependencies dep_dependencies_extracted dep_extracted_both prep_learning_from nn_c2008_PropBank nn_c2008_ pobj_from_c2008 ' +p,2575,'To overcome this problem , unsupervised learning methods using huge unlabeled data to boost the performance of rules learned by small labeled data have been proposed recently ',Yarowsky,'aux_overcome_To dep_unsupervised_overcome det_problem_this dobj_overcome_problem csubjpass_proposed_learning dobj_learning_methods partmod_methods_using amod_data_huge amod_data_unlabeled dobj_using_data aux_boost_to xcomp_using_boost det_performance_the dobj_boost_performance prep_performance_of pobj_of_rules partmod_rules_learned prep_learned_by amod_data_small amod_data_labeled pobj_by_data aux_proposed_have auxpass_proposed_been ccomp_unsupervised_proposed advmod_proposed_recently nn__ nn__ nn__ nsubj_unsupervised_ ' +o,2576,'Both Agichtein and Ganti and train a language model for each database column ',Canisius,'det__Both nn__Agichtein cc_Agichtein_and conj_Agichtein_Ganti cc__and amod_train_ conj__train det_model_a nn_model_language dep__model prep_model_for det_column_each nn_column_database pobj_for_column ' +p,2577,'Moreover , the deterministic dependency parser of Yamada and Matsumoto , when trained on the Penn Treebank , gives a dependency accuracy that is almost as good as that of and Charniak -LRB- 2000 -RRB- ',Collins,'advmod_gives_Moreover det_parser_the amod_parser_deterministic amod_parser_dependency nsubj_gives_parser prep_parser_of nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto pobj_of_ advmod_trained_when dep_parser_trained prep_trained_on det_Treebank_the nn_Treebank_Penn pobj_on_Treebank det_accuracy_a amod_accuracy_dependency dobj_gives_accuracy nsubj_good_that cop_good_is advmod_good_almost advmod_good_as rcmod_accuracy_good prep_good_as pobj_as_that prep_that_of pobj_of_ cc__and conj__Charniak dep_good_2000 ' +o,2578,'Related to this issue , we note that the head rules , which were nearly identical to those used in , have not been tuned at all to this task ',Collins,'dep_note_Related prep_Related_to det_issue_this pobj_to_issue nsubj_note_we complm_tuned_that det_rules_the nn_rules_head nsubjpass_tuned_rules nsubj_identical_which cop_identical_were advmod_identical_nearly rcmod_rules_identical prep_identical_to pobj_to_those partmod_those_used prep_used_in pobj_in_ aux_tuned_have neg_tuned_not auxpass_tuned_been ccomp_note_tuned prep_tuned_at pobj_at_all prep_all_to det_task_this pobj_to_task ' +o,2579,'The first is a baseline of sorts , our own version of the ` chunking as tagging \' approach introduced by Ramshaw and Marcus ',Ramshaw,'det_first_The nsubj_baseline_first cop_baseline_is det_baseline_a prep_baseline_of pobj_of_sorts poss_version_our amod_version_own appos_sorts_version prep_version_of det_approach_the dep_approach_chunking prep_chunking_as pobj_as_tagging pobj_of_approach partmod_approach_introduced prep_introduced_by pobj_by_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus ' +o,2580,'In most cases , supervised learning methods can perform well ',Pang,'prep_perform_In amod_cases_most pobj_In_cases amod_methods_supervised nn_methods_learning nsubj_perform_methods aux_perform_can advmod_perform_well ' +o,2581,'following our previous work ',Jiang,'poss_work_our amod_work_previous pobj_following_work ' +o,2582,'Then , we build a classier learned by training data , using a maximum entropy model and the features related to spelling variations in Table 3 ',Berger,'advmod_build_Then nsubj_build_we det_classier_a nsubj_learned_classier ccomp_build_learned prep_learned_by nn_data_training pobj_by_data xcomp_learned_using det__a amod__maximum amod__entropy nn__model dobj_using_ cc__and det_features_the conj__features partmod_features_related prep_related_to nn_variations_spelling pobj_to_variations prep_variations_in pobj_in_Table num_Table_3 ' +o,2583,'1 Introduction Sentiment analysis have been widely conducted in several domains such as movie reviews , product reviews , news and blog reviews ',Pang,'num_analysis_1 nn_analysis_Introduction nn_analysis_Sentiment nsubjpass_conducted_analysis aux_conducted_have auxpass_conducted_been advmod_conducted_widely prep_conducted_in amod_domains_several pobj_in_domains dep_as_such prep_domains_as nn_reviews_movie pobj_as_reviews nn_reviews_product conj_reviews_reviews conj_reviews_news cc_reviews_and amod_reviews_blog conj_reviews_reviews ' +o,2584,'1 Introduction Several approaches including statistical techniques , lexical techniques and hybrid techniques , have been pursued to design schemes for word alignment which aims at establishing links between words of a source language and a target language in a parallel corpus ',Brown,'num_approaches_1 nn_approaches_Introduction amod_approaches_Several nsubjpass_pursued_approaches prep_approaches_including amod__statistical nn__techniques nn__ dep__lexical nn__techniques pobj_including_ cc__and nn__hybrid nn__techniques conj__ aux_pursued_have auxpass_pursued_been aux_design_to xcomp_pursued_design dobj_design_schemes prep_schemes_for nn_alignment_word pobj_for_alignment nsubj_aims_which rcmod_schemes_aims prep_aims_at pcomp_at_establishing dobj_establishing_links prep_links_between pobj_between_words prep_words_of det_language_a nn_language_source pobj_of_language cc_words_and det_language_a nn_language_target conj_words_language prep_language_in det_corpus_a amod_corpus_parallel pobj_in_corpus ' +o,2585,'Under a phrase based translation model , this distinction is important and will be discussed in more detail ',Koehn,'prep_important_Under det_phrase_a pobj_Under_phrase partmod_phrase_based nn__translation nn__model dobj_based_ det_distinction_this nsubjpass_important_distinction cop_important_is cc_important_and aux_discussed_will auxpass_discussed_be conj_important_discussed prep_discussed_in amod_detail_more pobj_in_detail ' +o,2586,'32 Mapping Mapping the identified units -LRB- tokens or sequences -RRB- to their equivalents in the other language was achieved by training a new translation model -LRB- IBM 2 -RRB- using the EM algorithm as described in ',Brown,'nsubjpass_achieved_32 partmod_32_Mapping xcomp_Mapping_Mapping det_units_the amod_units_identified dobj_Mapping_units dep_units_tokens cc_tokens_or conj_tokens_sequences prep_Mapping_to poss_equivalents_their pobj_to_equivalents prep_Mapping_in det_language_the amod_language_other pobj_in_language auxpass_achieved_was prep_achieved_by pcomp_by_training det_model_a amod_model_new nn_model_translation nsubj_using_model appos_model_IBM num_IBM_2 dep_training_using det_algorithm_the nn_algorithm_EM nsubj_described_algorithm advmod_described_as xcomp_using_described prep_described_in ' +o,2587,'Feature-based methods use pre-defined feature sets to extract features to train classification models ',Jiang,'amod__Feature-based nn__methods nsubj_use_ amod_sets_pre-defined nn_sets_feature nsubj_extract_sets aux_extract_to xcomp_use_extract nsubj_train_features aux_train_to xcomp_extract_train nn_models_classification dobj_train_models ' +p,2588,'David McClosky , Eugene Charniak , and Mark Johnson Brown Laboratory for Linguistic Information Processing -LRB- BLLIP -RRB- Brown University Providence , RI 02912 -LCB- dmcc ec mj -RCB- @ csbrownedu Abstract Self-training has been shown capable of improving on state-of-the-art parser performance despite the conventional wisdom on the matter and several studies to the contrary ',McClosky,'nn_McClosky_David nn_Charniak_Eugene conj_McClosky_Charniak cc_McClosky_and nn_Laboratory_Mark nn_Laboratory_Johnson nn_Laboratory_Brown conj_McClosky_Laboratory dep_McClosky_for nn_Processing_Linguistic nn_Processing_Information pobj_for_Processing abbrev_Processing_BLLIP nn_RI_Brown nn_RI_University nn_RI_Providence dobj_csbrownedu_RI num_RI_02912 amod_mj_dmcc nn_mj_ec dep_RI_mj dep_RI_@ dep_Processing_csbrownedu nn_Self-training_Abstract nsubjpass_shown_Self-training aux_shown_has auxpass_shown_been ccomp_csbrownedu_shown acomp_shown_capable prep_capable_of pcomp_of_improving prep_improving_on amod__state-of-the-art nn__parser nn__performance pobj_on_ prep_improving_despite det_wisdom_the amod_wisdom_conventional pobj_despite_wisdom prep_wisdom_on det_matter_the pobj_on_matter cc_matter_and amod_studies_several conj_matter_studies prep_improving_to det__the nn__contrary pobj_to_ ' +o,2589,'Multiple translations of the same text , corresponding articles from multiple news sources , and bilingual corpus have been utilized ',Barzilay,'amod_translations_Multiple nsubj__translations prep_translations_of det_text_the amod_text_same pobj_of_text amod_articles_corresponding nsubj__articles prep_articles_from amod_sources_multiple nn_sources_news pobj_from_sources ccomp__ cc__and amod__bilingual nn__corpus nsubjpass_utilized_ aux_utilized_have auxpass_utilized_been conj__utilized ' +n,2590,'A maximum entropy approach has been applied to partof-speech tagging before , but the approach \'s ability to incorporate nonlocal and non-HMM-tagger-type evidence has not been fully explored ',Ratnaparkhi,'det_approach_A amod_approach_maximum nn_approach_entropy nsubjpass_applied_approach aux_applied_has auxpass_applied_been prep_applied_to amod_tagging_partof-speech pobj_to_tagging prep_applied_before pobj_before_ cc_applied_but det_approach_the poss_ability_approach possessive_approach_\'s nsubj_explored_ability aux_incorporate_to infmod_ability_incorporate amod_evidence_nonlocal cc_nonlocal_and conj_nonlocal_non-HMM-tagger-type dobj_incorporate_evidence aux_explored_has neg_explored_not dep_explored_been advmod_explored_fully conj_applied_explored ' +o,2591,'In contrast , the C&C tagger , which is based on that of , utilizes a wide range of features and a larger contextual window including the previous two tags and the two previous and two following words ',Ratnaparkhi,'prep_utilizes_In pobj_In_contrast det_tagger_the nn_tagger_C&C nsubj_utilizes_tagger nsubjpass_based_which auxpass_based_is rcmod_tagger_based prep_based_on pobj_on_that prep_that_of pobj_of_ det_range_a amod_range_wide dobj_utilizes_range prep_range_of pobj_of_features cc_features_and det_window_a amod_window_larger amod_window_contextual conj_features_window prep_window_including det_tags_the amod_tags_previous num_tags_two pobj_including_tags cc_tags_and dep_two_the conj_tags_two amod_two_previous cc_two_and num_words_two amod_words_following conj_two_words ' +o,2592,'Subjective phrases are used by and others in order to classify reviews or sentences as positive or negative ',Turney,'amod_phrases_Subjective nsubjpass_used_phrases auxpass_used_are prep_used_by pobj_by_ cc__and conj__others mark_classify_in dep_classify_order aux_classify_to purpcl_used_classify nsubj_positive_reviews cc_reviews_or conj_reviews_sentences advmod_positive_as xcomp_classify_positive cc_positive_or conj_positive_negative ' +o,2593,'42 String-Based Evaluation We evaluate the output of our generation system against the raw strings of Section 23 using the Simple String Accuracy and BLEU evaluation metrics ',Papineni,'num_Evaluation_42 amod_Evaluation_String-Based nsubj_evaluate_We dep_Evaluation_evaluate det_output_the dobj_evaluate_output prep_output_of poss_system_our nn_system_generation pobj_of_system prep_evaluate_against det_strings_the amod_strings_raw nsubj_using_strings prep_strings_of pobj_of_Section num_Section_23 pcomp_against_using det_Accuracy_the nn_Accuracy_Simple nn_Accuracy_String dobj_using_Accuracy cc_Accuracy_and nn_metrics_BLEU nn_metrics_ nn_metrics_evaluation conj_Accuracy_metrics ' +n,2594,'If we consider these probabilities as a vector , the similarities of two English words can be obtained by computing the dot product of their corresponding vectors2 The formula is described below : similarity -LRB- ei , ej -RRB- = Nsummationdisplay k = 1 p -LRB- ei fk -RRB- p -LRB- ej fk -RRB- -LRB- 3 -RRB- Paraphrasing methods based on monolingual parallel corpora such as can also be used to compute the similarity ratio of two words , but they dont have as rich training resources as the bilingual methods do ',Barzilay,'mark_consider_If nsubj_consider_we advcl_obtained_consider det_probabilities_these dobj_consider_probabilities prep_consider_as det_vector_a pobj_as_vector det_similarities_the nsubjpass_obtained_similarities prep_similarities_of num_words_two amod_words_English pobj_of_words aux_obtained_can auxpass_obtained_be prep_obtained_by pcomp_by_computing det_product_the amod_product_dot dobj_computing_product prep_product_of poss_vectors2_their amod_vectors2_corresponding pobj_of_vectors2 det_formula_The nsubjpass_described_formula auxpass_described_is rcmod_product_described advmod_described_below dobj_obtained_similarity appos_similarity_ei dep_ei_ej dep_similarity_= nn_k_Nsummationdisplay iobj_=_k number_1_= num_p_1 nn_p_p nn_fk_ei appos_p_fk dobj_=_p nn_fk_ej appos_p_fk dep_used_3 nn_methods_Paraphrasing nsubjpass_used_methods prep_methods_based dep_based_on amod_corpora_monolingual amod_corpora_parallel pobj_on_corpora dep_as_such prep_corpora_as pobj_as_ aux_used_can advmod_used_also auxpass_used_be dep_obtained_used aux_compute_to purpcl_used_compute det_ratio_the nn_ratio_similarity dobj_compute_ratio prep_ratio_of num_words_two pobj_of_words cc_used_but nsubj_dont_they conj_used_dont ccomp_dont_have advmod_rich_as amod_resources_rich amod_resources_training dobj_have_resources mark_do_as det_methods_the amod_methods_bilingual nsubj_do_methods advcl_have_do ' +o,2595,'Dagan , Church , and Gale expanded on this idea by replacing word alignment parameters , which were based on absolute word positions in aligned segments , with a much smaller set of relative offset parameters ',Brown,'nn_Gale_Dagan conj_Gale_Church cc_Gale_and dep_expanded_Gale advmod_expanded_ prep_expanded_on det_idea_this pobj_on_idea prep_expanded_by pcomp_by_replacing amod_parameters_ nn_parameters_word nn_parameters_alignment dobj_replacing_parameters nsubjpass_based_which auxpass_based_were rcmod_parameters_based prep_based_on amod_positions_absolute nn_positions_word pobj_on_positions prep_based_in amod_segments_aligned pobj_in_segments prep_based_with det_set_a advmod_smaller_much amod_set_smaller pobj_with_set prep_set_of amod_parameters_relative amod_parameters_offset pobj_of_parameters ' +o,2596,'Denote the global feature vector for segmented sentence y with (y) Rd, where d is the total number of features in the model; then Score(y) is computed by the dot product of vector (y) and a parameter vector Rd, where i is the weight for the ith feature: Score(y) = (y) 841 Inputs: training examples (xi,yi) Initialization: set = 0 Algorithm: for t = 1T, i = 1N calculate zi = argmaxyGEN(xi) (y) if zi negationslash= yi = + (yi)(zi) Outputs: Figure 1: the perceptron learning algorithm, adapted from Collins (2002) The perceptron training algorithm is used to determine the weight values .',Collins,'' +o,2597,'Thus , an orthogonal line of research can involve inducing classes for words which are more general than single categories , ie , something akin to ambiguity classes ',Goldberg,'advmod_involve_Thus det_line_an amod_line_orthogonal nsubj_involve_line prep_line_of pobj_of_research aux_involve_can amod_classes_inducing dobj_involve_classes prep_involve_for pobj_for_words nsubj_general_which cop_general_are advmod_general_more rcmod_words_general prep_general_than amod_categories_single pobj_than_categories appos_categories_ie nsubj_akin_something xcomp_general_akin prep_akin_to amod_classes_ambiguity pobj_to_classes ' +o,2598,'These tools are important in that the strongest collocational associations often represent different word senses , and thus ` they provide a powerful set of suggestions to the lexicographer for what needs to be accounted for in choosing a set of semantic tags \' ',Church,'det_tools_These nsubj_important_tools cop_important_are prep_important_in complm_represent_that det_associations_the amod_associations_strongest amod_associations_collocational nsubj_represent_associations advmod_represent_often pcomp_in_represent amod_senses_different nn_senses_word dobj_represent_senses cc_important_and advmod_important_thus nsubj_provide_they ccomp_important_provide det_set_a amod_set_powerful dobj_provide_set prep_set_of pobj_of_suggestions prep_provide_to det_lexicographer_the pobj_to_lexicographer prep_lexicographer_for nsubj_needs_what pcomp_for_needs aux_accounted_to auxpass_accounted_be xcomp_needs_accounted prep_accounted_for dep_for_in pcomp_in_choosing det_set_a dobj_choosing_set prep_set_of amod_tags_semantic pobj_of_tags ' +o,2599,'We see no good reason , however , why such text spans should necessarily be sentences , since the majority of tagging paradigms do not attempt to parse an entire sentence and operate only in the local window of two to three tokens ',Ratnaparkhi,'nsubj_see_We det_reason_no amod_reason_good dobj_see_reason advmod_see_however advmod_sentences_why amod_spans_such nn_spans_text nsubj_sentences_spans aux_sentences_should advmod_sentences_necessarily cop_sentences_be advcl_see_sentences prep_sentences_since det_majority_the pobj_since_majority prep_majority_of amod_paradigms_tagging pobj_of_paradigms nsubj_attempt_ aux_attempt_do neg_attempt_not dep_see_attempt aux_parse_to xcomp_attempt_parse det_sentence_an amod_sentence_entire dobj_parse_sentence cc_parse_and conj_parse_operate advmod_operate_only prep_operate_in det_window_the amod_window_local pobj_in_window prep_window_of number_three_two dep_three_to num_tokens_three pobj_of_tokens ' +o,2600,'The clusters were found automatically by attempting to minimize perplexity ',Brown,'det_clusters_The nsubjpass_found_clusters auxpass_found_were advmod_found_automatically prep_found_by pcomp_by_attempting aux_minimize_to xcomp_attempting_minimize advmod_minimize_perplexity ' +o,2601,'Many methods have been proposed to deal with this problem , including supervised learning algorithms , semi-supervised learning algorithms , and unsupervised learning algorithms ',Yarowsky,'amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been aux_deal_to xcomp_proposed_deal prep_deal_with det_problem_this pobj_with_problem prep_proposed_including amod_algorithms_supervised nn_algorithms_learning pobj_including_algorithms dep_algorithms_ conj__semi-supervised xcomp_semi-supervised_learning amod__algorithms dobj_learning_ cc__and conj__unsupervised xcomp_unsupervised_learning amod__algorithms dobj_learning_ ' +p,2602,'We conclude with some challenges that still remain in applying proactive learning for MT 2 Syntax Based Machine Translation In recent years , corpus based approaches to machine translation have become predominant , with Phrase Based Statistical Machine Translation -LRB- PBSMT -RRB- being the most actively progressing area ',Koehn,'nsubj_conclude_We prep_conclude_with det_challenges_some pobj_with_challenges nsubj_remain_that advmod_remain_still rcmod_challenges_remain prep_remain_in pcomp_in_applying dobj_applying_proactive partmod_proactive_learning prep_learning_for nn_Translation_MT num_Translation_2 nn_Translation_Syntax nn_Translation_Based nn_Translation_Machine pobj_for_Translation prep_predominant_In amod_years_recent pobj_In_years nsubj_predominant_corpus partmod_corpus_based dobj_based_approaches prep_based_to nn_translation_machine pobj_to_translation aux_predominant_have cop_predominant_become dep_conclude_predominant prep_predominant_with nn_Translation_Phrase nn_Translation_Based nn_Translation_Statistical nn_Translation_Machine pobj_with_Translation abbrev_Translation_PBSMT partmod_Translation_ cop_area_being det_area_the advmod_progressing_most advmod_progressing_actively amod_area_progressing xcomp__area ' +o,2603,'For each language pair , we use two development sets : one for Minimum Error Rate Training , and the other for tuning the scale factor for MBR decoding ',Och,'prep_use_For det_pair_each nn_pair_language pobj_For_pair nsubj_use_we num_development_two dobj_use_development dep_development_sets dep_development_one prep_one_for nn__Minimum nn__Error nn__Rate nn__Training pobj_for_ cc_development_and det_decoding_the amod_decoding_other prep_other_for pcomp_for_tuning det_factor_the nn_factor_scale dobj_tuning_factor prep_tuning_for pobj_for_MBR conj_development_decoding ' +o,2604,'The factored translation model combines features in a log-linear fashion ',Och,'det_model_The amod_model_factored nn_model_translation nsubj_combines_model dobj_combines_features prep_features_in det_fashion_a amod_fashion_log-linear pobj_in_fashion ' +o,2605,'The Maximum Entropy Markov Model used in POS-tagging is described in detail in and the LMR tagger here uses the same probability model ',Ratnaparkhi,'det_Model_The nn_Model_Maximum nn_Model_Entropy nn_Model_Markov nsubj_used_Model prep_used_in nsubjpass_described_POS-tagging auxpass_described_is pcomp_in_described prep_described_in pobj_in_detail prep_detail_in pobj_in_ cc_described_and det_tagger_the nn_tagger_LMR nsubj_uses_tagger advmod_uses_here conj_described_uses det_model_the amod_model_same nn_model_probability dobj_uses_model ' +o,2606,'3 Surface Realisation from f-Structures present a probabilistic surface generation model for LFG ',Cahill,'num_Realisation_3 nn_Realisation_Surface nsubj_present_Realisation prep_Realisation_from amod__f-Structures pobj_from_ det_model_a amod_model_probabilistic nn_model_surface nn_model_generation dobj_present_model prep_model_for nn__LFG pobj_for_ ' +o,2607,'We worked with an implementation of the log likelihood ratio -LRB- g-Score -RRB- as proposed by and two variants of the t-score , one considering all values -LRB- t-score -RRB- and one where only positive values -LRB- t-score + -RRB- are kept following the results of Curran and Moens ',Dunning,'nsubj_worked_We prep_worked_with det_implementation_an pobj_with_implementation prep_implementation_of det_ratio_the nn_ratio_log nn_ratio_likelihood pobj_of_ratio appos_ratio_g-Score advmod_proposed_as amod_ratio_proposed prep_proposed_by pobj_by_ cc_proposed_and num_variants_two conj_proposed_variants prep_variants_of det_t-score_the pobj_of_t-score appos_t-score_one partmod_one_considering det_values_all dobj_considering_values appos_values_t-score cc_variants_and conj_variants_one advmod_kept_where advmod_values_only amod_values_positive nsubjpass_kept_values amod_+_t-score appos_values_+ auxpass_kept_are rcmod_one_kept prep_kept_following det_results_the pobj_following_results prep_results_of nn__Curran cc_Curran_and conj_Curran_Moens pobj_of_ ' +o,2608,'See for details ',Yarowsky,'dobj_See_ prep__for pobj_for_details ' +o,2609,'To avoid this problem , we sample from a space of probable alignments , as is done in IBM models 3 and above , and weight counts based on the likelihood of each alignment sampled under the current probability model ',Brown,'aux_problem_To dep_problem_avoid det_problem_this nsubj_sample_we parataxis_problem_sample prep_sample_from det_space_a pobj_from_space prep_space_of amod_alignments_probable pobj_of_alignments mark_done_as auxpass_done_is advcl_problem_done prep_done_in nn_models_IBM pobj_in_models dep_models_3 dep_problem_and dep_problem_above pobj_above_ cc__and nn_counts_weight conj__counts partmod__based prep_based_on det_likelihood_the pobj_on_likelihood prep_likelihood_of det_alignment_each pobj_of_alignment partmod_alignment_sampled prep_sampled_under det_model_the amod_model_current nn_model_probability pobj_under_model ' +o,2610,'On the British National Corpus -LRB- BNC -RRB- , using similarity method , we retrieve the following neighbors for the first and second sense , respectively : 1 ',Lin,'prep_retrieve_On det_Corpus_the nn_Corpus_British nn_Corpus_National pobj_On_Corpus abbrev_Corpus_BNC partmod_Corpus_using amod_method_ nn_method_similarity dobj_using_method nsubj_retrieve_we det_neighbors_the amod_neighbors_following dobj_retrieve_neighbors prep_retrieve_for det_sense_the amod_sense_first cc_first_and conj_first_second pobj_for_sense advmod_sense_respectively dobj_retrieve_1 ' +o,2611,'We use Viterbi training but neighborhood estimation or pegging could also be used ',Brown,'nsubj_use_We nn__Viterbi nn__training dobj_use_ cc_use_but nn__neighborhood nn__estimation nsubjpass_used_ cc__or nn__pegging conj__ aux_used_could advmod_used_also auxpass_used_be conj_use_used ' +o,2612,' -RRB- , and distributional methods -LRB- eg , et al ',Bergsma,'nn_distributional_and dep__distributional nn_eg_methods dep_distributional_eg dobj__ dep__et advmod__al ' +o,2613,' , and et al ',Dredze,'cc__and conj__ nn_al_et dobj__al ' +o,2614,'The starting point is the log likelihood ratio ',Dunning,'det_point_The amod_point_starting nsubj_ratio_point cop_ratio_is det_ratio_the nn_ratio_log nn_ratio_likelihood ' +o,2615,'Some o1 \' l ; his research has treated the sentenees as unstructured word sequences to be aligned ; this work has primarily involved the acquisition of bilingual lexical correspondences , although there has also been a , n attempt to create a full MT system based on such trcat , ment ',Brown,'det_o1_Some poss_l_o1 poss_research_his nsubj_treated_research aux_treated_has dep_l_treated det_sentenees_the dobj_treated_sentenees prep_treated_as amod_sequences_unstructured nn_sequences_word pobj_as_sequences aux_aligned_to auxpass_aligned_be xcomp_treated_aligned det_work_this nsubj_involved_work aux_involved_has advmod_involved_primarily parataxis_treated_involved det_acquisition_the dobj_involved_acquisition prep_acquisition_of amod__bilingual amod__lexical nn__correspondences pobj_of_ mark_been_although expl_been_there aux_been_has advmod_been_also advcl_involved_been dep_attempt_a dep_attempt_n dep_l_attempt aux_create_to infmod_attempt_create det_system_a amod_system_full nn_system_MT dobj_create_system prep_attempt_based dep_based_on amod_trcat_such pobj_on_trcat appos_trcat_ment ' +o,2616,'There has been considerable use in the NLP community of both WordNet and LDOCE , but no one has merged the two in order to combine their strengths ',Brown,'expl_been_There aux_been_has amod_use_considerable dobj_been_use prep_use_in det_community_the nn_community_NLP pobj_in_community prep_community_of preconj__both nn__WordNet pobj_of_ cc__and nn__LDOCE conj__ cc_been_but advmod_merged_no dep_no_one aux_merged_has conj_been_merged dep_two_the dobj_merged_two prep_two_in pobj_in_order aux_combine_to xcomp_merged_combine poss_strengths_their dobj_combine_strengths ' +o,2617,'Classi er Training Set Precision Recall F-Measure Linear 10K pairs 0837 0774 0804 Maximum Entropy 10K pairs 0881 0851 0866 Maximum Entropy 450K pairs 0902 0944 0922 Table 4 : Performance of Alignment Classi er 32 Paraphrase Acquisition Much recent work on automatic paraphrasing has used relatively simple statistical techniques to identify text passages that contain the same information from parallel corpora ',Barzilay,'nn_pairs_Classi nn_pairs_er nn_pairs_Training nn_pairs_Set nn_pairs_Precision nn_pairs_Recall nn_pairs_F-Measure nn_pairs_Linear num_pairs_10K num_pairs_0837 number_0804_0774 num_pairs_0804 nn_pairs_Maximum nn_pairs_Entropy num_pairs_10K nsubj_used_pairs num_pairs_0881 number_0866_0851 num_pairs_0866 nn_pairs_Maximum nn_pairs_Entropy num_pairs_450K dep_pairs_pairs number_0944_0902 num_pairs_0944 num_Table_0922 dep_pairs_Table num_Table_4 dep_pairs_Performance prep_Performance_of nn_work_Alignment nn_work_Classi nn_work_er nn_work_32 nn_work_Paraphrase nn_work_Acquisition amod_work_Much amod_work_recent pobj_of_work prep_Performance_on amod__automatic nn__paraphrasing pobj_on_ aux_used_has dep_pairs_used advmod_simple_relatively amod_techniques_simple amod_techniques_statistical dobj_used_techniques aux_identify_to xcomp_used_identify nn_passages_text dobj_identify_passages nsubj_contain_that rcmod_passages_contain det_information_the amod_information_same dobj_contain_information prep_contain_from amod_corpora_parallel pobj_from_corpora ' +o,2618,'The window size may vary , used windows of size 2 and 5 ',Church,'det_size_The nn_size_window nsubj_vary_size aux_vary_may ccomp_used_vary advmod_used_ nsubj_used_windows prep_windows_of pobj_of_size num_size_2 cc_size_and conj_size_5 ' +o,2619,'We finally also include as alignment candidates those word pairs that are transliterations of each other to cover rare proper names , which is important for language pairs that dont share the same alphabet such as Arabic and English ',Hermjakob,'nsubj_include_We advmod_include_finally advmod_include_also prep_include_as amod_candidates_alignment pobj_as_candidates det_pairs_those nn_pairs_word dobj_include_pairs nsubj_transliterations_that cop_transliterations_are rcmod_pairs_transliterations prep_transliterations_of det__each amod__other aux_cover_to xcomp_other_cover amod_names_rare amod_names_proper dobj_cover_names pobj_of_ nsubj_important_which cop_important_is rcmod__important prep_important_for nn_pairs_language pobj_for_pairs det_dont_that nsubj_share_dont dep_include_share det_alphabet_the amod_alphabet_same dobj_share_alphabet dep_as_such prep_alphabet_as pobj_as_Arabic cc_Arabic_and conj_Arabic_English ' +n,2620,'32 Evaluation Metrics AER -LRB- Alignment Error Rate -RRB- is the most widely used metric of alignment quality , but requires gold-standard alignments labeled with sure\\/possible annotations to compute ; lacking such annotations , we can compute alignment fmeasure instead ',Och,'dep_Evaluation_32 amod__Evaluation nn__Metrics nn__AER nn_Rate_Alignment nn_Rate_Error appos__Rate nsubj_used_ cop_used_is det_used_the advmod_used_most advmod_used_widely amod_used_metric prep_used_of amod_quality_alignment pobj_of_quality cc_used_but conj_used_requires amod_alignments_gold-standard dobj_requires_alignments partmod_alignments_labeled prep_labeled_with amod_annotations_sure\\/possible pobj_with_annotations aux_compute_to xcomp_labeled_compute dep_compute_lacking amod_annotations_such dobj_lacking_annotations nsubj_compute_we aux_compute_can parataxis_used_compute amod_fmeasure_alignment dobj_compute_fmeasure advmod_compute_instead ' +o,2621,'The phrase-based approach developed for statistical machine translation is designed to overcome the restrictions on many-tomany mappings in word-based translation models ',Koehn,'det_approach_The amod_approach_phrase-based nsubjpass_designed_approach partmod_approach_developed prep_developed_for amod__statistical nn__machine nn__translation pobj_for_ auxpass_designed_is aux_overcome_to xcomp_designed_overcome det_restrictions_the dobj_overcome_restrictions prep_overcome_on amod_mappings_many-tomany pobj_on_mappings prep_mappings_in amod_models_word-based nn_models_translation pobj_in_models ' +o,2622,' used both supervised and unsupervised WSD for correct phonetizitation of words in speech synthesis ',Yarowsky,'nsubj_used_ det_WSD_both amod_WSD_supervised cc_supervised_and conj_supervised_unsupervised dobj_used_WSD prep_used_for amod_phonetizitation_correct pobj_for_phonetizitation prep_phonetizitation_of pobj_of_words prep_words_in nn_synthesis_speech pobj_in_synthesis ' +o,2623,' state that AER is derived from F-Measure ',Och,'nsubj_state_ complm_derived_that nsubjpass_derived_AER auxpass_derived_is ccomp_state_derived prep_derived_from pobj_from_F-Measure ' +p,2624,'A two-tier scheme where sentences are rst classi ed as subjective versus objective , and then applying the sentiment classi er on only the subjective sentences further improves performance ',Pang,'det_scheme_A amod_scheme_two-tier dep_scheme_ advmod_ed_where nsubj_ed_sentences cop_ed_are nn_ed_rst nn_ed_classi advcl__ed advmod_subjective_as amod_ed_subjective prep_ed_versus pobj_versus_objective cc__and advmod__then conj__applying det_sentiment_the dobj_applying_sentiment nn_er_classi nsubj_improves_er prep_er_on advmod_sentences_only det_sentences_the amod_sentences_subjective dep_further_sentences dep_on_further dep_scheme_improves dobj_improves_performance ' +o,2625,'Resources specifying the relations among lexical items such as WordNet and HowNet -LRB- among others -RRB- have inspired the work of many researchers in NLP ',Marcus,'dep_Resources_specifying det_relations_the dobj_specifying_relations prep_specifying_among amod_items_lexical pobj_among_items dep_as_such prep_items_as nn__WordNet pobj_as_ cc_specifying_and nn__HowNet nsubj_inspired_ dep_inspired_among pobj_among_others aux_inspired_have conj_specifying_inspired det_work_the dobj_inspired_work prep_work_of amod_researchers_many pobj_of_researchers prep_inspired_in nn__NLP pobj_in_ ' +o,2626,'Then , h -LRB- s -RRB- h -LRB- s -RRB- + Lmax , s S This epsilon1-admissible heuristic bounds our search error by Lmax3 3 Bitext Parsing In bitext parsing , one jointly infers a synchronous phrase structure tree over a sentence ws and its translation wt ',Wu,'advmod_bounds_Then nsubj_bounds_h dep_-LRB-_s nsubj__h dep_-LRB-_s nn_Lmax_+ dep_h_Lmax aux__s dep__S det__This amod__epsilon1-admissible amod__heuristic rcmod_h_ poss_error_our nn_error_search dobj_bounds_error prep_bounds_by nn_Parsing_Lmax3 num_Parsing_3 nn_Parsing_Bitext pobj_by_Parsing prep_infers_In amod_parsing_bitext pobj_In_parsing nsubj_infers_one advmod_infers_jointly dep_bounds_infers det_tree_a amod_tree_synchronous nn_tree_phrase nn_tree_structure dobj_infers_tree prep_tree_over det_ws_a nn_ws_sentence pobj_over_ws cc_ws_and poss_wt_its nn_wt_translation conj_ws_wt ' +n,2627,'Clustering algorithms have been previously shown to work fairly well for the classification of words into syntactic and semantic classes , but determining the optimum number of classes for a hierarchical cluster tree is an ongoing difficult problem , particularly without prior knowledge of the item classification ',Brown,'amod_algorithms_Clustering nsubjpass_shown_algorithms aux_shown_have auxpass_shown_been advmod_shown_previously aux_work_to xcomp_shown_work advmod_well_fairly advmod_work_well dep_well_for det_classification_the pobj_for_classification prep_classification_of pobj_of_words prep_work_into amod_classes_syntactic cc_syntactic_and conj_syntactic_semantic pobj_into_classes ccomp_work_ cc__but dep_number_determining det_number_the amod_number_optimum conj__number prep_number_of pobj_of_classes mark_problem_for det_tree_a amod_tree_hierarchical nn_tree_cluster nsubj_problem_tree cop_problem_is det_problem_an amod_problem_ongoing amod_problem_difficult advcl_number_problem advmod_problem_particularly prep_problem_without amod_knowledge_prior pobj_without_knowledge prep_knowledge_of det_classification_the nn_classification_item pobj_of_classification ' +o,2628,'Online discriminative training has already been studied by and Liang et al ',Tillmann,'amod_training_Online nn_training_discriminative nsubjpass_studied_training aux_studied_has advmod_studied_already auxpass_studied_been prep_studied_by pobj_by_ cc__and conj__Liang cc_Liang_et conj_Liang_al ' +p,2629,'There is also substantial work in the use of target-side syntax ',Shen,'expl_is_There advmod_is_also amod_work_substantial nsubj_is_work prep_work_in det_use_the pobj_in_use prep_use_of amod_syntax_target-side pobj_of_syntax ' +o,2630,'41 Experimental Set-up We used two different corpora : PropBank -LRB- wwwcisupennedu\\/ace -RRB- along with PennTree bank 2 and FrameNet ',Marcus,'num_Set-up_41 nn_Set-up_Experimental nsubj_used_We dep_Set-up_used num_corpora_two amod_corpora_different dobj_used_corpora dep_along_PropBank appos_PropBank_wwwcisupennedu\\/ace advmod__along dep_along_with nn_bank_PennTree pobj_with_bank num__2 dep_corpora_ cc__and conj__FrameNet ' +o,2631,'1 perform the following maximization : eI1 = argmax eI1 fPr -LRB- eI1 -RRB- Pr -LRB- fJ1 jeI1 -RRB- g -LRB- 2 -RRB- This approach is referred to as source-channel approach to statistical MT Sometimes , it is also referred to as the fundamental equation of statistical MT ',Brown,'dep_perform_1 det_maximization_the amod_maximization_following dobj_perform_maximization nsubj_=_eI1 parataxis_perform_= nn_fPr_argmax nn_fPr_eI1 dobj_=_fPr appos_fPr_eI1 nsubj_g_Pr amod_jeI1_fJ1 appos_Pr_jeI1 rcmod_fPr_g dep_referred_2 det_approach_This nsubjpass_referred_approach auxpass_referred_is dep_g_referred aux_as_to xcomp_referred_as amod_approach_source-channel pobj_as_approach prep_approach_to amod_MT_statistical pobj_to_MT advmod_as_Sometimes nsubjpass_referred_it auxpass_referred_is advmod_referred_also dep_referred_referred prep_referred_to advmod_equation_as det_equation_the amod_equation_fundamental pobj_to_equation prep_equation_of amod_MT_statistical pobj_of_MT ' +o,2632,' develop a prototype-driven approach , which requires just a few prototype examples for each POS tag and exploits these labeled words to constrain the labels of their distributionally similar words ',Haghighi,'nsubj_develop_ det_approach_a amod_approach_prototype-driven dobj_develop_approach nsubj_requires_which rcmod_approach_requires advmod_examples_just det_examples_a amod_examples_few nn_examples_prototype dobj_requires_examples prep_requires_for det_tag_each dep_tag_POS pobj_for_tag cc_tag_and conj_tag_exploits nsubj_labeled_these dep_develop_labeled dobj_labeled_words aux_constrain_to xcomp_labeled_constrain det_labels_the dobj_constrain_labels prep_labels_of poss_words_their advmod_similar_distributionally amod_words_similar pobj_of_words ' +o,2633,'PairClass is most similar to the algorithm of , but it differs in the following ways : PairClass does not use a lexicon to find synonyms for the input word pairs ',Turney,'nsubj_similar_PairClass cop_similar_is advmod_similar_most prep_similar_to det_algorithm_the pobj_to_algorithm prep_algorithm_of pobj_of_ cc_similar_but nsubj_differs_it conj_similar_differs prep_differs_in det_ways_the amod_ways_following pobj_in_ways nsubj_use_PairClass aux_use_does neg_use_not parataxis_similar_use det_lexicon_a dobj_use_lexicon aux_find_to infmod_lexicon_find dobj_find_synonyms prep_synonyms_for det_pairs_the nn_pairs_input nn_pairs_word pobj_for_pairs ' +o,2634,'scored with lowercased , tokenized NIST BLEU , and exact match METEOR ',Papineni,'prep_scored_with pobj_with_lowercased amod_BLEU_tokenized nn_BLEU_NIST conj_lowercased_BLEU cc_lowercased_and amod_METEOR_exact nn_METEOR_match conj_lowercased_METEOR ' +o,2635,'This allows us to compute the conditional probability as follows : P -LRB- flh -RRB- YIia -LRB- \' -LRB- n ` l -RRB- z ~ -LRB- h -RRB- -LRB- 2 -RRB- ~ , i -LRB- 3 -RRB- I i The maximum entropy estimation technique guarantees that for every feature gi , the expected value of gi according to the ME model will equal the empirical expectation of gi in the training corpus ',Berger,'nsubj_allows_This nsubj_compute_us aux_compute_to xcomp_allows_compute det_probability_the amod_probability_conditional dobj_compute_probability mark_follows_as advcl_compute_follows acomp_follows_ nsubj_guarantees_P appos_P_flh dep_P_YIia nn_~_n nn_~_l nn_~_z dep_YIia_~ appos_~_h appos_~_2 dep_~_~ dep_~_i dep_I_3 dep_~_I prep_I_i det_technique_The amod_technique_maximum amod_technique_entropy nn_technique_estimation pobj_i_technique parataxis_allows_guarantees complm_equal_that prep_equal_for det_gi_every nn_gi_feature pobj_for_gi det_value_the amod_value_expected nsubj_equal_value prep_value_of pobj_of_gi prep_gi_according dep_according_to det_model_the nn_model_ME pobj_to_model aux_equal_will ccomp_guarantees_equal det_expectation_the amod_expectation_empirical dobj_equal_expectation prep_expectation_of pobj_of_gi prep_equal_in det_corpus_the nn_corpus_training pobj_in_corpus ' +o,2636,'It has been observed that words close to each other in the source language tend to remain close to each other in the translation ',Berger,'nsubjpass_observed_It aux_observed_has auxpass_observed_been complm_tend_that nsubj_tend_words amod_words_close prep_close_to det_other_each pobj_to_other prep_other_in det_language_the nn_language_source pobj_in_language ccomp_observed_tend aux_remain_to xcomp_tend_remain advmod_remain_close dep_close_to det_other_each pobj_to_other prep_remain_in det__the nn__translation pobj_in_ ' +o,2637,'More specifically , we use a class-based bigram model from : -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- 11 = iiiiii ccPcwPwwP -LRB- 3 -RRB- In Equation -LRB- 3 -RRB- , c i is the class of the word w i , which could be a syntactic class or a semantic class ',Brown,'advmod_use_specifically nsubj_use_we dep_More_use det_bigram_a amod_bigram_class-based dobj_ccPcwPwwP_bigram dep_bigram_model prep_model_from pobj_from_ num_iiiiii_11 amod_iiiiii_= nsubj_ccPcwPwwP_iiiiii ccomp_use_ccPcwPwwP dep_ccPcwPwwP_3 prep_class_In pobj_In_Equation appos_Equation_3 dep_class_c nsubj_class_i cop_class_is det_class_the dep_More_class prep_class_of pobj_of_the nn_the_word nn_the_w dep_the_i nsubj_class_which aux_class_could cop_class_be det_class_a amod_class_syntactic rcmod_the_class cc_class_or det_class_a amod_class_semantic conj_class_class ' +o,2638,'2 Background 21 Hybrid Logic Dependency Semantics Hybrid Logic Dependency Semantics is an ontologically promiscuous -LRB- , 1985 -RRB- framework for representing the propositional content -LRB- or meaning -RRB- of an expression as an ontologically richly sorted , relational structure ',Hobbs,'num__2 nn__Background num__21 nn__Hybrid nn__Logic nn__Dependency nn__Semantics nn__Hybrid nn__Logic nn__Dependency nn__Semantics nsubj_framework_ cop_framework_is det_framework_an advmod_promiscuous_ontologically amod_framework_promiscuous dep_promiscuous_ dep__1985 prep_framework_for pcomp_for_representing det_content_the amod_content_propositional dobj_representing_content dep_meaning_or dep_framework_meaning prep_framework_of det_expression_an pobj_of_expression prep_expression_as det_structure_an advmod_richly_ontologically amod_structure_richly nn_structure_sorted amod_structure_relational pobj_as_structure ' +o,2639,'The core technology of the proposed method , ie , the automatic evaluation of translations , was developed in research aiming at the efficient development of Machine Translation -LRB- MT -RRB- technology ',Papineni,'det_technology_The amod_technology_core nsubjpass_developed_technology prep_technology_of det_method_the amod_method_proposed pobj_of_method appos_technology_ie det_evaluation_the amod_evaluation_automatic appos_technology_evaluation prep_evaluation_of pobj_of_translations auxpass_developed_was prep_developed_in pobj_in_research partmod_research_aiming prep_aiming_at det_development_the amod_development_efficient pobj_at_development prep_development_of nn_technology_Machine nn_technology_Translation nn_technology_MT pobj_of_technology ' +o,2640,'The weights of feature functions are optimized to maximize the scoring measure ',Och,'det_weights_The nsubjpass_optimized_weights prep_weights_of nn_functions_feature pobj_of_functions auxpass_optimized_are aux_maximize_to xcomp_optimized_maximize det_measure_the amod_measure_scoring dobj_maximize_measure ' +o,2641,'The problem is due to the assumption of normality in naive frequency based statistics according to ',Dunning,'det_problem_The nsubj_due_problem cop_due_is prep_due_to det_assumption_the pobj_to_assumption prep_assumption_of advmod_naive_normality dep_naive_in amod_frequency_naive pobj_of_frequency partmod_frequency_based dobj_based_statistics prep_based_according dep_according_to ' +n,2642,'Although , there are various manual\\/automatic evaluation methods for these systems , eg , BLEU , these methods are basically incapable of dealing with an MTsystem and a w\\/p-MT-system at the same time , as they have different output forms ',Papineni,'mark_are_Although expl_are_there advcl_are_are amod_methods_various amod_methods_manual\\/automatic nn_methods_evaluation nsubj_are_methods prep_methods_for det__these nn__systems dep__eg dep__BLEU pobj_for_ det_methods_these nsubj_are_methods advmod_incapable_basically acomp_are_incapable prep_incapable_of pcomp_of_dealing prep_dealing_with det_MTsystem_an pobj_with_MTsystem cc_MTsystem_and det_w\\/p-MT-system_a conj_MTsystem_w\\/p-MT-system prep_dealing_at det_time_the amod_time_same pobj_at_time mark_have_as nsubj_have_they dep_time_have amod_forms_different nn_forms_output dobj_have_forms ' +o,2643,'(2003), Pang and Lee (2004, 2005).',Pang,'' +o,2644,'They generally perform less well on low-frequency words ',Weeds,'nsubj_perform_They advmod_perform_generally advmod_well_less advmod_perform_well prep_perform_on amod_words_low-frequency pobj_on_words ' +o,2645,'The detailed algorithm can be found in ',Collins,'det_algorithm_The amod_algorithm_detailed nsubjpass_found_algorithm aux_found_can auxpass_found_be prep_found_in ' +o,2646,'Most work in the area of unknown words and tagging deals with predicting part-of-speech information based on word endings and affixation information , as shown by work in , , , and ',Marcus,'nsubj_work_Most prep_work_in det_area_the pobj_in_area prep_area_of amod_words_unknown pobj_of_words cc_area_and amod_deals_tagging conj_area_deals prep_deals_with pcomp_with_predicting amod_information_part-of-speech dobj_predicting_information prep_predicting_based dep_based_on nn_endings_word pobj_on_endings cc_endings_and nn_information_affixation conj_endings_information mark_shown_as dep_deals_shown prep_shown_by pobj_by_work prep_work_in nn__ conj__ conj__ cc__and pobj_in_ ' +o,2647,'Once this is accomplished , a variant of Powells algorithm is used to find weights that optimize BLEU score over these hypotheses , compared to reference translations ',Papineni,'mark_accomplished_Once nsubjpass_accomplished_this auxpass_accomplished_is advcl_used_accomplished det_variant_a nsubjpass_used_variant prep_variant_of nn_algorithm_Powells pobj_of_algorithm auxpass_used_is aux_find_to xcomp_used_find dobj_find_weights complm_score_that amod_BLEU_optimize nsubj_score_BLEU ccomp_find_score dobj_score_ prep__over det_hypotheses_these pobj_over_hypotheses prep_score_compared dep_compared_to nn_translations_reference pobj_to_translations ' +o,2648,'Automatic methods for this often make use of lexicons of words tagged with positive and negative semantic orientation ',Turney,'aux_make_Automatic nsubj_make_methods prep_methods_for pobj_for_this advmod_make_often dobj_make_use prep_use_of pobj_of_lexicons prep_lexicons_of pobj_of_words partmod_words_tagged prep_tagged_with amod_orientation_positive cc_positive_and conj_positive_negative amod_orientation_semantic pobj_with_orientation ' +o,2649,'4 Semantic Class Induction from Wikipedia Wikipedia has recently been used as a knowledge source for various language processing tasks , including taxonomy construction , coreference resolution , and English NER -LRB- eg , Bunescu and Pasca , Cucerzan , and Torisawa , Watanabe et al ',Kazama,'num_Induction_4 nn_Induction_Semantic nn_Induction_Class nsubjpass_used_Induction prep_Induction_from nn_Wikipedia_Wikipedia pobj_from_Wikipedia aux_used_has advmod_used_recently auxpass_used_been prep_used_as det_source_a nn_source_knowledge pobj_as_source prep_source_for amod_tasks_various nn_tasks_language nn_tasks_processing pobj_for_tasks prep_source_including amod__taxonomy nn__construction pobj_including_ amod__coreference nn__resolution conj__ cc__and nn_NER_English conj__NER dep_NER_eg nn__Bunescu cc_Bunescu_and conj_Bunescu_Pasca conj_eg_ nn__Cucerzan conj_eg_ conj_eg_ cc_eg_and nn__Torisawa conj_eg_ conj__Watanabe cc__et conj__al ' +o,2650,'Citation texts have also been used to create summaries of single scientific articles in Qazvinian and Radev and and Zhai ',Mei,'nn_texts_Citation nsubjpass_used_texts aux_used_have advmod_used_also auxpass_used_been aux_create_to xcomp_used_create dobj_create_summaries prep_summaries_of amod_articles_single amod_articles_scientific pobj_of_articles prep_articles_in nn__Qazvinian cc_Qazvinian_and conj_Qazvinian_Radev pobj_in_ cc_articles_and conj_articles_ cc__and nn__Zhai conj__ ' +o,2651,'C c C, p(C]v,r) is just the probability of the disjunction of the concepts in C; that is, = Zp(clv, r) cEC In order to see how p(clv,r) relates to the input data, note that given a concept c, verb v and argument position r, a noun can be generated according to the distribution p(n[c, v, r), where p(nlc, v, r) = 1 nEsyn(c) Now we have a model for the input data: p(n, v, r) = p(v,r)p(niv,r) = p(v,r) p(clv, rlp(ntc, v,r) cecn(n) Note that for c cn(n), p(nlc, v, r) = O. The association norm (and similar measures such as the mutual information score) have been criticised (Dunning, 1993) because these scores can be greatly over-estimated when frequency counts are low.',Dunning,'' +p,2652,'Averaged perceptron , which has been successfully applied to several tagging and parsing reranking tasks , was employed for training rerank267 CLANG GEOQUERY P R F P R F SCISSOR 895 737 808 985 744 848 SCISSOR + 870 780 823 955 772 854 Table 2 : The performance of the baseline model SCISSOR + compared with SCISSOR -LRB- with the best result in bold -RRB- , where P = precision , R = recall , and F = F-measure ',Collins,'amod__Averaged nn__perceptron nsubjpass_employed_ nsubjpass_applied_which aux_applied_has auxpass_applied_been advmod_applied_successfully rcmod__applied prep_applied_to amod_tagging_several pobj_to_tagging cc_tagging_and nn__parsing nn__reranking nn__tasks conj_tagging_ auxpass_employed_was prep_employed_for amod_SCISSOR_training nn_SCISSOR_rerank267 nn_SCISSOR_CLANG nn_SCISSOR_GEOQUERY nn_SCISSOR_P nn_SCISSOR_R nn_SCISSOR_F nn_SCISSOR_P nn_SCISSOR_R nn_SCISSOR_F pobj_for_SCISSOR number_854_895 cc_854_737 conj_854_808 number_854_985 number_854_744 number_854_848 number_854_SCISSOR cc_854_+ conj_854_870 number_854_780 number_854_823 number_854_955 number_854_772 dep_employed_854 cc_854_Table conj_854_2 det_performance_The dep_854_performance prep_performance_of det_model_the nn_model_baseline pobj_of_model nn_+_SCISSOR dep_854_+ dep_854_compared dep_compared_with pobj_with_SCISSOR dep_854_with det_result_the amod_result_best pobj_with_result prep_result_in pobj_in_bold advmod_P_where dep_precision_P dep_precision_= dep_recall_precision nn_=_R nsubj_recall_= ccomp_854_recall cc_854_and nn_=_F dep_F-measure_= conj_854_F-measure ' +o,2653,'Our hierarchical system is Hiero , modified to construct rules from a small sample of occurrences of each source phrase in training as described by ',Lopez,'poss_system_Our amod_system_hierarchical nsubj__system cop__is nn__Hiero partmod__modified aux_construct_to xcomp_modified_construct dobj_construct_rules prep_construct_from det_sample_a amod_sample_small pobj_from_sample prep_sample_of pobj_of_occurrences prep_occurrences_of det_phrase_each nn_phrase_source pobj_of_phrase prep_construct_in pcomp_in_training advmod_described_as acomp_training_described prep_described_by ' +o,2654,'So far , pivot features on the word level were used , eg Does the bigram not buy occur in this document ? ',Blitzer,'dep_far_So advmod_used_far amod_features_pivot nsubjpass_used_features prep_features_on det_level_the nn_level_word pobj_on_level auxpass_used_were ccomp_Does_used dep_used_ nsubj_Does_eg det_bigram_the nsubj_buy_bigram neg_buy_not ccomp_Does_buy dep_buy_occur prep_occur_in det_document_this pobj_in_document ' +o,2655,'43 Experiments results Our evaluation metric is BLEU , which are to perform case-insensitive matching of n-grams up to n = 4 ',Papineni,'num_Experiments_43 nsubj_results_Experiments poss_metric_Our nn_metric_evaluation nsubj__metric cop__is nn__BLEU ccomp_results_ nsubj_are_which rcmod__are aux_perform_to xcomp_are_perform amod_matching_case-insensitive dobj_perform_matching prep_matching_of pobj_of_n-grams advmod_perform_up dep_up_to pcomp_to_n dep_n_= dobj_n_4 ' +o,2656,'For example , alignments can be used to learn translation lexicons , transfer rules , and classifiers to find safe sentence segmentation points ',Berger,'prep_used_For pobj_For_example nsubjpass_used_alignments aux_used_can auxpass_used_be aux_learn_to purpcl_used_learn nn_lexicons_translation dobj_learn_lexicons advmod__ nn_rules_transfer nsubj__rules dep_used_ cc__and conj__classifiers aux_find_to xcomp_classifiers_find amod_points_safe nn_points_sentence nn_points_segmentation dobj_find_points ' +o,2657,'We were given around 15K sentences of labeled text from the Wall Street Journal -LRB- WSJ -RRB- as well as 200K unlabeled sentences ',Marcus,'nsubjpass_given_We auxpass_given_were prep_given_around num_sentences_15K pobj_around_sentences prep_sentences_of amod_text_labeled pobj_of_text prep_given_from det_Wall_the pobj_from_Wall nn__Street nn__Journal abbrev__WSJ dep_Wall_ dep_well_as cc_Wall_well dep_well_as num_sentences_200K amod_sentences_unlabeled conj_Wall_sentences ' +o,2658,'Further details are in the original paper ',Brown,'amod_details_Further nsubj_are_details prep_are_in det_paper_the amod_paper_original pobj_in_paper ' +o,2659,'Pooling the sets to form two large CE and AE test sets , the AE system improvements are significant at a 95 \% level ; the CE systems are only equivalent ',Och,'dep_significant_Pooling det_sets_the nsubj_form_sets aux_form_to xcomp_Pooling_form num_CE_two amod_CE_large dobj_form_CE cc_CE_and nn_sets_AE nn_sets_test conj_CE_sets det_improvements_the nn_improvements_AE nn_improvements_system nsubj_significant_improvements cop_significant_are prep_significant_at det__a dep_\%_95 amod__\% nn__level pobj_at_ det_systems_the nn_systems_CE nsubj_equivalent_systems cop_equivalent_are advmod_equivalent_only parataxis_significant_equivalent ' +o,2660,'3 Related work Word collocation Various collocation metrics have been proposed , including mean and variance , the t-test , the chi-square test , pointwise mutual information -LRB- MI -RRB- , and binomial loglikelihood ratio test -LRB- BLRT -RRB- ',Smadja,'num_work_3 amod_work_Related nsubj_Word_work amod_metrics_collocation amod_metrics_Various nn_metrics_collocation nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been ccomp_Word_proposed prep_proposed_including nn__mean cc_mean_and conj_mean_variance pobj_including_ det__the amod__t-test conj__ det_test_the amod_test_chi-square conj__test amod__pointwise amod__mutual nn__information abbrev__MI appos_test_ cc__and amod_test_binomial amod_test_loglikelihood nn_test_ratio conj__test abbrev_test_BLRT advmod_proposed_ ' +p,2661,'Substantial improvements have been made to parse western language such as English , and many powerful models have been proposed ',Collins,'amod_improvements_Substantial nsubjpass_made_improvements aux_made_have auxpass_made_been aux_parse_to xcomp_made_parse amod_language_western dobj_parse_language dep_as_such prep_language_as pobj_as_English cc_made_and amod_models_many amod_models_powerful nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been conj_made_proposed ' +p,2662,'Maximum entropy can be used to improve IBM-style translation probabilities by using features , such as improvements to P -LRB- f e -RRB- in ',Berger,'amod_entropy_Maximum nsubjpass_used_entropy aux_used_can auxpass_used_be aux_improve_to purpcl_used_improve amod_probabilities_IBM-style nn_probabilities_translation dobj_improve_probabilities prep_improve_by pcomp_by_using dobj_using_features dep_as_such prep_features_as pobj_as_improvements prep_improvements_to pobj_to_P dep_-LRB-_f dep_f_e prep_used_in ' +p,2663,'22 Statistical Translation Lexicon We use a statistical translation lexicon known as IBM Model-1 in for both efficiency and simplicity ',Brown,'num_Lexicon_22 nn_Lexicon_Statistical nn_Lexicon_Translation nsubj_use_We dep_Lexicon_use det_lexicon_a amod_lexicon_statistical nn_lexicon_translation dobj_use_lexicon partmod_lexicon_known prep_known_as nn_Model-1_IBM pobj_as_Model-1 prep_known_in pobj_in_ prep__for preconj_efficiency_both pobj_for_efficiency cc_efficiency_and conj_efficiency_simplicity ' +o,2664,'Examples of the latter include providing suggestions from a machine labeler and using extremely cheap human labelers , eg with the Amazon Mechanical Turk ',Snow,'nsubj_include_Examples prep_Examples_of det_latter_the pobj_of_latter ccomp_include_providing dobj_providing_suggestions prep_providing_from det_labeler_a nn_labeler_machine pobj_from_labeler cc_providing_and conj_providing_using advmod_cheap_extremely amod_labelers_cheap amod_labelers_human dobj_using_labelers appos_labelers_eg prep_eg_with det_Turk_the nn_Turk_Amazon nn_Turk_Mechanical pobj_with_Turk ' +o,2665,'2 Related Work WSD approaches can be classified as (a) knowledge-based approaches, which make use of linguistic knowledge, manually coded or extracted from lexical resources (Agirre and Rigau, 1996; Lesk 1986); (b) corpus-based approaches, which make use of shallow knowledge automatically acquired from corpus and statistical or machine learning algorithms to induce disambiguation models (Yarowsky, 1995; Schtze 1998); and (c) hybrid approaches, which mix characteristics from the two other approaches to automatically acquire disambiguation models from corpus supported by linguistic knowledge (Ng and Lee 1996; Stevenson and Wilks, 2001).',Yarowsky,'' +o,2666,'4 Experimental Set-up For the experiments , we use the WSJ portion of the Penn tree bank , using the standard train\\/development\\/test splits , viz 39,832 sentences from 2-21 sections , 2416 sentences from section 23 for testing and 1,700 sentences from section 22 for development ',Marcus,'num_Set-up_4 nn_Set-up_Experimental prep_use_For det_experiments_the pobj_For_experiments nsubj_use_we dep_Set-up_use det_portion_the nn_portion_WSJ nsubj_viz_portion prep_portion_of det__the nn__Penn nn__tree nn__bank pobj_of_ partmod_viz_using det_train\\/development\\/test_the amod_train\\/development\\/test_standard nsubj_splits_train\\/development\\/test ccomp_using_splits ccomp_use_viz num_sentences_39,832 dobj_viz_sentences prep_sentences_from num_sections_2-21 pobj_from_sections num_sentences_2416 conj_sentences_sentences prep_sentences_from pobj_from_section num_section_23 prep_section_for pobj_for_testing cc_sentences_and num_sentences_1,700 conj_sentences_sentences prep_sentences_from pobj_from_section num_section_22 prep_section_for pobj_for_development ' +o,2667,'Each model can represent an important feature for the translation , such as phrase-based , language , or lexical models ',Koehn,'det_model_Each nsubj_represent_model aux_represent_can det_feature_an amod_feature_important dobj_represent_feature prep_feature_for det_translation_the pobj_for_translation dep_as_such prep_feature_as pobj_as_phrase-based conj_phrase-based_language cc_phrase-based_or amod_models_lexical conj_phrase-based_models ' +o,2668,'models implement the intuition that the best model will be the one that is consistent with the set of constrains imposed by the evidence , but otherwise is as uniform as possible ',Berger,'nsubj_implement_models det_intuition_the dobj_implement_intuition complm_one_that det_model_the amod_model_best nsubj_one_model aux_one_will cop_one_be det_one_the ccomp_implement_one nsubj_consistent_that cop_consistent_is rcmod_one_consistent prep_consistent_with det_set_the pobj_with_set prep_set_of pobj_of_constrains partmod_constrains_imposed prep_imposed_by det_evidence_the pobj_by_evidence cc_one_but advmod_one_otherwise cop_uniform_is advmod_uniform_as conj_one_uniform prep_uniform_as pobj_as_possible ' +o,2669,'1 Introduction Much of statistical NLP research relies on some sort of manually annotated corpora to train their models , but these resources are extremely expensive to build , especially at a large scale , for example in treebanking ',Marcus,'num_Introduction_1 nsubj_relies_Much prep_Much_of amod_research_statistical nn_research_NLP pobj_of_research dep_Introduction_relies prep_relies_on det_sort_some pobj_on_sort prep_sort_of amod_corpora_manually amod_corpora_annotated pobj_of_corpora aux_train_to xcomp_relies_train poss_models_their dobj_train_models cc_relies_but det_resources_these nsubj_expensive_resources cop_expensive_are advmod_expensive_extremely conj_relies_expensive aux_build_to xcomp_expensive_build advmod_build_especially prep_build_at det_scale_a amod_scale_large pobj_at_scale prep_build_for pobj_for_example prep_example_in pobj_in_treebanking ' +o,2670,'33 Syntax based approach An alternative to the Window and Document-oriented approach is to use syntactical information ',Dunning,'num_Syntax_33 dep_Syntax_based dep_based_approach det_alternative_An nsubj_is_alternative prep_alternative_to det_Window_the pobj_to_Window cc_Window_and amod_approach_Document-oriented conj_Window_approach rcmod_approach_is aux_use_to xcomp_is_use amod__syntactical nn__information dobj_use_ ' +p,2671,'53 Comparison with System Combination We re-implemented a state-of-the-art system combination method ',Rosti,'num_Comparison_53 dep_Comparison_with nn_Combination_System pobj_with_Combination nsubj_re-implemented_We rcmod_Combination_re-implemented det_method_a amod_method_state-of-the-art nn_method_system nn_method_combination dobj_re-implemented_method ' +n,2672,'However , reordering models in traditional phrase-based systems are not sufficient to treat such complex cases when we translate long sentences ',Koehn,'advmod_sufficient_However nn_models_reordering nsubj_sufficient_models prep_models_in amod_systems_traditional amod_systems_phrase-based pobj_in_systems cop_sufficient_are neg_sufficient_not aux_treat_to xcomp_sufficient_treat amod_cases_such nn_cases_complex dobj_treat_cases advmod_translate_when nsubj_translate_we advcl_treat_translate amod_sentences_long dobj_translate_sentences ' +o,2673,'Turney reported that the NEAR operator outperformed simple page co-occurrence for his purposes ; our early experiments informally showed the same for this work ',Turney,'nn__Turney nsubj_reported_ complm_outperformed_that det_operator_the nn_operator_NEAR nsubj_outperformed_operator ccomp_reported_outperformed amod_co-occurrence_simple nn_co-occurrence_page dobj_outperformed_co-occurrence prep_outperformed_for poss_purposes_his pobj_for_purposes poss_experiments_our amod_experiments_early nsubj_showed_experiments advmod_showed_informally parataxis_reported_showed det_same_the dobj_showed_same prep_showed_for det_work_this pobj_for_work ' +p,2674,'-LRB- 3 -RRB- -LRB- -RRB- -LRB- -RRB- 0 log 2 log A LH LH = 1 Problems for an unscaled log approach Although log identifies collocations much better than competing approaches in terms of its recall , it suffers from its relatively poor precision rates ',Dunning,'dep_log_3 nsubj_log_0 nsubj_log_2 ccomp_log_log det_LH_A nn_LH_LH nsubj_Problems_LH dep_Problems_= num_Problems_1 ccomp_log_Problems prep_log_for det_approach_an amod_approach_unscaled nn_approach_log pobj_for_approach mark_identifies_Although nsubj_identifies_log advcl_suffers_identifies nsubj_better_collocations advmod_better_much xcomp_identifies_better mark__than amod_approaches_competing nsubj__approaches advcl_identifies_ prep__in pobj_in_terms prep_terms_of poss_recall_its pobj_of_recall nsubj_suffers_it dep_log_suffers prep_suffers_from poss_rates_its advmod_poor_relatively amod_rates_poor nn_rates_precision pobj_from_rates ' +o,2675,'41 Training The training procedure is identical to the factored phrase-based training described in ',Koehn,'nsubj_identical_41 partmod_41_Training det_procedure_The nn_procedure_training dobj_Training_procedure cop_identical_is prep_identical_to det_training_the amod_training_factored amod_training_phrase-based pobj_to_training partmod_training_described prep_described_in ' +o,2676,'We concatenate the lists and we learn a new combination of weights that maximizes the Bleu score of the combined nbest list using the same development corpus we used for tuning the individual systems ',Och,'nsubj_concatenate_We det_lists_the dobj_concatenate_lists cc_concatenate_and nsubj_learn_we conj_concatenate_learn det_combination_a amod_combination_new dobj_learn_combination prep_combination_of pobj_of_weights nsubj_maximizes_that rcmod_weights_maximizes det_score_the nn_score_Bleu dobj_maximizes_score prep_score_of det_list_the amod_list_combined nn_list_nbest pobj_of_list partmod_list_using det_corpus_the amod_corpus_same nn_corpus_development dobj_using_corpus nsubj_used_we dep_concatenate_used prep_used_for pcomp_for_tuning det_systems_the amod_systems_individual dobj_tuning_systems ' +o,2677,'1 Introduction Recently , researchers have developed algorithms that learn to map natural language sentences to representations of their underlying meaning ',Zettlemoyer,'num_Introduction_1 advmod_developed_Recently nsubj_developed_researchers aux_developed_have dep_Introduction_developed dobj_developed_algorithms nsubj_learn_that rcmod_algorithms_learn aux_map_to xcomp_learn_map amod_sentences_natural nn_sentences_language dobj_map_sentences prep_map_to pobj_to_representations prep_representations_of poss_meaning_their amod_meaning_underlying pobj_of_meaning ' +p,2678,'1 Introduction Phrase-based method and syntaxbased method represent the state-of-the-art technologies in statistical machine translation -LRB- SMT -RRB- ',Koehn,'num__1 nn__Introduction amod__Phrase-based nn__method nsubj_represent_ cc__and nn__syntaxbased nn__method conj__ det_technologies_the amod_technologies_state-of-the-art dobj_represent_technologies prep_technologies_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +n,2679,'Current tree-based models that integrate linguistics and statistics , such as GHKM , are not able to generalize well from a single phrase pair ',Galley,'amod_models_Current amod_models_tree-based nsubj_able_models nsubj_integrate_that rcmod_models_integrate dobj_integrate_linguistics cc_linguistics_and conj_linguistics_statistics dep_as_such prep_models_as nn__GHKM pobj_as_ cop_able_are neg_able_not aux_generalize_to xcomp_able_generalize advmod_generalize_well prep_generalize_from det_pair_a amod_pair_single nn_pair_phrase pobj_from_pair ' +o,2680,'The group of collocations and compounds should be delimited using statistical approaches , such as Xtract or LocalMax , so that only the most relevantthose of higher frequency are included in the database ',Smadja,'det_group_The nsubjpass_delimited_group prep_group_of pobj_of_collocations cc_collocations_and conj_collocations_compounds aux_delimited_should auxpass_delimited_be dep_approaches_using amod_approaches_statistical purpcl_delimited_approaches dep_as_such prep_approaches_as nn__Xtract pobj_as_ cc__or nn__LocalMax conj__ advmod_included_so dep_included_that advmod_relevantthose_only det_relevantthose_the amod_relevantthose_most nsubjpass_included_relevantthose prep_relevantthose_of amod_frequency_higher pobj_of_frequency auxpass_included_are dep_approaches_included prep_included_in det_database_the pobj_in_database ' +o,2681,'Expectation Maximization does surprisingly well on larger data sets and is competitive with the Bayesian estimators at least in terms of cross-validation accuracy , confirming the results reported by ',Johnson,'nn_Maximization_Expectation nsubj_does_Maximization advmod_well_surprisingly advmod_does_well prep_does_on amod_sets_larger nn_sets_data pobj_on_sets cc_does_and cop_competitive_is conj_does_competitive prep_competitive_with det_estimators_the nn_estimators_Bayesian pobj_with_estimators dep_in_at dep_at_least prep_competitive_in pobj_in_terms prep_terms_of amod_accuracy_cross-validation pobj_of_accuracy conj_does_confirming det_results_the dobj_confirming_results partmod_results_reported prep_reported_by ' +o,2682,' describes experiments on the same named-entity dataset as in this paper , but using explicit features rather than kernels ',Collins,'nsubj_describes_ dobj_describes_experiments prep_experiments_on det_dataset_the amod_dataset_same amod_dataset_named-entity pobj_on_dataset dep_in_as prep_describes_in det_paper_this pobj_in_paper prep_paper_but dep_but_using amod_features_explicit dobj_using_features cc_features_rather dep_rather_than conj_features_kernels ' +p,2683,'Annealing resembles the popular bootstrapping technique , which starts out aiming for high precision , and gradually improves coverage over time ',Yarowsky,'dep_Annealing_resembles det__the amod__popular amod__bootstrapping nn__technique nsubj_Annealing_ nsubj_starts_which rcmod__starts prt_starts_out xcomp_starts_aiming prep_aiming_for amod_precision_high pobj_for_precision cc_starts_and advmod_improves_gradually conj_starts_improves dobj_improves_coverage prep_improves_over pobj_over_time ' +o,2684,'Previous research has focused on classifying subjective-versus-objective expressions , and also on accurate sentiment polarity assignment ',Pang,'amod_research_Previous nsubj_focused_research aux_focused_has prep_focused_on pcomp_on_classifying amod_expressions_subjective-versus-objective dobj_classifying_expressions advmod_classifying_ cc_focused_and conj_focused_also prep_also_on amod_assignment_accurate nn_assignment_sentiment nn_assignment_polarity pobj_on_assignment ' +o,2685,'Corpora in various languages , such as the English Penn Treebank corpus , the Swedish Stockholm-Ume corpus , and the Icelandic Frequency Dictionary -LRB- IFD -RRB- corpus , have been used to train -LRB- in the case of data-driven methods -RRB- and develop -LRB- in the case of linguistic rule-based methods -RRB- different taggers , and to evaluate their accuracy , eg ',Marcus,'nsubjpass_used_Corpora prep_Corpora_in amod_languages_various pobj_in_languages dep_as_such prep_Corpora_as det__the nn__English nn__Penn nn__Treebank nn__corpus pobj_as_ det__the amod__Swedish amod__Stockholm-Ume nn__corpus conj__ cc__and det__the nn_Dictionary_Icelandic nn_Dictionary_Frequency nn__Dictionary abbrev_Dictionary_IFD nn__corpus conj__ aux_used_have auxpass_used_been aux_train_to ccomp_used_train dep_train_in det_case_the pobj_in_case prep_case_of amod_methods_data-driven pobj_of_methods cc_train_and conj_train_develop dep_taggers_in det_case_the pobj_in_case prep_case_of amod_methods_linguistic amod_methods_rule-based pobj_of_methods amod_taggers_different dobj_develop_taggers cc_train_and aux_evaluate_to conj_train_evaluate poss_accuracy_their dobj_evaluate_accuracy appos_accuracy_eg ' +o,2686,'Adapting a vectorbased approach reported by , the Task ID Frame Agent is domain-independent and automatically trained ',Chu-Carroll,'dep_domain-independent_Adapting det_approach_a amod_approach_vectorbased dobj_Adapting_approach partmod_approach_reported prep_reported_by pobj_by_ det_Agent_the nn_Agent_Task nn_Agent_ID nn_Agent_Frame nsubj_domain-independent_Agent cop_domain-independent_is cc_domain-independent_and advmod_trained_automatically conj_domain-independent_trained ' +o,2687,'5 Previous Work The LEAF model is inspired by the literature on generative modeling for statistical word alignment and particularly by Model 4 ',Brown,'num_Work_5 amod_Work_Previous det_model_The nn_model_LEAF nsubjpass_inspired_model auxpass_inspired_is dep_Work_inspired prep_inspired_by det_literature_the pobj_by_literature prep_literature_on amod_modeling_generative pobj_on_modeling prep_modeling_for amod_alignment_statistical nn_alignment_word pobj_for_alignment cc_by_and conj_by_particularly dep_by_by pobj_by_Model num_Model_4 ' +o,2688,'We scored systems and our own output using case-insensitive IBM-style BLEU 104 , METEOR 06 with all modules , and TER 5 ',Papineni,'nsubj_scored_We dobj_scored_systems cc_systems_and poss_output_our amod_output_own conj_systems_output partmod_output_using amod__case-insensitive amod__IBM-style number_104_BLEU num__104 dobj_using_ conj_systems_METEOR num_METEOR_06 partmod_METEOR_ prep__with det_modules_all pobj_with_modules cc_systems_and nn__TER num__5 conj_systems_ ' +o,2689,'3 Data The data consists of six sections of the Wall Street Journal part of the Penn Treebank , and follows the setting of past editions of the CoNLL shared task : training set -LRB- sections 15-18 -RRB- , development set -LRB- section 20 -RRB- and test set -LRB- section 21 -RRB- ',Marcus,'num_data_3 nn_data_Data det_data_The nsubj_consists_data prep_consists_of num_sections_six pobj_of_sections prep_sections_of det_part_the nn_part_Wall nn_part_Street nn_part_Journal pobj_of_part prep_part_of det__the nn__Penn nn__Treebank pobj_of_ cc_consists_and conj_consists_follows det_setting_the dobj_follows_setting prep_setting_of amod_editions_past pobj_of_editions prep_editions_of det_task_the nn_task_CoNLL amod_task_shared pobj_of_task nn_set_training dep_setting_set number_15-18_sections dep_set_15-18 nn_set_development conj_set_set appos_set_section num_section_20 cc_set_and nn_set_test conj_set_set appos_set_section num_section_21 ' +o,2690,'The rules are then treated as events in a relative frequency estimate4 We used Giza + + Model 4 to obtain word alignments , using the grow-diag-final-and heuristic to symmetrise the two directional predictions ',Koehn,'det_rules_The nsubjpass_treated_rules auxpass_treated_are advmod_treated_then prep_treated_as pobj_as_events prep_events_in det_estimate4_a amod_estimate4_relative nn_estimate4_frequency pobj_in_estimate4 nsubj_used_We dep_treated_used nn_Model_Giza nn_Model_+ nn_Model_+ nsubj_obtain_Model num_Model_4 aux_obtain_to xcomp_used_obtain nn_alignments_word dobj_obtain_alignments advmod_obtain_ xcomp_used_using det_heuristic_the amod_heuristic_grow-diag-final-and dobj_using_heuristic aux_symmetrise_to xcomp_using_symmetrise det_predictions_the num_predictions_two amod_predictions_directional dobj_symmetrise_predictions ' +o,2691,'It is an extension of Pharaoh , and supports factor training and decoding ',Koehn,'nsubj_extension_It cop_extension_is det_extension_an prep_extension_of nn__Pharaoh pobj_of_ cc_extension_and conj_extension_supports nn_training_factor dobj_supports_training cc_training_and conj_training_decoding ' +p,2692,'41 Complete ambiguity classes Ambiguity classes capture the relevant property we are interested in : words with the same category possibilities are grouped together4 And ambiguity classes have been shown to be successfully employed , in a variety of ways , to improve POS tagging ',Goldberg,'num_classes_41 amod_classes_Complete amod_classes_ambiguity nn_classes_Ambiguity nsubj_capture_classes dep_classes_capture det_property_the amod_property_relevant dobj_capture_property nsubj_interested_we cop_interested_are dep_classes_interested prep_interested_in nsubjpass_grouped_words prep_words_with det_possibilities_the amod_possibilities_same nn_possibilities_category pobj_with_possibilities auxpass_grouped_are parataxis_interested_grouped dobj_grouped_together4 cc_grouped_And amod_classes_ambiguity nsubjpass_shown_classes aux_shown_have auxpass_shown_been conj_grouped_shown aux_employed_to auxpass_employed_be advmod_employed_successfully xcomp_shown_employed prep_employed_in det_variety_a pobj_in_variety prep_variety_of pobj_of_ways aux_improve_to dep_employed_improve dep_tagging_POS dobj_improve_tagging ' +o,2693,'80 80 \% Positive child education Positive cost Negative SUBJECT increase Figure 3 : An example of a word-polarity lattice Various methods have already been proposed for sentiment polarity classification , ranging from the use of co-occurrence with typical positive and negative words to bag of words and dependency structure ',Turney,'number_80_80 num_\%_80 amod_Positive_Positive nn_Positive_child nn_Positive_education nsubj_cost_Positive dep_\%_cost amod_Figure_Negative nn_Figure_SUBJECT nn_Figure_increase iobj_cost_Figure dobj_cost_3 det_example_An dobj_cost_example prep_example_of det_lattice_a amod_lattice_word-polarity pobj_of_lattice amod_methods_Various nsubjpass_proposed_methods aux_proposed_have advmod_proposed_already auxpass_proposed_been rcmod_lattice_proposed prep_proposed_for nn_classification_sentiment nn_classification_polarity pobj_for_classification partmod_example_ranging prep_ranging_from det_use_the pobj_from_use prep_use_of amod__co-occurrence prep_co-occurrence_with amod_words_typical amod_words_positive cc_positive_and conj_positive_negative pobj_with_words amod_words_ prep__to pobj_to_bag prep_bag_of nn__words pobj_of_ cc__and conj__dependency nn__structure pobj_of_ ' +o,2694,'Researchers have mostly looked at representing words by their surrounding words and by their syntactical contexts ',Hindle,'nsubj_looked_Researchers aux_looked_have advmod_looked_mostly prep_looked_at pcomp_at_representing dobj_representing_words prep_representing_by poss_words_their amod_words_surrounding pobj_by_words dep_words_ cc__and conj__by poss_contexts_their amod_contexts_syntactical pobj_by_contexts ' +o,2695,'Recently , a number of machine learning approaches have been proposed ',Zettlemoyer,'advmod_proposed_Recently det_number_a nsubjpass_proposed_number prep_number_of nn_approaches_machine nn_approaches_learning pobj_of_approaches aux_proposed_have auxpass_proposed_been ' +o,2696,'Another motivation to evaluate the performance of a phrase translation model that contains only syntactic phrases comes from recent efforts to built syntactic translation models ',Wu,'det_motivation_Another nsubj_comes_motivation aux_evaluate_to infmod_motivation_evaluate det_performance_the dobj_evaluate_performance prep_performance_of det_model_a nn_model_phrase nn_model_translation pobj_of_model nsubj_contains_that rcmod_performance_contains advmod_phrases_only amod_phrases_syntactic dobj_contains_phrases prep_comes_from amod_efforts_recent pobj_from_efforts aux_built_to xcomp_comes_built amod_models_syntactic nn_models_translation dobj_built_models ' +o,2697,'Examples of formalisms using this approach include the work of Magerman , Charniak , , and Goodman ',Collins,'nsubj_include_Examples prep_Examples_of pobj_of_formalisms partmod_formalisms_using det_approach_this dobj_using_approach det_work_the dobj_include_work prep_work_of nn__Magerman pobj_of_ nn__Charniak conj__ nn__ conj__ cc__and nn__Goodman conj__ ' +o,2698,' was an implicit or selforganizing syntax model as it did not use a Treebank ',Wu,'nsubj_model_ cop_model_was det_model_an amod_model_implicit cc_implicit_or conj_implicit_selforganizing nn_model_syntax mark_use_as nsubj_use_it aux_use_did neg_use_not dep_model_use det_Treebank_a dobj_use_Treebank ' +o,2699,'The other approach is to estimate a single score or likelihood of a translation with rich features , for example , with the maximum entropy -LRB- MaxEnt -RRB- method as in ',He,'det_approach_The amod_approach_other nsubj_is_approach aux_estimate_to xcomp_is_estimate det_score_a amod_score_single dobj_estimate_score cc_score_or conj_score_likelihood prep_score_of det_translation_a pobj_of_translation prep_estimate_with amod_features_rich pobj_with_features prep_estimate_for pobj_for_example prep_estimate_with det_method_the amod_method_maximum amod_method_entropy appos_method_MaxEnt pobj_with_method advmod_estimate_as dep_as_in ' +o,2700,'We tune using Ochs algorithm to optimize weights for the distortion model , language model , phrase translation model and word penalty over the BLEU metric ',Och,'nsubj_tune_We xcomp_tune_using nn__Ochs nn__algorithm nsubj_optimize_ aux_optimize_to xcomp_using_optimize dobj_optimize_weights prep_weights_for det_model_the nn_model_distortion pobj_for_model nn_model_language conj_model_model nn_model_phrase nn_model_translation conj_model_model cc_model_and nn_penalty_word conj_model_penalty prep_optimize_over det__the amod__BLEU amod__metric pobj_over_ ' +o,2701,'We obtained word alignments of the training data by first running GIZA + + and then applying the refinement rule grow-diagfinal-and ',Koehn,'nsubj_obtained_We nn_alignments_word dobj_obtained_alignments prep_alignments_of det_data_the nn_data_training pobj_of_data prep_obtained_by advmod_running_first pcomp_by_running dobj_running_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_running_and advmod_applying_then conj_running_applying det_grow-diagfinal-and_the amod_grow-diagfinal-and_refinement nn_grow-diagfinal-and_rule dobj_applying_grow-diagfinal-and ' +o,2702,'31 Results for English We used sections 0 to 12 of the WSJ part of the Penn Treebank with a total of 24,618 sentences for our experiments ',Marcus,'num_Results_31 prep_Results_for pobj_for_English nsubj_used_We rcmod_English_used num_0_sections dobj_used_0 prep_used_to pobj_to_12 prep_12_of det_part_the nn_part_WSJ pobj_of_part prep_part_of det__the nn__Penn nn__Treebank pobj_of_ prep_used_with det_total_a pobj_with_total prep_total_of num_sentences_24,618 pobj_of_sentences prep_sentences_for poss_experiments_our pobj_for_experiments ' +o,2703,'2 Details of the SO-PMI Algorithm The SO-PMI algorithm is used to estimate the semantic orientation -LRB- SO -RRB- of a phrase by 1http : \\/ \\/ wwwepinionscom 189 References Peter D Turney ',Turney,'num_Details_2 nsubjpass_used_Details prep_Details_of det__the nn__SO-PMI nn__Algorithm nn__The nn__SO-PMI nn__algorithm pobj_of_ auxpass_used_is aux_estimate_to xcomp_used_estimate det_orientation_the amod_orientation_semantic dobj_estimate_orientation abbrev_orientation_SO prep_orientation_of det_phrase_a pobj_of_phrase prep_estimate_by pobj_by_1http nn_\\/_\\/ nsubj_wwwepinionscom_\\/ parataxis_used_wwwepinionscom amod_Turney_189 nn_Turney_References nn_Turney_Peter nn_Turney_D dobj_wwwepinionscom_Turney ' +o,2704,'We are given a source -LRB- Chinese -RRB- sentence f = fJ1 = f1 , , fj , , fJ , which is to be translated into a target -LRB- English -RRB- sentence e = eI1 = e1 , , ei , , eI Among all possible target sentences , we will choose the sentence with the highest probability : eI1 = argmax eI1 -LCB- Pr -LRB- eI1 fJ1 -RRB- -RCB- -LRB- 1 -RRB- As an alternative to the often used source-channel approach , we directly model the posterior probability Pr -LRB- eI1 fJ1 -RRB- using a log-linear combination of feature functions ',Brown,'nsubjpass_given_We auxpass_given_are det_source_a nsubj_sentence_source appos_source_Chinese dep_given_sentence nsubj_f1_f dep_f1_= amod_f1_fJ1 amod_f1_= ccomp_sentence_f1 dep_given_fj nsubj_eI_fJ nsubj_is_which rcmod_fJ_is aux_translated_to auxpass_translated_be xcomp_is_translated prep_translated_into det_sentence_a nn_sentence_target appos_sentence_English pobj_into_sentence purpcl_translated_e dep_e_= amod_e1_eI1 amod_e1_= dobj_e_e1 appos_fJ_ei dep_given_eI prep_choose_Among det_sentences_all amod_sentences_possible nn_sentences_target pobj_Among_sentences nsubj_choose_we aux_choose_will ccomp_eI_choose det_sentence_the dobj_choose_sentence prep_choose_with det_probability_the amod_probability_highest pobj_with_probability amod_eI1_eI1 amod_eI1_= nn_eI1_argmax dep_probability_eI1 dep_eI1_Pr amod_fJ1_eI1 appos_eI1_fJ1 dep_model_1 mark__As det_alternative_an nsubj__alternative prep_alternative_to det_approach_the advmod_used_often amod_approach_used amod_approach_source-channel pobj_to_approach dep_model_ nsubj_model_we advmod_model_directly dep_given_model det_Pr_the nn_Pr_posterior nn_Pr_probability nsubj__Pr amod_fJ1_eI1 appos_Pr_fJ1 ccomp_model_ xcomp__using det_combination_a amod_combination_log-linear dobj_using_combination prep_combination_of nn_functions_feature pobj_of_functions ' +o,2705,'The skip-chain CRFs model the long distance dependency between context and answer sentences and the 2D CRFs model the dependency between contiguous questions ',Galley,'det_model_The amod_model_skip-chain nn_model_CRFs nn_model_ det_dependency_the amod_dependency_long nn_dependency_distance dep_model_dependency prep_dependency_between nn_sentences_context cc_context_and conj_context_answer pobj_between_sentences cc_dependency_and det_model_the num_model_2D nn_model_CRFs nn_model_ conj_dependency_model det_questions_the amod_questions_dependency prep_dependency_between pobj_between_contiguous dep_model_questions ' +o,2706,'Recently , combined an MRD and a corpus in a bootstrapping process ',Yarowsky,'advmod_combined_Recently nsubj_combined_ det_MRD_an dobj_combined_MRD cc_MRD_and det_corpus_a conj_MRD_corpus prep_combined_in det_process_a amod_process_bootstrapping pobj_in_process ' +o,2707,'In parsing , the most relevant previous work is due to , who considered three binary features of the intervening material : did it contain -LRB- a -RRB- any word tokens at all , -LRB- b -RRB- any verbs , -LRB- c -RRB- any commas or colons ? ',Collins,'prep_due_In pobj_In_parsing det_work_the advmod_relevant_most amod_work_relevant amod_work_previous nsubj_due_work cop_due_is prep_due_to pobj_to_ nsubj_considered_who rcmod__considered num_features_three amod_features_binary dobj_considered_features prep_features_of det_material_the amod_material_intervening pobj_of_material aux_contain_did nsubj_contain_it dep_material_contain dep_-RRB-_a det_word_any nsubj_tokens_word dep_contain_tokens advmod_tokens_at dep_at_all dep_-RRB-_b nsubj_verbs_any ccomp_tokens_verbs dep_-RRB-_c nsubj_commas_any dep_tokens_commas cc_commas_or conj_commas_colons ' +o,2708,'Demonstrating the inadequacy of such approaches , Al-Onaizan and showed that even given the words in the reference translation , and their alignment to the source words , a decoder of this sort charged with merely rearranging them into the correct target-language order could achieve a BLEU score of at best 69 \% and that only when restricted to keep most words very close to their source positions ',Papineni,'dep_showed_Demonstrating det_inadequacy_the dobj_Demonstrating_inadequacy prep_inadequacy_of amod_approaches_such pobj_of_approaches nsubj_showed_Al-Onaizan cc_Al-Onaizan_and conj_Al-Onaizan_ complm_achieve_that advmod_achieve_even prep_achieve_given det_words_the dep_given_words prep_words_in det_translation_the nn_translation_reference pobj_in_translation cc_words_and poss_alignment_their conj_words_alignment prep_alignment_to det_words_the nn_words_source pobj_to_words det_decoder_a nsubj_achieve_decoder prep_decoder_of det_sort_this pobj_of_sort partmod_sort_charged prep_charged_with advmod_rearranging_merely pcomp_with_rearranging dobj_rearranging_them prep_rearranging_into det_order_the amod_order_correct amod_order_target-language pobj_into_order aux_achieve_could ccomp_showed_achieve det__a amod__BLEU nn__score dobj_achieve_ prep__of advmod_\%_at dep_at_best num_\%_69 pobj_of_\% cc_\%_and conj_\%_that advmod_restricted_only advmod_restricted_when dep_achieve_restricted aux_keep_to xcomp_restricted_keep amod_words_most nsubj_close_words advmod_close_very xcomp_keep_close prep_close_to poss_positions_their nn_positions_source pobj_to_positions ' +o,2709,'3 The statistical model We use the Xerox part-of-speech tagger , a statistical tagger made at the Xerox Palo Alto Research Center ',Cutting,'num_model_3 det_model_The amod_model_statistical nsubj_use_We dep_model_use det__the nn__Xerox amod__part-of-speech nn__tagger dobj_use_ det_tagger_a amod_tagger_statistical appos__tagger partmod_tagger_made prep_made_at det_Center_the nn_Center_Xerox nn_Center_Palo nn_Center_Alto nn_Center_Research pobj_at_Center ' +o,2710,'The accuracy of the derived model depends heavily on the initial bias , but with a good choice results are comparable to those of method three ',Cutting,'det_accuracy_The nsubj_depends_accuracy prep_accuracy_of det_model_the amod_model_derived pobj_of_model advmod_depends_heavily prep_depends_on det_bias_the amod_bias_initial pobj_on_bias cc_on_but conj_on_with det_choice_a amod_choice_good pobj_with_choice nsubj_comparable_results cop_comparable_are dep_depends_comparable prep_comparable_to pobj_to_those prep_those_of pobj_of_method num_method_three ' +o,2711,'Unlike MaxEnt training , the method used for estimating the weight vector for BLEU maximization are not computationally scalable for a large number of feature functions ',Och,'prep_scalable_Unlike nn_training_MaxEnt pobj_Unlike_training det__the nn__method nsubj_scalable_ partmod__used prep_used_for pcomp_for_estimating det_vector_the nn_vector_weight dobj_estimating_vector prep_estimating_for amod_maximization_BLEU pobj_for_maximization cop_scalable_are neg_scalable_not advmod_scalable_computationally prep_scalable_for det_number_a amod_number_large pobj_for_number prep_number_of nn_functions_feature pobj_of_functions ' +o,2712,'1 Introduction Sentence-aligned parallel bilingual corpora have been essential resources for statistical machine translation , and many other multi-lingual natural language processing applications ',Brown,'num_corpora_1 nn_corpora_Introduction amod_corpora_Sentence-aligned nn_corpora_parallel amod_corpora_bilingual nsubj_been_corpora aux_been_have amod_resources_essential dobj_been_resources prep_resources_for amod__statistical nn__machine nn__translation pobj_for_ cc_resources_and amod_applications_many amod_applications_other amod_applications_multi-lingual amod_applications_natural nn_applications_language nn_applications_processing conj_resources_applications ' +o,2713,'The third baseline , COMP is the document compression system developed by , which compresses documents by cutting out constituents in a combined syntax and discourse tree ','Daume III','det_baseline_The amod_baseline_third nsubj_system_baseline appos_baseline_COMP cop_system_is det_system_the nn_system_document nn_system_compression partmod_system_developed prep_developed_by pobj_by_ nsubj_compresses_which rcmod__compresses dobj_compresses_documents prep_compresses_by pcomp_by_cutting prt_cutting_out dobj_cutting_constituents prep_constituents_in det_syntax_a amod_syntax_combined pobj_in_syntax cc_syntax_and nn_tree_discourse conj_syntax_tree ' +o,2714,'During training each example is broken into elementary trees using head rules and argument\\/adjunct rules similar to those of ',Collins,'prep_broken_During pobj_During_training det_example_each nsubjpass_broken_example auxpass_broken_is prep_broken_into amod_trees_elementary pobj_into_trees xcomp_broken_using nn_rules_head dobj_using_rules cc_rules_and nn_rules_argument\\/adjunct conj_rules_rules dep_using_similar prep_similar_to pobj_to_those prep_those_of ' +o,2715,'For example , the sets of tags and rule labels have been clustered by our team gr ~ : mm ~ trian , while a vocabulary of about 60,000 words has been clustered by machine ',Brown,'prep_clustered_For pobj_For_example det_sets_the nsubjpass_clustered_sets prep_sets_of pobj_of_tags cc_tags_and nn_labels_rule conj_tags_labels aux_clustered_have auxpass_clustered_been prep_clustered_by poss_~_our nn_~_team nn_~_gr pobj_by_~ nn_trian_mm nn_trian_~ dep_~_trian mark_clustered_while det_vocabulary_a nsubjpass_clustered_vocabulary prep_vocabulary_of quantmod_60,000_about num_words_60,000 pobj_of_words aux_clustered_has auxpass_clustered_been dep_trian_clustered prep_clustered_by pobj_by_machine ' +o,2716,'Some of the early statistical terminology translation methods are ',Brown,'nsubj_are_Some prep_Some_of det_methods_the amod_methods_early amod_methods_statistical nn_methods_terminology nn_methods_translation pobj_of_methods ' +o,2717,'Freund and Schapire discuss how the theory for classification problems can be extended to deal with both of these questions ; describes how these results apply to NLP problems ',Collins,'nsubj_discuss_Freund cc_Freund_and conj_Freund_Schapire advmod_discuss_ advmod_extended_how det_theory_the nsubjpass_extended_theory prep_theory_for nn_problems_classification pobj_for_problems aux_extended_can auxpass_extended_be ccomp_discuss_extended aux_deal_to purpcl_extended_deal prep_deal_with pobj_with_both prep_both_of det_questions_these pobj_of_questions nsubj_describes_ parataxis_discuss_describes advmod_apply_how det_results_these nsubj_apply_results ccomp_describes_apply aux_NLP_to xcomp_apply_NLP dobj_NLP_problems ' +o,2718,'These joint counts are estimated using the phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Koehn,'det_counts_These amod_counts_joint nsubjpass_estimated_counts auxpass_estimated_are dep_generated_estimated xcomp_estimated_using det_algorithm_the nn_algorithm_phrase nn_algorithm_induction dobj_using_algorithm partmod_algorithm_described prep_described_in pobj_in_ prep_estimated_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments xcomp_generated_using nn_model_IBM dobj_using_model num__2 nsubj_generated_ ' +p,2719,'1 Introduction Phrase-based translation models , which go beyond the original IBM translation models 1 by modeling translations of phrases rather than individual words , have been suggested to be the state-of-theart in statistical machine translation by empirical evaluations ',Och,'num__1 nn__Introduction nn__Phrase-based nn__translation nn__models nsubjpass_suggested_ nsubj_go_which rcmod__go prep_go_beyond det_models_the amod_models_original nn_models_IBM nn_models_translation nsubj__models pcomp_beyond_ dobj__1 prep__by nn_translations_modeling pobj_by_translations prep_translations_of pobj_of_phrases cc_phrases_rather dep_rather_than amod_words_individual conj_phrases_words aux_suggested_have auxpass_suggested_been aux_state-of-theart_to cop_state-of-theart_be det_state-of-theart_the xcomp_suggested_state-of-theart prep_state-of-theart_in amod_translation_statistical nn_translation_machine pobj_in_translation prep_state-of-theart_by amod_evaluations_empirical pobj_by_evaluations ' +o,2720,' argue that these results show a pattern where discriminative probability models are inferior to generative probability models , but that improvements can be achieved by keeping a generative probability model and training according to a discriminative optimization criteria ',Klein,'nsubj_argue_ dep_show_that det_results_these nsubj_show_results dep_argue_show det_pattern_a dobj_show_pattern advmod_inferior_where amod_models_discriminative nn_models_probability nsubj_inferior_models cop_inferior_are rcmod_pattern_inferior aux_generative_to xcomp_inferior_generative nn_models_probability dobj_generative_models cc_show_but dep_achieved_that nsubjpass_achieved_improvements aux_achieved_can auxpass_achieved_be conj_show_achieved prep_achieved_by pcomp_by_keeping det_model_a amod_model_generative nn_model_probability dobj_keeping_model cc_keeping_and conj_keeping_training prep_training_according dep_according_to det_criteria_a amod_criteria_discriminative nn_criteria_optimization pobj_to_criteria ' +o,2721,'452 BLEU on NIST MT Test Sets We use MT02 as the development set4 for minimum error rate training -LRB- MERT -RRB- ',Och,'number_BLEU_452 dobj_use_BLEU prep_BLEU_on nn_Sets_NIST nn_Sets_MT nn_Sets_Test pobj_on_Sets nsubj_use_We dobj_use_MT02 prep_use_as det_set4_the nn_set4_development pobj_as_set4 prep_set4_for amod_training_minimum nn_training_error nn_training_rate pobj_for_training abbrev_training_MERT ' +o,2722,'2 Data Sets for the Experiments 21 Coordination Annotation in the PENN TREEBANK For our experiments , we used the WSJ part of the PENN TREEBANK ',Marcus,'num_Sets_2 nn_Sets_Data dep_Annotation_Sets prep_Sets_for det_Experiments_the pobj_for_Experiments dep_Annotation_21 nn_Annotation_Coordination dep_used_Annotation prep_Annotation_in det_TREEBANK_the nn_TREEBANK_PENN pobj_in_TREEBANK prep_used_For poss_experiments_our pobj_For_experiments nsubj_used_we det_part_the nn_part_WSJ dobj_used_part prep_part_of det_TREEBANK_the nn_TREEBANK_PENN pobj_of_TREEBANK ' +o,2723,'section 20 Majority voting (Mufioz et al. , 1999) (Tjong Kim Sang and Veenstra~ 1999) (Ramshaw and Marcus, 1995) (Argarnon et al. , 1998) accuracy precision O:98.10\% C:98.29\% 93.63\% O:98.1\% C:98.2\% 93.1\% 97.58\% 92.50\% 97.37\% 91.80\% 91.6\% recall FZ=I 92.89\% 93.26 92.4\% 92.8 92.25\% 92.37 92.27\% 92.03 91.6\% 91.6 section 00 accuracy precision Majority voting 0:98.59\% C:98.65\% 95.04\% r (Tjong Kim Sang and Veenstra, 1999) 98.04\% 93.71\% (Ramshaw and Marcus, 1995) 97.8\% 93.1\% recall FB=I 94.75\% 94.90 93.90\% 93.81 93.5\% 93.3 Table 3: The results of majority voting of different data representations applied to the two standard data sets put forward by (Ramshaw and Marcus, 1995) compared with earlier work.',Ramshaw,'' +o,2724,'PMI between two phrases is de ned as : log2 prob -LRB- ph1 is near ph2 -RRB- prob -LRB- ph 1 -RRB- prob -LRB- ph2 -RRB- PMI is positive when two phrases tend to co-occur and negative when they tend to be in a complementary distribution ',Church,'nn__PMI nsubj_prob_ prep__between num_phrases_two pobj_between_phrases cop_prob_is dep_ned_de dep_as_ned advmod_prob_as amod_prob_log2 nsubj_is_ph1 dep_prob_is prep_is_near pobj_near_ph2 amod_prob_prob number_1_ph dep_prob_1 nsubj_positive_prob appos_prob_ph2 advmod_positive_PMI cop_positive_is ccomp_prob_positive advmod_tend_when num_phrases_two nsubj_tend_phrases advcl_positive_tend aux_co-occur_to xcomp_tend_co-occur cc_prob_and conj_prob_negative advmod_tend_when nsubj_tend_they ccomp_negative_tend aux_be_to xcomp_tend_be prep_be_in det_distribution_a amod_distribution_complementary pobj_in_distribution ' +o,2725,'51 ExploringtheParameters Theparameterswhichhaveamajorinuenceonthe performance of a phrase-based SMT model are the alignment heuristics , the maximum phrase length -LRB- MPR -RRB- and the order of the language model ',Koehn,'num_performance_51 nn_performance_ExploringtheParameters nn_performance_Theparameterswhichhaveamajorinuenceonthe nsubj_heuristics_performance prep_performance_of det_model_a amod_model_phrase-based nn_model_SMT pobj_of_model cop_heuristics_are det_heuristics_the amod_heuristics_alignment det_length_the amod_length_maximum nn_length_phrase appos_heuristics_length abbrev_length_MPR cc_heuristics_and det_order_the conj_heuristics_order prep_order_of det_model_the nn_model_language pobj_of_model ' +o,2726,'Examples of such affinities include synonyms , verb similarities and word associations ',Rapp,'nsubj_include_Examples prep_Examples_of amod_affinities_such pobj_of_affinities amod__synonyms dobj_include_ amod__verb nn__similarities conj__ cc__and nn_associations_word conj__associations ' +o,2727,'Although Phramer provides decoding functionality equivalent to Pharaohs , we preferred to use Pharaoh for this task because it is much faster than Phramer between 2 and 15 times faster , depending on the configuration and preliminary tests showed that there is no noticeable difference between the output of these two in terms of BLEU score ',Papineni,'mark_provides_Although nsubj_provides_Phramer advcl_preferred_provides xcomp_provides_decoding nn_equivalent_functionality dobj_decoding_equivalent prep_decoding_to pobj_to_Pharaohs nsubj_preferred_we aux_Pharaoh_to dep_Pharaoh_use xcomp_preferred_Pharaoh prep_Pharaoh_for det_task_this pobj_for_task mark_faster_because nsubj_faster_it cop_faster_is advmod_faster_much advcl_Pharaoh_faster prep_faster_than pobj_than_Phramer prep_Phramer_between pobj_between_2 cc_faster_and num_times_15 measure_faster_times conj_faster_faster prep_faster_depending dep_depending_on det_configuration_the pobj_on_configuration cc_faster_and amod_tests_preliminary nsubj_showed_tests conj_faster_showed complm_is_that expl_is_there ccomp_showed_is det_difference_no amod_difference_noticeable nsubj_is_difference prep_difference_between det_output_the pobj_between_output prep_output_of dep_two_these pobj_of_two prep_difference_in pobj_in_terms prep_terms_of amod_score_BLEU amod_score_ pobj_of_score ' +o,2728,'In , the authors proposed a method to integrate the IBM translation model 2 with an ASR system ',Brown,'prep_proposed_In pobj_In_ det_authors_the nsubj_proposed_authors det_method_a dobj_proposed_method aux_integrate_to infmod_method_integrate det_model_the nn_model_IBM nn_model_translation nsubj__model num__2 xcomp_integrate_ prep__with det_system_an nn_system_ASR pobj_with_system ' +o,2729,'Other research has been conducted in analyzing sentiment at a sentence level using bootstrapping techniques , finding strength of opinions , summing up orientations of opinion words in a sentence , and identifying opinion holders ',Stoyanov,'amod_research_Other nsubjpass_conducted_research aux_conducted_has auxpass_conducted_been prep_conducted_in pcomp_in_analyzing dobj_analyzing_sentiment prep_analyzing_at det_level_a nn_level_sentence pobj_at_level partmod_level_using xcomp_using_bootstrapping nn__techniques dobj_bootstrapping_ conj_analyzing_finding dobj_finding_strength prep_strength_of nn__opinions pobj_of_ conj_analyzing_summing prt_summing_up dobj_summing_orientations prep_orientations_of nn_words_opinion pobj_of_words prep_words_in det__a nn__sentence pobj_in_ cc_analyzing_and conj_analyzing_identifying nn_holders_opinion dobj_identifying_holders ' +o,2730,'3 Experiments We tested our methods experimentally on the English Penn Treebank and on the Czech Prague Dependency Treebank ',Marcus,'num_Experiments_3 nsubj_tested_We dep_Experiments_tested poss_methods_our dobj_tested_methods advmod_tested_experimentally prep_tested_on det__the nn__English nn__Penn nn__Treebank pobj_on_ cc_on_and conj_on_on det__the nn__Czech nn__Prague nn__Dependency nn__Treebank pobj_on_ ' +o,2731,'Much research is also being directed at acquiring affect lexica automatically ',Turney,'amod_research_Much nsubjpass_directed_research aux_directed_is advmod_directed_also auxpass_directed_being prep_directed_at pcomp_at_acquiring nn_lexica_affect dobj_acquiring_lexica advmod_acquiring_automatically ' +o,2732,'We ran GIZA + + on the training corpus in both directions using its default setting , and then applied the refinement rule diagand described in to obtain a single many-to-many word alignment for each sentence pair ',Koehn,'nsubj_ran_We nsubj_using_GIZA cc__+ advmod__+ dep_using_ prep__on det_corpus_the nn_corpus_training pobj_on_corpus prep__in det_directions_both pobj_in_directions xcomp_ran_using poss_setting_its nn_setting_default dobj_using_setting cc_ran_and advmod_applied_then conj_ran_applied det_diagand_the amod_diagand_refinement nn_diagand_rule dobj_applied_diagand partmod_diagand_described prep_described_in pobj_in_ aux_obtain_to xcomp_described_obtain det_alignment_a amod_alignment_single amod_alignment_many-to-many nn_alignment_word dobj_obtain_alignment prep_alignment_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,2733,'In this data set the 4-tuples of the test and training sets were extracted from Penn Treebank Wall Street Journal ',Marcus,'prep_set_In det_data_this pobj_In_data det_4-tuples_the nsubjpass_extracted_4-tuples prep_4-tuples_of det_sets_the nn_sets_test cc_test_and conj_test_training pobj_of_sets auxpass_extracted_were ccomp_set_extracted prep_extracted_from nn_Wall_Penn nn_Wall_Treebank pobj_from_Wall nn_Journal_Street nsubj_set_Journal ' +o,2734,'3 Previous Work on Subjectivity Tagging In previous work , a corpus of sentences from the Wall Street Journal Treebank Corpus was manually annotated with subjectivity classi cations bymultiplejudges ',Marcus,'num_Work_3 amod_Work_Previous nsubjpass_annotated_Work prep_Work_on nn_Tagging_Subjectivity pobj_on_Tagging prep_annotated_In amod__previous nn__work pobj_In_ det_corpus_a nsubjpass_annotated_corpus prep_corpus_of pobj_of_sentences prep_sentences_from det__the nn__Wall nn__Street nn__Journal nn__Treebank nn__Corpus pobj_from_ auxpass_annotated_was advmod_annotated_manually prep_annotated_with amod_bymultiplejudges_subjectivity nn_bymultiplejudges_classi nn_bymultiplejudges_cations pobj_with_bymultiplejudges ' +o,2735,'Our method is thus related to previous work based on Harris s distributional hypothesis2 It has been used to determine both word and syntactic path similarity ',Hindle,'poss_method_Our nsubjpass_related_method auxpass_related_is advmod_related_thus prep_related_to amod_work_previous pobj_to_work partmod_work_based prep_based_on nn_hypothesis2_Harris nn_hypothesis2_ nn_hypothesis2_s amod_hypothesis2_distributional pobj_on_hypothesis2 nsubjpass_used_It aux_used_has auxpass_used_been dep_related_used aux_determine_to xcomp_used_determine preconj_word_both dobj_determine_word cc_word_and amod_similarity_syntactic nn_similarity_path conj_word_similarity ' +o,2736,'4 Training This section discusses how to extract our translation rules given a triple nullnull , null null , nullnull As we know , the traditional tree-to-string rules can be easily extracted from nullnull , null null , nullnull using the algorithm of Mi and Huang 2 We would like 2 Mi and Huang extend the tree-based rule extraction algorithm to forest-based by introducing non-deterministic mechanism ',Zhang,'num_Training_4 det_section_This nsubj_discusses_section dep_Training_discusses advmod_extract_how aux_extract_to ccomp_discusses_extract poss_rules_our nn_rules_translation dobj_extract_rules prep_extract_given det_nullnull_a amod_nullnull_triple nsubj_nullnull_nullnull amod_nullnull_null dep_null_null dep_given_nullnull mark_know_As nsubj_know_we advcl_extracted_know det_rules_the amod_rules_traditional amod_rules_tree-to-string nsubjpass_extracted_rules aux_extracted_can auxpass_extracted_be advmod_extracted_easily ccomp_nullnull_extracted prep_extracted_from amod_null_nullnull amod_null_null pobj_from_null advmod_using_nullnull ccomp_nullnull_using det_algorithm_the dobj_using_algorithm prep_algorithm_of pobj_of_Mi cc_using_and conj_using_Huang number_2_ dobj_Huang_2 nsubj_like_We aux_like_would dep_Training_like num_Mi_2 dobj_like_Mi cc_like_and nn__Huang nsubj_extend_ conj_like_extend det_algorithm_the amod_algorithm_tree-based nn_algorithm_rule nn_algorithm_extraction dobj_extend_algorithm amod_algorithm_ dep__to dep__forest-based prep_extend_by pcomp_by_introducing amod_mechanism_non-deterministic dobj_introducing_mechanism ' +p,2737,' demonstrated that semi-supervised WSD could be successful ',Yarowsky,'nsubj_demonstrated_ complm_successful_that amod_WSD_semi-supervised nsubj_successful_WSD aux_successful_could cop_successful_be ccomp_demonstrated_successful ' +p,2738,'Second , benefits for sentiment analysis can be realized by decomposing the problem into S\\/O -LRB- or neutral versus polar -RRB- and polarity classification ',Pang,'advmod_realized_Second nsubjpass_realized_benefits prep_benefits_for nn_analysis_sentiment pobj_for_analysis aux_realized_can auxpass_realized_be prep_realized_by pcomp_by_decomposing det_problem_the dobj_decomposing_problem prep_decomposing_into pobj_into_S\\/O dep_neutral_or dep_S\\/O_neutral cc_neutral_versus conj_neutral_polar cc_S\\/O_and amod_classification_polarity conj_S\\/O_classification ' +o,2739,'Some other researchers also work on detecting negative cases , ie contradiction , instead of entailment ',Marneffe,'det_researchers_Some amod_researchers_other advmod_work_also dep_researchers_work prep_work_on pcomp_on_detecting amod_cases_negative dobj_detecting_cases nn_contradiction_ie appos_cases_contradiction dep_of_instead prep_cases_of pobj_of_entailment ' +p,2740,'This similarity score is computed as a max over a number of component scoring functions, some based on external lexical resources, including: various string similarity functions, of which most are applied to word lemmas measures of synonymy, hypernymy, antonymy, and semantic relatedness, including a widelyused measure due to Jiang and Conrath (1997), based on manually constructed lexical resources such as WordNet and NomBank a function based on the well-known distributional similarity metric of Lin (1998), which automatically infers similarity of words and phrases from their distributions in a very large corpus of English text The ability to leverage external lexical resources both manually and automatically constructedis critical to the success of MANLI.',Lin,'' +o,2741,'The output of GIZA + + is then post-processed using the three symmetrization heuristics described in ',Och,'det_output_The nsubj_post-processed_output prep_output_of nn_+_GIZA nn_+_+ pobj_of_+ cop_post-processed_is advmod_post-processed_then xcomp_post-processed_using det_heuristics_the num_heuristics_three nn_heuristics_symmetrization dobj_using_heuristics partmod_heuristics_described prep_described_in ' +o,2742,'Another technique used was to filter sentences of the out-of-domain corpus based on their similarity to the target domain , as predicted by a classifier ',Dredze,'det_technique_Another nsubj_was_technique partmod_technique_used aux_filter_to xcomp_was_filter dobj_filter_sentences prep_sentences_of det_corpus_the amod_corpus_out-of-domain pobj_of_corpus partmod_corpus_based prep_based_on poss_similarity_their pobj_on_similarity prep_based_to det_domain_the nn_domain_target pobj_to_domain mark_predicted_as advcl_based_predicted prep_predicted_by det_classifier_a pobj_by_classifier ' +o,2743,'Evaluation We evaluate translation output using three automatic evaluation measures : BLEU , NIST , and METEOR 5 All measures used were the case-sensitive , corpuslevel versions ',Papineni,'aux_evaluate_Evaluation nsubj_evaluate_We nn_output_translation dobj_evaluate_output partmod_output_using num_measures_three amod_measures_automatic nn_measures_evaluation dobj_using_measures nn__BLEU dep_output_ nn__NIST appos__ cc_output_and conj_output_METEOR number_5_ num_measures_5 det_measures_All nsubj_versions_measures partmod_measures_used cop_versions_were det_versions_the amod_versions_case-sensitive amod_versions_corpuslevel rcmod_METEOR_versions ' +o,2744,'This set of 800 sentences was used for Minimum Error Rate Training to tune the weights of our system with respect to BLEU score ',Och,'det_set_This nsubjpass_used_set prep_set_of num_sentences_800 pobj_of_sentences auxpass_used_was prep_used_for nn__Minimum nn__Error nn__Rate nn__Training pobj_for_ aux_tune_to xcomp_used_tune det_weights_the dobj_tune_weights prep_weights_of poss_system_our pobj_of_system prep_tune_with pobj_with_respect prep_tune_to amod_score_BLEU pobj_to_score ' +o,2745,'So far research in automatic opinion recognition has primarily addressed learning subjective language , identifying opinionated documents and sentences , and discriminating between positive and negative language ',Pang,'advmod_far_So advmod_research_far nsubj_addressed_research prep_research_in amod_recognition_automatic nn_recognition_opinion pobj_in_recognition aux_addressed_has advmod_addressed_primarily ccomp_addressed_learning amod__subjective nn__language dobj_learning_ conj_learning_identifying amod_documents_opinionated dobj_identifying_documents nsubj__ cc__and conj__sentences rcmod_documents_ cc_learning_and conj_learning_discriminating prep_discriminating_between amod_language_positive cc_positive_and conj_positive_negative pobj_between_language ' +o,2746,'These findings are in line with results with incremental parsing with perceptrons , where it is suggested that a generative baseline feature provides the perceptron algorithm with a much better starting point for learning ',Collins,'det_findings_These nsubj_are_findings prep_are_in pobj_in_line prep_are_with amod_results_ pobj_with_results prep_results_with amod_parsing_incremental pobj_with_parsing prep_are_with pobj_with_perceptrons advmod_suggested_where nsubjpass_suggested_it auxpass_suggested_is rcmod_perceptrons_suggested complm_provides_that det_feature_a amod_feature_generative nn_feature_baseline nsubj_provides_feature ccomp_suggested_provides det_algorithm_the nn_algorithm_perceptron dobj_provides_algorithm prep_provides_with det_point_a advmod_better_much amod_point_better amod_point_starting pobj_with_point prep_point_for pcomp_for_learning ' +p,2747,'The search across a dimension uses the efficient method of ',Och,'det_search_The nsubj_uses_search prep_search_across det_dimension_a pobj_across_dimension det_method_the amod_method_efficient dobj_uses_method prep_method_of ' +p,2748,'In the hierarchical phrase-based model , and an inversion transduction grammar -LRB- ITG -RRB- , the problem is resolved by restricting to a binarized form where at most two non-terminals are allowed in the righthand side ',Wu,'prep_resolved_In det__the amod__hierarchical amod__phrase-based nn__model pobj_In_ cc__and det_grammar_an nn_grammar_inversion nn_grammar_transduction conj__grammar abbrev__ITG advmod_resolved_ det_problem_the nsubjpass_resolved_problem auxpass_resolved_is prep_resolved_by pcomp_by_restricting prep_restricting_to det_form_a amod_form_binarized pobj_to_form advmod_allowed_where quantmod_two_at dep_at_most num_non-terminals_two nsubjpass_allowed_non-terminals auxpass_allowed_are rcmod_form_allowed prep_allowed_in det_side_the amod_side_righthand pobj_in_side ' +o,2749,'As two examples , and give good overviews of the techniques and equations used for Markov models and part-ofspeech tagging , but they are not very explicit in the details that are needed for their application ',Marcus,'prep_give_As num_examples_two pobj_As_examples nsubj_give_ cc__and conj__ amod_overviews_good nsubj_used_overviews prep_overviews_of det_techniques_the pobj_of_techniques cc_techniques_and conj_techniques_equations dep_give_used prep_used_for nn_models_Markov pobj_for_models cc_models_and amod_tagging_part-ofspeech conj_models_tagging cc_used_but nsubj_explicit_they cop_explicit_are neg_explicit_not advmod_explicit_very conj_used_explicit prep_explicit_in det_details_the pobj_in_details nsubjpass_needed_that auxpass_needed_are rcmod_details_needed prep_needed_for poss_application_their pobj_for_application ' +o,2750,'41 Data Sets Our results are based on syntactic data drawn from the Penn Treebank , specifically the portion used by CoNLL 2000 shared task ',Marcus,'num_Sets_41 nn_Sets_Data poss_results_Our nsubjpass_based_results auxpass_based_are dep_Sets_based prep_based_on amod_data_syntactic pobj_on_data partmod_data_drawn prep_drawn_from det__the nn__Penn nn__Treebank pobj_from_ advmod_drawn_specifically det_portion_the nsubj_shared_portion partmod_portion_used prep_used_by pobj_by_CoNLL num_CoNLL_2000 dep_Sets_shared nn__task dobj_shared_ ' +o,2751,' Marcus , M , Santorini , B , and Malvinkiewicz , MA ',Marcus,'nn_B_ nn_B_Marcus appos_B_M appos_B_Santorini cc_B_and nn_MA_Malvinkiewicz conj_B_MA ' +o,2752,'According to the statistical machine translation formalism , the translation process is to search for the best sentence bE such that bE = arg max E P -LRB- EjJ -RRB- = arg maxE P -LRB- JjE -RRB- P -LRB- E -RRB- where P -LRB- JjE -RRB- is a translation model characterizing the correspondence between E and J ; P -LRB- E -RRB- , the English language model probability ',Brown,'prep_is_According dep_According_to det__the amod__statistical nn__machine nn__translation nn__formalism pobj_to_ det_process_the nn_process_translation nsubj_is_process aux_search_to xcomp_is_search prep_search_for det_bE_the amod_bE_best nn_bE_sentence pobj_for_bE predet_bE_such det_bE_that nsubj_=_bE dep_P_= nn_P_arg nn_P_max nn_P_E rcmod_bE_P appos_bE_EjJ rcmod_bE_= amod_P_arg nn_P_maxE nn_P_P appos_P_JjE dobj_=_P appos_P_E advmod_model_where nsubj_model_P appos_P_JjE cop_model_is det_model_a nn_model_translation rcmod_P_model partmod_model_characterizing det_correspondence_the dobj_characterizing_correspondence prep_correspondence_between pobj_between_E cc_E_and conj_E_J dep_bE_P appos_P_E det_probability_the amod_probability_English nn_probability_language nn_probability_model appos_P_probability ' +o,2753,'3 Results and Analysis shows that the oracle parsing accuracy of a k-best edge-factored MST parser is considerably higher than the one-best score of the same parser , even when k is small ',Hall,'num_Results_3 nsubj_shows_Results cc_Results_and nn__Analysis conj_Results_ dep_higher_that det_accuracy_the amod_accuracy_oracle nn_accuracy_parsing nsubj_higher_accuracy prep_accuracy_of det_parser_a amod_parser_k-best amod_parser_edge-factored nn_parser_MST pobj_of_parser cop_higher_is advmod_higher_considerably dep_shows_higher prep_higher_than det_score_the amod_score_one-best pobj_than_score prep_score_of det_parser_the amod_parser_same pobj_of_parser advmod_higher_even advmod_small_when nsubj_small_k cop_small_is dep_higher_small ' +p,2754,'In syntactic parse re-ranking supersenses have been used to build useful latent semantic features ',Koo,'prep_used_In amod_parse_syntactic pobj_In_parse amod_supersenses_re-ranking nsubjpass_used_supersenses aux_used_have auxpass_used_been aux_build_to xcomp_used_build amod_features_useful dep_useful_latent amod_features_semantic dobj_build_features ' +o,2755,'To examine the effects of including some known AMs on the performance , the following AMs had a 50 \% chance of being included in the initial population : pointwise mutual information , the Dice coefficient , and the heuristic measure defined in : H -LRB- a , b , c -RRB- = 2log f -LRB- abc -RRB- f -LRB- a -RRB- f -LRB- c -RRB- if POS -LRB- b -RRB- = X , log f -LRB- abc -RRB- f -LRB- a -RRB- f -LRB- b -RRB- f -LRB- c -RRB- otherwise ',Church,'aux_examine_To dep_had_examine det_effects_the dobj_examine_effects prep_effects_of pcomp_of_including det_AMs_some amod_AMs_known dobj_including_AMs prep_including_on det_performance_the pobj_on_performance det_AMs_the amod_AMs_following nsubj_had_AMs det_chance_a dep_\%_50 amod_chance_\% dobj_had_chance prep_chance_of auxpass_included_being pcomp_of_included prep_included_in det_population_the amod_population_initial pobj_in_population amod__pointwise amod__mutual nn__information dep_chance_ det_coefficient_the nn_coefficient_Dice appos__coefficient cc_chance_and det_measure_the amod_measure_heuristic conj_chance_measure partmod_measure_defined prep_defined_in pobj_in_ nsubj_=_H dep_H_a appos_a_b appos_a_c parataxis_had_= iobj_=_2log dobj_=_f appos_f_abc dobj_=_f dep_-LRB-_a dobj_=_f dep_-LRB-_c mark_=_if nsubj_=_POS dep_-LRB-_b advcl_=_= nsubj_log_X ccomp_=_log dobj_log_f appos_f_abc dep_f_f dep_-RRB-_a nsubj_f_f dep_-LRB-_b dep_had_f dep_f_c advmod_c_otherwise ' +o,2756,'41 Judging Rule Correctness Following the spirit of the fine-grained human evaluation in , we randomly sampled 800 rules from our rule-base and presented them to an annotator who judged them for correctness , according to the lexical reference notion specified above ',Snow,'dep_sampled_Judging nn_Correctness_Rule dobj_Judging_Correctness prep_Correctness_Following det_spirit_the pobj_Following_spirit prep_spirit_of det_evaluation_the amod_evaluation_fine-grained amod_evaluation_human pobj_of_evaluation prep_evaluation_in pobj_in_ nsubj_sampled_we advmod_sampled_randomly dep_41_sampled num_rules_800 dobj_sampled_rules prep_sampled_from poss_rule-base_our pobj_from_rule-base cc_sampled_and conj_sampled_presented dobj_presented_them prep_presented_to det_annotator_an pobj_to_annotator nsubj_judged_who rcmod_annotator_judged dobj_judged_them prep_judged_for pobj_for_correctness prep_judged_according dep_according_to det_notion_the amod_notion_lexical nn_notion_reference pobj_to_notion partmod_notion_specified prep_specified_above ' +o,2757,'From this LFG annotated treebank , large-scale unification grammar resources were automatically extracted and used in parsing and generation ',Cahill,'prep_extracted_From det_treebank_this nn_treebank_LFG amod_treebank_annotated pobj_From_treebank amod_resources_large-scale nn_resources_unification nn_resources_grammar nsubjpass_extracted_resources auxpass_extracted_were advmod_extracted_automatically cc_extracted_and conj_extracted_used prep_extracted_in pcomp_in_parsing dobj_parsing_ cc__and conj__generation ' +o,2758,'This approach is usually referred to as the noisy sourcechannel approach in SMT ',Brown,'det_approach_This nsubjpass_referred_approach auxpass_referred_is advmod_referred_usually prep_referred_to advmod_approach_as det_approach_the advmod_sourcechannel_noisy amod_approach_sourcechannel pobj_to_approach prep_approach_in pobj_in_SMT ' +o,2759,'In many applications , it has been shown that sentences with subjective meanings are paid more attention than factual ones ',Pang,'prep_shown_In amod_applications_many pobj_In_applications nsubjpass_shown_it aux_shown_has auxpass_shown_been complm_paid_that nsubjpass_paid_sentences prep_sentences_with amod_meanings_subjective pobj_with_meanings auxpass_paid_are ccomp_shown_paid amod_attention_more dobj_paid_attention prep_paid_than amod_ones_factual pobj_than_ones advmod__ amod_ones_ ' +o,2760,'Our approach was to identify a parallel corpus of manually and automatically transcribed documents , the TDT2 corpus , and then use a statistical approach to identify tokens with significantly Table 5 : Impact of recall and precision enhancing devices ',Dunning,'poss_approach_Our nsubj_was_approach aux_identify_to xcomp_was_identify det_corpus_a amod_corpus_parallel dobj_identify_corpus prep_corpus_of dep_documents_manually cc_manually_and advmod_transcribed_automatically conj_manually_transcribed pobj_of_documents det_corpus_the nn_corpus_TDT2 appos_documents_corpus cc_identify_and advmod_identify_then conj_identify_use det__a amod__statistical nn__approach dobj_use_ aux_identify_to xcomp_use_identify dobj_identify_tokens prep_identify_with advmod_Table_significantly amod_Impact_Table dep_Impact_5 pobj_with_Impact prep_Impact_of nn_devices_recall cc_recall_and conj_recall_precision nn_devices_enhancing pobj_of_devices ' +o,2761,'Most previous research in translation knowledge acquisition is based on parallel corpora ',Brown,'amod_research_Most amod_research_previous nsubjpass_based_research prep_research_in nn_acquisition_translation nn_acquisition_knowledge pobj_in_acquisition auxpass_based_is prep_based_on amod_corpora_parallel pobj_on_corpora ' +o,2762,'This text was part-of-speech tagged using the Xerox HMM tagger ',Cutting,'det_text_This nsubjpass_tagged_text auxpass_tagged_was advmod_tagged_part-of-speech xcomp_tagged_using det_tagger_the nn_tagger_Xerox nn_tagger_HMM dobj_using_tagger ' +o,2763,'GIZA + + toolkit is used to perform word alignment in both directions with default settings , and the intersect-diag-grow method is used to generate symmetric word alignment refinement ',Och,'amod__+ cc_+_+ conj_+_toolkit nsubjpass_used_ auxpass_used_is dep_GIZA_used aux_perform_to xcomp_used_perform nn_alignment_word dobj_perform_alignment prep_perform_in det_directions_both pobj_in_directions prep_directions_with nn_settings_default pobj_with_settings cc_used_and det_method_the amod_method_intersect-diag-grow nsubjpass_used_method auxpass_used_is conj_used_used aux_generate_to xcomp_used_generate amod_refinement_symmetric nn_refinement_word nn_refinement_alignment dobj_generate_refinement ' +o,2764,'Word-aligned corpora have been found to be an excellent source for translation-related knowledge , not only for phrase-based models , but also for syntax-based models -LRB- eg , -RRB- ',Shen,'amod_corpora_Word-aligned nsubjpass_found_corpora aux_found_have auxpass_found_been aux_source_to cop_source_be det_source_an amod_source_excellent xcomp_found_source prep_source_for amod_knowledge_translation-related pobj_for_knowledge cc_for_not dep_not_only prep_source_for amod__phrase-based nn__models pobj_for_ dep_also_but cc_for_also conj_for_for amod_models_syntax-based pobj_for_models appos_models_eg dep_eg_ ' +o,2765,'As a solution , a given amount of labeled training data is divided into two distinct sets , ie , 4\\/5 for estimating , and the 667 remaining 1\\/5 for estimating ',Suzuki,'prep_divided_As det_solution_a pobj_As_solution det_amount_a amod_amount_given nsubjpass_divided_amount prep_amount_of amod_data_labeled nn_data_training pobj_of_data auxpass_divided_is prep_divided_into num_sets_two amod_sets_distinct pobj_into_sets conj_sets_ie conj_sets_4\\/5 prep_4\\/5_for pobj_for_estimating cc_sets_and det_1\\/5_the amod_1\\/5_667 amod_1\\/5_remaining conj_sets_1\\/5 prep_1\\/5_for pobj_for_estimating ' +o,2766,'In this paper it is shown that the synchronous grammars used in , Zhang et al ',Wu,'prep_shown_In det_paper_this pobj_In_paper nsubjpass_shown_it auxpass_shown_is complm_used_that det_grammars_the amod_grammars_synchronous nsubj_used_grammars ccomp_shown_used prep_used_in pobj_in_ conj__Zhang cc__et conj__al ' +o,2767,'Like the data used by , this data was retagged by the Brill tagger in order to obtain realistic part-of-speech -LRB- POS -RRB- tags 3 ',Ramshaw,'prep_retagged_Like det_data_the pobj_Like_data partmod_data_used prep_used_by pobj_by_ det_data_this nsubjpass_retagged_data auxpass_retagged_was prep_retagged_by det_tagger_the nn_tagger_Brill pobj_by_tagger mark_obtain_in dep_obtain_order aux_obtain_to purpcl_retagged_obtain amod_tags_realistic nn_tags_part-of-speech appos_tags_POS dobj_obtain_tags dep_tags_3 ' +o,2768,'Purely syntactic categories lead to a smaller number of tags which also improves the accuracy of manual tagging 2 ',Marcus,'advmod_syntactic_Purely amod_categories_syntactic nsubj_lead_categories prep_lead_to det_number_a amod_number_smaller pobj_to_number prep_number_of pobj_of_tags nsubj_improves_which advmod_improves_also rcmod_tags_improves det_accuracy_the dobj_improves_accuracy prep_accuracy_of pobj_of_manual partmod_manual_tagging dobj_tagging_2 ' +o,2769,'In our experiments , we used the Hidden Markov Model -LRB- HMM -RRB- tagging method described in ',Cutting,'prep_used_In poss_experiments_our pobj_In_experiments nsubj_used_we det_Model_the nn_Model_Hidden nn_Model_Markov dobj_used_Model abbrev_Model_HMM partmod_Model_tagging dobj_tagging_method partmod_method_described prep_described_in ' +o,2770,'However , at the short term , the incorporation of these type of features will force us to either build a new decoder or extend an existing one , or to move to a new MT architecture , for instance , in the fashion of the architectures suggested by Tillmann and Zhang or et al ',Liang,'advmod_force_However prep_force_at det_term_the amod_term_short pobj_at_term det_incorporation_the nsubj_force_incorporation prep_incorporation_of det_type_these pobj_of_type prep_type_of pobj_of_features aux_force_will nsubj_build_us aux_build_to dep_build_either ccomp_force_build det_decoder_a amod_decoder_new dobj_build_decoder cc_build_or conj_build_extend det_one_an amod_one_existing dobj_extend_one cc_build_or aux_move_to conj_build_move prep_move_to det_architecture_a amod_architecture_new nn_architecture_MT pobj_to_architecture prep_move_for pobj_for_instance prep_move_in det_fashion_the pobj_in_fashion prep_fashion_of det_architectures_the pobj_of_architectures partmod_architectures_suggested prep_suggested_by nn__Tillmann cc_Tillmann_and conj_Tillmann_Zhang pobj_by_ cc__or conj__ cc__et conj__al ' +o,2771,'The f are optimized by Minimum-Error Training -LRB- MERT -RRB- ',Och,'det_f_The nsubjpass_optimized_f auxpass_optimized_are prep_optimized_by nn_Training_Minimum-Error pobj_by_Training abbrev_Training_MERT ' +o,2772,'Recent work by pro514 poses factored translation models that combine feature functions to handle syntactic , morphological , and other linguistic information in a log-linear model ',Koehn,'amod_work_Recent nsubj_poses_work prep_work_by amod_pro514_ pobj_by_pro514 amod_models_factored nn_models_translation dobj_poses_models nsubj_combine_that rcmod_models_combine nn_functions_feature nsubj_handle_functions aux_handle_to xcomp_combine_handle amod_information_syntactic conj_syntactic_morphological cc_syntactic_and conj_syntactic_other dep_syntactic_linguistic dobj_handle_information prep_handle_in det_model_a amod_model_log-linear pobj_in_model ' +o,2773,'Another application of hard clustering methods -LRB- in particular bottom-up variants -RRB- is that they can also produce a binary tree , which can be used for decision-tree based systems such as the SPATTER parser or the ATR Decision-Tree Part-OfSpeech Tagger ',Brown,'det_application_Another nsubj_is_application prep_application_of amod_methods_hard dep_hard_clustering pobj_of_methods dep_methods_in amod_variants_particular amod_variants_bottom-up pobj_in_variants complm_produce_that nsubj_produce_they aux_produce_can advmod_produce_also ccomp_is_produce det_tree_a amod_tree_binary dobj_produce_tree nsubjpass_used_which aux_used_can auxpass_used_be rcmod_tree_used prep_used_for pobj_for_decision-tree partmod_decision-tree_based dobj_based_systems dep_as_such prep_systems_as det__the amod__SPATTER nn__parser pobj_as_ cc__or det__the nn__ATR nn__Decision-Tree nn__Part-OfSpeech nn__Tagger conj__ ' +o,2774,'The parser is trained on dependencies extracted from the English Penn Treebank version 30 by using the head-percolation rules of ',Marcus,'det_parser_The nsubjpass_trained_parser auxpass_trained_is prep_trained_on pobj_on_dependencies partmod_dependencies_extracted prep_extracted_from det__the nn__English nn__Penn nn__Treebank nn__version num__30 pobj_from_ prep_extracted_by pcomp_by_using det_rules_the amod_rules_head-percolation dobj_using_rules prep_rules_of pobj_of_ ' +o,2775,'After each step the annotations were compared using the ~ statistic as reliability measure for all classification tasks ',Carletta,'prep_compared_After det_step_each pobj_After_step det_annotations_the nsubjpass_compared_annotations auxpass_compared_were xcomp_compared_using det_statistic_the amod_statistic_~ dobj_using_statistic prep_using_as amod_measure_reliability pobj_as_measure prep_measure_for det_tasks_all nn_tasks_classification pobj_for_tasks ' +o,2776,'Metrics in the Rouge family allow for skip n-grams ; Kauchak and Barzilay -LRB- 2006 -RRB- take paraphrasing into account ; metrics such as METEOR and GTM calculate both recall and precision ; METEOR is also similar to SIA in that word class information is used ',Och,'nsubj_allow_Metrics prep_Metrics_in det_family_the nn_family_Rouge pobj_in_family prep_allow_for amod__skip amod__n-grams pobj_for_ nsubj_take_Kauchak cc_Kauchak_and conj_Kauchak_Barzilay appos_Kauchak_2006 parataxis_allow_take xcomp_take_paraphrasing prep_paraphrasing_into pobj_into_account nsubj_calculate_metrics dep_as_such prep_metrics_as nn__METEOR pobj_as_ cc__and nn__GTM conj__ parataxis_take_calculate preconj_recall_both dobj_calculate_recall cc_recall_and conj_recall_precision nsubj_similar_METEOR cop_similar_is advmod_similar_also parataxis_take_similar prep_similar_to nn__SIA pobj_to_ rel_used_in pobj_in_that nn_information_word nn_information_class nsubjpass_used_information auxpass_used_is rcmod__used ' +o,2777,'To set the weights , m , we performed minimum error rate training on the development set using Bleu as the objective function ',Papineni,'aux_set_To dep_performed_set det_weights_the dobj_set_weights appos_weights_m nsubj_performed_we amod__minimum nn__error nn__rate nn__training dobj_performed_ prep_performed_on det_development_the pobj_on_development partmod_development_set xcomp_set_using nn__Bleu dobj_using_ prep__as det_function_the amod_function_objective pobj_as_function ' +o,2778,'Parsing research has also begun to adopt discriminative methods from the Machine Learning literature , such as the perceptron and the largemargin methods underlying Support Vector Machines ',Collins,'amod_research_Parsing nsubj_begun_research aux_begun_has advmod_begun_also aux_adopt_to xcomp_begun_adopt amod_methods_discriminative dobj_adopt_methods prep_adopt_from det_literature_the nn_literature_Machine nn_literature_Learning pobj_from_literature dep_as_such prep_literature_as det__the nn__perceptron pobj_as_ cc__and det__the amod__largemargin amod__methods amod__underlying nn__Support nn__Vector nn__Machines conj__ ' +o,2779,'Furthermore , it is not possible to apply the powerful ` one sense per discourse \' property because there is no discourse in dictionaries ',Yarowsky,'advmod_possible_Furthermore nsubj_possible_it cop_possible_is neg_possible_not aux_apply_to xcomp_possible_apply det_sense_the amod_sense_powerful nn_sense_one dobj_apply_sense prep_sense_per poss__discourse nn__property pobj_per_ mark_is_because expl_is_there advcl_possible_is det_discourse_no nsubj_is_discourse prep_discourse_in pobj_in_dictionaries ' +n,2780,'Our syntactic-relation-based thesaurus is based on the method proposed by , although Hindle did not apply it to information retrieval ',Hindle,'poss_thesaurus_Our amod_thesaurus_syntactic-relation-based nsubjpass_based_thesaurus auxpass_based_is prep_based_on det_method_the pobj_on_method partmod_method_proposed prep_proposed_by pobj_by_ mark_apply_although nsubj_apply_Hindle aux_apply_did neg_apply_not advcl_proposed_apply dobj_apply_it prep_apply_to nn_retrieval_information pobj_to_retrieval ' +n,2781,'Both Charniak and Bikel were trained using the goldstandard tags , as this produced higher accuracy on the development set than using s tags ',Ratnaparkhi,'preconj__Both nn__Charniak nsubjpass_trained_ cc__and nn__Bikel conj__ auxpass_trained_were xcomp_trained_using det_tags_the amod_tags_goldstandard dobj_using_tags mark_produced_as nsubj_produced_this advcl_trained_produced amod_accuracy_higher dobj_produced_accuracy prep_produced_on det_development_the pobj_on_development partmod_development_set prep_set_than pcomp_than_using amod_tags_ nn_tags_s dobj_using_tags ' +o,2782,'We might find better suited metrics , such as METEOR , which is oriented towards word selection8 ',Banerjee,'nsubj_find_We aux_find_might amod_metrics_better nn_metrics_suited dobj_find_metrics dep_as_such prep_metrics_as nn__METEOR pobj_as_ nsubjpass_oriented_which auxpass_oriented_is rcmod__oriented prep_oriented_towards nn_selection8_word pobj_towards_selection8 ' +o,2783,'Language models , such as N-gram class models and Ergodic Hidden Markov Models were proposed and used in applications such as syntactic class -LRB- POS -RRB- tagging for English , clustering and scoring of recognizer sentence hypotheses ',Brown,'amod_models_Language dep_as_such prep_models_as amod_models_N-gram nn_models_class pobj_as_models nn__ cc__and conj__Ergodic nn__Hidden nn__Markov nn__Models nsubjpass_proposed_ auxpass_proposed_were rcmod_models_proposed cc_proposed_and conj_proposed_used prep_proposed_in nsubj_tagging_applications dep_as_such prep_applications_as amod_class_syntactic pobj_as_class appos_class_POS pcomp_in_tagging prep_tagging_for nn__English pobj_for_ conj_tagging_clustering cc_tagging_and conj_tagging_scoring prep_scoring_of nn_hypotheses_recognizer nn_hypotheses_sentence pobj_of_hypotheses ' +o,2784,'The data used for all our experiments is extracted from the PENN \' WSJ Treebank by the program provided by Sabine Buchholz from Tilbug University ',Marcus,'det_data_The nsubjpass_extracted_data partmod_data_used prep_used_for predet_experiments_all poss_experiments_our pobj_for_experiments auxpass_extracted_is prep_extracted_from det_PENN_the poss__PENN nn__WSJ nn__Treebank pobj_from_ prep_extracted_by det_program_the pobj_by_program partmod_program_provided prep_provided_by nn_Buchholz_Sabine pobj_by_Buchholz prep_Buchholz_from nn_University_Tilbug pobj_from_University ' +o,2785,'He then goes on to adapt the conventional noisy channel MT model of to NLU , where extracting a semantic representation from an input text corresponds to finding : argmax -LRB- Sem -RRB- -LCB- p -LRB- Input Sem -RRB- p -LRB- Sem -RRB- -RCB- , where p -LRB- Sem -RRB- is a model for generating semantic representations , and p -LRB- Input Sem -RRB- is a model for the relation between semantic representations and corresponding texts ',Brown,'nsubj_goes_He advmod_goes_then prep_goes_on aux_adapt_to dep_argmax_adapt det_model_the amod_model_conventional amod_model_noisy nn_model_channel nn_model_MT dobj_adapt_model prep_model_of pobj_of_ prep_adapt_to pobj_to_NLU advmod_corresponds_where csubj_corresponds_extracting det_representation_a amod_representation_semantic dobj_extracting_representation prep_extracting_from det_text_an nn_text_input pobj_from_text advcl_adapt_corresponds prep_corresponds_to pobj_to_finding pobj_on_argmax dep_argmax_Sem measure_p_p nn_Sem_Input appos_p_Sem amod_argmax_p dep_p_Sem advmod_model_where nsubj_model_p appos_p_Sem cop_model_is det_model_a rcmod_argmax_model prep_model_for amod_representations_generating amod_representations_semantic pobj_for_representations cc_goes_and nsubj_model_p nn_Sem_Input appos_p_Sem cop_model_is det_model_a conj_goes_model prep_model_for det_relation_the pobj_for_relation prep_relation_between amod_representations_semantic pobj_between_representations cc_representations_and amod_texts_corresponding conj_representations_texts ' +o,2786,'We use data from the CoNLL-2004 shared taskthe PropBank annotations of the Penn Treebank , with sections 1518 as the training set and section 20 as the development set ',Marcus,'nsubj_use_We nsubj_shared_data prep_data_from det_CoNLL-2004_the pobj_from_CoNLL-2004 ccomp_use_shared amod_annotations_taskthe nn_annotations_PropBank nn_annotations_ dobj_shared_annotations prep_annotations_of det__the nn__Penn nn__Treebank pobj_of_ prep_shared_with number_1518_sections pobj_with_1518 prep_1518_as det_training_the pobj_as_training partmod_training_set cc_1518_and conj_1518_section num_section_20 prep_section_as det_set_the nn_set_development pobj_as_set ' +n,2787,'The utility of ITG as a reordering constraint for most language pairs , is well-known both empirically and analytically , howeverITGsstraight -LRB- monotone -RRB- andinverted -LRB- reverse -RRB- rules exhibit strong cohesiveness , which is inadequate to express orientations that require gaps ',Wu,'det_utility_The prep_utility_of pobj_of_ITG dep_utility_as det_constraint_a nn_constraint_reordering pobj_as_constraint prep_constraint_for amod_pairs_most nn_pairs_language pobj_for_pairs cop_well-known_is dep_constraint_well-known det_howeverITGsstraight_both advmod__empirically amod_howeverITGsstraight_ cc__and advmod__analytically conj__ nsubj_andinverted_howeverITGsstraight appos_howeverITGsstraight_monotone ccomp_well-known_andinverted nn_rules_reverse nsubj_exhibit_rules ccomp_andinverted_exhibit amod_cohesiveness_strong dobj_exhibit_cohesiveness nsubj_inadequate_which cop_inadequate_is rcmod_constraint_inadequate aux_express_to xcomp_inadequate_express dobj_express_orientations nsubj_require_that rcmod_orientations_require dobj_require_gaps ' +o,2788,'Here , we use the hidden Markov model -LRB- HMM -RRB- alignment model and Model 4 of Brown et al ',Berger,'advmod_use_Here nsubj_use_we det_model_the amod_model_hidden nn_model_Markov dobj_use_model abbrev_model_HMM amod__alignment nn__model dep_model_ cc_model_and conj_model_Model num_Model_4 prep_Model_of pobj_of_Brown cc_Brown_et conj_Brown_al ' +o,2789,'As reported previously , the standard left-corner grmninar embeds sufficient non-local infornlation in its productions to significantly improve the labeled precision and recall of its MLPs with respect to MLPs of the PCFG estimated from the untransfornmd trees ',Collins,'mark_reported_As advcl_embeds_reported advmod_reported_previously det_grmninar_the amod_grmninar_standard amod_grmninar_left-corner nsubj_embeds_grmninar amod_infornlation_sufficient amod_infornlation_non-local dobj_embeds_infornlation prep_infornlation_in poss_productions_its pobj_in_productions aux_improve_to advmod_improve_significantly infmod_productions_improve det_precision_the amod_precision_labeled dobj_improve_precision cc_precision_and conj_precision_recall prep_precision_of poss_MLPs_its pobj_of_MLPs prep_improve_with pobj_with_respect prep_improve_to pobj_to_MLPs prep_MLPs_of det_PCFG_the pobj_of_PCFG partmod_PCFG_estimated prep_estimated_from det__the amod__untransfornmd nn__trees pobj_from_ ' +o,2790,'Self-training is a form of semi-supervised learning ',Yarowsky,'amod__Self-training nsubj_form_ cop_form_is det_form_a prep_form_of amod_learning_semi-supervised pobj_of_learning ' +o,2791,'A more recent bootstrapping approach is described in ',Yarowsky,'det_approach_A advmod_recent_more amod_approach_recent amod_approach_bootstrapping nsubjpass_described_approach auxpass_described_is prep_described_in ' +o,2792,'31 Collocation Features The collocation features were inspired by the one-sense-per-collocation heuristic proposed by ',Yarowsky,'num_Features_31 nn_Features_Collocation det_features_The nn_features_collocation nsubjpass_inspired_features auxpass_inspired_were dep_Features_inspired prep_inspired_by det_heuristic_the amod_heuristic_one-sense-per-collocation pobj_by_heuristic partmod_heuristic_proposed prep_proposed_by ' +o,2793,'1 Introduction In the past few years , there has been an increasing interest in mining opinions from product reviews ',Pang,'num_Introduction_1 prep_been_In det_years_the amod_years_past amod_years_few pobj_In_years expl_been_there aux_been_has dep_Introduction_been det_interest_an amod_interest_increasing dobj_been_interest prep_been_in nn_opinions_mining pobj_in_opinions prep_been_from nn_reviews_product pobj_from_reviews ' +o,2794,'For more detail , see ',Chen,'prep_see_For amod_detail_more pobj_For_detail ' +o,2795,'62 Experiment 2 : Yarowskys Words We also conducted translation on seven of the twelve English words studied in ',Yarowsky,'num_Experiment_62 num_Experiment_2 nn_Words_Yarowskys dep_Experiment_Words nsubj_conducted_We advmod_conducted_also rcmod_Words_conducted dobj_conducted_translation prep_conducted_on pobj_on_seven prep_seven_of det_words_the num_words_twelve amod_words_English pobj_of_words partmod_words_studied prep_studied_in ' +p,2796,'In agreement with recent resuits on parsing with lexicalised probabilistic grammars , we find that statistics over lexical , as opposed to structural , features best correspond to human intuitivejudgments and to experimental findings ',Collins,'prep_find_In pobj_In_agreement prep_agreement_with amod_resuits_recent pobj_with_resuits prep_resuits_on pcomp_on_parsing prep_parsing_with amod__lexicalised amod__probabilistic nn__grammars pobj_with_ nsubj_find_we complm_features_that nsubj_features_statistics prep_statistics_over pobj_over_lexical advmod_opposed_as ccomp_features_opposed prep_opposed_to pobj_to_structural ccomp_find_features advmod_features_best dep_features_correspond prep_correspond_to amod_intuitivejudgments_human pobj_to_intuitivejudgments cc_to_and conj_to_to amod_findings_experimental pobj_to_findings ' +o,2797,'One of the earliest attempts at extracting interrupted collocations \' -LRB- ie non-contiguous collocations , including VPCs -RRB- , was that of Smadja ',Dunning,'nsubj_that_One prep_One_of det_attempts_the amod_attempts_earliest pobj_of_attempts prep_attempts_at pcomp_at_extracting dep_extracting_interrupted dep_interrupted_collocations dep_collocations_ie amod_collocations_non-contiguous dobj_ie_collocations prep_collocations_including pobj_including_VPCs cop_that_was prep_that_of nn__Smadja pobj_of_ ' +n,2798,'This latter point is a critical difference that contrasts to the major weakness of the work of which uses a top-N list of translations to select the maximum BLEU sentence as a target for training -LRB- so called local update -RRB- ',Liang,'det_point_This amod_point_latter nsubj_difference_point cop_difference_is det_difference_a amod_difference_critical nsubj_contrasts_that rcmod_difference_contrasts prep_contrasts_to det_weakness_the amod_weakness_major pobj_to_weakness prep_weakness_of det_work_the pobj_of_work prep_work_of pobj_of_ nsubj_uses_which rcmod_weakness_uses det_list_a amod_list_top-N nsubj_select_list prep_list_of pobj_of_translations aux_select_to xcomp_uses_select det_sentence_the amod_sentence_maximum amod_sentence_BLEU dobj_select_sentence prep_select_as det_target_a pobj_as_target prep_target_for pobj_for_training advmod_called_so dep_target_called amod_update_local dobj_called_update ' +p,2799,'1 Introduction In recent years , statistical machine translation have experienced a quantum leap in quality thanks to automatic evaluation and errorbased optimization ',Och,'num_Introduction_1 prep_experienced_In amod_years_recent pobj_In_years amod_translation_statistical nn_translation_machine nsubj_experienced_translation aux_experienced_have dep_Introduction_experienced det_leap_a nn_leap_quantum dobj_experienced_leap prep_leap_in nn_thanks_quality pobj_in_thanks prep_experienced_to amod__automatic nn__evaluation pobj_to_ cc__and amod_optimization_errorbased conj__optimization ' +o,2800,'In this work , we employ a syntax-based model that applies a series of tree\\/string -LRB- xRS -RRB- rules to a source language string to produce a target language phrase structure tree ',Galley,'prep_employ_In det_work_this pobj_In_work nsubj_employ_we det_model_a amod_model_syntax-based dobj_employ_model nsubj_applies_that rcmod_model_applies det_series_a dobj_applies_series prep_series_of pcomp_of_tree\\/string appos_rules_xRS dobj_tree\\/string_rules amod_rules_ prep__to det_string_a nn_string_source nn_string_language pobj_to_string aux_produce_to xcomp_tree\\/string_produce det_tree_a nn_tree_target nn_tree_language nn_tree_phrase nn_tree_structure dobj_produce_tree ' +o,2801,'We first added sister-head dependencies for NPs -LRB- following original proposal -RRB- and then for PPs , which are flat in Negra , and thus similar in structure to NPs -LRB- see Section 22 -RRB- ',Collins,'nsubj_added_We advmod_added_first amod_dependencies_sister-head dobj_added_dependencies prep_added_for pobj_for_NPs dep_added_following amod_proposal_ amod_proposal_original pobj_following_proposal cc_added_and conj_added_then prep_then_for pobj_for_PPs nsubj_flat_which cop_flat_are rcmod_PPs_flat prep_flat_in pobj_in_Negra cc_Negra_and advmod_similar_thus conj_Negra_similar prep_similar_in pobj_in_structure prep_structure_to pobj_to_NPs dep_structure_see dobj_see_Section num_Section_22 ' +o,2802,'BABAR uses the log-likelihood statistic to evaluate the strength of a co-occurrence relationship ',Dunning,'nsubj_uses_BABAR det__the amod__log-likelihood nn__statistic dobj_uses_ aux_evaluate_to xcomp_uses_evaluate det_strength_the dobj_evaluate_strength prep_strength_of det_relationship_a amod_relationship_co-occurrence pobj_of_relationship ' +p,2803,'1 Introduction Automatic Metrics for machine translation -LRB- MT -RRB- evaluation have been receiving significant attention in the past two years , since IBM \'s BLEU metric was proposed and made available ',Papineni,'num_Metrics_1 nn_Metrics_Introduction nn_Metrics_Automatic nsubj_attention_Metrics prep_Metrics_for nn_evaluation_machine nn_evaluation_translation nn_evaluation_MT pobj_for_evaluation aux_attention_have aux_attention_been dep_attention_receiving amod_attention_significant prep_attention_in det_years_the amod_years_past num_years_two pobj_in_years mark_proposed_since poss_metric_IBM possessive_IBM_\'s amod_metric_BLEU nsubjpass_proposed_metric auxpass_proposed_was advcl_attention_proposed cc_proposed_and conj_proposed_made dep_made_available ' +n,2804,'Even the creators of BLEU point out that it may not correlate particularly well with human judgment at the sentence level ',Papineni,'advmod_creators_Even det_creators_the nsubj_point_creators prep_creators_of pobj_of_BLEU prt_point_out complm_correlate_that nsubj_correlate_it aux_correlate_may neg_correlate_not ccomp_point_correlate advmod_well_particularly advmod_correlate_well prep_correlate_with amod_judgment_human pobj_with_judgment prep_judgment_at det_level_the nn_level_sentence pobj_at_level ' +o,2805,'While earlier approaches for text compression were based on symbolic reduction rules , more recent approaches use an aligned corpus of documents and their human written summaries to determine which constituents can be reduced ',Jing,'mark_based_While amod_approaches_earlier nsubjpass_based_approaches prep_approaches_for nn_compression_text pobj_for_compression auxpass_based_were advcl_use_based prep_based_on amod__symbolic nn__reduction nn__rules pobj_on_ advmod_approaches_more amod_approaches_recent nsubj_use_approaches det_corpus_an amod_corpus_aligned nsubj_determine_corpus prep_corpus_of pobj_of_documents cc_documents_and poss_summaries_their amod_summaries_human amod_summaries_written conj_documents_summaries aux_determine_to xcomp_use_determine dobj_reduced_which nsubjpass_reduced_constituents aux_reduced_can auxpass_reduced_be ccomp_determine_reduced ' +o,2806,'One kind is the Penn Treebank ',Marcus,'num_kind_One nsubj_Treebank_kind cop_Treebank_is det_Treebank_the nn_Treebank_Penn ' +o,2807,'Table 2: Corpora and Modalities CORPUS MODALITY ACE asserted, or other TIMEML must, may, should, would, or could Prasad et al., 2006 assertion, belief, facts or eventualities Saur et al., 2007 certain, probable, possible, or other Inui et al., 2008 affirm, infer, doubt, hear, intend, ask, recommend, hypothesize, or other THIS STUDY S/O, necessity, hope, possible, recommend, intend Table 3: Markup Scheme (Tags and Definitions) Tag Definition (Examples) R Remedy, Medical operation (e.g. radiotherapy) T Medical test, Medical examination (e.g., CT, MRI) D Deasese, Symptom (e.g., Endometrial cancer, headache) M Medication, administration of a drug (e.g., Levofloxacin, Flexeril) A patient action (e.g., admitted to a hospital) V Other verb (e.g., cancer spread to ) 2 Related Works 2.1 Previous Markup Schemes In the NLP field, fact identification has not been studied well to date.',Prasad,'' +o,2808,'Including about 14 million sentence pairs extracted from the Gigaword data , we obtain a statistically significant improvement from 423 to 456 in BLEU ',Papineni,'prep_obtain_Including quantmod_million_about number_million_14 num_pairs_million nn_pairs_sentence pobj_Including_pairs partmod_pairs_extracted prep_extracted_from det_data_the nn_data_Gigaword pobj_from_data nsubj_obtain_we det_improvement_a advmod_significant_statistically amod_improvement_significant dobj_obtain_improvement prep_improvement_from number_456_423 dep_456_to pobj_from_456 prep_456_in pobj_in_BLEU ' +o,2809,'For example , in IBM Model 1 the lexicon probability of source word f given target word e is calculated as : p -LRB- f e -RRB- = summationtext k c -LRB- f e ; e k , fk -RRB- summationtext k , f c -LRB- f e ; e k , fk -RRB- -LRB- 1 -RRB- c -LRB- f e ; ek , fk -RRB- = summationdisplay ek , fk P -LRB- ek , fk -RRB- summationdisplay a P -LRB- a ek , fk -RRB- -LRB- 2 -RRB- summationdisplay j -LRB- f , fkj -RRB- -LRB- e , ekaj -RRB- Therefore , the distribution of P -LRB- ek , fk -RRB- will affect the alignment results ',Och,'prep_p_For pobj_For_example prep_example_in nn_Model_IBM pobj_in_Model dep_probability_1 det_probability_the nn_probability_lexicon dep_Model_probability prep_probability_of nn_word_source pobj_of_word nsubjpass_calculated_f prep_f_given nn_word_target dep_given_word dep_word_e auxpass_calculated_is rcmod_word_calculated prep_calculated_as pobj_as_ dep_-LRB-_f dep_f_e dep_k_= amod_k_summationtext ccomp_p_k nsubj_p_c nn_e_f dep_c_e amod_k_e dep_e_k appos_k_fk amod_k_summationtext dep_c_k appos_k_f dobj_affect_c dep_-LRB-_f dep_f_e dep_f_e dep_e_k dep_e_fk dep_summationdisplay_1 nsubj_ek_c nn_e_f dep_c_e dep_e_ek appos_ek_fk dep_ek_= amod_ek_summationdisplay dep_summationdisplay_ek nn_P_fk nsubj_summationdisplay_P appos_P_ek dep_ek_fk dep_affect_summationdisplay det_P_a nsubj_summationdisplay_P det_ek_a dep_P_ek appos_ek_fk appos_P_2 ccomp_summationdisplay_summationdisplay dobj_summationdisplay_j dep_j_f appos_f_fkj appos_j_e dep_e_ekaj advmod_summationdisplay_Therefore det_distribution_the nsubj_affect_distribution prep_distribution_of pobj_of_P appos_P_ek dep_ek_fk aux_affect_will rcmod_f_affect det_results_the amod_results_alignment dobj_affect_results ' +o,2810,' proposed dealing with the sparseness problem by estimating the likelihood of unseen events from that of ` similar \' events that have been seen ',Hindle,'nsubj_proposed_ xcomp_proposed_dealing prep_dealing_with det_problem_the amod_problem_sparseness pobj_with_problem prep_dealing_by pcomp_by_estimating det_likelihood_the dobj_estimating_likelihood prep_likelihood_of amod_events_unseen pobj_of_events prep_estimating_from pobj_from_that prep_that_of amod_events_similar pobj_of_events nsubjpass_seen_that aux_seen_have auxpass_seen_been rcmod_events_seen ' +n,2811,'1 Introduction In recent years , various phrase translation approaches have been shown to outperform word-to-word translation models ',Brown,'num_Introduction_1 prep_approaches_In amod_years_recent pobj_In_years amod_translation_various nn_translation_phrase nsubj_approaches_translation dep_Introduction_approaches nsubjpass_shown_ aux_shown_have auxpass_shown_been ccomp_approaches_shown aux_outperform_to xcomp_shown_outperform amod_models_word-to-word nn_models_translation dobj_outperform_models ' +o,2812,'In all experiments that follow , each system configuration was independently optimized on the NIST 2003 Chinese-English test set -LRB- 919 sentences -RRB- using minimum error rate training and tested on the NIST 2005 Chinese-English task -LRB- 1082 sentences -RRB- ',Och,'prep_optimized_In det_experiments_all pobj_In_experiments nsubj_follow_that rcmod_experiments_follow det_configuration_each nn_configuration_system nsubjpass_optimized_configuration auxpass_optimized_was advmod_optimized_independently prep_optimized_on det_set_the nn_set_NIST num_set_2003 amod_set_Chinese-English nn_set_test pobj_on_set num_sentences_919 appos_set_sentences xcomp_optimized_using amod__minimum nn__error nn__rate nn__training dobj_using_ cc_optimized_and conj_optimized_tested prep_tested_on det_task_the nn_task_NIST num_task_2005 amod_task_Chinese-English pobj_on_task num_sentences_1082 appos_task_sentences ' +o,2813,'As a common strategy , POS guessers examine the endings of unknown words along with their capitalization , or consider the distribution of unknown words over specific parts-of-speech ',Cutting,'prep_examine_As det_strategy_a amod_strategy_common pobj_As_strategy dep_guessers_POS nsubj_examine_guessers det_endings_the dobj_examine_endings prep_endings_of amod_words_unknown pobj_of_words partmod_words_ prt__along prep__with poss_capitalization_their pobj_with_capitalization cc_examine_or conj_examine_consider det_distribution_the dobj_examine_distribution prep_distribution_of amod_words_unknown pobj_of_words prep_examine_over amod__specific amod__parts-of-speech pobj_over_ ' +o,2814,'Many statistical taggers and parsers have been trained on it , eg , Srinivas -LRB- 1997 -RRB- and Alshawi and Carter -LRB- 1994 -RRB- ',Ramshaw,'amod_taggers_Many amod_taggers_statistical nsubjpass_trained_taggers cc_taggers_and conj_taggers_parsers aux_trained_have auxpass_trained_been prep_trained_on pobj_on_it nn__eg appos_it_ appos__Srinivas appos_Srinivas_1997 cc_Srinivas_and conj_Srinivas_Alshawi cc_Alshawi_and conj_Alshawi_Carter appos_Alshawi_1994 ' +o,2815,'The XEROX tagger comes with a list of built-in ending guessing rules ',Cutting,'det_tagger_The nn_tagger_XEROX nsubj_comes_tagger prep_comes_with det_list_a pobj_with_list prep_list_of pobj_of_built-in partmod_built-in_ending xcomp_ending_guessing dobj_guessing_rules ' +o,2816,'The translation output is measured using BLEU ',Papineni,'det_output_The nn_output_translation nsubjpass_measured_output auxpass_measured_is xcomp_measured_using dobj_using_BLEU ' +o,2817,'There exist many different string similarity measures : word overlap , longest common subsequence , Levenshteinedit distance , word n-gramoverlap etc Semantic similarity measures are obtained by first computing the semantic similarity of the words containedin the sentencesbeing compared ',Barzilay,'expl_exist_There amod_measures_many amod_measures_different nn_measures_string nn_measures_similarity dobj_exist_measures nsubj_overlap_word parataxis_exist_overlap nn__ appos__longest amod__common nn__subsequence amod__ dep__Levenshteinedit nn__distance nsubj_compared_ nn_Semantic_word nn_Semantic_n-gramoverlap nn_Semantic_ nn_Semantic_etc tmod_compared_Semantic nn_measures_similarity nsubjpass_obtained_measures auxpass_obtained_are dep_Semantic_obtained prep_obtained_by advmod_computing_first pcomp_by_computing det_similarity_the amod_similarity_semantic dobj_computing_similarity prep_similarity_of det_containedin_the nn_containedin_words pobj_of_containedin det_sentencesbeing_the nsubj_compared_sentencesbeing ccomp_overlap_compared ' +o,2818,'Minor variants support voted perceptron and MEMMs with the same ef cient feature encoding ',Collins,'nn_support_Minor nn_support_variants nsubj_voted_support nn__perceptron dobj_voted_ cc__and nn__MEMMs conj__ prep_voted_with det_feature_the amod_feature_same nn_feature_ef nn_feature_cient pobj_with_feature partmod_feature_encoding ' +o,2819,'In addition to the manual alignment supplied with these data , we create an automatic word alignment for them using GIZA + + and the grow-diagfinal -LRB- GDF -RRB- symmetrization algorithm ',Koehn,'prep_create_In pobj_In_addition prep_addition_to det_alignment_the amod_alignment_manual pobj_to_alignment partmod_alignment_supplied prep_supplied_with det_data_these pobj_with_data nsubj_create_we det_alignment_an amod_alignment_automatic nn_alignment_word dobj_create_alignment prep_create_for nsubj_using_them pcomp_for_using dobj_using_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_GIZA_and det_algorithm_the amod_algorithm_grow-diagfinal abbrev_algorithm_GDF nn_algorithm_symmetrization conj_GIZA_algorithm ' +o,2820,'Adopting the SCF acquisition system of Briscoe and Carroll , we have experimented with an alternative hypothesis test , the binomial log-likelihood ratio -LRB- LLR -RRB- test ',Dunning,'dep_experimented_Adopting det_system_the nn_system_SCF nn_system_acquisition dobj_Adopting_system prep_system_of pobj_of_Briscoe cc_Briscoe_and conj_Briscoe_Carroll nsubj_experimented_we aux_experimented_have prep_experimented_with det_test_an amod_test_alternative nn_test_hypothesis pobj_with_test det_ratio_the amod_ratio_binomial amod_ratio_log-likelihood appos_test_ratio abbrev_test_LLR dep_test_test ' +o,2821,'Similarities are captured from different viewpoints : DP-HWC -LRB- i -RRB- - l This metric corresponds to the HWC metric presented by ',Liu,'nsubjpass_captured_Similarities auxpass_captured_are prep_captured_from amod_viewpoints_different pobj_from_viewpoints dep_viewpoints_DP-HWC appos_viewpoints_i tmod_corresponds_l det_metric_This nsubj_corresponds_metric parataxis_captured_corresponds prep_corresponds_to det_metric_the nn_metric_HWC pobj_to_metric partmod_metric_presented prep_presented_by ' +o,2822,'Model Overall Unknown Word Accuracy Accuracy Baseline , 9672 \% 845 \% J Table 3 Baseline model performance This table also shows the results reported in Ratnaparkhi for COnvenience ',Ratnaparkhi,'nsubj_Unknown_Model advmod_Unknown_Overall nn_Baseline_Word nn_Baseline_Accuracy nn_Baseline_Accuracy nsubj_performance_Baseline num_\%_9672 appos_Baseline_\% num_\%_845 dep_\%_\% nn_performance_J nn_performance_ nn_performance_Table num_performance_3 nn_performance_Baseline nn_performance_model xcomp_Unknown_performance det_table_This nsubj_shows_table advmod_shows_also dep_Unknown_shows det_results_the nsubj_reported_results ccomp_shows_reported prep_reported_in nn__Ratnaparkhi pobj_in_ prep__for pobj_for_COnvenience ' +o,2823,'1142 We show that by using a variant of SVM Anchored SVM Learning with a polynomial kernel , one can learn accurate models for English NP-chunking , base-phrase chunking , and Dutch Named Entity Recognition , on a heavily pruned feature space ',Ramshaw,'nsubj_show_We dep_1142_show complm_learn_that prep_learn_by pcomp_by_using det_variant_a dobj_using_variant prep_variant_of nn__SVM nn__Anchored nn__SVM nn__Learning pobj_of_ prep__with det_kernel_a amod_kernel_polynomial pobj_with_kernel nsubj_learn_one aux_learn_can ccomp_show_learn amod_models_accurate dobj_learn_models prep_learn_for nn__English nn__NP-chunking pobj_for_ amod__base-phrase nn__chunking conj__ cc__and nn__Dutch nn__Named nn__Entity nn__Recognition conj__ prep_learn_on det_space_a advmod_pruned_heavily amod_space_pruned nn_space_feature pobj_on_space ' +o,2824,'Our method is similar to , , and in the use of dependency relationships as the word features ',Hindle,'poss_method_Our nsubj_similar_method cop_similar_is prep_similar_to nn__ conj__ cc__and pobj_to_ prep_similar_in det_use_the pobj_in_use prep_use_of amod_relationships_dependency pobj_of_relationships mark_features_as det_word_the nsubj_features_word advcl_similar_features ' +o,2825,'There are also research work on automatically classifying movie or product reviews as positive or negative ',Pang,'expl_are_There advmod_are_also nn_work_research nsubj_are_work prep_work_on advmod_classifying_automatically pcomp_on_classifying nn_reviews_movie cc_movie_or conj_movie_product nsubj_positive_reviews advmod_positive_as xcomp_classifying_positive cc_positive_or conj_positive_negative ' +o,2826,'As described in Section 3 we retrieved neighbors using similarity measure on a RASP parsed version of the BNC ',Lin,'mark_described_As prep_retrieved_in pobj_in_Section num_Section_3 nsubj_retrieved_we ccomp_described_retrieved dobj_retrieved_neighbors xcomp_retrieved_using amod_measure_ nn_measure_similarity dobj_using_measure prep_using_on det_version_a nn_version_RASP amod_version_parsed nn_version_ pobj_on_version prep_version_of det_BNC_the pobj_of_BNC ' +o,2827,'This curve plots the average labeled attachment score over Basque , Chinese , English , and Turkish as a function of parsing time per token4 Accuracy of only 1 \% below the maximum can be achieved with average processing time of 17 ms per token , or 60 tokens per second5 We also refer the reader to for more detailed analysis of the ISBN dependency parser results , where , among other things , it was shown that the ISBN model is especially accurate at modeling long dependencies ',Titov,'det_plots_This nn_plots_curve det_average_the nsubj_labeled_average dep_plots_labeled nn_score_attachment dobj_labeled_score prep_labeled_over nn_English_Basque appos_English_Chinese pobj_over_English cc_labeled_and nsubjpass_achieved_Turkish prep_Turkish_as det_function_a pobj_as_function prep_function_of amod_time_parsing pobj_of_time prep_time_per amod_Accuracy_token4 pobj_per_Accuracy prep_time_of quantmod_1_only num_\%_1 pobj_of_\% prep_\%_below det_maximum_the pobj_below_maximum aux_achieved_can auxpass_achieved_be conj_labeled_achieved prep_achieved_with amod_time_average amod_time_processing pobj_with_time prep_time_of num_ms_17 pobj_of_ms prep_ms_per pobj_per_token cc_ms_or num_tokens_60 conj_ms_tokens prep_tokens_per pobj_per_second5 nsubj_refer_We advmod_refer_also dep_plots_refer det_reader_the nsubj__reader aux__to xcomp_refer_ prep__for amod_analysis_more amod_analysis_detailed pobj_for_analysis prep_analysis_of det_results_the nn_results_ISBN nn_results_dependency nn_results_parser pobj_of_results advmod_shown_where prep_shown_among amod_things_other pobj_among_things nsubjpass_shown_it auxpass_shown_was advcl_refer_shown complm_accurate_that det_model_the nn_model_ISBN nsubj_accurate_model cop_accurate_is advmod_accurate_especially ccomp_shown_accurate prep_accurate_at amod_dependencies_modeling amod_dependencies_long pobj_at_dependencies ' +o,2828,' extracted hyponymyrelationsfromtherstsentences -LRB- ie , dening sentences -RRB- of Wikipedia articles and then used them as a gazetteer for NER ',Kazama,'nsubj_extracted_ dobj_extracted_hyponymyrelationsfromtherstsentences dep_hyponymyrelationsfromtherstsentences_ie amod_sentences_dening appos_ie_sentences prep_hyponymyrelationsfromtherstsentences_of nn_articles_Wikipedia pobj_of_articles cc_extracted_and advmod_used_then conj_extracted_used dobj_used_them prep_used_as det_gazetteer_a pobj_as_gazetteer prep_gazetteer_for pobj_for_NER ' +o,2829,'The feature functions are combined under a log-linear framework , andtheweights aretuned bytheminimum-error-rate training using BLEU as the optimization metric ',Och,'det_functions_The nn_functions_feature nsubjpass_combined_functions auxpass_combined_are prep_combined_under det_framework_a amod_framework_log-linear pobj_under_framework appos_framework_andtheweights partmod_andtheweights_aretuned amod__bytheminimum-error-rate nn__training dobj_aretuned_ xcomp_aretuned_using nn__BLEU dobj_using_ prep__as det_metric_the nn_metric_optimization pobj_as_metric ' +o,2830,'We extracted tagged sentences from the parse trees5 We split the data into training , development , and test sets as in ',Collins,'nsubj_extracted_We amod_sentences_tagged dobj_extracted_sentences prep_extracted_from det_trees5_the amod_trees5_parse pobj_from_trees5 nsubj_split_We dep_extracted_split det_data_the dobj_split_data prep_split_into nn_development_training pobj_into_development cc_development_and nn_sets_test conj_development_sets prep_split_as dep_as_in ' +o,2831,'-LRB- RM95 -RRB- Lance A ',Ramshaw,'nn_A_Lance dep_RM95_A ' +p,2832,'In the classic work on SMT , Brownandhiscolleagues atIBMintroduced the notion of alignment between a sentence f and its translation e and used it in the development of translation models ',Brown,'prep_atIBMintroduced_In det_work_the amod_work_classic pobj_In_work prep_work_on pobj_on_SMT nsubj_atIBMintroduced_Brownandhiscolleagues det_notion_the dobj_atIBMintroduced_notion prep_notion_of pobj_of_alignment prep_alignment_between det_sentence_a pobj_between_sentence dep_notion_f cc_notion_and poss_translation_its conj_notion_translation dep_notion_e cc_atIBMintroduced_and conj_atIBMintroduced_used dobj_used_it prep_used_in det_development_the pobj_in_development prep_development_of nn_models_translation pobj_of_models ' +o,2833,'Statistical techniques developed for lexicalized grammars , readily apply to CCG to improve the average parsing performance in large-scale practical applications ',Collins,'nn_techniques_Statistical nsubj_developed_techniques dep_apply_developed prep_developed_for amod__lexicalized nn__grammars pobj_for_ advmod_developed_readily aux_CCG_to xcomp_apply_CCG aux_improve_to xcomp_CCG_improve det_performance_the amod_performance_average nn_performance_parsing dobj_improve_performance prep_improve_in amod_applications_large-scale amod_applications_practical pobj_in_applications nsubj_apply_ ' +o,2834,'4An adaptation of the averaged perceptron algorithm is used to tune the model parameters ',Collins,'amod_adaptation_4An nsubjpass_used_adaptation prep_adaptation_of det__the amod__averaged nn__perceptron nn__algorithm pobj_of_ auxpass_used_is aux_tune_to xcomp_used_tune det_parameters_the nn_parameters_model dobj_tune_parameters ' +o,2835,' , -RRB- ',Pang,'appos__ ' +o,2836,'In the SMT research community , the second step has been well studied and many methods have been proposed to speed up the decoding process , such as node-based or span-based beam search with different pruning strategies and cube pruning ',Zhang,'prep_studied_In det_community_the nn_community_SMT nn_community_research pobj_In_community det_step_the amod_step_second nsubjpass_studied_step aux_studied_has auxpass_studied_been advmod_studied_well cc_studied_and conj_studied_many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been ccomp_many_proposed aux_speed_to xcomp_proposed_speed prt_speed_up det_process_the amod_process_decoding dobj_speed_process dep_as_such prep_process_as amod_search_node-based cc_node-based_or conj_node-based_span-based nn_search_beam pobj_as_search prep_search_with amod__different nn__pruning nn__strategies pobj_with_ cc__and nn__cube nn__pruning conj__ ' +o,2837,'In this method , each training sentence is decoded and weights are updated at every iteration ',Liang,'prep_decoded_In det_method_this pobj_In_method det_sentence_each nn_sentence_training nsubjpass_decoded_sentence auxpass_decoded_is cc_decoded_and nsubj_updated_weights aux_updated_are conj_decoded_updated prep_updated_at det_iteration_every pobj_at_iteration ' +o,2838,'Distance from a target word is used for this purpose and it is calculated by the assumption that the target words in the context window have the same sense ',Yarowsky,'dep_Distance_from det_word_a nn_word_target nsubjpass_used_word auxpass_used_is pcomp_from_used prep_used_for det_purpose_this pobj_for_purpose cc_used_and nsubjpass_calculated_it auxpass_calculated_is conj_used_calculated prep_calculated_by det_assumption_the pobj_by_assumption complm_have_that det_words_the nn_words_target nsubj_have_words prep_words_in det_window_the nn_window_context pobj_in_window ccomp_calculated_have det_sense_the amod_sense_same dobj_have_sense ' +o,2839,'5 Phrase Pair Induction A common approach to phrase-based translation is to extract an inventory of phrase pairs -LRB- PPI -RRB- from bitext , For example , in the phraseextract algorithm , a word alignment am1 is generated over the bitext , and all word subsequences ei2i1 and fj2j1 are found that satisfy : am1 : aj -LRB- i1 , i2 -RRB- iff j -LRB- j1 , j2 -RRB- ',Koehn,'num_Induction_5 nn_Induction_Phrase nn_Induction_Pair det_approach_A amod_approach_common nsubj_is_approach prep_approach_to amod_translation_phrase-based pobj_to_translation rcmod_Induction_is aux_extract_to xcomp_is_extract det_inventory_an dobj_extract_inventory prep_inventory_of nn_pairs_phrase pobj_of_pairs abbrev_pairs_PPI prep_extract_from amod__bitext pobj_from_ prep_generated_For pobj_For_example prep_generated_in det__the amod__phraseextract nn__algorithm pobj_in_ det_am1_a nn_am1_word nn_am1_alignment nsubjpass_generated_am1 auxpass_generated_is ccomp_is_generated prep_generated_over det_bitext_the pobj_over_bitext cc_is_and det_ei2i1_all nn_ei2i1_word nn_ei2i1_subsequences nsubjpass_found_ei2i1 cc_ei2i1_and conj_ei2i1_fj2j1 auxpass_found_are conj_is_found nsubj_satisfy_that dep_aj_satisfy advmod_satisfy_am1 dobj_found_aj appos_aj_i1 dep_i1_i2 nn_j_iff dep_Induction_j appos_j_j1 dep_j1_j2 ' +o,2840,'Most approaches inherently extract semantic knowledge in the abstracted form of semantic clusters ',Brown,'amod_approaches_Most nsubj__approaches advmod__inherently dep__extract amod_knowledge_semantic dobj_extract_knowledge prep_extract_in det_form_the amod_form_abstracted pobj_in_form prep_form_of amod_clusters_semantic pobj_of_clusters ' +o,2841,'Besides being linguistically motivated , the need for EDL is also supported by empirical findings in MT that one-level rules are often inadequate ',Galley,'prep_supported_Besides cop_motivated_being advmod_motivated_linguistically pcomp_Besides_motivated det_need_the nsubjpass_supported_need prep_need_for pobj_for_EDL auxpass_supported_is advmod_supported_also prep_supported_by amod_findings_empirical pobj_by_findings prep_findings_in pobj_in_MT det_rules_that amod_rules_one-level nsubj_inadequate_rules cop_inadequate_are advmod_inadequate_often dep_supported_inadequate ' +o,2842,'The class-based approaches calculate co-occurrence data of words belonging to different classes , ~ rather than individual words , to enhance the co-occurrence data collected and to cover words which have low occurrence frequencies ',Brown,'det_approaches_The amod_approaches_class-based nsubj__approaches xcomp__calculate amod_data_co-occurrence dobj_calculate_data prep_data_of pobj_of_words partmod_words_belonging prep_belonging_to amod_classes_different pobj_to_classes appos_classes_~ cc_classes_rather dep_rather_than amod_words_individual conj_classes_words aux_enhance_to ccomp_belonging_enhance det_data_the amod_data_co-occurrence dobj_enhance_data partmod_data_collected cc_enhance_and aux_cover_to conj_enhance_cover dobj_cover_words nsubj_have_which rcmod_words_have amod_frequencies_low nn_frequencies_occurrence dobj_have_frequencies ' +o,2843,'The model scaling factors M1 are trained with respect to the final translation quality measured by an error criterion ',Och,'det_M1_The nn_M1_model nn_M1_scaling nn_M1_factors nsubjpass_trained_M1 auxpass_trained_are prep_trained_with pobj_with_respect prep_trained_to det_quality_the amod_quality_final nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_by det_criterion_an nn_criterion_error pobj_by_criterion ' +o,2844,' first introduced the machine learning techniques to chunking problem ',Ramshaw,'nsubj_introduced_ advmod_introduced_first det_machine_the dobj_introduced_machine partmod_machine_learning dobj_learning_techniques prep_learning_to amod_problem_chunking pobj_to_problem ' +o,2845,'This is in contrast to standard summarization models that look to promote sentence diversity in order to cover as many important topics as possible ',Goldstein,'nsubj_is_This prep_is_in dep_in_contrast dep_in_to amod_models_standard nn_models_summarization pobj_in_models nsubj_look_that rcmod_models_look aux_promote_to xcomp_look_promote nn_diversity_sentence dobj_promote_diversity mark_cover_in dep_cover_order aux_cover_to purpcl_promote_cover prep_cover_as amod_topics_many amod_topics_important pobj_as_topics advmod_possible_as amod_topics_possible ' +o,2846,'The parameters of the NIST systems were tuned using Ochs algorithm to maximize BLEU on the MT02 test set ',Och,'det_parameters_The nsubjpass_tuned_parameters prep_parameters_of det_systems_the amod_systems_NIST pobj_of_systems auxpass_tuned_were dep_tuned_using nn_algorithm_Ochs dobj_using_algorithm aux_maximize_to infmod_algorithm_maximize dobj_maximize_BLEU prep_maximize_on det_set_the nn_set_MT02 nn_set_test pobj_on_set ' +o,2847,'The NP chunks in the shared task data are base-NP chunks which are non-recursive NPs , a definition first proposed by ',Ramshaw,'det_chunks_The nn_chunks_NP nsubj_chunks_chunks prep_chunks_in det_data_the amod_data_shared nn_data_task pobj_in_data cop_chunks_are amod_chunks_base-NP nsubj_NPs_which cop_NPs_are amod_NPs_non-recursive rcmod_chunks_NPs det_definition_a appos_NPs_definition advmod_proposed_first partmod_definition_proposed prep_proposed_by ' +o,2848,'Heuristic approaches obtain word alignments by using various similarity functions between the types of the two languages ',Berger,'amod_approaches_Heuristic nsubj_obtain_approaches nn_alignments_word dobj_obtain_alignments prep_obtain_by pcomp_by_using amod_functions_various nn_functions_similarity dobj_using_functions prep_functions_between det_types_the pobj_between_types prep_types_of det_languages_the num_languages_two pobj_of_languages advmod_using_ ' +o,2849,'Proceedings of EACL \'99 Determinants of Adjective-Noun Plausibility Maria Lapata and Scott McDonald and Frank Keller School of Cognitive Science Division of Informatics, University of Edinburgh 2 Buccleuch Place, Edinburgh EH8 9LW, UK {mlap, scottm, keller} @cogsci.ed.ac.uk Abstract This paper explores the determinants of adjective-noun plausibility by using correlation analysis to compare judgements elicited from human subjects with five corpus-based variables: co-occurrence frequency of the adjective-noun pair, noun frequency, conditional probability of the noun given the adjective, the log-likelihood ratio, and Resnik\'s (1993) selectional association measure.',Dunning,'' +o,2850,' et al 1993 -RRB- ',Brown,'nn_1993_ nn_1993_et nn_1993_al ' +o,2851,'We used minimum error rate training and the A \\* beam search decoder implemented by Koehn ',Koehn,'partmod_We_used amod__minimum nn__error nn__rate nn__training dobj_used_ cc__and det_A_the conj__A dep_decoder_\\* nn_decoder_beam nn_decoder_search dep_We_decoder partmod_decoder_implemented prep_implemented_by pobj_by_Koehn ' +o,2852,'In cases like , unsupervised methods offer accuracy results than rival supervised methods while requiring only a fraction of the data preparation effort ',Yarowsky,'prep_offer_In pobj_In_cases prep_cases_like pobj_like_ amod_methods_unsupervised nsubj_offer_methods nn_results_accuracy dobj_offer_results prep_offer_than amod_methods_rival amod_methods_supervised pobj_than_methods partmod_methods_ prep__while pcomp_while_requiring advmod_fraction_only det_fraction_a dobj_requiring_fraction prep_fraction_of det_effort_the nn_effort_data nn_effort_preparation pobj_of_effort ' +o,2853,'The most popular non-data-splitting methods for predicting test set cross-entropy -LRB- or likelihood -RRB- are AIC and variants such as AICc , quasi-AIC -LRB- QAIC -RRB- , and QAICc ',Brown,'det_methods_The advmod_popular_most amod_methods_popular amod_methods_non-data-splitting nsubj_AIC_methods prep_methods_for pcomp_for_predicting dobj_predicting_test partmod_test_set advmod_set_cross-entropy dep_-LRB-_or dep_-LRB-_likelihood cop_AIC_are cc_AIC_and conj_AIC_variants dep_as_such prep_AIC_as pobj_as_AICc amod_QAIC_quasi-AIC conj_AICc_QAIC cc_AICc_and nn__QAICc conj_AICc_ ' +o,2854,'We use as our English corpus the Wall Street Journal -LRB- WSJ -RRB- portion of the Penn Treebank ',Marcus,'nsubj_use_We prep_use_as poss_corpus_our amod_corpus_English pobj_as_corpus det_portion_the nn_Journal_Wall nn_Journal_Street nn_portion_Journal abbrev_Journal_WSJ dobj_use_portion prep_portion_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,2855,'Though inter-rater reliability using the kappa statistic may be calculated for each group , the distribution of categories in the contribution group was highly skewed and warrants further discussion ',Carletta,'mark_calculated_Though amod_reliability_inter-rater nsubjpass_calculated_reliability partmod_reliability_using det__the nn__kappa nn__statistic dobj_using_ aux_calculated_may auxpass_calculated_be advcl_skewed_calculated prep_calculated_for det_group_each pobj_for_group det_distribution_the nsubj_skewed_distribution prep_distribution_of pobj_of_categories prep_categories_in det_group_the nn_group_contribution pobj_in_group cop_skewed_was advmod_skewed_highly cc_skewed_and conj_skewed_warrants amod_discussion_further dobj_warrants_discussion ' +o,2856,'Next , using our feature vector , we applied five different linear classifiers to extract PPI from AIMed : L2-SVM , 1-norm soft-margin SVM -LRB- L1-SVM -RRB- , logistic regression -LRB- LR -RRB- , averaged perceptron -LRB- AP -RRB- , and confidence weighted linear classification -LRB- CW -RRB- ',Collins,'advmod_applied_Next dep_applied_using poss_vector_our nn_vector_feature dobj_using_vector nsubj_applied_we num_classifiers_five amod_classifiers_different amod_classifiers_linear dobj_applied_classifiers aux_extract_to xcomp_applied_extract dobj_extract_PPI prep_extract_from pobj_from_AIMed nsubj_averaged_L2-SVM amod_SVM_1-norm amod_SVM_soft-margin appos_L2-SVM_SVM appos_SVM_L1-SVM amod__logistic nn__regression abbrev__LR appos_SVM_ parataxis_applied_averaged amod__perceptron abbrev__AP dobj_averaged_ cc_applied_and nn_classification_confidence amod_classification_weighted amod_classification_linear nsubj__classification abbrev_classification_CW conj_applied_ ' +o,2857,'6 Coding reliability The reliability of the annotation was evaluated using the kappa statistic ',Carletta,'num_Coding_6 dep_evaluated_Coding advmod_evaluated_reliability det_reliability_The nsubjpass_evaluated_reliability prep_reliability_of det_annotation_the pobj_of_annotation auxpass_evaluated_was xcomp_evaluated_using det_statistic_the nn_statistic_kappa dobj_using_statistic ' +o,2858,'We implement this algorithm using the perceptron framework , as it can be easily modified for structured prediction while preserving convergence guarantees ','Daume III','nsubj_implement_We det_algorithm_this dobj_implement_algorithm partmod_algorithm_using det_framework_the nn_framework_perceptron dobj_using_framework mark_modified_as nsubjpass_modified_it aux_modified_can auxpass_modified_be advmod_modified_easily advcl_using_modified prep_modified_for amod_prediction_structured pobj_for_prediction prep_modified_while pcomp_while_preserving amod_guarantees_convergence dobj_preserving_guarantees ' +o,2859,'The word alignment used in GHKM is usually computed independent ofthesyntacticstructure,andasDeNeroandKlein (2007) and May and Knight (2007) have noted, Ch-En En-Ch Union Heuristic 28.6\% 33.0\% 45.9\% 20.1\% Table 1: Percentage of corpus used to generate big templates, based on different word alignments 9-12 13-20 21 Ch-En 18.2\% 17.4\% 64.4\% En-Ch 15.9\% 20.7\% 63.4\% Union 9.8\% 15.1\% 75.1\% Heuristic 24.6\% 27.9\% 47.5\% Table 2: In the selected big templates, the distribution of words in the templates of different sizes, which are measured based on the number of symbols in their RHSs is not the best for SSMT systems.',DeNero,'' +o,2860,'21 Likelihood Ratios in the Type-based Stage The log-likelihood ratio by tests whether the probability of a word is dependent on the occurrence of the preceding word type ',Dunning,'aux_ratio_21 nn_Ratios_Likelihood nsubj_ratio_Ratios prep_Ratios_in det_Stage_the amod_Stage_Type-based pobj_in_Stage det_ratio_The amod_ratio_log-likelihood prep_ratio_by amod_tests_ pobj_by_tests dep_dependent_whether det_probability_the nsubj_dependent_probability prep_probability_of det_word_a pobj_of_word cop_dependent_is ccomp_ratio_dependent prep_dependent_on det_occurrence_the pobj_on_occurrence prep_occurrence_of det_type_the amod_type_preceding nn_type_word pobj_of_type ' +o,2861,'Word alignment models were first introduced in statistical machine translation ',Brown,'amod_models_Word nn_models_alignment nsubjpass_introduced_models auxpass_introduced_were advmod_introduced_first prep_introduced_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,2862,'Combining statistical and parsing methods has been done by and ',Hindle,'nsubjpass_done_Combining dep_Combining_statistical cc_Combining_and amod_methods_parsing conj_Combining_methods aux_done_has auxpass_done_been prep_done_by pobj_by_ cc__and conj__ ' +p,2863,' work is perhaps one of the most notable examples of unsupervised polarity classification ',Turney,'amod_work_ nsubj_one_work cop_one_is advmod_one_perhaps prep_one_of det_examples_the advmod_notable_most amod_examples_notable pobj_of_examples prep_examples_of amod_classification_unsupervised amod_classification_polarity pobj_of_classification ' +o,2864,'The other is the self-training which first parses and reranks the NANC corpus , and then use them as additional training data to retrain the model ',McClosky,'det_other_The nsubj__other cop__is det__the amod__self-training nsubj_parses_which advmod_parses_first rcmod__parses cc_parses_and conj_parses_reranks det_corpus_the nn_corpus_NANC dobj_reranks_corpus cc_parses_and advmod_parses_then conj_parses_use dobj_use_them mark_retrain_as amod_data_additional nn_data_training nsubj_retrain_data aux_retrain_to advcl_use_retrain det_model_the dobj_retrain_model ' +o,2865,'3 The Effect of Training Corpus Size A number of past research work on WSD , such as , were tested on a small number of words like ` line \' and ` interest \' ',Marcus,'dep_Effect_3 det_Effect_The nsubjpass_tested_Effect prep_Effect_of nn_Size_Training nn_Size_Corpus pobj_of_Size det_number_A dep_Size_number prep_Effect_of amod_work_past nn_work_research pobj_of_work prep_work_on pobj_on_WSD dep_as_such prep_Effect_as pobj_as_ auxpass_tested_were prep_tested_on det_number_a amod_number_small pobj_on_number prep_number_of pobj_of_words prep_words_like pobj_like_line cc_line_and conj_line_interest ' +o,2866,'Among various language modeling approaches , ngram modeling has been widely used in many applications , such as speech recognition , machine translation ',Brown,'prep_used_Among amod_approaches_various nn_approaches_language nn_approaches_modeling pobj_Among_approaches nn_modeling_ngram nsubjpass_used_modeling aux_used_has auxpass_used_been advmod_used_widely prep_used_in amod_applications_many pobj_in_applications dep_as_such prep_applications_as nn_recognition_speech pobj_as_recognition nn_translation_machine appos_recognition_translation ' +o,2867,'Annotation was highly reliable with a kappa of 3https : \\/ \\/ wwwciagov\\/cia\\/publications \\/ factbook\\/indexhtml 4Given that the task is not about standard Named Entity Recognition , we assume that the general semantic class of the name is already known ',Carletta,'nsubj_reliable_Annotation cop_reliable_was advmod_reliable_highly prep_reliable_with det__a nn__kappa pobj_with_ prep__of pobj_of_3https nn_\\/_\\/ nsubj_4Given_\\/ dep_4Given_wwwciagov\\/cia\\/publications amod_4Given_\\/ amod_4Given_factbook\\/indexhtml ccomp_assume_4Given complm_is_that det_task_the nsubj_is_task ccomp_4Given_is neg_is_not prep_is_about amod_Recognition_standard nn_Recognition_Named nn_Recognition_Entity pobj_about_Recognition nsubj_assume_we parataxis_reliable_assume complm_known_that det_class_the amod_class_general amod_class_semantic nsubjpass_known_class prep_class_of det_name_the pobj_of_name auxpass_known_is advmod_known_already ccomp_assume_known ' +o,2868,'When we have a junction tree for each document , we can efficiently perform belief propagation in order to compute argmax in Equation -LRB- 1 -RRB- , or the marginal probabilities of cliques and labels , necessary for the parameter estimation of machine learning classifiers , including perceptrons , and maximum entropy models ',Berger,'advmod_have_When nsubj_have_we dep_perform_have det_tree_a nn_tree_junction dobj_have_tree prep_tree_for det_document_each pobj_for_document nsubj_perform_we aux_perform_can advmod_perform_efficiently nn_propagation_belief dobj_perform_propagation mark_compute_in dep_compute_order aux_compute_to purpcl_perform_compute dobj_compute_argmax prep_compute_in pobj_in_Equation appos_Equation_1 cc_Equation_or det_probabilities_the amod_probabilities_marginal conj_Equation_probabilities prep_probabilities_of pobj_of_cliques cc_cliques_and conj_cliques_labels amod_Equation_necessary prep_necessary_for det_estimation_the nn_estimation_parameter pobj_for_estimation prep_estimation_of nn_classifiers_machine nn_classifiers_learning pobj_of_classifiers prep_Equation_including amod__perceptrons pobj_including_ cc__and amod_models_maximum nn_models_entropy conj__models ' +o,2869,'Standard SMT alignment models are used to align letter-pairs within named entity pairs for transliteration ',Brown,'nn__Standard nn__SMT nn__alignment nn__models nsubjpass_used_ auxpass_used_are prep_used_to nn_letter-pairs_align pobj_to_letter-pairs prep_used_within amod_pairs_named nn_pairs_entity pobj_within_pairs prep_pairs_for pobj_for_transliteration ' +o,2870,'There has been a large interest in recognizing non-overlapping noun phrases -LRB- and follow-up papers -RRB- but relatively little has been written about identifying phrases of other syntactic categories ',Ramshaw,'expl_been_There aux_been_has det_interest_a amod_interest_large dobj_been_interest prep_interest_in pcomp_in_recognizing amod_phrases_non-overlapping nn_phrases_noun dobj_recognizing_phrases dep_phrases_ cc__and amod_papers_follow-up conj__papers cc_been_but advmod_little_relatively nsubjpass_written_little aux_written_has auxpass_written_been conj_been_written prep_written_about pcomp_about_identifying dobj_identifying_phrases prep_phrases_of amod_categories_other amod_categories_syntactic pobj_of_categories ' +o,2871,'Several approaches for learning from both labeled and unlabeled data have been proposed where the unlabeled data is utilised to boost the performance of the algorithm ',Yarowsky,'amod_approaches_Several nsubjpass_proposed_approaches prep_approaches_for pcomp_for_learning prep_learning_from preconj_labeled_both amod_data_labeled cc_labeled_and conj_labeled_unlabeled pobj_from_data aux_proposed_have auxpass_proposed_been xcomp_proposed_ advmod_utilised_where det_data_the amod_data_unlabeled nsubjpass_utilised_data auxpass_utilised_is advcl__utilised aux_boost_to xcomp_utilised_boost det_performance_the dobj_boost_performance prep_performance_of det_algorithm_the pobj_of_algorithm ' +o,2872,'96 Research on DA classification initially focused on two-party conversational speech and , more recently , has extended to multi-party audio recordings like the ICSI corpus ',Berger,'num_Research_96 dep_Research_on nn_classification_DA pobj_on_classification advmod_focused_initially dep_classification_focused prep_focused_on amod__two-party amod__conversational nn__speech pobj_on_ cc_focused_and advmod_recently_more dep_focused_recently aux_extended_has conj_focused_extended prep_extended_to amod_recordings_multi-party amod_recordings_audio pobj_to_recordings prep_recordings_like det__the nn__ICSI nn__corpus pobj_like_ ' +o,2873,'Charniak 1996 , 1997 -RRB- , while most current stochastic parsing models use a ` markov grammar \' ',Collins,'nsubj_Charniak_1996 appos_1996_1997 mark_use_while advmod_models_most amod_models_current nn_models_stochastic nn_models_parsing nsubj_use_models dep_1997_use det_grammar_a amod_grammar_markov dobj_use_grammar ' +o,2874,'The typical practice of preprocessing distributional data is to remove rare word co-occurrences , thus aiming to reduce noise from idiosyncratic word uses and linguistic processing errors and at the same time form more compact word representations ',Grefenstette,'det_practice_The amod_practice_typical nsubj_is_practice prep_practice_of pcomp_of_preprocessing amod_data_distributional dobj_preprocessing_data aux_remove_to xcomp_is_remove amod_co-occurrences_rare nn_co-occurrences_word dobj_remove_co-occurrences advmod_aiming_thus xcomp_remove_aiming aux_reduce_to xcomp_aiming_reduce dobj_reduce_noise prep_reduce_from amod_uses_idiosyncratic nn_uses_word pobj_from_uses cc_uses_and amod_errors_linguistic nn_errors_processing conj_uses_errors cc_from_and conj_from_at det_form_the amod_form_same nn_form_time pobj_at_form advmod_representations_more amod_representations_compact nn_representations_word dep_from_representations ' +o,2875,'4 Related Work The automatic extraction of English subcategorization frames has been considered in , where a procedure is presented that takes untamed text as input and generates a list of verbal subcategorization frames ',Dunning,'num_Work_4 amod_Work_Related det_extraction_The amod_extraction_automatic nsubjpass_considered_extraction prep_extraction_of amod_frames_English nn_frames_subcategorization pobj_of_frames aux_considered_has auxpass_considered_been dep_Work_considered prep_considered_in pobj_in_ advmod_presented_where det_procedure_a nsubjpass_presented_procedure auxpass_presented_is rcmod__presented complm_takes_that ccomp_presented_takes amod_text_untamed dobj_takes_text prep_takes_as pobj_as_input cc_takes_and conj_takes_generates det_list_a dobj_generates_list prep_list_of amod_frames_verbal nn_frames_subcategorization pobj_of_frames ' +p,2876,'a2 Maximum-entropy method The maximum-entropy method is useful with sparse data conditions and has been used by many researchers ',Ratnaparkhi,'amod_method_a2 amod_method_Maximum-entropy det_method_The amod_method_maximum-entropy nsubjpass_useful_method cop_useful_is dep_method_useful prep_useful_with amod_conditions_sparse nn_conditions_data pobj_with_conditions cc_useful_and aux_used_has auxpass_used_been conj_useful_used prep_used_by amod_researchers_many pobj_by_researchers ' +o,2877,'The algorithm is similar to the perceptron algorithm described in ',Collins,'det_algorithm_The nsubj_similar_algorithm cop_similar_is prep_similar_to det_algorithm_the nn_algorithm_perceptron pobj_to_algorithm partmod_algorithm_described prep_described_in ' +o,2878,'In our experiments we use standard methods in phrase-based systems to define the set of phrase entries for each sentence in training data ',Koehn,'prep_use_In poss_experiments_our pobj_In_experiments nsubj_use_we amod_methods_standard nsubj__methods prep_methods_in amod_systems_phrase-based pobj_in_systems ccomp_use_ aux_define_to xcomp__define det_set_the dobj_define_set prep_set_of nn_entries_phrase pobj_of_entries prep_define_for det_sentence_each pobj_for_sentence prep_sentence_in nn_data_training pobj_in_data ' +o,2879,'In additioil , , point ou ; that there is a st , rent tenden -LRB- : y for words 1 ; O occur in -LRB- -RCB- Ile sense within any given dis : ourse -LRB- ` one sense pe , r dis : ourse \' -RRB- ',Yarowsky,'amod_ou_additioil dep_additioil_ dep_additioil_ nn_ou_point pobj_In_ou dep_is_that expl_is_there dep_ou_is dep_sense_a dep_sense_st nn_tenden_rent nn_sense_tenden dep_-LRB-_y dep_y_for pobj_for_words dep_y_1 nn_occur_O dep_1_occur prep_occur_in nn_sense_Ile nsubj_is_sense prep_sense_within det_dis_any amod_dis_given pobj_within_dis dep_sense_ourse nsubj_sense_one dep_ourse_sense dobj_sense_pe appos_pe_r dep_sense_dis dep_dis_ourse ' +o,2880,'This is a particularly exciting area in computational linguistics as evidenced by the large number of contributions in these special issues : Biber , Brent , Hindle and Rooth -LRB- this issue -RRB- , Pustejovsky et al ',Brown,'nsubj_area_This cop_area_is det_area_a advmod_exciting_particularly amod_area_exciting prep_area_in amod_linguistics_computational pobj_in_linguistics dep_evidenced_as dep_linguistics_evidenced prep_evidenced_by det_number_the amod_number_large pobj_by_number prep_number_of pobj_of_contributions prep_contributions_in det_issues_these amod_issues_special pobj_in_issues nn__Biber dep_evidenced_ nn__Brent appos_linguistics_ conj__Hindle cc__and conj__Rooth det_issue_this appos__issue appos__Pustejovsky cc_Pustejovsky_et conj_Pustejovsky_al ' +o,2881,'Treebank , six of which are errors ',Marcus,'num_Treebank_ num_Treebank_six prep_Treebank_of nsubj_errors_which cop_errors_are pcomp_of_errors ' +o,2882,'Stochastic ITGs are parameterized like their PCFG counterparts ; productions A X are assigned probability Pr -LRB- X A -RRB- ',Wu,'nn_ITGs_Stochastic nsubjpass_parameterized_ITGs auxpass_parameterized_are prep_parameterized_like poss__their nn__PCFG nn__counterparts pobj_like_ nn_X_productions nn_X_A nsubjpass_assigned_X auxpass_assigned_are parataxis_parameterized_assigned nn_Pr_probability dobj_assigned_Pr nn_A_X appos_Pr_A ' +o,2883,'Introduction Word sense disambiguation has long been one of the major concerns in natural language processing area , whose aim is to identify the correct sense of a word in a particular context , among all of its senses defined in a dictionary or a thesaurus ',Yarowsky,'nn_disambiguation_Introduction nn_disambiguation_Word nn_disambiguation_sense aux_been_has advmod_been_long dep_disambiguation_been dobj_been_one prep_one_of det_concerns_the amod_concerns_major pobj_of_concerns prep_concerns_in amod__natural nn__language nn__processing nn__area pobj_in_ poss_aim_whose nsubj_is_aim rcmod__is aux_identify_to xcomp_is_identify det_sense_the amod_sense_correct dobj_identify_sense prep_sense_of det_word_a pobj_of_word prep_identify_in det_context_a amod_context_particular pobj_in_context prep_identify_among pobj_among_all prep_all_of poss_senses_its pobj_of_senses partmod_senses_defined prep_defined_in det_dictionary_a pobj_in_dictionary cc_dictionary_or det_thesaurus_a conj_dictionary_thesaurus ' +o,2884,'As a learning algorithm for our classification model , we used Maximum Entropy ',Berger,'prep_used_As det_algorithm_a nn_algorithm_learning pobj_As_algorithm prep_algorithm_for poss_model_our nn_model_classification pobj_for_model nsubj_used_we nn_Entropy_Maximum dobj_used_Entropy ' +o,2885,'Their weights are calculated by deleted interpolation ',Brown,'poss_weights_Their nsubjpass_calculated_weights auxpass_calculated_are prep_calculated_by amod_interpolation_deleted pobj_by_interpolation ' +o,2886,'Many NLP systems use the output of supervised parsers -LRB- eg , for QA , for IE , for SRL , for Textual Inference and for MT -RRB- ',Srikumar,'amod_systems_Many nn_systems_NLP nsubj_use_systems det_output_the dobj_use_output prep_output_of amod_parsers_supervised pobj_of_parsers dep_parsers_eg conj_eg_ prep__for pobj_for_QA conj_eg_ prep__for pobj_for_IE conj_eg_ prep__for pobj_for_SRL conj_eg_ prep__for nn_Inference_Textual pobj_for_Inference cc_eg_and conj_eg_ prep__for pobj_for_MT ' +o,2887,'Our evaluation metric is BLEU-4 , as calculated by the script mteval-v11bpl with its default setting except that we used case-sensitive matching of n-grams ',Papineni,'poss_metric_Our nn_metric_evaluation nsubj__metric cop__is nn__BLEU-4 mark_calculated_as dep__calculated prep_calculated_by det_mteval-v11bpl_the nn_mteval-v11bpl_script pobj_by_mteval-v11bpl prep_calculated_with poss_default_its pobj_with_default partmod_default_setting prep_setting_except complm_used_that nsubj_used_we pcomp_except_used amod_matching_case-sensitive dobj_used_matching prep_matching_of pobj_of_n-grams ' +o,2888,'Yarowsky proposed the unsupervised learning method for WS ',Yarowsky,'advmod_proposed_Yarowsky det_method_the amod_method_unsupervised nn_method_learning nsubj_proposed_method prep_method_for pobj_for_WS ' +o,2889,'We also combine our basic algorithm -LRB- Section 42 -RRB- with s algorithm in order to resolve the modifier-function traces ',Cahill,'nsubj_combine_We advmod_combine_also poss_algorithm_our amod_algorithm_basic dobj_combine_algorithm appos_algorithm_Section num_Section_42 prep_combine_with amod_algorithm_ nn_algorithm_s pobj_with_algorithm mark_resolve_in dep_resolve_order aux_resolve_to purpcl_combine_resolve det_traces_the amod_traces_modifier-function dobj_resolve_traces ' +o,2890,'7An alternative framework that formally describes some dependency parsers is that of transition systems ',McDonald,'amod_framework_7An amod_framework_alternative nsubj_that_framework nsubj_describes_that advmod_describes_formally rcmod_framework_describes det_parsers_some amod_parsers_dependency dobj_describes_parsers cop_that_is prep_that_of nn_systems_transition pobj_of_systems ' +o,2891,'Words surrounding the current word have been occasionally used in taggers , such as , Brills transformation based tagger , and the HMM model of Lee et al ',Ratnaparkhi,'nsubjpass_used_Words partmod_Words_surrounding det_word_the amod_word_current dobj_surrounding_word aux_used_have auxpass_used_been advmod_used_occasionally prep_used_in pobj_in_taggers dep_as_such prep_taggers_as amod_transformation_ amod_transformation_Brills pobj_as_transformation partmod_transformation_based nn__tagger dobj_based_ cc_taggers_and det_model_the nn_model_HMM conj_taggers_model prep_model_of pobj_of_Lee cc_Lee_et conj_Lee_al ' +o,2892,'Recent work emphasizes a corpus-based unsupervised approach that avoids the need for costly truthed training data ',Yarowsky,'amod_work_Recent nsubj_emphasizes_work det__a amod__corpus-based amod__unsupervised nn__approach dobj_emphasizes_ nsubj_avoids_that rcmod__avoids det_need_the dobj_avoids_need prep_need_for amod_data_costly amod_data_truthed nn_data_training pobj_for_data ' +p,2893,'NJ 08903 USA suzanne ~ ruccs , rutgers , edu Empirically-induced models that learn a linguistically meaningflll grammar seem to give tile best practical results in statistical natural language processing ',Collins,'nn_USA_NJ num_USA_08903 nsubj_suzanne_USA amod_ruccs_~ nsubj_edu_ruccs appos_ruccs_rutgers ccomp_suzanne_edu amod_models_Empirically-induced dobj_edu_models nsubj_learn_that rcmod_models_learn det__a advmod_meaningflll_linguistically amod__meaningflll nn__grammar nsubj_seem_ ccomp_learn_seem aux_give_to xcomp_seem_give iobj_give_tile dep_practical_best amod_results_practical dobj_give_results prep_results_in amod_processing_statistical amod_processing_natural nn_processing_language pobj_in_processing ' +p,2894,'Freund and Schapire originally proposed the averaged parameter method ; it was shown to give substantial improvements in accuracy for tagging tasks in ',Collins,'nn__Freund cc_Freund_and conj_Freund_Schapire nsubj_proposed_ advmod_proposed_originally det_method_the amod_method_averaged nn_method_parameter dobj_proposed_method nsubjpass_shown_it auxpass_shown_was parataxis_proposed_shown aux_give_to xcomp_shown_give amod_improvements_substantial dobj_give_improvements prep_improvements_in pobj_in_accuracy prep_give_for pcomp_for_tagging dobj_tagging_tasks prep_tagging_in ' +o,2895,'1 Introduction Word alignment is a critical component in training statistical machine translation systems and has received a significant amount of research , for example , , including work leveraging syntactic parse trees , eg , ',DeNero,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubj_component_alignment cop_component_is det_component_a amod_component_critical prep_component_in amod_systems_training amod_systems_statistical nn_systems_machine nn_systems_translation pobj_in_systems cc_component_and aux_received_has conj_component_received det_amount_a amod_amount_significant dobj_received_amount prep_amount_of pobj_of_research prep_received_for pobj_for_example conj_component_ conj_component_including pobj_including_work partmod_work_leveraging amod_trees_syntactic nn_trees_parse dobj_leveraging_trees appos_trees_eg ' +p,2896,'Phrase-based decoding is a dominant formalism in statistical machine translation ',Koehn,'amod__Phrase-based nn__decoding nsubj_formalism_ cop_formalism_is det_formalism_a amod_formalism_dominant prep_formalism_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,2897,'Recently have sought to acquire automatically some semantic patterns that can be used as contextual information to improve reference resolution , using techniques adapted from information extraction ',Bean,'advmod_sought_Recently advmod_sought_ aux_sought_have aux_acquire_to xcomp_sought_acquire advmod_acquire_automatically det_patterns_some amod_patterns_semantic nsubj_sought_patterns nsubjpass_used_that aux_used_can auxpass_used_be rcmod_patterns_used prep_used_as amod_information_contextual pobj_as_information aux_improve_to purpcl_used_improve nn_resolution_reference dobj_improve_resolution xcomp_improve_using dobj_using_techniques partmod_techniques_adapted prep_adapted_from nn_extraction_information pobj_from_extraction ' +o,2898,'A standard solution is to use a weighted linear mixture of N-gram models , 1 n N , ',Brown,'det_solution_A amod_solution_standard nsubj_is_solution aux_use_to xcomp_is_use det_mixture_a amod_mixture_weighted amod_mixture_linear dobj_use_mixture prep_mixture_of amod_models_N-gram pobj_of_models num_N_1 nn_N_n appos_models_N ' +o,2899,'Recently , confusion-network-based system combination algorithms have been developed to combine outputs of multiple machine translation -LRB- MT -RRB- systems to form a consensus output ',Rosti,'advmod_developed_Recently amod_algorithms_confusion-network-based nn_algorithms_system nn_algorithms_combination nsubjpass_developed_algorithms aux_developed_have auxpass_developed_been aux_combine_to xcomp_developed_combine dobj_combine_outputs prep_outputs_of amod_translation_multiple nn_translation_machine pobj_of_translation abbrev_outputs_MT dep_outputs_systems aux_form_to infmod_systems_form det_output_a nn_output_consensus dobj_form_output ' +p,2900,'However , other types of nonlocal information have also been shown to be effective and we will examine the effectiveness of other non-local information which can be embedded into label information ',Finkel,'advmod_shown_However amod_types_other nsubjpass_shown_types prep_types_of amod_information_nonlocal pobj_of_information aux_shown_have advmod_shown_also auxpass_shown_been aux_effective_to cop_effective_be xcomp_shown_effective advmod_effective_ cc_shown_and nsubj_examine_we aux_examine_will conj_shown_examine det_effectiveness_the dobj_examine_effectiveness prep_effectiveness_of amod_information_other amod_information_non-local pobj_of_information nsubjpass_embedded_which aux_embedded_can auxpass_embedded_be rcmod_effectiveness_embedded prep_embedded_into nn_information_label pobj_into_information ' +o,2901,'There are several basic methods for evaluating associations between words : based on frequency counts , information theoretic and statistical significance ',Smadja,'expl_are_There amod_methods_several amod_methods_basic nsubj_are_methods prep_methods_for pcomp_for_evaluating dobj_evaluating_associations prep_associations_between pobj_between_words prep_methods_based dep_based_on nn__frequency nn__counts pobj_on_ nn__information amod__theoretic conj__ cc__and amod_significance_statistical conj__significance ' +o,2902,'A possible solution to this problem is to directly estimate p -LRB- A w -RRB- by applying a maximum entropy model ',Berger,'det_solution_A amod_solution_possible nsubj_is_solution prep_solution_to det_problem_this pobj_to_problem aux_estimate_to advmod_estimate_directly xcomp_is_estimate dobj_estimate_p det_w_A appos_p_w prep_estimate_by pcomp_by_applying det_model_a amod_model_maximum nn_model_entropy dobj_applying_model ' +o,2903,'We used the preprocessed data to train the phrase-based translation model by using GIZA + + and the Pharaoh tool kit ',Koehn,'nsubj_used_We det_data_the amod_data_preprocessed dobj_used_data aux_train_to xcomp_used_train det_model_the amod_model_phrase-based nn_model_translation dobj_train_model prep_train_by pcomp_by_using nn__GIZA nn__+ nn__+ dobj_using_ cc__and det_kit_the nn_kit_Pharaoh nn_kit_tool conj__kit ' +o,2904,'We estimate loss gradients -LRB- Equation 13 -RRB- using a sample of the inference set , which gives a 100-fold increase in training speed ',Turian,'nsubj_estimate_We nn_gradients_loss nsubj_using_gradients appos_gradients_Equation num_Equation_13 dep_estimate_using det_sample_a dobj_using_sample prep_sample_of det_set_the amod_set_inference pobj_of_set nsubj_gives_which rcmod_set_gives det_increase_a amod_increase_100-fold dobj_gives_increase prep_increase_in nn_speed_training pobj_in_speed ' +o,2905,'To quickly -LRB- and approximately -RRB- evaluate this phenomenon , we trained the statistical IBM wordalignment model 4 ,1 using the GIZA + + software for the following language pairs : ChineseEnglish , Italian English , and DutchEnglish , using the IWSLT-2006 corpus for the first two language pairs , and the Europarl corpus for the last one ',Och,'aux_evaluate_To advmod_evaluate_quickly dep_-LRB-_and dep_-LRB-_approximately dep_trained_evaluate det_phenomenon_this dobj_evaluate_phenomenon nsubj_trained_we det_model_the amod_model_statistical nn_model_IBM nn_model_wordalignment iobj_trained_model number__4 dobj_trained_ dobj_trained_,1 xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__software dobj_using_ prep__for det_pairs_the amod_pairs_following nn_pairs_language pobj_for_pairs nn_English_ChineseEnglish amod_English_Italian dep__English cc_English_and conj_English_DutchEnglish partmod_English_using det__the nn__IWSLT-2006 nn__corpus dobj_using_ prep__for det_pairs_the amod_pairs_first num_pairs_two nn_pairs_language pobj_for_pairs cc__and det__the nn__Europarl nn__corpus conj__ prep__for det_one_the amod_one_last pobj_for_one ' +o,2906,'Perhaps the most related is 86 learning as search optimization -LRB- LASO -RRB- ','Daume III','dep_related_Perhaps det_related_the advmod_related_most cop_86_is dep_related_86 xcomp_86_learning prep_learning_as nn_optimization_search pobj_as_optimization abbrev_optimization_LASO ' +o,2907,'Experiments We have conducted a series of lexical acquisition experiments with the above algorithm on largescale English corpora , eg , the Brown corpus and the PTB WSJ corpus ',Marcus,'nsubj_conducted_We aux_conducted_have dep_Experiments_conducted det_series_a dobj_conducted_series prep_series_of amod_experiments_lexical nn_experiments_acquisition pobj_of_experiments prep_conducted_with det_algorithm_the amod_algorithm_above pobj_with_algorithm prep_conducted_on amod_corpora_largescale amod_corpora_English pobj_on_corpora conj_corpora_eg det__the nn__Brown nn__corpus conj_corpora_ cc_corpora_and det_corpus_the nn_corpus_PTB nn_corpus_WSJ conj_corpora_corpus ' +o,2908,'To avoid this problem , generative models for NLP tasks have often been manually designed to achieve an appropriate representation of the joint distribution , such as in the parsing models of ',Collins,'aux_avoid_To dep_designed_avoid det_problem_this dobj_avoid_problem amod_models_generative nsubjpass_designed_models prep_models_for nn_tasks_NLP pobj_for_tasks aux_designed_have advmod_designed_often auxpass_designed_been advmod_designed_manually aux_achieve_to xcomp_designed_achieve det_representation_an amod_representation_appropriate dobj_achieve_representation prep_representation_of det_distribution_the amod_distribution_joint pobj_of_distribution dep_in_such dep_in_as prep_designed_in det_models_the amod_models_parsing pobj_in_models prep_models_of ' +o,2909,'Although grammatical function and empty nodes annotation expressing long-distance dependencies are provided in Treebanks such as the Penn Treebank , most statistical Treebank trained parsers fully or largely ignore them 1 , which entails two problems : first , the training can not profit from valuable annotation data ',Marcus,'mark_provided_Although amod_function_grammatical nsubjpass_provided_function cc_function_and amod_annotation_empty nn_annotation_nodes conj_function_annotation partmod_function_expressing amod_dependencies_long-distance dobj_expressing_dependencies auxpass_provided_are advcl_ignore_provided prep_provided_in pobj_in_Treebanks dep_as_such prep_Treebanks_as det__the nn__Penn nn__Treebank pobj_as_ advmod_parsers_most amod_parsers_statistical nn_parsers_Treebank amod_parsers_trained nsubj_ignore_parsers advmod_ignore_fully cc_fully_or conj_fully_largely nsubj_1_them xcomp_ignore_1 nsubj_entails_which rcmod_1_entails num_problems_two dobj_entails_problems advmod_profit_first det_training_the nsubj_profit_training aux_profit_can neg_profit_not parataxis_ignore_profit prep_profit_from amod_data_valuable nn_data_annotation pobj_from_data ' +o,2910,'In the tagging domain , compared log-linear and perceptron training for HMM-style tagging based on dynamic programming ',Collins,'prep_compared_In det_domain_the amod_domain_tagging pobj_In_domain nsubj_compared_ nn_training_log-linear cc_log-linear_and conj_log-linear_perceptron dobj_compared_training prep_compared_for amod_tagging_HMM-style pobj_for_tagging prep_compared_based dep_based_on amod_programming_dynamic pobj_on_programming ' +o,2911,'413 Letter Lexical Transliteration Similar to IBM Model-1 , we use a bag-of-letter generative model within a block to approximate the lexical transliteration equivalence : P -LRB- fj + lj ei + ki -RRB- = j + lproductdisplay jprime = j i + ksummationdisplay iprime = i P -LRB- fjprime eiprime -RRB- P -LRB- eiprime ei + ki -RRB- , -LRB- 10 -RRB- where P -LRB- eiprime ei + ki -RRB- similarequal 1 \\/ -LRB- k +1 -RRB- is approximated by a bagof-word unigram ',Brown,'dep_use_413 nn_Transliteration_Letter nn_Transliteration_Lexical nsubj_Similar_Transliteration xcomp_413_Similar prep_Similar_to nn__IBM nn__Model-1 pobj_to_ nsubj_use_we det_model_a amod_model_bag-of-letter nn_model_generative dobj_use_model prep_use_within det_block_a pobj_within_block aux_approximate_to infmod_block_approximate det_equivalence_the amod_equivalence_lexical nn_equivalence_transliteration dobj_approximate_equivalence nsubj_=_P nsubj_ki_fj dep_ki_+ nn_ki_lj nn_ki_ei nn_ki_+ dep_P_ki rcmod_equivalence_= nsubj_=_j cc_jprime_+ advmod_jprime_lproductdisplay amod_j_jprime prep_jprime_= pobj_=_j dep_j_i cc_jprime_+ advmod_iprime_ksummationdisplay conj_jprime_iprime ccomp_=_= dobj_=_i dep_i_P amod_eiprime_fjprime appos_i_eiprime partmod_i_P amod_ki_eiprime nn_ki_ei nn_ki_+ appos_i_ki appos_equivalence_10 advmod_similarequal_where nsubj_similarequal_P amod_ki_eiprime nn_ki_ei nn_ki_+ appos_P_ki rcmod_equivalence_similarequal num_\\/_1 nsubjpass_approximated_\\/ number_+1_k dep_\\/_+1 auxpass_approximated_is ccomp_similarequal_approximated prep_approximated_by det_unigram_a amod_unigram_bagof-word pobj_by_unigram ' +o,2912,'The weights for the various components of the model -LRB- phrase translation model , language model , distortion model etc -RRB- are set by minimum error rate training ',Och,'det_weights_The nsubjpass_set_weights prep_weights_for det_components_the amod_components_various pobj_for_components prep_components_of det_model_the pobj_of_model nn_model_phrase nn_model_translation dep_model_model nn_model_language appos_model_model nn_etc_distortion nn_etc_model appos_model_etc auxpass_set_are prep_set_by amod_training_minimum nn_training_error nn_training_rate pobj_by_training ' +o,2913,'291 31 Level of Analysis Research on sentiment annotation is usually conducted at the text or at the sentence levels ',Pang,'num_Level_291 num_Level_31 nsubjpass_conducted_Level prep_Level_of nn_Research_Analysis pobj_of_Research prep_Research_on nn_annotation_sentiment pobj_on_annotation auxpass_conducted_is advmod_conducted_usually prep_conducted_at det__the nn__text pobj_at_ cc_at_or conj_at_at det__the nn__sentence nn__levels pobj_at_ ' +o,2914,'4 Structural Correspondence Learning SCL -LRB- Structural Correspondence Learning -RRB- is a recently proposed domain adaptation technique which uses unlabeled data from both source and target domain to learn correspondences between features from different domains ',Blitzer,'dep_Structural_4 amod__Structural nn__Correspondence nn__Learning nn__SCL nn_Learning_Structural nn_Learning_Correspondence appos__Learning nsubj_technique_ cop_technique_is det_technique_a advmod_proposed_recently amod_technique_proposed nn_technique_domain nn_technique_adaptation nsubj_uses_which rcmod_technique_uses amod_data_unlabeled dobj_uses_data prep_uses_from preconj_domain_both nn_domain_source cc_source_and conj_source_target pobj_from_domain aux_learn_to xcomp_uses_learn dobj_learn_correspondences prep_correspondences_between pobj_between_features prep_learn_from amod_domains_different pobj_from_domains ' +p,2915,'The translation quality was evaluated using a well-established automatic measure : BLEU score ',Papineni,'det_quality_The nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_was xcomp_evaluated_using det_score_a amod_score_well-established amod_score_automatic nn_score_measure amod_score_BLEU dobj_using_score ' +p,2916,'Penn Treebank was also used to induce part-of-speech -LRB- POS -RRB- taggers because the corpus contains very precise and detailed POS markers as well as bracket , annotations ',Marcus,'nn__Penn nn__Treebank nsubjpass_used_ auxpass_used_was advmod_used_also aux_induce_to xcomp_used_induce amod_taggers_part-of-speech appos_taggers_POS dobj_induce_taggers mark_contains_because det_corpus_the nsubj_contains_corpus advcl_induce_contains advmod_precise_very amod_markers_precise cc_precise_and conj_precise_detailed dep_markers_POS dobj_contains_markers dep_well_as cc_markers_well dep_well_as conj_markers_bracket appos_bracket_annotations ' +o,2917,'Uses for k-best lists include minimum Bayes risk decoding , discriminative reranking , and discriminative training ',McClosky,'nsubj_include_Uses prep_Uses_for amod_lists_k-best pobj_for_lists amod__minimum nn__Bayes nn__risk amod__decoding dobj_include_ amod__discriminative dep_discriminative_reranking conj__ cc__and amod_training_discriminative conj__training ' +o,2918,'The measure simHinate is the same as the similarity measure proposed in , except that it does not use dependency triples with negative mutual information ',Hindle,'det_simHinate_The nn_simHinate_measure nsubj_same_simHinate cop_same_is det_same_the prep_same_as det_measure_the nn_measure_similarity pobj_as_measure amod_measure_proposed prep_measure_in pobj_in_ prep_measure_except complm_use_that nsubj_use_it aux_use_does neg_use_not pcomp_except_use amod_triples_dependency dobj_use_triples prep_use_with amod_information_negative amod_information_mutual pobj_with_information ' +p,2919,'5 Related Work Discriminative models have recently been proved to be more effective than generative models in some NLP tasks , eg , parsing , POS tagging and LM for speech recognition ',Collins,'num_models_5 amod_models_Related nn_models_Work nn_models_Discriminative nsubjpass_proved_models aux_proved_have advmod_proved_recently auxpass_proved_been aux_effective_to cop_effective_be advmod_effective_more xcomp_proved_effective prep_effective_than amod_models_generative pobj_than_models prep_models_in det_tasks_some nn_tasks_NLP pobj_in_tasks conj_models_eg amod__parsing nn__ appos__POS nn__tagging conj_models_ cc_models_and conj_models_LM prep_LM_for nn__speech nn__recognition pobj_for_ ' +o,2920,'Kappa is defined as K = P -LRB- A -RRB- P -LRB- E -RRB- 1P -LRB- E -RRB- , where P -LRB- A -RRB- is the proportion of times that the labels agree , and P -LRB- E -RRB- is the proportion of times that they may agree by chance ',Carletta,'nsubjpass_P_Kappa auxpass_P_is dep_P_defined quantmod_K_as num_P_K nn_P_= nn_P_P appos_P_A appos_P_E num_E_1P tmod_P_E advmod_E_ advmod_proportion_where nsubj_proportion_P appos_P_A cop_proportion_is det_proportion_the advcl_P_proportion prep_proportion_of pobj_of_times complm_agree_that det_labels_the nsubj_agree_labels ccomp_proportion_agree cc_P_and nsubj_proportion_P appos_P_E cop_proportion_is det_proportion_the conj_P_proportion prep_proportion_of pobj_of_times complm_agree_that nsubj_agree_they aux_agree_may ccomp_proportion_agree prep_agree_by pobj_by_chance ' +o,2921,'Various clustering techniques have been proposed which perform automatic word clustering optimizing a maximum-likelihood criterion with iterative clustering algorithms ',Brown,'amod_techniques_Various nn_techniques_clustering nsubj_been_techniques aux_been_have amod__proposed dobj_been_ nsubj_perform_which rcmod__perform amod_word_automatic dobj_perform_word partmod_word_clustering xcomp_clustering_optimizing det_criterion_a amod_criterion_maximum-likelihood dobj_optimizing_criterion prep_optimizing_with amod_algorithms_iterative amod_algorithms_clustering pobj_with_algorithms ' +o,2922,' compare two tagging frameworks for tagging French , one that is statistical , built upon the Xerox tagger , and another based on linguistic constraints only ',Cutting,'nsubj_compare_ num_frameworks_two amod_frameworks_tagging dobj_compare_frameworks prep_compare_for pcomp_for_tagging dobj_tagging_French appos_French_one nsubj_statistical_that cop_statistical_is rcmod_one_statistical partmod_French_built prep_built_upon det__the nn__Xerox nn__tagger pobj_upon_ cc_French_and conj_French_another partmod_another_based prep_based_on amod_constraints_linguistic pobj_on_constraints advmod_based_only ' +n,2923,'Sentence-level approximations to B exist , but we found it most effective to perform B computations in the context of a setOof previously-translated sentences , following Watanabe et al ',Liang,'amod_approximations_Sentence-level nsubj_exist_approximations prep_approximations_to pobj_to_B acomp_exist_ cc_exist_but nsubj_found_we conj_exist_found nsubj_perform_it advmod_effective_most dep_perform_effective aux_perform_to xcomp_found_perform nn_computations_B dobj_perform_computations prep_perform_in det_context_the pobj_in_context prep_context_of det_sentences_a amod_sentences_setOof amod_sentences_previously-translated pobj_of_sentences prep_context_following pobj_following_Watanabe cc_Watanabe_et conj_Watanabe_al ' +o,2924,'So far , these techniques have focused on phrasebased models using contiguous phrases ',Koehn,'dep_far_So advmod_focused_far det_techniques_these nsubj_focused_techniques aux_focused_have prep_focused_on amod_models_phrasebased pobj_on_models xcomp_focused_using amod_phrases_contiguous dobj_using_phrases ' +o,2925,' propose a model that encodes how likely it is that different sized spans of text are skipped to reach words and phrases to recycle ','Daume III','advmod_propose_ det_model_a dobj_propose_model nsubj_encodes_that rcmod_model_encodes advmod_is_how dep_how_likely nsubj_is_it ccomp_encodes_is complm_skipped_that amod_spans_different amod_spans_sized nsubj_skipped_spans prep_spans_of pobj_of_text cop_skipped_are ccomp_is_skipped aux_reach_to xcomp_skipped_reach dobj_reach_words cc_words_and conj_words_phrases aux_recycle_to xcomp_reach_recycle ' +o,2926,' used collocation with excellent or poor to obtain positive and negative clues for document classification ',Turney,'nsubj_used_ dobj_used_collocation prep_used_with pobj_with_excellent cc_excellent_or conj_excellent_poor aux_obtain_to xcomp_used_obtain amod_clues_positive cc_positive_and conj_positive_negative dobj_obtain_clues prep_clues_for nn_classification_document pobj_for_classification ' +o,2927,'These constituent matching\\/violation counts are used as a feature in the decoders log-linear model and their weights are tuned via minimal error rate training -LRB- MERT -RRB- ',Och,'det_counts_These amod_counts_constituent nn_counts_matching\\/violation nsubjpass_used_counts auxpass_used_are mark_tuned_as det_feature_a nsubjpass_tuned_feature prep_feature_in det_model_the amod_model_decoders amod_model_log-linear pobj_in_model cc_model_and poss_weights_their conj_model_weights auxpass_tuned_are advcl_used_tuned prep_tuned_via amod_training_minimal nn_training_error nn_training_rate pobj_via_training abbrev_training_MERT ' +n,2928,'In such a process , original phrase-based decoding does not take advantage of any linguistic analysis , which , however , is broadly used in rule-based approaches ',Koehn,'prep_take_In predet_process_such det_process_a pobj_In_process amod__original amod__phrase-based nn__decoding nsubj_take_ aux_take_does neg_take_not dobj_take_advantage prep_advantage_of det_analysis_any amod_analysis_linguistic pobj_of_analysis nsubjpass_used_which advmod_used_however auxpass_used_is advmod_used_broadly rcmod_analysis_used prep_used_in amod_approaches_rule-based pobj_in_approaches ' +o,2929,'Some of the differences between our approach and those of are mentioned below : ? ? objectives : aims at binary text classification , while our objective is six class classification of one-liner headlines ',Turney,'nsubj_objectives_Some prep_Some_of det_differences_the pobj_of_differences prep_differences_between poss_approach_our pobj_between_approach cc_differences_and conj_differences_those prep_those_of pobj_of_ cop_objectives_are dep_objectives_mentioned dep_mentioned_below nsubj_aims_ parataxis_objectives_aims prep_aims_at amod_classification_binary nn_classification_text pobj_at_classification mark_classification_while poss_objective_our nsubj_classification_objective cop_classification_is num_classification_six nn_classification_class advcl_aims_classification prep_classification_of amod_headlines_one-liner pobj_of_headlines ' +o,2930,'Subjective phrases are used by and others in order to classify reviews or sentences as positive or negative ',Pang,'amod_phrases_Subjective nsubjpass_used_phrases auxpass_used_are prep_used_by pobj_by_ cc__and conj__others mark_classify_in dep_classify_order aux_classify_to purpcl_used_classify nsubj_positive_reviews cc_reviews_or conj_reviews_sentences advmod_positive_as xcomp_classify_positive cc_positive_or conj_positive_negative ' +o,2931,'For evaluation we use ROUGE SU4 recall metric1 , which was among the official automatic evaluation metrics for DUC ',Lin,'prep_use_For pobj_For_evaluation nsubj_use_we nn_metric1_ROUGE nn_metric1_ nn_metric1_SU4 nn_metric1_recall dobj_use_metric1 nsubj_was_which rcmod_metric1_was prep_was_among det_metrics_the amod_metrics_official amod_metrics_automatic nn_metrics_evaluation pobj_among_metrics prep_metrics_for pobj_for_DUC ' +o,2932,'Specifically , identify which political candidate is predicted to win by an opinion posted on a message board and aggregate opinions to correctly predict an election result ',Kim,'advmod_identify_Specifically nsubj_identify_ dobj_predicted_which amod_candidate_political nsubjpass_predicted_candidate auxpass_predicted_is ccomp_identify_predicted aux_win_to xcomp_predicted_win prep_win_by det_opinion_an pobj_by_opinion partmod_opinion_posted prep_posted_on det_board_a nn_board_message pobj_on_board cc_board_and amod_opinions_aggregate conj_board_opinions aux_predict_to advmod_predict_correctly xcomp_posted_predict det_result_an nn_result_election dobj_predict_result ' +o,2933,'Inversion Transduction Grammar -LRB- ITG -RRB- is the model of , Tree-to-String is the model of Yamada and Knight -LRB- 2001 -RRB- , and Tree-to-String , Clone allows the node cloning operation described above ',Wu,'nn_Grammar_Inversion nn_Grammar_Transduction nsubj_model_Grammar abbrev_Grammar_ITG cop_model_is det_model_the prep_model_of pobj_of_ nsubj_model_Tree-to-String cop_model_is det_model_the ccomp_model_model prep_model_of pobj_of_Yamada cc_Yamada_and conj_Yamada_Knight appos_model_2001 cc_model_and ccomp_allows_Tree-to-String nsubj_allows_Clone conj_model_allows det_operation_the amod_operation_node nn_operation_cloning dobj_allows_operation partmod_operation_described prep_described_above ' +o,2934,' also pointed out that due to the limited references for every MT output , using the overlapping ratio of n-grams longer than 2 did not improve sentence level evaluation performance of BLEU ',Liu,'advmod_pointed_ advmod_pointed_also prt_pointed_out dep_improve_that dep_to_due prep_improve_to det_references_the amod_references_limited pobj_to_references prep_references_for det_output_every nn_output_MT pobj_for_output csubj_improve_using det_ratio_the amod_ratio_overlapping dobj_using_ratio prep_ratio_of pobj_of_n-grams advmod_n-grams_longer dep_longer_than pobj_than_2 aux_improve_did neg_improve_not dep_pointed_improve nn_performance_sentence nn_performance_level nn_performance_evaluation dobj_improve_performance prep_performance_of pobj_of_BLEU ' +o,2935,'2 Related Work The most commonly used similarity measures are based on the WordNet lexical database and a number of such measures have been made publicly available ',Hughes,'num_Work_2 amod_Work_Related det_measures_The advmod_used_most advmod_used_commonly amod_measures_used nn_measures_similarity nsubjpass_based_measures auxpass_based_are dep_Work_based prep_based_on det__the nn__WordNet amod__lexical nn__database pobj_on_ cc_based_and det_number_a nsubjpass_made_number prep_number_of amod_measures_such pobj_of_measures aux_made_have auxpass_made_been conj_based_made advmod_available_publicly amod__available dobj_made_ ' +o,2936,'decades like n-gram back-off word models , class models , structured language models or maximum entropy language models ',Brown,'dep__decades prep_decades_like amod__n-gram amod__back-off nn__word nn__models nn__ dep__class nn__models nn__ amod__structured nn__language nn__models pobj_like_ cc__or amod_models_maximum nn_models_entropy nn_models_language conj__models ' +o,2937,'For instance , instead of representing the polarity of a term using a binary value , Mullen and Collier use method to assign a real value to represent term polarity and introduce a variety of numerical features that are aggregate measures of the polarity values of terms selected from the document under consideration ',Turney,'prep_use_For pobj_For_instance dep_of_instead prep_use_of pcomp_of_representing det_polarity_the dobj_representing_polarity prep_polarity_of det_term_a pobj_of_term partmod_term_using det_value_a amod_value_binary dobj_using_value nn__Mullen cc_Mullen_and conj_Mullen_Collier nsubj_use_ amod_method_ dobj_use_method aux_assign_to xcomp_use_assign det_value_a amod_value_real dobj_assign_value aux_represent_to xcomp_assign_represent dobj_represent_term advmod_represent_polarity cc_represent_and conj_represent_introduce det_variety_a dobj_introduce_variety prep_variety_of amod_features_numerical pobj_of_features nsubj_measures_that cop_measures_are amod_measures_aggregate rcmod_variety_measures prep_measures_of det_values_the amod_values_polarity pobj_of_values prep_values_of pobj_of_terms partmod_terms_selected prep_selected_from det_document_the pobj_from_document prep_selected_under pobj_under_consideration ' +o,2938,'Wed like to learn the number of paradigm classes from the data , but doing this would probably require extending adaptor grammars to incorporate the kind of adaptive statesplitting found in the iHMM and iPCFG ',Liang,'nsubj_like_Wed aux_learn_to xcomp_like_learn det_number_the dobj_learn_number prep_number_of nn_classes_paradigm pobj_of_classes prep_learn_from det_data_the pobj_from_data cc_doing_but dep_learn_doing nsubj_require_this aux_require_would advmod_require_probably ccomp_doing_require xcomp_require_extending nn_grammars_adaptor dobj_extending_grammars aux_incorporate_to xcomp_extending_incorporate det_kind_the dobj_incorporate_kind prep_kind_of amod_statesplitting_adaptive pobj_of_statesplitting partmod_statesplitting_found prep_found_in det_iHMM_the pobj_in_iHMM cc_doing_and conj_doing_iPCFG ' +o,2939,'This approach is similar to conventional techniques for automatic thesaurus construction ',Lin,'det_approach_This nsubj_similar_approach cop_similar_is prep_similar_to amod_techniques_conventional pobj_to_techniques prep_techniques_for amod_construction_automatic nn_construction_thesaurus pobj_for_construction ' +o,2940,'51 CoNLL named entities presence feature We use Stanford named entity recognizer -LRB- NER -RRB- to identify CoNLL style NEs7 as possible answer strings in a candidate sentence for a given type of question ',Finkel,'num_entities_51 nn_entities_CoNLL amod_entities_named prep_entities_presence dep_presence_feature nsubj_use_We dep_entities_use nsubj_named_Stanford ccomp_use_named nn_recognizer_entity nsubj__recognizer abbrev_recognizer_NER ccomp_named_ aux_identify_to xcomp__identify amod_NEs7_CoNLL nn_NEs7_style dobj_identify_NEs7 prep_identify_as amod_strings_possible nn_strings_answer pobj_as_strings prep_strings_in det_sentence_a nn_sentence_candidate pobj_in_sentence prep_identify_for det_type_a amod_type_given pobj_for_type prep_type_of pobj_of_question ' +o,2941,'Work on learning with hidden variables can be used for both CRFs and for inference based learning algorithms like those used in this work ',Liang,'nsubjpass_used_Work prep_Work_on pcomp_on_learning prep_learning_with amod_variables_hidden pobj_with_variables aux_used_can auxpass_used_be prep_used_for det__both nn__CRFs pobj_for_ cc_for_and conj_for_for pobj_for_inference partmod_inference_based xcomp_based_learning dobj_learning_algorithms prep_learning_like pobj_like_those partmod_those_used prep_used_in det_work_this pobj_in_work ' +o,2942,'Meanwhile , translation grammars have grown in complexity from simple inversion transduction grammars to general tree-to-string transducers and have increased in size by including more synchronous tree fragments ',Wu,'advmod_grown_Meanwhile nn_grammars_translation nsubj_grown_grammars aux_grown_have prep_grown_in pobj_in_complexity prep_grown_from amod_grammars_simple nn_grammars_inversion nn_grammars_transduction pobj_from_grammars partmod_grammars_ prep__to amod__general amod__tree-to-string nn__transducers pobj_to_ cc_grown_and aux_increased_have conj_grown_increased prep_increased_in pobj_in_size prep_increased_by pcomp_by_including advmod_synchronous_more amod_fragments_synchronous nn_fragments_tree dobj_including_fragments advmod_including_ ' +o,2943,'Schtze , 1993 -RRB- is not suited to highly skewed distributions omni-present in natural language ',Dunning,'advmod_suited_Schtze nsubjpass_suited_1993 auxpass_suited_is neg_suited_not prep_suited_to advmod_skewed_highly pobj_to_skewed nsubj_omni-present_distributions dep_suited_omni-present prep_omni-present_in amod_language_natural pobj_in_language ' +o,2944,'Several other measures like Log-Likelihood , Pearsons a2a4a3 , Z-Score , Cubic Association Ratio -LRB- MI3 -RRB- , etc , have been also proposed ',Dunning,'amod_measures_Several amod_measures_other dep_measures_like amod__Log-Likelihood dobj_like_ nn__Pearsons nn__a2a4a3 appos__ nn__Z-Score appos__ nn_Ratio_Cubic nn_Ratio_Association appos__Ratio abbrev_Ratio_MI3 dep_like_etc aux_proposed_have auxpass_proposed_been advmod_proposed_also dep_like_proposed ' +o,2945,'The parser implementation in was used in this experiment and it was run in a mode which emulated the parser ',Collins,'det_implementation_The nn_implementation_parser dep_implementation_in nsubjpass_used_ auxpass_used_was pcomp_in_used prep_used_in det_experiment_this pobj_in_experiment cc_used_and nsubjpass_run_it auxpass_run_was conj_used_run prep_run_in det_mode_a pobj_in_mode nsubj_emulated_which rcmod_mode_emulated det_parser_the amod_parser_ dobj_emulated_parser ' +o,2946,'Similarly to classical NLP tasks such as text chunking and named entity recognition , we formulate mention detection as a sequence classification problem , by assigning a label to each token in the text , indicating whether it starts a specific mention , is inside a specific mention , or is outside any mentions ',Ramshaw,'prep_Similarly_to amod_tasks_classical nn_tasks_NLP pobj_to_tasks dep_as_such prep_tasks_as nn__text nn__chunking pobj_as_ cc_Similarly_and conj_Similarly_named nn_recognition_entity nsubj_indicating_recognition dep_indicating_ nsubj_formulate_we dep_indicating_formulate nn_detection_mention dobj_formulate_detection prep_formulate_as det_problem_a nn_problem_sequence nn_problem_classification pobj_as_problem prep_formulate_by pcomp_by_assigning det_label_a dobj_assigning_label prep_assigning_to det_token_each pobj_to_token prep_token_in det_text_the pobj_in_text xcomp_named_indicating complm_starts_whether nsubj_starts_it ccomp_indicating_starts det_mention_a amod_mention_specific dobj_starts_mention conj_starts_is prep_is_inside det_mention_a amod_mention_specific pobj_inside_mention cc_starts_or cop_mentions_is advmod_mentions_outside dep_outside_any conj_starts_mentions ' +o,2947,'= = = = = m aj j m j aj l i i l i ii m j j mlajdeft en pp m ap 0:1 11 1 2 0 0 0 -RRB- , -LRB- -RRB- , , -LRB- -RRB- -LRB- ! -RRB- -LRB- -RRB- , Pr -LRB- -RRB- , -LRB- 00 eef -LRB- 3 -RRB- 1 A cept is defined as the set of target words connected to a source word ',Dunning,'nsubj_=_= dobj_=_= punct_=_= nn_=_= nn_=_m nn_=_aj nn_=_j nn_=_m nn_=_j nn_=_aj nn_=_l nn_=_i nn_=_i nn_=_l nn_=_i nn_=_ii nn_=_m nn_=_j nn_=_j nn_=_mlajdeft nn_=_en nn_=_pp nn_=_m dep_=_ap number_11_0:1 amod_=_11 number_2_1 dep_11_2 prep_2_0 num_0_0 pobj_0_0 nn_-LRB-_Pr advmod_eef_00 parataxis_-LRB-_eef dobj_eef_3 dep_defined_1 det_cept_A nsubjpass_defined_cept auxpass_defined_is dep_=_defined prep_defined_as det_set_the pobj_as_set prep_set_of nn_words_target pobj_of_words partmod_words_connected prep_connected_to det__a nn__source nn__word pobj_to_ ' +o,2948,'Recently , several solutions to the problem of tagging unknown words have been presented ',Marcus,'advmod_presented_Recently amod_solutions_several nsubjpass_presented_solutions prep_solutions_to det_problem_the pobj_to_problem prep_problem_of pcomp_of_tagging amod_words_unknown dobj_tagging_words aux_presented_have auxpass_presented_been dep_presented_ ' +o,2949,'We also do not require a newly added feature to be either atomic or a collocation of an atomic feature with a feature already included into the model as it was proposed in ',Berger,'nsubj_require_We advmod_require_also aux_require_do neg_require_not det_feature_a advmod_feature_newly amod_feature_added nsubj_be_feature aux_be_to xcomp_require_be advmod_be_either amod_collocation_atomic cc_atomic_or conj_atomic_a nsubj_included_collocation prep_collocation_of det_feature_an amod_feature_atomic pobj_of_feature prep_feature_with det_feature_a pobj_with_feature advmod_included_already dep_require_included prep_included_into det_model_the pobj_into_model mark_proposed_as nsubjpass_proposed_it auxpass_proposed_was advcl_included_proposed prep_proposed_in pobj_in_ ' +o,2950,'32 Domain Adaptation Track As mentioned previously , the source data is drawn from a corpus of news , specifically the Wall Street Journal section of the Penn Treebank ',Marcus,'num_Track_32 nn_Track_Domain nn_Track_Adaptation mark_mentioned_As advcl_drawn_mentioned advmod_mentioned_previously det_data_the nn_data_source nsubjpass_drawn_data auxpass_drawn_is dep_Track_drawn prep_drawn_from det_corpus_a pobj_from_corpus prep_corpus_of pobj_of_news advmod_corpus_specifically det_section_the nn_section_Wall nn_section_Street nn_section_Journal dep_corpus_section prep_section_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +p,2951,'Usually in 1 In our experiments , we set negative PMI values to 0 , because , in their seminal paper on word association ratio , show that negative PMI values are not expected to be accurate unless co-occurrence counts are made from an extremely large corpus ',Church,'advmod_set_Usually prep_set_in pobj_in_1 prep_set_In poss_experiments_our pobj_In_experiments nsubj_set_we amod_values_negative nn_values_PMI dobj_set_values prep_set_to pobj_to_0 mark_show_because nsubj_show_ prep_show_in poss_paper_their amod_paper_seminal pobj_in_paper prep_paper_on nn_ratio_word nn_ratio_association pobj_on_ratio advcl_set_show complm_expected_that amod_values_negative nn_values_PMI nsubjpass_expected_values auxpass_expected_are neg_expected_not ccomp_show_expected aux_accurate_to cop_accurate_be xcomp_expected_accurate mark_made_unless amod_counts_co-occurrence nsubjpass_made_counts auxpass_made_are advcl_accurate_made prep_made_from det_corpus_an advmod_large_extremely amod_corpus_large pobj_from_corpus ' +o,2952,'Our technique of generating negative examples is similar to the approach of ',Okanohara,'poss_technique_Our nsubj_similar_technique prep_technique_of amod_examples_generating amod_examples_negative pobj_of_examples cop_similar_is prep_similar_to det_approach_the pobj_to_approach prep_approach_of ' +o,2953,'31 Binarizable segmentations -LRB- a -RRB- Following , every sequence of phrase alignments can be viewed 1For example , if the cut-off on phrase pairs is ten words , all sentence pairs smaller than ten words in the training data will be included as phrase pairs as well ',Zhang,'num_segmentations_31 amod_segmentations_Binarizable nsubjpass_viewed_segmentations dep_-LRB-_a prep_segmentations_Following pobj_Following_ det_sequence_every appos__sequence prep_sequence_of nn_alignments_phrase pobj_of_alignments aux_viewed_can auxpass_viewed_be amod_example_1For dobj_viewed_example mark_words_if det_cut-off_the nsubj_words_cut-off prep_cut-off_on nn_pairs_phrase pobj_on_pairs cop_words_is amod_words_ten dep_example_words det_pairs_all nn_pairs_sentence measure_smaller_pairs dep_included_smaller quantmod_ten_than num_words_ten nsubjpass_included_words prep_words_in det_data_the nn_data_training pobj_in_data aux_included_will auxpass_included_be dep_viewed_included prep_included_as nn_pairs_phrase pobj_as_pairs advmod_well_as advmod_included_well ' +o,2954,'31 Phrase-Based Models According to the translation model presented in , given a source sentence f , the best target translation can be obtained using the following model best e 288 -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- maxarg -RRB- -LRB- maxarg e e e eef fee length LM best pp p = = -LRB- 1 -RRB- Where the translation model can be decomposed into -RRB- -LRB- efp = = I i i iii i i II aefpbadef efp 1 1 1 1 -RRB- , -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- w -LRB- 2 -RRB- Where -RRB- -LRB- i i ef is the phrase translation probability ',Koehn,'num_Phrase-Based_31 nsubj_Models_Phrase-Based prep_Models_According dep_According_to det_model_the nn_model_translation pobj_to_model dep_model_presented prep_presented_in pobj_in_ dep_presented_given det_sentence_a nn_sentence_source iobj_given_sentence dep_translation_f det_translation_the amod_translation_best nn_translation_target dobj_given_translation aux_obtained_can auxpass_obtained_be rcmod_translation_obtained purpcl_obtained_using det_model_the amod_model_following dobj_using_model amod_model_best dep_best_e dep_e_288 dep_best_maxarg dep_best_maxarg dep_=_e dep_LM_e dep_LM_e nn_length_eef nn_length_fee nsubj_LM_length csubj_=_LM amod_p_best nn_p_pp nsubj_=_p ccomp_LM_= ccomp_maxarg_= dep_=_1 advmod_decomposed_Where det_model_the nn_model_translation nsubjpass_decomposed_model aux_decomposed_can auxpass_decomposed_be advcl_=_decomposed prep_decomposed_into nn_=_efp nsubj_=_= dep_translation_= dep_efp_I nn_efp_i nn_efp_i nn_efp_iii nn_efp_i nn_efp_i nn_efp_II nn_efp_aefpbadef dobj_=_efp number_1_1 num_1_1 number_1_1 dep_efp_1 measure_Where_w appos_w_2 dep_-LRB-_Where nn_ef_i nn_ef_i nsubj_probability_ef cop_probability_is det_probability_the nn_probability_phrase nn_probability_translation dep_presented_probability ' +o,2955,'The model is composed of three parts (Collins, 2002a): a set of candidate SAPTs GEN, which is the top n SAPTs of a sentence from SCISSOR; a function that maps a sentence Inputs: A set of training examples (xi,yi ), i = 1n, where xi is a sentence, and yi is a candidate SAPT that has the highest similarity score with the gold-standard SAPT Initialization: Set W = 0 Algorithm: For t = 1T,i = 1n Calculate yi = argmaxyGEN(xi) (xi,y) W If (yi negationslash= yi ) then W = W +(xi,yi ) (xi,yi) Output: The parameter vector W Figure 2: The perceptron training algorithm.',Collins,'' +o,2956,'This is the same complexity as the ITG alignment algorithm used by and others , meaning complete Viterbi decoding is possible without pruning for realistic-length sentences ',Wu,'nsubj_complexity_This cop_complexity_is det_complexity_the amod_complexity_same prep_complexity_as det_algorithm_the nn_algorithm_ITG nn_algorithm_alignment pobj_as_algorithm partmod_complexity_used prep_used_by pobj_by_ cc__and conj__others partmod_complexity_meaning acomp_meaning_complete nn_decoding_Viterbi nsubj_possible_decoding cop_possible_is dep_complexity_possible prep_possible_without pobj_without_pruning prep_pruning_for amod_sentences_realistic-length pobj_for_sentences ' +o,2957,'One can also examine the distribution of character or word ngrams , eg Language Modeling , phrases , and so on ',Church,'nsubj_examine_One aux_examine_can advmod_examine_also det_distribution_the dobj_examine_distribution prep_distribution_of nn_ngrams_character cc_character_or conj_character_word pobj_of_ngrams nn__eg nn__Language nn__Modeling appos_distribution_ nn__phrases appos_distribution_ cc_distribution_and conj_distribution_so prep_so_on ' +n,2958,'However , many of these models are not applicable to parallel treebanks because they assume translation units where either the source text , the target text or both are represented as word sequences without any syntactic structure ',Galley,'advmod_applicable_However nsubj_applicable_many prep_many_of det_models_these pobj_of_models cop_applicable_are neg_applicable_not aux_parallel_to xcomp_applicable_parallel dobj_parallel_treebanks mark_units_because nsubj_units_they dep_units_assume nn_units_translation advcl_parallel_units advmod_represented_where preconj_text_either det_text_the nn_text_source nsubjpass_represented_text det_text_the nn_text_target conj_text_text cc_text_or conj_text_both auxpass_represented_are advcl_units_represented prep_represented_as nn_sequences_word pobj_as_sequences prep_represented_without det_structure_any amod_structure_syntactic pobj_without_structure ' +p,2959,'1 Introduction Phrase-based Statistical MT -LRB- PB-SMT -RRB- has become the predominant approach to Machine Translation in recent years ',Koehn,'num_Introduction_1 measure_Phrase-based_Introduction amod__Phrase-based nn__Statistical nn__MT abbrev__PB-SMT nsubj_approach_ aux_approach_has cop_approach_become det_approach_the amod_approach_predominant prep_approach_to nn_Translation_Machine pobj_to_Translation prep_Translation_in amod_years_recent pobj_in_years ' +o,2960,'The key difference is that , instead of using the delta rule of Equation -LRB- 8 -RRB- -LRB- as shown in line 5 of Figure 4 -RRB- , updates parameters using the rule : t +1 d t d + f d -LRB- w R i -RRB- f d -LRB- w i -RRB- ',Collins,'det_difference_The amod_difference_key nsubj_is_difference complm_parameters_that dep_of_instead prep_parameters_of pcomp_of_using det_rule_the amod_rule_delta dobj_using_rule prep_rule_of pobj_of_Equation appos_Equation_8 mark_shown_as dep_Equation_shown prep_shown_in pobj_in_line num_line_5 prep_line_of pobj_of_Figure num_Figure_4 amod_updates_ nsubj_parameters_updates ccomp_is_parameters xcomp_parameters_using det_rule_the dobj_using_rule number_+1_t dep_rule_+1 dep_+_d dep_d_t dep_+_d dep_f_+ dep_+_f dep_f_d nn_i_w nn_i_R appos_d_i dep_d_f dep_rule_d number_i_w dep_rule_i ' +o,2961,'Statistics on co-occurrence of words in a local context were used recently for monolingual word sense disambiguation ',Smadja,'nsubjpass_used_Statistics prep_Statistics_on pobj_on_co-occurrence prep_co-occurrence_of pobj_of_words prep_words_in det_context_a amod_context_local pobj_in_context auxpass_used_were advmod_used_recently prep_used_for amod__monolingual nn__word nn__sense nn__disambiguation pobj_for_ ' +p,2962,'4 Evaluation The purpose of our evaluation is to contrast our proposed feature based approach with a state-ofthe-art sequential learning technique ',Collins,'num_Evaluation_4 det_purpose_The nsubj_is_purpose prep_purpose_of poss_evaluation_our pobj_of_evaluation dep_Evaluation_is aux_contrast_to xcomp_is_contrast poss_feature_our amod_feature_proposed dobj_contrast_feature prep_contrast_based dep_based_approach prep_approach_with det_sequential_a amod_sequential_state-ofthe-art pobj_with_sequential partmod_sequential_learning dobj_learning_technique ' +o,2963,'That some model structures work better than others at real NLP tasks was discussed by Johnson and ',Klein,'complm_work_That det_structures_some nn_structures_model nsubj_work_structures advmod_work_better mark_discussed_than nsubjpass_discussed_others prep_others_at amod_tasks_real nn_tasks_NLP pobj_at_tasks auxpass_discussed_was dep_better_discussed prep_discussed_by nn_and_Johnson nn_and_ pobj_by_and ' +n,2964,'2 Statistical Word Alignment Statistical translation models only allow word to word and multi-word to word alignments ',Brown,'num_models_2 nn_models_Statistical nn_models_Word nn_models_Alignment nn_models_Statistical nn_models_translation nsubj__models advmod__only dep__allow dobj_allow_word prep_allow_to pobj_to_word cc_word_and conj_word_multi-word prep_allow_to nn_alignments_word pobj_to_alignments ' +o,2965,'This shows that hypothesis features are either not discriminative enough , or that the reranking model is too weak This performance gap can be mainly attributed to two problems : optimization error and modeling error -LRB- see Figure 1 -RRB- 1 Much work has focused on developing better algorithms to tackle the optimization problem -LRB- eg MERT -RRB- , since MT evaluation metrics such as BLEU and PER are riddled with local minima and are difficult to differentiate with respect to re-ranker parameters ',Och,'nsubj_shows_This dep_discriminative_that nn_features_hypothesis nsubj_discriminative_features cop_discriminative_are dep_discriminative_either neg_discriminative_not dep_shows_discriminative advmod_discriminative_enough cc_discriminative_or dep_weak_that det_model_the amod_model_reranking nsubj_weak_model cop_weak_is advmod_weak_too conj_discriminative_weak det_gap_This nn_gap_performance nsubjpass_attributed_gap aux_attributed_can auxpass_attributed_be advmod_attributed_mainly ccomp_weak_attributed prep_attributed_to num_problems_two pobj_to_problems nn_error_optimization nsubj_focused_error cc_error_and nn_error_modeling conj_error_error dep_error_see dobj_see_Figure num_Figure_1 num_work_1 amod_work_Much dep_error_work aux_focused_has parataxis_shows_focused prep_focused_on pcomp_on_developing amod_algorithms_better dobj_developing_algorithms aux_tackle_to xcomp_developing_tackle det_problem_the nn_problem_optimization dobj_tackle_problem nn__eg nn__MERT abbrev_problem_ mark_riddled_since nn_metrics_MT nn_metrics_evaluation nsubj_riddled_metrics dep_as_such prep_metrics_as pobj_as_BLEU cc_BLEU_and conj_BLEU_PER aux_riddled_are advcl_focused_riddled prep_riddled_with amod_minima_local pobj_with_minima cc_focused_and cop_difficult_are conj_focused_difficult aux_differentiate_to xcomp_difficult_differentiate prep_differentiate_with pobj_with_respect prep_differentiate_to amod_parameters_re-ranker pobj_to_parameters ' +o,2966,'Figure 1 -LRB- b -RRB- shows several orders of the sentence which violate this constraint1 Previous studies have shown that if both the source and target dependency trees represent linguistic constituency , the alignment between subtrees in the two languages is very complex ',Wellington,'nsubj_shows_Figure num_Figure_1 dep_-LRB-_b amod_orders_several dobj_shows_orders prep_orders_of det_sentence_the pobj_of_sentence nsubj_violate_which rcmod_sentence_violate det_studies_this amod_studies_constraint1 amod_studies_Previous nsubj_shown_studies aux_shown_have ccomp_violate_shown complm_complex_that mark_represent_if preconj_source_both det_source_the nsubj_represent_source cc_source_and nn_trees_target nn_trees_dependency conj_source_trees advcl_complex_represent amod_constituency_linguistic dobj_represent_constituency det_alignment_the nsubj_complex_alignment prep_alignment_between pobj_between_subtrees prep_subtrees_in det_languages_the num_languages_two pobj_in_languages cop_complex_is advmod_complex_very ccomp_shown_complex ' +o,2967,'1 Introduction Word alignments were first introduced as an intermediate result of statistical machine translation systems ',Brown,'num_alignments_1 nn_alignments_Introduction nn_alignments_Word nsubjpass_introduced_alignments auxpass_introduced_were advmod_introduced_first prep_introduced_as det_result_an amod_result_intermediate pobj_as_result prep_result_of amod_systems_statistical nn_systems_machine nn_systems_translation pobj_of_systems ' +o,2968,'2 Related work Our work is closest in spirit to the two papers that inspired us and ',Barzilay,'num_work_2 amod_work_Related poss_work_Our nsubj_closest_work cop_closest_is dep_work_closest prep_closest_in pobj_in_spirit prep_spirit_to det_papers_the num_papers_two pobj_to_papers nsubj_inspired_that rcmod_closest_inspired nsubj__us xcomp_inspired_ cc__and conj__ ' +o,2969,'1 Introduction The importance of learning to manipulate monolingual paraphrase relationships for applications like summarization , search , and dialog has been highlighted by a number of recent efforts ',Barzilay,'num_Introduction_1 det_importance_The nsubjpass_highlighted_importance prep_importance_of pcomp_of_learning aux_manipulate_to xcomp_learning_manipulate amod_relationships_monolingual nn_relationships_paraphrase dobj_manipulate_relationships prep_manipulate_for pobj_for_applications prep_applications_like pobj_like_summarization appos_summarization_search cc_learning_and conj_learning_dialog aux_highlighted_has auxpass_highlighted_been dep_Introduction_highlighted prep_highlighted_by det_number_a pobj_by_number prep_number_of amod_efforts_recent pobj_of_efforts ' +o,2970,'32 ROUGE Version 155 of the ROUGE scoring algorithm is also used for evaluating results ',Lin,'num_Version_32 nn_Version_ROUGE nsubjpass_used_Version num_Version_155 prep_Version_of det__the nn__ROUGE nn__scoring nn__algorithm pobj_of_ auxpass_used_is advmod_used_also prep_used_for pcomp_for_evaluating dobj_evaluating_results ' +o,2971,'Normally, one would eliminate the redundant structures produced by the grammar in (1) by replacing it with the canonical form grammar (Wu, 1997), which has the following form: S A | B | C A [AB] | [BB] | [CB] | [AC] | [BC] | [CC] B AA |BA|CA| AC |BC|CC C e/f (2) By design, this grammar allows only one struc147 a0 a1 a2 a0 a3 a4 a2 a5 a1 a6 a7 a8 a6 a8 a9 a8 a2 a8 a10 a8 a1 a2 a3 a6 a8 a4 a7 a8 a6 a8 a9 a8 a8 a11 a12 a11 a0 a1 a2 a0 a3 a4 a2 a5 a1 a6 a7 a8 a6 a8 a9 a8 a0 a1 a2 a0 a3 a4 a2 a5 a1 a6 a7 a8 a6 a8 a9 a8 a0 a1 a2 a0 a3 a4 a2 a5 a1 a6 a7 a8 a6 a8 a9 a8 a13 a11 Figure 3: An example of how dependency trees interact with ITGs.',Wu,'' +o,2972,'The Bloomier filter LM has a precomputed matching of keys shared between a constant number of cells in the filter array ',Talbot,'det__The nn__Bloomier nn__filter nn__LM nsubj_has_ det_matching_a amod_matching_precomputed dobj_has_matching prep_matching_of pobj_of_keys partmod_keys_shared prep_shared_between det_number_a amod_number_constant pobj_between_number prep_number_of pobj_of_cells prep_cells_in det_array_the nn_array_filter pobj_in_array ' +o,2973,'From multilingual texts , translation lexica can be generated ',Dunning,'prep_generated_From amod_texts_multilingual pobj_From_texts nn_lexica_translation nsubjpass_generated_lexica aux_generated_can auxpass_generated_be dep_generated_ ' +o,2974,'There are several basic methods for evaluating associations between words : based on frequency counts , information theoretic and statistical significance ',Church,'expl_are_There amod_methods_several amod_methods_basic nsubj_are_methods prep_methods_for pcomp_for_evaluating dobj_evaluating_associations prep_associations_between pobj_between_words prep_methods_based dep_based_on nn__frequency nn__counts pobj_on_ nn__information amod__theoretic conj__ cc__and amod__statistical nn__significance conj__ ' +o,2975,'For -LRB- 1 -RRB- , the morphemes and labels for our task are : -LRB- 2 -RRB- kita NEG tINC inE1S chabe VT - j SC laj PREP inA1S yol S - j SC iin PRON We also consider POS-tagging for Danish , Dutch , English , and Swedish ; the English is from sections 00-05 -LRB- as training set -RRB- and 19-21 -LRB- as development set -RRB- of the Penn Treebank , and the other languages are from the CoNLL-X dependency parsing shared task 1 We split the original training data into training and development sets ',Marcus,'prep_consider_For dep_morphemes_1 det_morphemes_the pobj_For_morphemes cc_morphemes_and conj_morphemes_labels prep_morphemes_for poss_task_our pobj_for_task dep_task_are dep_NEG_2 amod_NEG_kita dep_morphemes_NEG amod_VT_tINC nn_VT_inE1S nn_VT_chabe dep_NEG_VT nn_laj_j nn_laj_SC dep_morphemes_laj nn_S_PREP nn_S_inA1S nn_S_yol dep_laj_S dep_morphemes_j nn_PRON_SC nn_PRON_iin dep_j_PRON nsubj_consider_We advmod_consider_also dobj_consider_POS-tagging prep_POS-tagging_for nn_Swedish_Danish conj_Swedish_Dutch conj_Swedish_English cc_Swedish_and pobj_for_Swedish det_English_the nsubj_is_English parataxis_consider_is prep_is_from pobj_from_sections dep_sections_00-05 mark_set_as nsubj_set_training dep_sections_set cc_sections_and conj_sections_19-21 mark_set_as nsubj_set_development dep_19-21_set prep_19-21_of det__the nn__Penn nn__Treebank pobj_of_ cc_consider_and det_languages_the amod_languages_other nsubj_are_languages conj_consider_are prep_are_from det_dependency_the amod_dependency_CoNLL-X pobj_from_dependency partmod_dependency_parsing amod_task_shared dobj_parsing_task xcomp_parsing_ dobj__1 nsubj_split_We rcmod_1_split det_data_the amod_data_original nn_data_training dobj_split_data prep_split_into nn_sets_training cc_training_and conj_training_development pobj_into_sets ' +o,2976,'Almost all recent work in developing automatically trained part-of-speech taggers has been on further exploring Markovmodel based tagging ',Marcus,'advmod_work_Almost det_work_all amod_work_recent nsubj_been_work prep_work_in amod_taggers_developing amod_taggers_automatically amod_taggers_trained amod_taggers_part-of-speech pobj_in_taggers aux_been_has prep_been_on advmod_exploring_further pcomp_on_exploring dobj_exploring_Markovmodel prep_exploring_based amod__tagging dep_based_ ' +o,2977,'The evaluation metric is casesensitive BLEU-4 ',Papineni,'det_metric_The nn_metric_evaluation nsubj_BLEU-4_metric cop_BLEU-4_is amod_BLEU-4_casesensitive ' +o,2978,'The first of these nonstructural problems with Model 1 , as standardly trained , is that rare words in the source language tend to act as garbage collectors , aligning to too many words in the target language ',Brown,'det_first_The nsubj_is_first prep_first_of det_problems_these amod_problems_nonstructural pobj_of_problems prep_problems_with pobj_with_Model num_Model_1 advmod_trained_as advmod_trained_standardly amod_first_trained complm_tend_that amod_words_rare nsubj_tend_words prep_words_in det_language_the nn_language_source pobj_in_language ccomp_is_tend aux_act_to xcomp_tend_act mark__as nn_collectors_garbage nsubj__collectors advcl_act_ xcomp__aligning prep_aligning_to advmod_many_too amod_words_many pobj_to_words prep_words_in det_language_the nn_language_target pobj_in_language ' +o,2979,'42 Interpreting reliability results It has been argued elsewhere that since the amount of agreement one would expect by chance depends on the number and relative frequencies of the categories under test , reliability for category classifications should be measured using the kappa coefficient ',Carletta,'num_results_42 nn_results_Interpreting amod_results_reliability nsubj_depends_results nsubjpass_argued_It aux_argued_has auxpass_argued_been rcmod_results_argued advmod_argued_elsewhere xcomp_argued_ complm_expect_that dep_expect_since det_amount_the nsubj_expect_amount prep_amount_of pobj_of_agreement num_agreement_one aux_expect_would ccomp__expect prep_expect_by pobj_by_chance prep_depends_on det_number_the pobj_on_number cc_number_and amod_frequencies_relative conj_number_frequencies prep_frequencies_of det_categories_the pobj_of_categories prep_categories_under pobj_under_test advmod_depends_reliability mark_measured_for nn_classifications_category nsubjpass_measured_classifications aux_measured_should auxpass_measured_be advcl_depends_measured purpcl_measured_using det_coefficient_the nn_coefficient_kappa dobj_using_coefficient ' +o,2980,'31 Exhaustive search by tree fragments This method generates all possible tree fragments rooted by each node in the source parse tree or forest , and then matches all the generated tree fragments against the source parts -LRB- left hand side -RRB- of translation rules to extract the useful rules ',Zhang,'num_search_31 nn_search_Exhaustive nsubj_fragments_search prep_search_by pobj_by_tree det_method_This nsubj_generates_method ccomp_fragments_generates det_fragments_all amod_fragments_possible nn_fragments_tree dobj_generates_fragments partmod_fragments_rooted prep_rooted_by det_node_each pobj_by_node prep_node_in det_tree_the nn_tree_source nn_tree_parse pobj_in_tree cc_tree_or conj_tree_forest cc_generates_and advmod_matches_then conj_generates_matches predet_fragments_all det_fragments_the nn_fragments_generated nn_fragments_tree dobj_matches_fragments prep_matches_against det_parts_the nn_parts_source pobj_against_parts amod_side_left nn_side_hand appos_parts_side prep_parts_of nn_rules_translation pobj_of_rules aux_extract_to xcomp_matches_extract det_rules_the amod_rules_useful dobj_extract_rules ' +o,2981,'The prime public domain examples of such implementations include the TrigramsnTags tagger , Xerox tagger and LT POS tagger ',Cutting,'det_examples_The amod_examples_prime amod_examples_public nn_examples_domain nsubj_include_examples prep_examples_of amod_implementations_such pobj_of_implementations det__the nn__TrigramsnTags nn__tagger dobj_include_ nn__Xerox nn__tagger conj__ cc__and nn__LT dep__POS nn__tagger conj__ ' +o,2982,'The coreference resolution system employs a variety of lexical , semantic , distance and syntactic feature ',Luo,'det_system_The amod_system_coreference nn_system_resolution nsubj_employs_system det_variety_a dobj_employs_variety prep_variety_of amod_feature_lexical conj_lexical_semantic conj_lexical_distance cc_lexical_and conj_lexical_syntactic pobj_of_feature ' +o,2983,'4 Experiments The Penn Treebank is used as the testing corpus ',Marcus,'num_Experiments_4 det__The nn__Penn nn__Treebank nsubjpass_used_ auxpass_used_is dep_Experiments_used prep_used_as det_corpus_the nn_corpus_testing pobj_as_corpus ' +o,2984,'24 Syntactic Similarity We have incorporated , with minor modifications , some of the syntactic metrics described by and Amigo et al ',Liu,'num_Similarity_24 nn_Similarity_Syntactic nsubj_incorporated_We aux_incorporated_have dep_Similarity_incorporated prep_incorporated_with amod_modifications_minor pobj_with_modifications appos_modifications_some prep_some_of det_metrics_the amod_metrics_syntactic pobj_of_metrics partmod_metrics_described prep_described_by pobj_by_ cc__and conj__Amigo cc_Amigo_et conj_Amigo_al ' +p,2985,'This combination of the perceptron algorithm with beam-search is similar to that described by 5 The perceptron algorithm is a convenient choice because it converges quickly usually taking only a few iterations over the training set ',Collins,'det_combination_This nsubj_similar_combination prep_combination_of det_algorithm_the nn_algorithm_perceptron pobj_of_algorithm prep_algorithm_with pobj_with_beam-search cop_similar_is prep_similar_to pobj_to_that partmod_that_described prep_described_by pobj_by_ num__5 det_algorithm_The nn_algorithm_perceptron nsubj_choice_algorithm cop_choice_is det_choice_a amod_choice_convenient dep_similar_choice mark_converges_because nsubj_converges_it advcl_choice_converges advmod_converges_quickly advmod_converges_usually xcomp_converges_taking quantmod_few_only quantmod_few_a num_iterations_few dobj_taking_iterations prep_iterations_over det_set_the nn_set_training pobj_over_set ' +o,2986,'CIT -RRB- ',Koehn,'nn_-RRB-_CIT ' +o,2987,'There have been many studies on POS guessing of unknown words ',Berger,'expl_been_There aux_been_have amod_studies_many dobj_been_studies prep_studies_on dep_guessing_POS pobj_on_guessing prep_guessing_of amod__unknown nn__words pobj_of_ ' +o,2988,'-LRB- levelopment of cor1 -RRB- ora with morl -RRB- ho-synta -LRB- : ti -LRB- : and syntacti -LRB- : mmotation , ',Marcus,'number_cor1_levelopment quantmod_cor1_of punct_morl_ora dep_morl_with dep_cor1_morl dep_cor1_ho-synta dep_ho-synta_ti cc_ti_and conj_ti_syntacti dep__mmotation dep_syntacti_ appos__ ' +n,2989,'Even the 3 A demo of the parser can be found at http://lfgdemocomputingdcuie/lfgparserhtml creators of BLEU point out that it may not correlate particularly well with human judgment at the sentence level ',Papineni,'advmod_found_Even det_demo_the num_demo_3 nn_demo_A nsubjpass_found_demo prep_demo_of det_parser_the pobj_of_parser aux_found_can auxpass_found_be prep_found_at amod_creators_http://lfgdemocomputingdcuie/lfgparserhtml pobj_at_creators prep_creators_of amod_point_BLEU pobj_of_point dep_point_out complm_correlate_that nsubj_correlate_it aux_correlate_may neg_correlate_not ccomp_found_correlate advmod_well_particularly advmod_correlate_well prep_correlate_with amod_judgment_human pobj_with_judgment prep_judgment_at det_level_the nn_level_sentence pobj_at_level ' +o,2990,'Reference-based metrics such as BLEU have rephrased this subjective task as a somewhat more objective question : how closely does the translation resemble sentences that are known to be good translations for the same source ? ',Papineni,'amod_metrics_Reference-based nsubj_rephrased_metrics dep_as_such prep_metrics_as nn__BLEU pobj_as_ aux_rephrased_have det_task_this amod_task_subjective dobj_rephrased_task prep_rephrased_as det_question_a advmod_more_somewhat amod_question_more nn_question_objective pobj_as_question advmod_sentences_how advmod_how_closely aux_sentences_does det_translation_the nsubj_sentences_translation cop_sentences_resemble dep_question_sentences nsubjpass_known_that auxpass_known_are rcmod_sentences_known aux_translations_to cop_translations_be amod_translations_good xcomp_known_translations prep_translations_for det_source_the amod_source_same pobj_for_source ' +o,2991,'This method uses mutual information and loglikelihood , which used to calculate the dependency value between words ',Dunning,'det_method_This nsubj_uses_method amod_information_mutual dobj_uses_information cc_information_and conj_information_loglikelihood nsubj_used_which advmod_used_ rcmod_information_used aux_calculate_to xcomp_used_calculate det_value_the amod_value_dependency dobj_calculate_value prep_value_between pobj_between_words ' +p,2992,'1 Motivation Phrase-based statistical machine translation has emerged as the dominant paradigm in machine translation research ',Koehn,'num__1 nn__Motivation nn__Phrase-based amod__statistical nn__machine nn__translation nsubj_emerged_ aux_emerged_has prep_emerged_as det_paradigm_the amod_paradigm_dominant pobj_as_paradigm prep_paradigm_in nn_research_machine nn_research_translation pobj_in_research ' +o,2993,'In both cases there 1Alternatively , decisions from the sentence classifier can guide which input is seen by the document level classifier ',Pang,'prep_guide_In det_cases_both pobj_In_cases advmod_cases_there dep_there_1Alternatively nsubj_guide_decisions prep_decisions_from det_classifier_the nn_classifier_sentence pobj_from_classifier aux_guide_can dobj_seen_which nsubjpass_seen_input auxpass_seen_is ccomp_guide_seen prep_seen_by det_classifier_the nn_classifier_document nn_classifier_level pobj_by_classifier ' +o,2994,'We use BLEU scores to measure translation accuracy ',Papineni,'nsubj_use_We amod_scores_BLEU nsubj__scores ccomp_use_ aux_measure_to xcomp__measure nn_accuracy_translation dobj_measure_accuracy ' +o,2995,'To search for the most probable parse , we use the heuristic search algorithm described in , which is a form of beam search ',Titov,'aux_search_To dep_use_search prep_search_for det_parse_the advmod_probable_most amod_parse_probable pobj_for_parse nsubj_use_we det_algorithm_the amod_algorithm_heuristic nn_algorithm_search dobj_use_algorithm partmod_algorithm_described prep_described_in pobj_in_ nsubj_form_which cop_form_is det_form_a rcmod__form prep_form_of nn_search_beam pobj_of_search ' +o,2996,'Maximum entropy -LRB- ME -RRB- models have been used in bilingual sense disambiguation , word reordering , and sentence segmentation , parsing , POS tagging and PP attachment , machine translation , and FrameNet classification ',Berger,'amod_entropy_Maximum nn_models_entropy abbrev_models_ME nsubjpass_used_models aux_used_have auxpass_used_been prep_used_in amod_disambiguation_bilingual nn_disambiguation_sense pobj_in_disambiguation nn_reordering_word conj_disambiguation_reordering cc_disambiguation_and nn__sentence nn__segmentation conj_disambiguation_ conj__parsing dep_tagging_POS conj__tagging cc_tagging_and nn__PP nn__attachment conj_tagging_ nn__machine nn__translation conj__ cc__and nn__FrameNet nn__classification conj__ ' +o,2997,'Furthermore , these systems have tackled the problem at different levels of granularity , from the document level , sentence level , phrase level , as well as the speaker level in debates ',Pang,'advmod_tackled_Furthermore det_systems_these nsubj_tackled_systems aux_tackled_have det_problem_the dobj_tackled_problem prep_tackled_at amod_levels_different pobj_at_levels prep_levels_of pobj_of_granularity prep_tackled_from det__the nn__document nn__level nn__ nn__sentence nn__level appos__ nn__phrase nn__level pobj_from_ dep_well_as cc__well dep_well_as det_level_the nn_level_speaker conj__level prep_level_in pobj_in_debates partmod_debates_ ' +o,2998,'ca and , in mining relationships between named entities , or in extracting useful facet terms from news articles ',Cucerzan,'aux__ca cc__and advmod__ conj__in nn_relationships_mining pobj_in_relationships prep_relationships_between amod_entities_named pobj_between_entities cc_in_or conj_in_in pcomp_in_extracting amod_terms_useful nn_terms_facet dobj_extracting_terms prep_extracting_from nn_articles_news pobj_from_articles nsubj__ ' +o,2999,'The translation quality is measured by three MT evaluation metrics : TER , BLEU , and METEOR ',Papineni,'det_quality_The nn_quality_translation auxpass_measured_is dep_quality_measured prep_measured_by num_metrics_three nn_metrics_MT nn_metrics_evaluation pobj_by_metrics nn__TER dep_quality_ nn__BLEU conj__ cc__and nn__METEOR conj__ ' +o,3000,'Research have also been made into alternatives to the current log-linear scoring model such as discriminative models with millions of features , or kernel based models ',Liang,'nsubjpass_made_Research aux_made_have advmod_made_also auxpass_made_been prep_made_into pobj_into_alternatives prep_made_to det_model_the amod_model_current amod_model_log-linear amod_model_scoring pobj_to_model dep_as_such prep_model_as amod_models_discriminative pobj_as_models prep_models_with pobj_with_millions prep_millions_of nn__features pobj_of_ cc_model_or conj_model_kernel partmod_kernel_based nsubj__models xcomp_based_ ' +o,3001,'It has been implemented in the TACITUS System and has been applied to several varieties of text ',Hobbs,'nsubjpass_implemented_It aux_implemented_has auxpass_implemented_been prep_implemented_in det__the nn__TACITUS nn__System pobj_in_ cc_implemented_and aux_applied_has auxpass_applied_been conj_implemented_applied prep_applied_to amod_varieties_several pobj_to_varieties prep_varieties_of pobj_of_text ' +o,3002,' also includes a brief discussion of crossing constraints that can be derived from phrase structure correspondences ',Wu,'advmod__also dep__includes det_discussion_a amod_discussion_brief nsubj__discussion prep_discussion_of pcomp_of_crossing dobj_crossing_constraints nsubjpass_derived_that aux_derived_can auxpass_derived_be rcmod_constraints_derived prep_derived_from nn_correspondences_phrase nn_correspondences_structure pobj_from_correspondences ' +p,3003,'1 Introduction Treebank-based probabilistic parsing has been the subject of intensive research over the past few years , resulting in parsing models that achieve both broad coverage and high parsing accuracy ',Collins,'num_parsing_1 nn_parsing_Introduction nn_parsing_Treebank-based amod_parsing_probabilistic nsubj_been_parsing aux_been_has det_subject_the dobj_been_subject prep_subject_of amod_research_intensive pobj_of_research prep_research_over det_years_the amod_years_past amod_years_few pobj_over_years xcomp_been_resulting prep_resulting_in pcomp_in_parsing dobj_parsing_models nsubj_achieve_that rcmod_models_achieve det_coverage_both amod_coverage_broad dobj_achieve_coverage cc_coverage_and amod_accuracy_high amod_accuracy_parsing conj_coverage_accuracy ' +o,3004,'Various methods have been proposed for synonym acquisition ',Hindle,'amod__Various nn__methods nsubjpass_proposed_ aux_proposed_have auxpass_proposed_been prep_proposed_for nn_acquisition_synonym pobj_for_acquisition ' +o,3005,'Acknowledgments I want to thank my fellow organizers of the shared task , Johan Hall , Sandra Kubler , Ryan , Jens Nilsson , Sebastian Riedel , and Deniz Yuret , whoarealsoco-authorsofthelongerpaperonwhich this paper is partly based ',McDonald,'nsubj_want_I ccomp_based_want aux_thank_to xcomp_want_thank poss_organizers_my amod_organizers_fellow dobj_thank_organizers prep_organizers_of det_task_the amod_task_shared pobj_of_task nn_Hall_Johan conj_task_Hall nn_Kubler_Sandra conj_task_Kubler nn__Ryan conj_task_ nn_Nilsson_Jens conj_task_Nilsson nn_Riedel_Sebastian conj_task_Riedel cc_task_and nn_Yuret_Deniz conj_task_Yuret amod_paper_whoarealsoco-authorsofthelongerpaperonwhich det_paper_this nsubjpass_based_paper auxpass_based_is advmod_based_partly dep_Acknowledgments_based dep_based_ ' +o,3006,'We retrained the parser on lowercased Penn Treebank II , to match the lowercased output of the MT decoder ',Marcus,'nsubj_retrained_We det_parser_the dobj_retrained_parser prep_retrained_on amod__lowercased nn__Penn nn__Treebank nn__II pobj_on_ aux_match_to xcomp_retrained_match det_output_the amod_output_lowercased dobj_match_output prep_output_of det_decoder_the nn_decoder_MT pobj_of_decoder ' +o,3007,'The senses are : 1 material from cellulose 2 report 3 publication 4 medium for writing 5 scientific 6 publishing firm 7 physical object inventory is suitable for which application , other than cross-lingual applications where the inventory can be determined from parallel data ',Carpuat,'det_senses_The nsubj_material_senses cop_material_are num_material_1 prep_material_from pobj_from_cellulose num_cellulose_2 dep_cellulose_report num_publication_3 dobj_report_publication dep_medium_4 amod_publication_medium prep_report_for pcomp_for_writing num_6_5 amod_6_scientific dobj_writing_6 xcomp_writing_publishing dobj_publishing_firm num_inventory_7 amod_inventory_physical nn_inventory_object nsubj_suitable_inventory cop_suitable_is dep_material_suitable prep_suitable_for dep_application_which pobj_for_application acomp_suitable_other prep_other_than amod_applications_cross-lingual pobj_than_applications advmod_determined_where det_inventory_the nsubjpass_determined_inventory aux_determined_can auxpass_determined_be advcl_suitable_determined prep_determined_from amod_data_parallel pobj_from_data ' +o,3008,'Mutual information MI -LRB- x , y -RRB- is defined as following : -RRB- -LRB- -RRB- -LRB- -RRB- , -LRB- log -RRB- -LRB- -RRB- -LRB- -RRB- , -LRB- log -RRB- , -LRB- 22 yfxf yxfN ypxp yxp yxMI = = -LRB- 4 -RRB- where f -LRB- x -RRB- and f -LRB- y -RRB- are frequency of term x and term y , respectively ',Church,'amod_MI_Mutual nn_MI_information nsubjpass_defined_MI dep_MI_x appos_x_y auxpass_defined_is ccomp_=_defined prep_defined_as dep_as_following dep_following_ appos_-LRB-_log nsubj_=_log num_yxMI_22 amod_yxMI_yfxf nn_yxMI_yxfN nn_yxMI_ypxp nn_yxMI_yxp nsubj_=_yxMI advmod_=_= dep_=_4 advmod_frequency_where nsubj_frequency_f dep_f_x cc_f_and conj_f_f appos_f_y cop_frequency_are advcl_=_frequency prep_frequency_of nn_x_term pobj_of_x cc_x_and nn_y_term conj_x_y advmod_frequency_respectively ' +o,3009,'Pr -LRB- pi , F , A -RRB- = summationdisplay i , c -LRB- -RRB- = -LRB- pi , F , A -RRB- productdisplay rji p -LRB- rj -RRB- -LRB- 4 -RRB- In order to acquire the rules specific to our model and to induce their probabilities , we parse the English side of our corpus with an in-house implementation of Collins parsing models and we word-align the parallel corpus with the Giza + +2 implementation of the IBM models ',Brown,'nsubj_summationdisplay_pi dep_pi_F dep_pi_A dep_summationdisplay_= ccomp_Pr_summationdisplay nsubj_Pr_i nsubj_rji_c num_F_= nn_F_pi dep_c_F appos_F_A advmod_rji_productdisplay rcmod_i_rji dobj_rji_p appos_p_rj dep_parse_4 mark_acquire_In dep_acquire_order aux_acquire_to dep_parse_acquire det_rules_the nsubj_specific_rules xcomp_acquire_specific prep_specific_to poss_model_our pobj_to_model cc_acquire_and aux_induce_to conj_acquire_induce poss_probabilities_their dobj_induce_probabilities nsubj_parse_we rcmod_p_parse det_side_the amod_side_English dobj_parse_side prep_side_of poss_corpus_our pobj_of_corpus prep_parse_with det__an amod__in-house nn__implementation pobj_with_ prep__of nn__Collins amod__parsing nn__models pobj_of_ cc_parse_and nsubj_word-align_we conj_parse_word-align det_corpus_the amod_corpus_parallel dobj_word-align_corpus prep_word-align_with det_implementation_the nn_implementation_Giza nn_implementation_+ num_implementation_+2 pobj_with_implementation prep_implementation_of det_models_the nn_models_IBM pobj_of_models ' +o,3010,'The data sets used are the standard data sets for this problem taken from the Wall Street Journal corpus in the Penn Treebank ',Marcus,'det_data_The nsubj_sets_data csubj_sets_used cop_sets_are det_sets_the amod_sets_standard nn_sets_data ccomp_sets_sets prep_sets_for det__this nn__problem pobj_for_ partmod__taken prep_taken_from det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_from_corpus prep_corpus_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +o,3011,'We achieve competitive performance in comparison to alternate model families , in particular generative models such as MRFs trained with EM and HMMs trained with soft constraints ',Haghighi,'nsubj_achieve_We amod_performance_competitive dobj_achieve_performance prep_performance_in pobj_in_comparison prep_achieve_to amod_families_alternate nn_families_model pobj_to_families prep_achieve_in amod_models_particular nn_models_generative pobj_in_models dep_as_such prep_models_as pobj_as_MRFs partmod_MRFs_trained prep_trained_with nn__EM pobj_with_ cc__and conj__HMMs partmod__trained prep_trained_with amod__soft nn__constraints pobj_with_ ' +o,3012,'23 ITG Constraints The Inversion Transduction Grammar -LRB- ITG -RRB- , a derivative of the Syntax Directed Transduction Grammars , constrains the possible permutations of the input string by defining rewrite rules that indicate permutations of the string ',Wu,'nsubj_constrains_23 nn__ITG nn__Constraints nn__The nn__Inversion nn__Transduction nn__Grammar abbrev__ITG dep_23_ det_derivative_a appos_23_derivative prep_derivative_of det__the nn__Syntax nn__Directed nn__Transduction nn__Grammars pobj_of_ det_permutations_the amod_permutations_possible dobj_constrains_permutations prep_permutations_of det_string_the nn_string_input pobj_of_string prep_constrains_by pcomp_by_defining amod_rules_rewrite dobj_defining_rules nsubj_indicate_that rcmod_rules_indicate dobj_indicate_permutations prep_permutations_of det_string_the pobj_of_string ' +o,3013,'Following , 765 Feature Sets Templates Error \% A Ratnaparkhis 305 B A + -LRB- t0 , t1 -RRB- , -LRB- t0 , t1 , t1 -RRB- , -LRB- t0 , t1 , t2 -RRB- 292 C B + -LRB- t0 , t2 -RRB- , -LRB- t0 , t2 -RRB- , -LRB- t0 , t2 , w0 -RRB- , -LRB- t0 , t1 , w0 -RRB- , -LRB- t0 , t1 , w0 -RRB- , -LRB- t0 , t2 , w0 -RRB- , -LRB- t0 , t2 , t1 , w0 -RRB- , -LRB- t0 , t1 , t1 , w0 -RRB- , -LRB- t0 , t1 , t2 , w0 -RRB- 284 D C + -LRB- t0 , w1 , w0 -RRB- , -LRB- t0 , w1 , w0 -RRB- 278 E D + -LRB- t0 , X = prefix or suffix of w0 -RRB- ,4 -LRB- X 9 272 Table 2 : Experiments on the development data with beam width of 3 we cut the PTB into the training , development and test sets as shown in Table 1 ',Collins,'prep_Templates_Following pobj_Following_ nn_Sets_765 nn_Sets_Feature appos__Sets num_\%_Error dobj_Templates_\% det_B_A nn_B_Ratnaparkhis num_B_305 nsubj_Templates_B det_+_A dep_B_+ appos_+_t0 dep_t0_t1 nn_t1_t0 dep_t1_t1 appos_+_t1 nn_t2_t0 dep_t2_t1 appos_+_t2 nn_D_292 nn_D_C nn_D_B nn_D_+ appos_D_t0 dep_t0_t2 appos_D_t0 dep_t0_t2 nn_w0_t0 dep_w0_t2 appos_D_w0 nn_w0_t0 dep_w0_t1 appos_D_w0 nn_w0_t0 dep_w0_t1 appos_D_w0 nn_w0_t0 dep_w0_t2 appos_D_w0 appos_D_t0 nn_w0_t2 dep_w0_t1 dep_t0_w0 appos_D_t0 nn_w0_t1 dep_w0_t1 dep_t0_w0 appos_D_t0 nn_w0_t1 dep_w0_t2 dep_t0_w0 nn_D_284 measure_E_D nn_+_C dep_D_+ nn_w0_t0 dep_w0_w1 appos_D_w0 nn_w0_t0 dep_w0_w1 appos_D_w0 num_D_278 amod_D_E nsubj_+_D rcmod_+_+ dep_=_t0 nsubj_=_X dep_Experiments_= dobj_=_prefix cc_prefix_or conj_prefix_suffix prep_prefix_of pobj_of_w0 nsubj_Experiments_,4 partmod_,4_X number_272_9 num_Table_272 dobj_X_Table num_Table_2 dep_cut_Experiments prep_Experiments_on det_data_the nn_data_development pobj_on_data prep_Experiments_with nn_width_beam pobj_with_width prep_width_of pobj_of_3 nsubj_cut_we ccomp_+_cut det_PTB_the dobj_cut_PTB prep_cut_into det_sets_the nn_sets_training conj_training_development cc_training_and conj_training_test pobj_into_sets mark_shown_as advcl_cut_shown prep_shown_in pobj_in_Table num_Table_1 ' +o,3014,'Another current topic of machine translation is automatic evaluation of MT quality ',Papineni,'det_topic_Another amod_topic_current nsubj_evaluation_topic prep_topic_of nn_translation_machine pobj_of_translation cop_evaluation_is amod_evaluation_automatic prep_evaluation_of nn_quality_MT pobj_of_quality ' +o,3015,'This model can be seen as an extension of the standard Maximum Entropy Markov Model -LRB- MEMM , see -RRB- with an extra dependency on the predicate label , we will henceforth refer to this model as MEMM + pred ',Ratnaparkhi,'det_model_This nsubjpass_seen_model aux_seen_can auxpass_seen_be prep_seen_as det_extension_an pobj_as_extension prep_extension_of det_Model_the amod_Model_standard nn_Model_Maximum nn_Model_Entropy nn_Model_Markov pobj_of_Model parataxis_seen_MEMM dep_MEMM_see dobj_MEMM_ prep_seen_with det_dependency_an amod_dependency_extra pobj_with_dependency prep_refer_on det_label_the amod_label_predicate pobj_on_label nsubj_refer_we aux_refer_will advmod_refer_henceforth dep_seen_refer prep_refer_to det_model_this pobj_to_model mark_pred_as nn_+_MEMM nsubj_pred_+ advcl_refer_pred ' +o,3016,'Obviously , all these semantic resources have been acquiredusing a very differentset of processes , tools and corpora ',Snow,'advmod_acquiredusing_Obviously predet_resources_all det_resources_these amod_resources_semantic nsubj_acquiredusing_resources aux_acquiredusing_have aux_acquiredusing_been det_differentset_a amod_differentset_very dobj_acquiredusing_differentset prep_differentset_of nn__processes pobj_of_ conj__tools cc__and conj__corpora ' +o,3017,'Following and other work on general-purpose generators , we adopt BLEU score , average simple string accuracy -LRB- SSA -RRB- and percentage of exactly matched sentences for accuracy evaluation6 For coverage evaluation , we measure the percentage of input fstructures that generate a sentence ',Papineni,'prep_adopt_Following amod_work_ cc__and conj__other pobj_Following_work prep_work_on amod_generators_general-purpose pobj_on_generators nsubj_adopt_we amod__BLEU nn__score dobj_adopt_ amod_accuracy_average amod_accuracy_simple nn_accuracy_string conj__accuracy abbrev_accuracy_SSA cc__and conj__percentage prep_percentage_of advmod_sentences_exactly amod_sentences_matched pobj_of_sentences prep_sentences_for nn_evaluation6_accuracy pobj_for_evaluation6 prep_measure_For nn_evaluation_coverage pobj_For_evaluation nsubj_measure_we dep_adopt_measure det_percentage_the dobj_measure_percentage prep_percentage_of nn_fstructures_input pobj_of_fstructures nsubj_generate_that rcmod_percentage_generate det_sentence_a dobj_generate_sentence ' +o,3018,'Therefore , including a model based on surface forms , as suggested , is also necessary ',Koehn,'advmod_necessary_Therefore dep_necessary_including det_model_a pobj_including_model partmod_model_based prep_based_on nn_forms_surface pobj_on_forms mark_suggested_as dep_necessary_suggested acomp_suggested_ cop_necessary_is advmod_necessary_also ' +o,3019,'It has shown promise in improving the performance of many tasks such as name tagging , semantic class extraction , chunking , coreference resolution and text classification ',Bean,'nsubj_shown_It aux_shown_has dobj_shown_promise prep_shown_in pcomp_in_improving det_performance_the dobj_improving_performance prep_performance_of amod_tasks_many pobj_of_tasks dep_as_such prep_tasks_as pobj_as_name partmod_name_tagging amod__ amod__semantic nn__class nn__extraction dobj_tagging_ amod__chunking conj__ amod__coreference nn__resolution conj__ cc__and nn__text nn__classification conj__ ' +o,3020,' argue for CL on grounds of accuracy , but see also Johnson -LRB- 2001 -RRB- ',Klein,'ccomp__argue prep_argue_for pobj_for_CL prep_CL_on pobj_on_grounds prep_grounds_of pobj_of_accuracy cc_argue_but conj_argue_see advmod_see_also nsubj__Johnson appos_Johnson_2001 ' +o,3021,'We evaluated the translation quality using the BLEU metric , as calculated by mteval-v11bpl with its default setting except that we used case-sensitive matching of n-grams ',Papineni,'nsubj_evaluated_We det_quality_the nn_quality_translation dobj_evaluated_quality partmod_quality_using det__the amod__BLEU amod__metric dobj_using_ mark_calculated_as advcl_evaluated_calculated prep_calculated_by pobj_by_mteval-v11bpl prep_calculated_with poss_default_its pobj_with_default partmod_default_setting prep_setting_except complm_used_that nsubj_used_we pcomp_except_used amod_matching_case-sensitive dobj_used_matching prep_matching_of pobj_of_n-grams ' +o,3022,'We should note , however , that most other stochastic parsers do include counts of single nonheadwords : they appear in the backed-off statistics of these parsers ',Collins,'nsubj_note_We aux_note_should advmod_note_however complm_include_that amod_parsers_most amod_parsers_other amod_parsers_stochastic nsubj_include_parsers aux_include_do ccomp_note_include dobj_include_counts prep_counts_of amod_nonheadwords_single pobj_of_nonheadwords nsubj_appear_they parataxis_note_appear prep_appear_in det_statistics_the amod_statistics_backed-off pobj_in_statistics prep_statistics_of det_parsers_these pobj_of_parsers ' +n,3023,'Our method , extending this line of research with the use of labeled LFG dependencies , partial matching , and n-best parses , allows us to considerably outperform highest correlations with human judgement -LRB- they report 0144 for the correlation with human fluency judgement , 0202 for the correlation with human overall judgement -RRB- , although it has to be kept in mind that such comparison is only tentative , as their correlation is calculated on a different test set ',Liu,'poss_method_Our nsubj_allows_method partmod_method_extending det_line_this dobj_extending_line prep_line_of pobj_of_research prep_extending_with det_use_the pobj_with_use prep_use_of amod_dependencies_labeled nn_dependencies_LFG pobj_of_dependencies amod_matching_partial conj_dependencies_matching cc_dependencies_and amod_parses_n-best conj_dependencies_parses nsubj_outperform_us aux_outperform_to advmod_outperform_considerably xcomp_allows_outperform amod_correlations_ dep__highest dobj_outperform_correlations prep_outperform_with amod_judgement_human pobj_with_judgement nsubj_report_they dep_judgement_report dobj_report_0144 prep_0144_for det_correlation_the pobj_for_correlation prep_report_with amod_judgement_human nn_judgement_fluency pobj_with_judgement appos_judgement_0202 prep_0202_for det_correlation_the pobj_for_correlation prep_correlation_with amod_judgement_human amod_judgement_overall pobj_with_judgement mark_has_although nsubj_has_it advcl_outperform_has aux_kept_to auxpass_kept_be xcomp_has_kept prep_kept_in pobj_in_mind complm_tentative_that amod_comparison_such nsubj_tentative_comparison cop_tentative_is advmod_tentative_only ccomp_kept_tentative mark_calculated_as poss_correlation_their nsubjpass_calculated_correlation auxpass_calculated_is advcl_tentative_calculated prep_calculated_on det_set_a amod_set_different nn_set_test pobj_on_set ' +o,3024,'Our method is based on the ones described in , The objective of this paper is to dynamically rank speakers or participants in a discussion ',Mihalcea,'poss_method_Our nsubjpass_based_method auxpass_based_is prep_based_on det_ones_the pobj_on_ones partmod_ones_described prep_is_in pobj_in_ det_objective_The nsubj_is_objective prep_objective_of det_paper_this pobj_of_paper dep_based_is aux_rank_to advmod_rank_dynamically xcomp_is_rank dobj_rank_speakers cc_speakers_or conj_speakers_participants prep_rank_in det_discussion_a pobj_in_discussion ' +p,3025,'The best examples of this approach has been the resent work of Yarowsky , , ',Yarowsky,'det_examples_The amod_examples_best nsubj_been_examples prep_examples_of det_approach_this pobj_of_approach aux_been_has det_work_the amod_work_resent dobj_been_work prep_work_of pobj_of_Yarowsky num_Yarowsky_ num_Yarowsky_ ' +o,3026,'2 Incremental Parsing This section gives a description of Collins and Roarks incremental parser and discusses its problem ',Collins,'num_Parsing_2 nn_Parsing_Incremental det_section_This nsubj_gives_section dep_Parsing_gives det_description_a dobj_gives_description prep_description_of nn__Collins cc_Collins_and conj_Collins_Roarks amod__incremental nn__parser pobj_of_ cc_gives_and conj_gives_discusses poss_problem_its dobj_discusses_problem ' +o,3027,'Hockenmaier et al , although to some extent following the approach of Xia where LTAGs are extracted , have pursued an alternative by extracting Combinatory Categorial Grammar -LRB- CCG -RRB- lexicons from the Penn Treebank ',Marcus,'nn__Hockenmaier cc_Hockenmaier_et conj_Hockenmaier_al nsubj_pursued_ mark_extent_although aux_extent_to det_extent_some infmod__extent prep_extent_following det_approach_the pobj_following_approach prep_approach_of nn__Xia pobj_of_ advmod_extracted_where nsubjpass_extracted_LTAGs auxpass_extracted_are rcmod__extracted aux_pursued_have det_alternative_an dobj_pursued_alternative prep_pursued_by pcomp_by_extracting nn_Grammar_Combinatory nn_Grammar_Categorial dobj_extracting_Grammar abbrev_Grammar_CCG advmod_lexicons_ amod_Grammar_lexicons prep_extracting_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3028,'The collocations have been calculated according to the method described in by moving a window on the texts ',Church,'det_collocations_The nsubjpass_calculated_collocations aux_calculated_have auxpass_calculated_been prep_calculated_according dep_according_to det_method_the pobj_to_method partmod_method_described prep_described_in pobj_in_ prep_described_by pcomp_by_moving det_window_a dobj_moving_window prep_moving_on det_texts_the pobj_on_texts ' +o,3029,' first introduced an iterative method for increasing a small set of seed data used to disambiguate dual word senses by exploiting the constraint that in a segment of discourse only one sense of a word is used ',Yarowsky,'advmod_introduced_ advmod_introduced_first det_method_an amod_method_iterative dobj_introduced_method prep_method_for pcomp_for_increasing det_set_a amod_set_small dobj_increasing_set prep_set_of nn_data_seed pobj_of_data partmod_data_used aux_disambiguate_to xcomp_used_disambiguate amod_senses_dual nn_senses_word dobj_disambiguate_senses prep_disambiguate_by pcomp_by_exploiting det_constraint_the dobj_exploiting_constraint dep_used_that prep_used_in det_segment_a pobj_in_segment prep_segment_of pobj_of_discourse advmod_sense_only num_sense_one nsubjpass_used_sense prep_sense_of det_word_a pobj_of_word auxpass_used_is dep_introduced_used ' +o,3030,'4 Using vector-based models of semantic representation to account for the systematic variances in neural activity 41 Lexical Semantic Representation Computational linguists have demonstrated that a words meaning is captured to some extent by the distribution of words and phrases with which it commonly co-occurs ',Church,'partmod_4_Using amod_models_vector-based dobj_Using_models prep_models_of amod_representation_semantic pobj_of_representation aux_account_to xcomp_Using_account prep_account_for det_variances_the amod_variances_systematic pobj_for_variances prep_variances_in amod_activity_neural pobj_in_activity num_linguists_41 nn_linguists_Lexical nn_linguists_Semantic nn_linguists_Representation nn_linguists_Computational nsubj_demonstrated_linguists aux_demonstrated_have dep_4_demonstrated complm_captured_that det_meaning_a nn_meaning_words nsubjpass_captured_meaning auxpass_captured_is ccomp_demonstrated_captured prep_captured_to det_extent_some pobj_to_extent prep_captured_by det_distribution_the pobj_by_distribution prep_distribution_of pobj_of_words cc_words_and conj_words_phrases rel_co-occurs_with pobj_with_which nsubj_co-occurs_it advmod_co-occurs_commonly rcmod_distribution_co-occurs ' +o,3031,'The Stanford parser is representative of a large number of PTB parsers , exemplified by and Charniak -LRB- 2000 -RRB- ',Collins,'det_parser_The nn_parser_Stanford nsubj_representative_parser cop_representative_is prep_representative_of det_number_a amod_number_large pobj_of_number prep_number_of nn_parsers_PTB pobj_of_parsers partmod_number_exemplified prep_exemplified_by pobj_by_ cc__and conj__Charniak dep_exemplified_2000 ' +o,3032,'23 Previous Randomized LMs Recent work has used lossy encodings based on Bloom filters to represent logarithmically quantized corpus statistics for language modeling ',Talbot,'num__23 amod__Previous nn__Randomized nn__LMs amod__Recent nn__work nsubj_used_ aux_used_has amod_encodings_lossy dobj_used_encodings prep_used_based dep_based_on nn_filters_Bloom pobj_on_filters partmod_filters_ aux_represent_to xcomp__represent advmod_quantized_logarithmically amod_statistics_quantized nn_statistics_corpus dobj_represent_statistics prep_statistics_for nn_modeling_language pobj_for_modeling ' +o,3033,'The disambiguation model of Enju is based on a feature forest model , which is a log-linear model on packed forest structure ',Berger,'det_model_The nn_model_disambiguation nsubjpass_based_model prep_model_of pobj_of_Enju auxpass_based_is prep_based_on det__a nn__feature nn__forest nn__model pobj_on_ nsubj__which cop__is det__a amod__log-linear nn__model rcmod__ prep__on amod_structure_packed nn_structure_forest pobj_on_structure ' +o,3034,'All formats 2The data described in (Ramshaw and Marcus, 1995) is available from ftp://ftp.cis.upenn.edu/pub/chunker/ 175 Proceedings of EACL \'99 word/POS context chunk tag context IOB1 L=2/R=I IOB2 L--2/R=I IOE1 L=I/R=2 IOE2 L=I/R=2 [ +] L=2/R=I + L=0/R=2 [ + IO L=2/R=0 + L=I/R=I IO +] L=I/R=I+L=0/R=2 F~=I 1/2 90.12 1/0 89.30 1/2 89.55 0/1 89.73 0/0 + 0/0 89.32 0/0 + I/I 89.78 1/1 + 0/0 89.86 Table 3: Results second experiment series: the best F~=I scores for different left (L) and right (R) chunk tag context sizes for the seven representation formats using 5-fold cross-validation on section 15 of the WSJ corpus.',Ramshaw,'' +p,3035,'The IBM models 1-5 produce word alignments with increasing algorithmic complexity and performance ',Brown,'det_1-5_The nn_1-5_IBM nn_1-5_models nsubj__1-5 xcomp__produce nn_alignments_word dobj_produce_alignments prep_produce_with pcomp_with_increasing amod_complexity_algorithmic dobj_increasing_complexity cc_complexity_and conj_complexity_performance ' +o,3036,'Part-of-Speech -LRB- POS -RRB- annotation for example can be seen as the task of choosing the appropriate tag for a word from an ontology of word categories -LRB- compare for example the Penn Treebank POS tagset as described in -RRB- ',Marcus,'appos_Part-of-Speech_POS nsubjpass_seen_annotation prep_annotation_for pobj_for_example aux_seen_can auxpass_seen_be rcmod_Part-of-Speech_seen prep_seen_as det_task_the pobj_as_task prep_task_of pcomp_of_choosing det_tag_the amod_tag_appropriate dobj_choosing_tag prep_tag_for det_word_a pobj_for_word prep_choosing_from det_ontology_an pobj_from_ontology prep_ontology_of nn_categories_word pobj_of_categories dep_Part-of-Speech_compare prep_compare_for pobj_for_example det_tagset_the nn_tagset_Penn nn_tagset_Treebank dep_tagset_POS dep_example_tagset advmod_described_as amod_example_described prep_example_in pobj_in_ ' +n,3037,'Unlike well-known bootstrapping approaches , EM and CE have the possible advantage of maintaining posteriors over hidden labels -LRB- or structure -RRB- throughout learning ; bootstrapping either chooses , for each example , a single label , or remains completely agnostic ',Yarowsky,'prep_have_Unlike amod_approaches_well-known nn_approaches_bootstrapping pobj_Unlike_approaches nsubj_have_ conj__EM cc__and conj__CE det_advantage_the amod_advantage_possible dobj_have_advantage prep_advantage_of pcomp_of_maintaining dobj_maintaining_posteriors prep_posteriors_over amod_labels_hidden pobj_over_labels cc_structure_or dep_labels_structure prep_maintaining_throughout pcomp_throughout_learning nsubj_chooses_bootstrapping advmod_chooses_either parataxis_have_chooses prep_label_for det_example_each pobj_for_example det_label_a amod_label_single conj_chooses_label cc_chooses_or cop_agnostic_remains advmod_agnostic_completely conj_chooses_agnostic ' +o,3038,'We have also implemented a Bloom Filter LM in Joshua , following ',Talbot,'nsubj_implemented_We aux_implemented_have advmod_implemented_also det_LM_a nn_LM_Bloom nn_LM_Filter dobj_implemented_LM prep_LM_in pobj_in_Joshua xcomp_implemented_following ' +o,3039,'pointwise mutual information , 3 ',Church,'amod__pointwise amod__mutual nn__information appos__3 ' +o,3040,'72 Minimum-Risk Training Adjusting or changes the distribution p Minimum error rate training -LRB- MERT -RRB- tries to tune to minimize the BLEU loss of a decoder that chooses the most probable output according to p ',Och,'nn_Adjusting_Minimum-Risk nn_Adjusting_Training tmod_tries_Adjusting dep_Adjusting_or dep_Adjusting_changes det_training_the nn_training_distribution nn_training_p nn_training_Minimum nn_training_error nn_training_rate nsubj_tries_training abbrev_training_MERT advmod_tries_ rcmod_72_tries prep_tries_to pobj_to_tune aux_minimize_to xcomp_tries_minimize det_loss_the amod_loss_BLEU dobj_minimize_loss prep_72_of det_decoder_a pobj_of_decoder nsubj_chooses_that rcmod_decoder_chooses det_output_the advmod_probable_most amod_output_probable dobj_chooses_output prep_chooses_according dep_according_to pobj_to_p ' +n,3041,'1 Introduction Recent approaches to statistical machine translation -LRB- SMT -RRB- piggyback on the central concepts of phrasebased SMT and at the same time attempt to improve some of its shortcomings by incorporating syntactic knowledge in the translation process ',Och,'num_approaches_1 nn_approaches_Introduction amod_approaches_Recent nsubj_attempt_approaches prep_approaches_to amod_piggyback_statistical nn_piggyback_machine nn_piggyback_translation abbrev_piggyback_SMT pobj_to_piggyback prep_piggyback_on det_concepts_the amod_concepts_central pobj_on_concepts prep_concepts_of amod__phrasebased nn__SMT pobj_of_ cc_to_and conj_to_at det_time_the amod_time_same pobj_at_time aux_improve_to xcomp_attempt_improve dobj_improve_some prep_some_of poss_shortcomings_its pobj_of_shortcomings prep_improve_by pcomp_by_incorporating amod_knowledge_syntactic dobj_incorporating_knowledge prep_incorporating_in det_process_the nn_process_translation pobj_in_process ' +o,3042,'We use a hand-written competence grammar , combined with performance-driven disambiguation obtained from the Penn Treebank ',Marcus,'nsubj_use_We det_grammar_a amod_grammar_hand-written nn_grammar_competence dobj_use_grammar prep_use_combined dep_combined_with amod_disambiguation_performance-driven pobj_with_disambiguation partmod_disambiguation_obtained prep_obtained_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3043,'In Section 2 , we examine aggregate Markov models , or class-based bigram models in which the mapping from words to classes 81 is probabilistic ',Brown,'prep_examine_In pobj_In_Section num_Section_2 nsubj_examine_we amod_models_aggregate nn_models_Markov dobj_examine_models cc_examine_or amod_models_class-based nn_models_bigram nsubj__models conj_examine_ dep_probabilistic_in pobj_in_which det_mapping_the nsubj_probabilistic_mapping prep_mapping_from pobj_from_words prep_words_to pobj_to_classes num_classes_81 cop_probabilistic_is ccomp__probabilistic ' +o,3044,'For example , in the context of syntactic disambiguation , Black and Magerman proposed statistical parsing models based-on decisiontree learning techniques , which incorporated not only syntactic but also lexical\\/semantic information in the decision-trees ',Dunning,'prep_decisiontree_For pobj_For_example prep_decisiontree_in det_context_the pobj_in_context prep_context_of amod_disambiguation_syntactic pobj_of_disambiguation nn__Black nsubj_decisiontree_ cc__and nn_models_Magerman nn_models_ amod_models_proposed amod_models_statistical amod_models_parsing conj__models amod_models_based-on xcomp_decisiontree_learning dobj_learning_techniques nsubj_incorporated_which rcmod_techniques_incorporated cc_techniques_not dep_not_only amod_information_syntactic dep_also_but cc_syntactic_also conj_syntactic_lexical\\/semantic conj_techniques_information prep_information_in det_decision-trees_the pobj_in_decision-trees ' +o,3045,'We propose a method similar to to generalize beyond the training set ',Yarowsky,'nsubj_propose_We det_method_a nsubj_similar_method xcomp_propose_similar aux__to xcomp_similar_ aux_generalize_to xcomp__generalize prep_generalize_beyond det_set_the nn_set_training pobj_beyond_set ' +o,3046,'SMT has evolved from the original word-based approach into phrase-based approaches and syntax-based approaches ',Brown,'nsubj_evolved_SMT aux_evolved_has prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep_evolved_into amod__phrase-based nn__approaches pobj_into_ cc__and amod__syntax-based nn__approaches conj__ ' +o,3047,'So we will engineer more such features , especially with lexicalization and soft alignments , and study the impact of alignment quality on parsing improvement ',Liang,'cc_engineer_So nsubj_engineer_we aux_engineer_will amod_features_more amod_features_such dobj_engineer_features conj_engineer_especially prep_especially_with nn__lexicalization cc_lexicalization_and conj_lexicalization_soft nn__alignments pobj_with_ cc_engineer_and conj_engineer_study det_impact_the dobj_study_impact prep_impact_of amod_quality_alignment pobj_of_quality prep_study_on pcomp_on_parsing dobj_parsing_improvement ' +o,3048,' , the BBN parser builds augmented parse trees according to a process similar to that described in ',Collins,'advmod_builds_ det_parser_the nn_parser_BBN nsubj_builds_parser amod_trees_augmented nn_trees_parse dobj_builds_trees prep_builds_according dep_according_to det_process_a pobj_to_process amod_process_similar prep_similar_to pobj_to_that partmod_that_described prep_described_in ' +o,3049,'Starting with bilingualphrasepairsextractedfromautomatically aligned parallel text , these PSCFG approaches augment each contiguous -LRB- in source and target words -RRB- phrase pair with a left-hand-side symbol -LRB- like the VP in the example above -RRB- , and perform a generalization procedure to form rules that include nonterminal symbols ',Koehn,'prep_approaches_Starting dep_Starting_with advmod_aligned_bilingualphrasepairsextractedfromautomatically amod__aligned amod__parallel nn__text pobj_with_ det_PSCFG_these nsubj_approaches_PSCFG xcomp_approaches_augment det_pair_each amod_pair_contiguous dep_contiguous_in nn_words_source cc_source_and conj_source_target pobj_in_words nn_pair_phrase dobj_augment_pair prep_augment_with det_symbol_a amod_symbol_left-hand-side pobj_with_symbol dep_symbol_like det_VP_the pobj_like_VP prep_VP_in det_example_the pobj_in_example advmod_example_above cc_augment_and conj_augment_perform det_procedure_a nn_procedure_generalization dobj_augment_procedure aux_form_to xcomp_augment_form dobj_form_rules nsubj_include_that rcmod_rules_include amod_symbols_nonterminal dobj_include_symbols ' +o,3050,'The current approach does not use specialized probability features as in in any stage during decoder parameter training ',Och,'det_approach_The amod_approach_current nsubj_use_approach aux_use_does neg_use_not amod_features_specialized nn_features_probability dobj_use_features dep_in_as prep_use_in pcomp_in_ prep__in det_stage_any pobj_in_stage prep__during nn_training_decoder nn_training_parameter pobj_during_training ' +o,3051,'The simplest model of compound noun disambiguation compares the frequencies of the two competing analyses and opts for the most frequent one (Pustejovsky et al. , Model Alta BNC Baseline 63.93 63.93 f (n1;n2) : f (n2;n3) 77.86 66.39 f (n1;n2) : f (n1;n3) 78.68# 65.57 f (n1;n2)= f (n1) : f (n2;n3)= f (n2) 68.85 65.57 f (n1;n2)= f (n2) : f (n2;n3)= f (n3) 70.49 63.11 f (n1;n2)= f (n2) : f (n1;n3)= f (n3) 80.32 66.39 f (n1;n2) : f (n2;n3) (NEAR) 68.03 63.11 f (n1;n2) : f (n1;n3) (NEAR) 71.31 67.21 f (n1;n2)= f (n1) : f (n2;n3)= f (n2) (NEAR) 61.47 62.29 f (n1;n2)= f (n2) : f (n2;n3)= f (n3) (NEAR) 65.57 57.37 f (n1;n2)= f (n2) : f (n1;n3)= f (n3) (NEAR) 75.40 68.03# Table 8: Performance of Altavista counts and BNC counts for compound bracketing (data from Lauer 1995) Model Accuracy Baseline 63.93 Best BNC 68.036 Lauer (1995): adjacency 68.90 Lauer (1995): dependency 77.50 Best Altavista 78.686 Lauer (1995): tuned 80.70 Upper bound 81.50 Table 9: Performance comparison with the literature for compound bracketing 1993).',Marcus,'' +o,3052,'It was also included in the DUC 2004 evaluation plan where summary quality was automatically judged using a set of n-gram word overlap metrics called ROUGE ',Lin,'nsubjpass_included_It auxpass_included_was advmod_included_also prep_included_in det_plan_the nn_plan_DUC num_plan_2004 nn_plan_evaluation pobj_in_plan advmod_judged_where nn_quality_summary nsubjpass_judged_quality auxpass_judged_was advmod_judged_automatically advcl_included_judged xcomp_judged_using det_set_a nsubj_overlap_set prep_set_of amod_word_n-gram pobj_of_word ccomp_using_overlap nsubj_called_metrics ccomp_overlap_called advmod_called_ROUGE ' +o,3053,'In a test set containing 26 repairs Dowding et al 1993 , they obtained a detection recall rate of 42 \% with a precision of 85 \% , and a correction recall rate of 31 \% with a precision of 62 \% ',Marcus,'prep_obtained_In det_test_a pobj_In_test partmod_test_set xcomp_set_containing num_repairs_26 dobj_containing_repairs partmod_repairs_Dowding nn_al_et dobj_Dowding_al num_al_1993 nsubj_obtained_they det_rate_a nn_rate_detection nn_rate_recall dobj_obtained_rate prep_rate_of num_\%_42 pobj_of_\% prep_\%_with det_precision_a pobj_with_precision prep_precision_of num_\%_85 pobj_of_\% cc_rate_and det_rate_a nn_rate_correction nn_rate_recall conj_rate_rate prep_rate_of num_\%_31 pobj_of_\% prep_obtained_with det_precision_a pobj_with_precision prep_precision_of num_\%_62 pobj_of_\% ' +o,3054,'A la , they represent the words as a sequence of labeled words with IOB annotations , where the B marks a word at the beginning of a chunk , I marks a word inside a chunk , and O marks those words -LRB- and punctuation -RRB- that are outside chunks ',Ramshaw,'det_la_A nsubj__la nsubj_represent_they ccomp__represent det_words_the dobj_represent_words prep_represent_as det_sequence_a pobj_as_sequence prep_sequence_of amod_words_labeled pobj_of_words prep_words_with nn_annotations_IOB pobj_with_annotations advmod_marks_where det_B_the nsubj_marks_B dep_marks_marks det_word_a dobj_marks_word prep_marks_at det_beginning_the pobj_at_beginning prep_beginning_of det_chunk_a pobj_of_chunk nsubj_marks_I ccomp__marks det_word_a dobj_marks_word prep_word_inside det_chunk_a pobj_inside_chunk cc__and nsubj_marks_O conj__marks det_words_those dobj_marks_words dep_punctuation_and dep_words_punctuation nsubj_chunks_that cop_chunks_are amod_chunks_outside ccomp_words_chunks ' +o,3055,'Since we also adopt a linear scoring function in Equation -LRB- 3 -RRB- , the feature weights of our combination model can also be tuned on a development data set to optimize the specified evaluation metrics using the standard Minimum Error Rate Training -LRB- MERT -RRB- algorithm ',Och,'mark_adopt_Since nsubj_adopt_we advmod_adopt_also advcl_tuned_adopt det_function_a amod_function_linear amod_function_scoring dobj_adopt_function prep_function_in pobj_in_Equation appos_Equation_3 det_weights_the nn_weights_feature nsubjpass_tuned_weights prep_weights_of poss_model_our nn_model_combination pobj_of_model aux_tuned_can advmod_tuned_also auxpass_tuned_be prep_tuned_on det_data_a nn_data_development pobj_on_data partmod_data_set aux_optimize_to xcomp_set_optimize det_metrics_the amod_metrics_specified nn_metrics_evaluation dobj_optimize_metrics partmod_metrics_using det_algorithm_the amod_Training_standard nn_Training_Minimum nn_Training_Error nn_Training_Rate nn_algorithm_Training abbrev_Training_MERT dobj_using_algorithm ' +o,3056,'Based on annotation differences in the datasets and a bug in their system , their results are inconclusive ',Dredze,'prep_inconclusive_Based dep_Based_on nn_differences_annotation pobj_on_differences prep_differences_in det__the amod__datasets pobj_in_ cc_differences_and det_bug_a conj_differences_bug prep_bug_in poss__their nn__system pobj_in_ poss_results_their nsubj_inconclusive_results cop_inconclusive_are ' +o,3057,'While this approach exploits only syntactic and lexical information , also rely on cohesion information , derived from word distribution in a text : Phrases that are linked to a local context are retained , while phrases that have no such links are dropped ',Jing,'mark_information_While det_exploits_this nn_exploits_approach dep_information_exploits advmod_information_only amod_information_syntactic cc_syntactic_and conj_syntactic_lexical advcl_rely_information nsubj_rely_ advmod_rely_also prep_rely_on nn_information_cohesion pobj_on_information partmod_information_derived prep_derived_from nn_distribution_word pobj_from_distribution prep_distribution_in det_text_a pobj_in_text nsubjpass_retained_Phrases nsubjpass_linked_that auxpass_linked_are rcmod_Phrases_linked prep_linked_to det_context_a amod_context_local pobj_to_context auxpass_retained_are parataxis_rely_retained mark_dropped_while nsubjpass_dropped_phrases nsubj_have_that rcmod_phrases_have det_links_no amod_links_such dobj_have_links auxpass_dropped_are advcl_retained_dropped ' +p,3058,'Probably the most widely used feature weighting function is (point-wise) Mutual Information (MI) (Church and Patrick 1990; Hindle 1990; Luk 1995; Lin 1998; Gauch, Wang, and Rachakonda 1999; Dagan 2000; Baroni and Vegnaduzzo 2004; Chklovski and Pantel 2004; Pantel and Ravichandran 2004; Pantel, Ravichandran, and Hovy 2004; Weeds, Weir, and McCarthy 2004), dened by: weight MI (w,f)=log 2 P(w,f) P(w)P(f) (1) We calculate the MI weights by the following statistics in the space of co-occurrence instances S: weight MI (w,f)=log 2 count(w,f) nrels count(w) count(f) (2) where count(w,f) is the frequency of the co-occurrence pair w,f in S, count(w)and count(f) are the independent frequencies of w and f in S,andnrels is the size of S.High MI weights are assumed to correspond to strong wordfeature associations.',Church,'' +o,3059,'In the rest of the paper we use the following notation , adapted from ',Collins,'prep_use_In det_rest_the pobj_In_rest prep_rest_of det_paper_the pobj_of_paper nsubj_use_we det_notation_the amod_notation_following dobj_use_notation partmod_notation_adapted prep_adapted_from ' +o,3060,'The TRIPS structure generally has more levels of structure -LRB- roughly corresponding to levels in X-bar theory -RRB- than the Penn Treebank analyses , in particular for base noun phrases ',Marcus,'det_structure_The nn_structure_TRIPS nsubj_has_structure advmod_has_generally amod_levels_more dobj_has_levels prep_levels_of pobj_of_structure advmod_corresponding_roughly dep_structure_corresponding prep_corresponding_to pobj_to_levels prep_levels_in amod_theory_X-bar pobj_in_theory prep_structure_than det__the nn__Penn nn__Treebank nn__analyses pobj_than_ prep_levels_in amod_phrases_particular prep_particular_for pobj_for_base nn_phrases_noun pobj_in_phrases ' +o,3061,'Independently , in artificial intelligence an effort arose to encode large amounts of commonsense knowledge ',Hobbs,'prep_Independently_in amod_intelligence_artificial pobj_in_intelligence det_effort_an nsubj_arose_effort rcmod_intelligence_arose aux_encode_to xcomp_arose_encode amod_amounts_large dobj_encode_amounts prep_amounts_of amod_knowledge_commonsense pobj_of_knowledge ' +o,3062,'We extracted all examples of each word from the 14-million-word English portion of the Hansards8 Note that this is considerably smaller than corpus of 460 million words , so bootstrapping will not perform as well , and may be more sensitive to the choice of seed ',Yarowsky,'nsubj_extracted_We det_examples_all nsubj_Note_examples prep_examples_of det_word_each pobj_of_word prep_word_from det_portion_the amod_portion_14-million-word amod_portion_English pobj_from_portion prep_portion_of det_Hansards8_the pobj_of_Hansards8 ccomp_extracted_Note complm_smaller_that nsubj_smaller_this cop_smaller_is advmod_smaller_considerably ccomp_Note_smaller mark_perform_than amod_corpus_ nsubj_perform_corpus prep_corpus_of number_million_460 num_words_million pobj_of_words advmod_bootstrapping_so amod_words_bootstrapping aux_perform_will neg_perform_not advcl_smaller_perform advmod_well_as advmod_perform_well cc_perform_and aux_sensitive_may cop_sensitive_be advmod_sensitive_more conj_perform_sensitive prep_sensitive_to det_choice_the pobj_to_choice prep_choice_of pobj_of_seed ' +o,3063,'Since one of these filters restricts the number of nonterminal symbols to two , our extracted grammar is equivalent to an inversion transduction grammar ',Wu,'mark_restricts_Since nsubj_restricts_one prep_one_of det_filters_these pobj_of_filters advcl_equivalent_restricts det_number_the dobj_restricts_number prep_number_of amod_symbols_nonterminal pobj_of_symbols prep_restricts_to pobj_to_two poss_grammar_our amod_grammar_extracted nsubj_equivalent_grammar cop_equivalent_is prep_equivalent_to det_grammar_an nn_grammar_inversion nn_grammar_transduction pobj_to_grammar ' +o,3064,'More recently , there have been many proposals to introduce syntactic knowledge into SMT models ',Wu,'advmod_recently_More advmod_been_recently expl_been_there aux_been_have amod_proposals_many dobj_been_proposals aux_introduce_to infmod_proposals_introduce amod_knowledge_syntactic dobj_introduce_knowledge prep_introduce_into nn_models_SMT pobj_into_models ' +o,3065,'Sentiment classification at the document level investigates ways to classify each evaluative document -LRB- eg , product review -RRB- as positive or negative ',Pang,'nn_classification_Sentiment nsubj_investigates_classification prep_classification_at det_level_the nn_level_document pobj_at_level nsubj_classify_ways aux_classify_to xcomp_investigates_classify det_document_each amod_document_evaluative dobj_classify_document appos_document_eg nn_review_product dep_eg_review advmod_positive_as amod_document_positive cc_positive_or conj_positive_negative ' +o,3066,'For example , the feature 1 On the ATR English Grammar , see below ; for a detailed description of a precursor to the Gr-r ~ raar , see ',Marcus,'prep_see_For pobj_For_example det_feature_the ccomp_see_feature dep_On_1 dep_feature_On det_Grammar_the nn_Grammar_ATR nn_Grammar_English pobj_On_Grammar advmod_see_below prep_see_for det_description_a amod_description_detailed pobj_for_description prep_description_of det_precursor_a pobj_of_precursor prep_precursor_to det_raar_the amod_raar_Gr-r amod_raar_~ pobj_to_raar parataxis_see_see acomp_see_ ' +o,3067,'The supervised training described in uses manually annotated data for the estimation of the weight coefficients ',Collins,'det_training_The amod_training_supervised nsubj_annotated_training partmod_training_described prep_described_in amod_uses_ pobj_in_uses advmod_annotated_manually dobj_annotated_data prep_annotated_for det_estimation_the pobj_for_estimation prep_estimation_of det_coefficients_the nn_coefficients_weight pobj_of_coefficients ' +o,3068,'4 Training This section discusses how to extract our translation rules given a triple nullnull , null null , nullnull As we know , the traditional tree-to-string rules can be easily extracted from nullnull , null null , nullnull using the algorithm of Mi and Huang 2 We would like 2 Mi and Huang extend the tree-based rule extraction algorithm to forest-based by introducing non-deterministic mechanism ',Galley,'num_Training_4 det_section_This nsubj_discusses_section dep_Training_discusses advmod_extract_how aux_extract_to ccomp_discusses_extract poss_rules_our nn_rules_translation dobj_extract_rules prep_extract_given det_nullnull_a amod_nullnull_triple nsubj_nullnull_nullnull amod_nullnull_null dep_null_null dep_given_nullnull mark_know_As nsubj_know_we advcl_extracted_know det_rules_the amod_rules_traditional amod_rules_tree-to-string nsubjpass_extracted_rules aux_extracted_can auxpass_extracted_be advmod_extracted_easily ccomp_nullnull_extracted prep_extracted_from amod_null_nullnull amod_null_null pobj_from_null advmod_using_nullnull ccomp_nullnull_using det_algorithm_the dobj_using_algorithm prep_algorithm_of pobj_of_Mi cc_using_and conj_using_Huang number_2_ dobj_Huang_2 nsubj_like_We aux_like_would dep_Training_like num_Mi_2 dobj_like_Mi cc_like_and nn__Huang nsubj_extend_ conj_like_extend det_algorithm_the amod_algorithm_tree-based nn_algorithm_rule nn_algorithm_extraction dobj_extend_algorithm amod_algorithm_ dep__to dep__forest-based prep_extend_by pcomp_by_introducing amod_mechanism_non-deterministic dobj_introducing_mechanism ' +o,3069,'We use evaluations similar to those used before ',Rapp,'nsubj_use_We nsubj_similar_evaluations xcomp_use_similar prep_similar_to pobj_to_those partmod_those_used advmod_used_before ' +o,3070,'Syntactic Score -LRB- SC -RRB- Some erroneous sentences often contain words and concepts that are locally correct but can not form coherent sentences ',Liu,'amod_Score_Syntactic abbrev_Score_SC det_sentences_Some amod_sentences_erroneous nsubj_contain_sentences advmod_contain_often dep_Score_contain dobj_contain_words cc_words_and conj_words_concepts nsubj_correct_that cop_correct_are advmod_correct_locally rcmod_words_correct cc_correct_but aux_form_can neg_form_not conj_correct_form amod_sentences_coherent dobj_form_sentences ' +o,3071,'44 Corpora We ran the three syntactic preprocessors over a total of three corpora , of varying size : the Brown corpus -LRB- 460K tokens -RRB- and Wall Street Journal corpus -LRB- 12M tokens -RRB- , both derived from the Penn Treebank , and the written component of the British National Corpus -LRB- 98M tokens : Burnard -RRB- ',Marcus,'num_Corpora_44 nsubj_ran_We dep_Corpora_ran det_preprocessors_the num_preprocessors_three amod_preprocessors_syntactic dobj_ran_preprocessors prep_preprocessors_over det_total_a pobj_over_total prep_total_of num_corpora_three pobj_of_corpora prep_total_of amod_size_varying pobj_of_size det_corpus_the nn_corpus_Brown dep_preprocessors_corpus num_tokens_460K appos_corpus_tokens cc_corpus_and nn_corpus_Wall nn_corpus_Street nn_corpus_Journal conj_corpus_corpus num_tokens_12M appos_corpus_tokens nsubj_derived_both rcmod_corpus_derived prep_derived_from det__the nn__Penn nn__Treebank pobj_from_ cc_preprocessors_and det_component_the amod_component_written conj_preprocessors_component prep_component_of det_Corpus_the nn_Corpus_British nn_Corpus_National pobj_of_Corpus num_tokens_98M dep_Corpus_tokens nn__Burnard dep_tokens_ ' +o,3072,'We briefly describe the tagger -LRB- see for more details -RRB- , a Hidden Markov Model trained with the perceptron algorithm introduced in ',Collins,'nsubj_describe_We advmod_describe_briefly det_tagger_the dobj_describe_tagger dep_tagger_see dobj_see_ prep__for amod_details_more pobj_for_details det_Model_a nn_Model_Hidden nn_Model_Markov appos_tagger_Model partmod_Model_trained prep_trained_with det_algorithm_the nn_algorithm_perceptron pobj_with_algorithm partmod_algorithm_introduced prep_introduced_in ' +o,3073,'3 Data The data consists of sections of the Wall Street Journal part of the Penn TreeBank , with information on predicate-argument structures extracted from the PropBank corpus ',Marcus,'num_data_3 nn_data_Data det_data_The nsubj_consists_data prep_consists_of pobj_of_sections prep_sections_of det_part_the nn_part_Wall nn_part_Street nn_part_Journal pobj_of_part prep_part_of det__the nn__Penn nn__TreeBank pobj_of_ prep_consists_with pobj_with_information prep_information_on amod_structures_predicate-argument pobj_on_structures partmod_structures_extracted prep_extracted_from det__the nn__PropBank nn__corpus pobj_from_ ' +o,3074,'The Kappa statistic is typically used to measure the human interrater agreement ',Carletta,'det__The nn__Kappa nn__statistic nsubjpass_used_ auxpass_used_is advmod_used_typically aux_measure_to xcomp_used_measure det_agreement_the amod_agreement_human nn_agreement_interrater dobj_measure_agreement ' +o,3075,'Method Prec Rec F-measure GIZA + + Intersect 967 530 685 GIZA + + Union 825 690 751 GIZA + + GDF 840 682 752 Phrasal ITG 507 803 622 Phrasal ITG + NCC 754 780 767 Following the lead of , we hand-aligned the first 100 sentence pairs of our training set according to the Blinker annotation guidelines ',Fraser,'ccomp_hand-aligned_Method nn_Intersect_Prec nn_Intersect_Rec nn_Intersect_F-measure nn_Intersect_GIZA nn_Intersect_+ nn_Intersect_+ dobj_Method_Intersect num_GIZA_967 number_685_530 num_GIZA_685 nn_GIZA_GIZA nn_GIZA_+ nn_GIZA_+ nn_GIZA_Union num_GIZA_825 number_751_690 num_GIZA_751 prep_Intersect_GIZA amod_GIZA_+ cc_+_+ conj_+_GDF num_767_840 number_752_682 num_767_752 nn_767_Phrasal nn_767_ITG nn_767_507 nn_767_803 nn_767_622 nn_767_Phrasal nn_767_ITG nn_767_+ nn_767_NCC num_767_754 amod_767_780 dep_GIZA_767 prep_767_Following det_lead_the pobj_Following_lead prep_lead_of pobj_of_ nsubj_hand-aligned_we det_pairs_the amod_pairs_first num_pairs_100 nn_pairs_sentence dobj_hand-aligned_pairs prep_pairs_of poss_training_our pobj_of_training partmod_training_set prep_set_according dep_according_to det__the nn__Blinker nn__annotation nn__guidelines pobj_to_ ' +o,3076,'Motivated by the fact that non-syntactic phrases make non-trivial contribution to phrase-based SMT , the tree sequencebased translation model is proposed that uses tree sequence as the basic translation unit , rather than using single sub-tree as in the STSG ',Zhang,'ccomp_sequencebased_Motivated prep_Motivated_by det_fact_the pobj_by_fact complm_make_that amod_phrases_non-syntactic nsubj_make_phrases ccomp_Motivated_make amod_contribution_non-trivial dobj_make_contribution prep_make_to amod_SMT_phrase-based pobj_to_SMT det_tree_the nsubj_sequencebased_tree nn_model_translation nsubj__model cop__is amod__proposed ccomp_sequencebased_ nsubj_uses_that rcmod__uses nn_sequence_tree dobj_uses_sequence prep_uses_as det_unit_the amod_unit_basic nn_unit_translation pobj_as_unit dep_as_rather dep_as_than pcomp_than_using amod_sub-tree_single dobj_using_sub-tree prep_using_as dep_as_in det_STSG_the pobj_in_STSG ' +o,3077,'Recent work explored the task of part-of-speech tagging -LRB- PoS -RRB- using unsupervised Hidden Markov Models -LRB- HMMs -RRB- with encouraging results ',Johnson,'amod__Recent nn__work nsubj_explored_ det_task_the dobj_explored_task prep_task_of amod_tagging_part-of-speech pobj_of_tagging appos_tagging_PoS partmod_tagging_using amod_Models_unsupervised nn_Models_Hidden nn_Models_Markov dobj_using_Models abbrev_Models_HMMs prep_Models_with amod_results_encouraging pobj_with_results ' +p,3078,'Even robust parsers using linguistically sophisticated formalisms , such as TAG , CCG , HPSG and LFG , often use training data derived from the Penn Treebank ',Cahill,'advmod_parsers_Even amod_parsers_robust partmod_parsers_using advmod_sophisticated_linguistically amod_formalisms_sophisticated dobj_using_formalisms dep_as_such dep_formalisms_as nn__TAG pobj_as_ nn__CCG conj__ nn__HPSG conj__ cc__and nn__LFG conj__ advmod_use_often dep_parsers_use nn_data_training dobj_use_data partmod_data_derived prep_derived_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3079,' trained their feature set using an online discriminative algorithm ',Tillmann,'nsubj_trained_ poss_feature_their dobj_trained_feature partmod_feature_set xcomp_set_using det_algorithm_an amod_algorithm_online nn_algorithm_discriminative dobj_using_algorithm ' +o,3080,'We evaluated the translation quality using case-insensitive BLEU metric ',Papineni,'nsubj_evaluated_We det_quality_the nn_quality_translation dobj_evaluated_quality partmod_quality_using amod_metric_case-insensitive amod_metric_BLEU dobj_using_metric ' +o,3081,'To generate phrase pairs from a parallel corpus , we use the ` diag-and \' phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Koehn,'aux_generate_To dep_use_generate nn_pairs_phrase dobj_generate_pairs prep_generate_from det_corpus_a amod_corpus_parallel pobj_from_corpus nsubj_use_we dep_described_use det_algorithm_the amod_algorithm_diag-and nn_algorithm_phrase nn_algorithm_induction dobj_use_algorithm prep_described_in pobj_in_ prep_described_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments partmod_alignments_generated xcomp_generated_using nn_model_IBM dobj_using_model num__2 nsubj_described_ ' +o,3082,'At one extreme are those , exemplified by that of , that have no dependence on syntactic theory beyond the idea that natural language is hierarchical ',Wu,'prep_those_At pobj_At_one nsubj_those_extreme cop_those_are partmod_those_exemplified prep_exemplified_by pobj_by_that prep_that_of pobj_of_ nsubj_have_that rcmod_those_have det_dependence_no dobj_have_dependence prep_dependence_on amod_theory_syntactic pobj_on_theory prep_have_beyond det_idea_the pobj_beyond_idea complm_hierarchical_that amod_language_natural nsubj_hierarchical_language cop_hierarchical_is dep_idea_hierarchical ' +p,3083,'The NIST MT03 set is used to tune model weights -LRB- eg those of -LRB- 16 -RRB- -RRB- and the scaling factor 17We have also experimented with MERT , and found that the deterministic annealing gave results that were more consistent across runs and often better ',Och,'det_set_The nn_set_NIST nn_set_MT03 nsubjpass_used_set auxpass_used_is prep_used_to nn_weights_tune nn_weights_model pobj_to_weights dep_weights_eg dep_eg_those prep_those_of pobj_of_16 cc_weights_and det_factor_the amod_factor_scaling conj_weights_factor nsubj_experimented_17We aux_experimented_have advmod_experimented_also dep_used_experimented prep_experimented_with nn__MERT pobj_with_ cc_experimented_and conj_experimented_found complm_gave_that det_annealing_the amod_annealing_deterministic nsubj_gave_annealing ccomp_found_gave dobj_gave_results nsubj_consistent_that cop_consistent_were advmod_consistent_more rcmod_results_consistent prep_consistent_across pobj_across_runs cc_consistent_and advmod_better_often conj_consistent_better ' +p,3084,'-LRB- 2 -RRB- We note that these posterior probabilities can be computed efficiently for some alignment models such as the HMM , Models 1 and 2 ',Brown,'dep_note_2 nsubj_note_We dobj_note_that det_probabilities_these nn_probabilities_posterior nsubjpass_computed_probabilities aux_computed_can auxpass_computed_be dep_note_computed advmod_computed_efficiently prep_computed_for det_models_some amod_models_alignment pobj_for_models dep_as_such prep_models_as det__the nn__HMM pobj_as_ appos__Models num_Models_1 cc_Models_and conj_Models_2 ' +o,3085,'We refer to a3a16a5a7 as the source language string and a10 a11a7 as the target language string in accordance with the noisy channel terminology used in the IBM models of ',Brown,'nsubj_refer_We aux_a3a16a5a7_to xcomp_refer_a3a16a5a7 prep_a3a16a5a7_as det_string_the nn_string_source nn_string_language pobj_as_string cc_string_and nn_a11a7_a10 conj_string_a11a7 prep_string_as det_string_the nn_string_target nn_string_language pobj_as_string prep_string_in pobj_in_accordance prep_a3a16a5a7_with det_channel_the amod_channel_noisy pobj_with_channel advmod_used_terminology partmod_channel_used prep_used_in det_models_the nn_models_IBM pobj_in_models prep_models_of ' +o,3086,' proposed Model 6 , a log-linear combination of IBM translation models and HMM model ',Och,'advmod_proposed_ nsubj_proposed_Model num_Model_6 det_combination_a amod_combination_log-linear appos_Model_combination prep_combination_of nn_models_IBM nn_models_translation pobj_of_models cc_models_and nn_model_HMM conj_models_model ' +o,3087,'For instance , we may find metrics which compute similarities over shallow syntactic structures\\/sequences , constituency trees and dependency trees ',Liu,'prep_find_For pobj_For_instance nsubj_find_we aux_find_may dobj_find_metrics nsubj_compute_which rcmod_metrics_compute dobj_compute_similarities prep_compute_over nn__shallow nn__syntactic nn__structures\\/sequences nn__ dep__constituency nn__trees pobj_over_ cc__and nn_trees_dependency conj__trees ' +o,3088,'c2009 Association for Computational Linguistics Semi-supervised Training for the Averaged Perceptron POS Tagger Drahomra johanka Spoustova Jan Hajic Jan Raab Miroslav Spousta Institute of Formal and Applied Linguistics Faculty of Mathematics and Physics, Charles University Prague, Czech Republic {johanka,hajic,raab,spousta}@ ufal.mff.cuni.cz Abstract This paper describes POS tagging experiments with semi-supervised training as an extension to the (supervised) averaged perceptron algorithm, first introduced for this task by (Collins, 2002).',Collins,'' +o,3089,'1 Introduction A recent development in data-driven parsing is the use of discriminative training methods ',Collins,'num_development_1 nn_development_Introduction nn_development_A amod_development_recent nsubj_use_development prep_development_in amod_parsing_data-driven pobj_in_parsing cop_use_is det_use_the prep_use_of amod_methods_discriminative nn_methods_training pobj_of_methods ' +o,3090,' invented heuristic symmetriza57 FRENCH\\/ENGLISH ARABIC\\/ENGLISH SYSTEM F-MEASURE -LRB- = 04 -RRB- BLEU F-MEASURE -LRB- = 01 -RRB- BLEU GIZA + + 735 3063 758 5155 741 3140 791 5289 LEAF UNSUPERVISED 745 723 LEAF SEMI-SUPERVISED 763 3186 845 5434 Table 3 : Experimental Results tion of the output of a 1-to-N model and a M-to-1 model resulting in a M-to-N alignment , this was extended in ',Och,'advmod_invented_ amod_symmetriza57_heuristic dobj_invented_symmetriza57 nn_F-MEASURE_FRENCH\\/ENGLISH nn_F-MEASURE_ARABIC\\/ENGLISH nn_F-MEASURE_SYSTEM nsubj_invented_F-MEASURE number_04_= dep_F-MEASURE_04 amod_F-MEASURE_BLEU dep_F-MEASURE_F-MEASURE number_01_= dep_F-MEASURE_01 nn_GIZA_BLEU nsubjpass_extended_GIZA amod_GIZA_+ number_5434_+ cc_5434_735 conj_5434_3063 number_5434_758 number_5434_5155 number_5434_ number_5434_741 number_5434_3140 number_5434_791 number_5434_5289 number_5434_LEAF number_5434_UNSUPERVISED number_5434_745 cc_5434_723 conj_5434_LEAF number_5434_SEMI-SUPERVISED number_5434_763 number_5434_3186 number_5434_845 dep_+_5434 cc_5434_Table conj_5434_3 nn_tion_Experimental nn_tion_Results dep_5434_tion prep_tion_of det_output_the pobj_of_output prep_output_of det_model_a amod_model_1-to-N pobj_of_model cc_output_and det_model_a amod_model_M-to-1 conj_output_model partmod_model_resulting prep_resulting_in det_alignment_a amod_alignment_M-to-N pobj_in_alignment appos_alignment_this auxpass_extended_was rcmod_F-MEASURE_extended prep_extended_in pobj_in_ ' +o,3091,'Several algorithms have been proposed in the literature that try to find the best splits , see for instance ',Berger,'amod_algorithms_Several nsubjpass_proposed_algorithms aux_proposed_have auxpass_proposed_been prep_proposed_in det_literature_the pobj_in_literature nsubj_try_that rcmod_literature_try aux_find_to xcomp_try_find det_splits_the amod_splits_best dobj_find_splits dep_find_see prep_see_for pobj_for_instance ' +p,3092,'The averaged perceptron is a variant which averages the w across all iterations ; it has demonstrated good generalization especially with data that is not linearly separable , as in many natural language processing problems ',Collins,'det__The amod__averaged nn__perceptron nsubj_variant_ cop_variant_is det_variant_a nsubj_averages_which rcmod_variant_averages det_w_the dobj_averages_w prep_averages_across det_iterations_all pobj_across_iterations nsubj_demonstrated_it aux_demonstrated_has parataxis_variant_demonstrated amod_generalization_good dobj_demonstrated_generalization dep_with_especially prep_demonstrated_with pobj_with_data nsubj_separable_that cop_separable_is neg_separable_not advmod_separable_linearly rcmod_data_separable dep_with_as dep_with_in amod_problems_many amod_problems_natural nn_problems_language amod_problems_processing pobj_in_problems ' +o,3093,'4 Related Work and tackle the problem of segmenting Chinese while aligning it to English ',Wu,'num__4 amod__Related nn__Work nsubj_tackle_ cc__and conj__ det_problem_the dobj_tackle_problem prep_problem_of pcomp_of_segmenting amod_while_Chinese dobj_segmenting_while xcomp_segmenting_aligning dobj_aligning_it prep_aligning_to pobj_to_English ' +o,3094,'Syntagmatic strategies for determining similarity have often been based on statistical analyses of large corpora that yield clusters of words occurring in similar bigram and trigram contexts , as well as in similar predicateargument structure contexts ',Brown,'amod_strategies_Syntagmatic nsubjpass_based_strategies prep_strategies_for pcomp_for_determining dobj_determining_similarity aux_based_have advmod_based_often auxpass_based_been prep_based_on amod_analyses_statistical pobj_on_analyses prep_analyses_of amod_corpora_large pobj_of_corpora complm__that nn_clusters_yield nsubj__clusters prep_clusters_of pobj_of_words partmod_words_occurring prep_occurring_in amod_bigram_similar pobj_in_bigram cc_bigram_and nn__trigram nn__contexts conj_bigram_ dep_well_as cc_in_well dep_well_as conj_in_in amod_contexts_similar nn_contexts_predicateargument nn_contexts_structure pobj_in_contexts ccomp_based_ ' +p,3095,'1 Introduction In this paper , we show how discriminative training with averaged perceptron models can be used to substantially improve surface realization with Combinatory Categorial Grammar ',Collins,'num_Introduction_1 prep_show_In det_paper_this pobj_In_paper nsubj_show_we dep_Introduction_show advmod_used_how amod_training_discriminative nsubjpass_used_training prep_training_with amod_models_averaged nn_models_perceptron pobj_with_models advmod_used_ aux_used_can auxpass_used_be ccomp_show_used aux_improve_to advmod_improve_substantially purpcl_used_improve nn_realization_surface dobj_improve_realization prep_improve_with nn__Combinatory nn__Categorial nn__Grammar pobj_with_ ' +o,3096,' developed a training procedure that incorporates various MT evaluation criteria in the training procedure of log-linear MT models ',Och,'nsubj_developed_ det_procedure_a nn_procedure_training dobj_developed_procedure nsubj_incorporates_that rcmod_procedure_incorporates amod_criteria_various nn_criteria_MT nn_criteria_evaluation dobj_incorporates_criteria prep_criteria_in det_procedure_the nn_procedure_training pobj_in_procedure prep_procedure_of amod_models_log-linear nn_models_MT pobj_of_models ' +p,3097,'1 Introduction Statistical machine translation has seen many improvements in recent years , most notably the transition from wordto phrase-based models ',Koehn,'num__1 nn__Introduction nn__Statistical nn__machine nn__translation nsubj_seen_ aux_seen_has amod_improvements_many dobj_seen_improvements prep_seen_in amod_years_recent pobj_in_years advmod_notably_most advmod_years_notably det_transition_the dep_years_transition prep_transition_from amod_models_wordto amod_models_phrase-based pobj_from_models ' +o,3098,'Previous studies defined a probabilistic model of unification-based grammars including HPSG as a log-linear model or maximum entropy model ',Berger,'amod__Previous nn__studies nsubj_defined_ det_model_a amod_model_probabilistic dobj_defined_model prep_model_of amod_grammars_unification-based pobj_of_grammars prep_grammars_including pcomp_including_HPSG prep_HPSG_as det_model_a amod_model_log-linear pobj_as_model cc_model_or amod_model_maximum nn_model_entropy conj_model_model ' +n,3099,'For the results in this paper , we have used Pointwise Mutual Information -LRB- PMI -RRB- instead of IBM Model 1 , since found it to be as effective on Springer , but faster to compute ',Brown,'prep_used_For det_results_the pobj_For_results prep_results_in det_paper_this pobj_in_paper nsubj_used_we aux_used_have nn_Information_Pointwise nn_Information_Mutual dobj_used_Information abbrev_Information_PMI dep_of_instead prep_Information_of nn__IBM nn__Model num__1 pobj_of_ mark_found_since nsubj_found_ advcl_used_found nsubj_effective_it aux_effective_to cop_effective_be advmod_effective_as xcomp_found_effective prep_effective_on pobj_on_Springer cc_effective_but conj_effective_faster aux_compute_to xcomp_effective_compute ' +o,3100,'The elements of this set are pairs -LRB- x , y -RRB- where y is a possible translation for x 4 IBMs model 1 IBMs model 1 is the simplest of a hierarchy of five statistical models introduced in ',Brown,'det_elements_The nsubj_pairs_elements prep_elements_of det_set_this pobj_of_set cop_pairs_are dep_pairs_x appos_x_y advmod_translation_where nsubj_translation_y cop_translation_is det_translation_a amod_translation_possible rcmod_pairs_translation prep_translation_for nn_model_x num_model_4 nn_model_IBMs pobj_for_model num_model_1 nn_model_IBMs nsubj_simplest_model num_model_1 cop_simplest_is det_simplest_the dep_pairs_simplest prep_simplest_of det_hierarchy_a pobj_of_hierarchy prep_hierarchy_of num_models_five amod_models_statistical pobj_of_models partmod_models_introduced prep_introduced_in ' +p,3101,'Point-wise mutual information and Relative Feature Focus are well-known examples ',Lin,'amod__Point-wise amod__mutual nn__information nsubj_examples_ cc__and nn__Relative nn__Feature nn__Focus conj__ cop_examples_are amod_examples_well-known ' +n,3102,'Numbers in the table correspond to the percentage of experiments in which the condition at the head of the column was true -LRB- for example figure in the first row and first column means that for 989 percent of the language pairs the BLEU score for the bidirectional decoder was better than that of the forward decoder -RRB- proach -RRB- ',Brown,'prep_Numbers_in det_correspond_the nn_correspond_table pobj_in_correspond prep_correspond_to det_percentage_the pobj_to_percentage prep_percentage_of pobj_of_experiments rel_true_in pobj_in_which det_condition_the nsubj_true_condition prep_condition_at det_head_the pobj_at_head prep_head_of det_column_the pobj_of_column cop_true_was rcmod_experiments_true mark_means_for nn_figure_example nsubj_means_figure prep_figure_in det_row_the amod_row_first pobj_in_row cc_row_and amod_column_first conj_row_column dep_Numbers_means complm_better_that prep_better_for num_percent_989 pobj_for_percent prep_percent_of det_pairs_the nn_pairs_language pobj_of_pairs det_score_the amod_score_BLEU nsubj_better_score prep_score_for det_decoder_the amod_decoder_bidirectional pobj_for_decoder cop_better_was ccomp_means_better prep_better_than pobj_than_that prep_that_of det__the amod__forward nn__decoder nn__proach pobj_of_ ' +o,3103,'For this paper , we train the parameter vector using the perceptron algorithm ',Collins,'prep_train_For det_paper_this pobj_For_paper nsubj_train_we det_vector_the nn_vector_parameter nsubj_using_vector dep_train_using det_algorithm_the nn_algorithm_perceptron dobj_using_algorithm ' +p,3104,'A solution that leverages the complementary strengths of these two approachesdescribed in detail by was recently and successfully explored by Nivre and McDonald -LRB- 2008 -RRB- ',McDonald,'det_solution_A nsubj_was_solution nsubj_leverages_that rcmod_solution_leverages det_strengths_the amod_strengths_complementary dobj_leverages_strengths prep_strengths_of dep_two_these pobj_of_two partmod_two_approachesdescribed prep_approachesdescribed_in pobj_in_detail prep_approachesdescribed_by pobj_by_ advmod_was_recently cc_was_and advmod_explored_successfully conj_was_explored prep_explored_by pobj_by_Nivre cc_Nivre_and conj_Nivre_McDonald appos_Nivre_2008 ' +p,3105,'A detailed description of the popular translation\\/alignment models IBM-1 to IBM-5 , as well as the Hidden-Markov alignment model -LRB- HMM -RRB- can be found in ',Brown,'det_description_A amod_description_detailed nsubjpass_found_description prep_description_of det_IBM-1_the amod_IBM-1_popular amod_IBM-1_translation\\/alignment nn_IBM-1_models pobj_of_IBM-1 prep_IBM-1_to nn__IBM-5 pobj_to_ dep_well_as cc__well dep_well_as det__the advmod_alignment_Hidden-Markov amod__alignment nn__model abbrev__HMM conj__ aux_found_can auxpass_found_be prep_found_in pobj_in_ ' +o,3106,'In the supervised condition , we used just 2 additional task instances , plant and tank , each with 4000 handannotated instances drawn from a large balanced corpus ',Yarowsky,'prep_used_In det_condition_the amod_condition_supervised pobj_In_condition nsubj_used_we quantmod_2_just num_instances_2 amod_instances_additional nn_instances_task dobj_used_instances appos_instances_plant cc_plant_and conj_plant_tank dep_instances_each prep_used_with num_instances_4000 amod_instances_handannotated pobj_with_instances partmod_instances_drawn prep_drawn_from det_corpus_a amod_corpus_large amod_corpus_balanced pobj_from_corpus ' +o,3107,'MI is defined in general as follows : y -RRB- I ix y -RRB- = log2 P -LRB- x -RRB- P -LRB- y -RRB- We can use this definition to derive an estimate of the connectedness between words , in terms of collocations , but also in terms of phrases and grammatical relations ',Smadja,'nsubjpass_defined_MI auxpass_defined_is prep_defined_in pobj_in_general prep_general_as dep_as_follows dep_follows_y num_P_I nn_P_ix nn_P_y nn_P_= nn_P_log2 pobj_as_P dep__x nsubj__P appos_P_y nsubj_use_We aux_use_can rcmod_P_use det_definition_this dobj_use_definition aux_derive_to infmod_definition_derive det_estimate_an dobj_derive_estimate prep_estimate_of det_connectedness_the pobj_of_connectedness prep_connectedness_between pobj_between_words prep_derive_in pobj_in_terms prep_terms_of pobj_of_collocations dep_defined_ cc__but advmod__also conj__in pobj_in_terms prep_terms_of pobj_of_phrases cc_phrases_and amod__grammatical nn__relations conj_phrases_ ' +o,3108,'Future work will include : -LRB- i -RRB- applying the method to retrieve other types of collocations , and -LRB- ii -RRB- evaluating the method using Internet directories ',Smadja,'amod_work_Future nsubj_include_work aux_include_will dep_applying_i dep_include_applying det_method_the dobj_applying_method aux_retrieve_to xcomp_applying_retrieve amod_types_other dobj_retrieve_types prep_types_of amod__collocations pobj_of_ cc_retrieve_and dep_evaluating_ii conj_retrieve_evaluating det_method_the dobj_evaluating_method xcomp_evaluating_using nn_directories_Internet dobj_using_directories ' +o,3109,'1 Introduction and Motivation Detecting contradictory statements is an important and challenging NLP task with a wide range of potential applications including analysis of political discourse , of scientific literature , and more ',Marneffe,'nn_statements_Introduction cc_Introduction_and conj_Introduction_Motivation nn_statements_Detecting amod_statements_contradictory nsubj_task_statements cop_task_is det_task_an amod_task_important cc_important_and conj_important_challenging nn_task_NLP rcmod_1_task prep_task_with det_range_a amod_range_wide pobj_with_range prep_task_of amod_applications_potential pobj_of_applications prep_task_including pobj_including_analysis prep_analysis_of amod_discourse_political pobj_of_discourse prep_task_of amod_literature_scientific pobj_of_literature cc_1_and conj_1_more ' +o,3110,'Performance is also measured by the BLEU score , which measures similarity to the reference translation taken from the English side of the parallel corpus ',Papineni,'nsubjpass_measured_Performance auxpass_measured_is advmod_measured_also prep_measured_by det__the amod__BLEU nn__score pobj_by_ nsubj_measures_which rcmod__measures dobj_measures_similarity prep_measures_to det_translation_the nn_translation_reference pobj_to_translation partmod_translation_taken prep_taken_from det_side_the amod_side_English pobj_from_side prep_side_of det_corpus_the amod_corpus_parallel pobj_of_corpus ' +p,3111,'Our model exploits the same kind of tag-n-gram information that forms the core of many successful tagging models , for example , , , ',Ratnaparkhi,'poss_exploits_Our nn_exploits_model nsubj_kind_exploits det_kind_the amod_kind_same prep_kind_of amod_information_tag-n-gram pobj_of_information nsubj_forms_that rcmod_kind_forms det_core_the dobj_forms_core prep_core_of amod_models_many amod_models_successful nn_models_tagging pobj_of_models prep_forms_for pobj_for_example nn__ dep_kind_ ' +o,3112,'Some researchers use manually designed rules to take into account the grammatical role of the antecedent candidates as well as the governing relations between the candidate and the pronoun , while others use features determined over the parse tree in a machine-learning approach ',Luo,'det_use_Some nn_use_researchers nn_use_ nsubj_designed_use advmod_designed_manually dobj_designed_rules aux_take_to infmod_rules_take prep_take_into pobj_into_account det_role_the amod_role_grammatical dobj_take_role prep_role_of det_candidates_the amod_candidates_antecedent pobj_of_candidates dep_well_as cc_role_well dep_well_as det_relations_the amod_relations_governing conj_role_relations prep_relations_between det_candidate_the pobj_between_candidate cc_candidate_and det_pronoun_the conj_candidate_pronoun mark_use_while nsubj_use_others advcl_take_use dobj_use_features partmod_features_determined prep_determined_over det_tree_the amod_tree_parse pobj_over_tree prep_tree_in det_approach_a amod_approach_machine-learning pobj_in_approach ' +o,3113,'41 Extraction from Definition Sentences Definition sentences in the Wikipedia article were used for acquiring hyponymy relations by for named entity recognition ',Kazama,'num_Extraction_41 nsubjpass_used_Extraction prep_Extraction_from nn_sentences_Definition nn_sentences_Sentences nn_sentences_Definition pobj_from_sentences prep_sentences_in det_article_the nn_article_Wikipedia pobj_in_article auxpass_used_were prep_used_for pcomp_for_acquiring amod_relations_hyponymy dobj_acquiring_relations prep_acquiring_by pobj_by_ prep__for amod_recognition_named nn_recognition_entity pobj_for_recognition ' +o,3114,'1 Introduction Word alignment was first proposed as an intermediate result of statistical machine translation ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubjpass_proposed_alignment auxpass_proposed_was advmod_proposed_first prep_proposed_as det_result_an amod_result_intermediate pobj_as_result prep_result_of amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,3115,'As a model learning method , we adopt the maximum entropy model learning method ',Berger,'prep_adopt_As det_model_a nsubj_learning_model pcomp_As_learning dobj_learning_method nsubj_adopt_we det_model_the amod_model_maximum nn_model_entropy dobj_adopt_model partmod_model_learning dobj_learning_method ' +o,3116,'1 Introduction Recent work in learning semantics has focused on mapping sentences to meaning representations -LRB- eg , some logical form -RRB- given aligned sentence\\/meaning pairs as training data ',Zettlemoyer,'num_work_1 amod_work_Introduction amod_work_Recent nsubj_focused_work prep_work_in pcomp_in_learning dobj_learning_semantics aux_focused_has prep_focused_on nn_sentences_mapping pobj_on_sentences prep_focused_to amod_representations_meaning pobj_to_representations dep_representations_eg det_form_some amod_form_logical appos_eg_form partmod_representations_given dep_given_aligned xcomp_aligned_sentence\\/meaning dobj_sentence\\/meaning_pairs prep_sentence\\/meaning_as nn_data_training pobj_as_data ' +o,3117,'3 MaltParser MaltParser is a languageindependent system for data-driven dependency parsing , based on a transition-based parsing model ',McDonald,'num__3 nn__MaltParser nn__MaltParser nsubj_system_ cop_system_is det_system_a amod_system_languageindependent prep_system_for amod_parsing_data-driven amod_parsing_dependency pobj_for_parsing prep_system_based dep_based_on det_model_a amod_model_transition-based nn_model_parsing pobj_on_model ' +o,3118,'Following recent research about disambiguation models on linguistic grammars , we apply a log-linear model or maximum entropy model on HPSG derivations ',Berger,'prep_apply_Following amod_research_recent pobj_Following_research prep_research_about nn_models_disambiguation pobj_about_models prep_models_on amod__linguistic nn__grammars pobj_on_ nsubj_apply_we det_model_a amod_model_log-linear dobj_apply_model cc_model_or nn__maximum nn__entropy nn__model conj_model_ prep_apply_on pcomp_on_HPSG dobj_HPSG_derivations ' +n,3119,'This provides a compelling advantage over previous dependency language models for MT , whichusea5-gramLMonlyduringreranking ',Shen,'nsubj_provides_This det_advantage_a amod_advantage_compelling dobj_provides_advantage prep_advantage_over amod_models_previous amod_models_dependency nn_models_language pobj_over_models prep_models_for pobj_for_MT num_MT_ num_MT_whichusea5-gramLMonlyduringreranking ' +o,3120,'Extensions to Hiero Several authors describe extensions to Hiero , to incorporate additional syntactic information , or to combine it with discriminative latent models ',Zollmann,'nsubj_describe_Extensions prep_Extensions_to nn_authors_Hiero amod_authors_Several pobj_to_authors dobj_describe_extensions prep_describe_to pobj_to_Hiero aux_incorporate_to dep_describe_incorporate amod__additional amod__syntactic nn__information dobj_incorporate_ cc_incorporate_or aux_combine_to conj_incorporate_combine dobj_combine_it prep_combine_with amod__discriminative amod__latent nn__models pobj_with_ ' +o,3121,'Daume III divided features into three classes : domainindependent features , source-domain features and target-domain features ','Daume III','nn__Daume nn__III partmod__divided dobj_divided_features prep_divided_into num_classes_three pobj_into_classes amod_features_domainindependent dep__features amod_features_source-domain conj_features_features cc_features_and amod_features_target-domain conj_features_features ' +o,3122,'The algorithm we implemented is inspired by the work of on word sense disambiguation ',Yarowsky,'det_algorithm_The nsubjpass_inspired_algorithm nsubj_implemented_we rcmod_algorithm_implemented auxpass_inspired_is prep_inspired_by det_work_the pobj_by_work prep_work_of pobj_of_ prep_inspired_on nn_disambiguation_word nn_disambiguation_sense pobj_on_disambiguation ' +o,3123,'There have been considerable amount of efforts to improve the reordering model in SMT systems , ranging from the fundamental distance-based distortion model , flat reordering model , to lexicalized reordering model , hierarchical phrase-based model , and maximum entropy-based phrase reordering model ',Och,'expl_been_There aux_been_have amod_amount_considerable dobj_been_amount prep_amount_of pobj_of_efforts aux_improve_to infmod_efforts_improve det_model_the nn_model_reordering dobj_improve_model prep_model_in nn_systems_SMT pobj_in_systems xcomp_improve_ranging prep_ranging_from det__the amod__fundamental amod__distance-based nn__distortion nn__model pobj_from_ amod__flat nn__reordering nn__model appos__ prep_ranging_to amod__lexicalized nn__reordering nn__model pobj_to_ amod__hierarchical amod__phrase-based nn__model conj__ cc__and amod__maximum amod__entropy-based nn__phrase nn__reordering nn__model conj__ ' +o,3124,'MT output was evaluated using the standard evaluation metric BLEU 2 The parameters of the MT System were optimized for BLEU metric on NIST MTEval2002 test sets using minimum error rate training , and the systems were tested on NIST MTEval2003 test sets for both languages ',Papineni,'nn_output_MT nsubjpass_evaluated_output auxpass_evaluated_was xcomp_evaluated_using det_evaluation_the amod_evaluation_standard dobj_using_evaluation amod_BLEU_metric nsubj__BLEU dep_evaluated_ dep_optimized_2 det_parameters_The nsubjpass_optimized_parameters prep_parameters_of det_System_the nn_System_MT pobj_of_System auxpass_optimized_were dep__optimized prep_optimized_for amod_metric_BLEU pobj_for_metric prep_metric_on amod_sets_NIST nn_sets_MTEval2002 nn_sets_test pobj_on_sets xcomp_optimized_using amod__minimum nn__error nn__rate nn__training dobj_using_ cc_optimized_and det_systems_the nsubjpass_tested_systems auxpass_tested_were conj_optimized_tested prep_tested_on amod_sets_NIST nn_sets_MTEval2003 nn_sets_test pobj_on_sets prep_sets_for det_languages_both pobj_for_languages ' +o,3125,'We thus propose to adapt the statistical machine translation model for SMS text normalization ',Brown,'nsubj_propose_We advmod_propose_thus aux_adapt_to xcomp_propose_adapt det__the amod__statistical nn__machine nn__translation nn__model dobj_adapt_ prep_adapt_for nn_normalization_SMS nn_normalization_text pobj_for_normalization ' +o,3126,'These linguistically-motivated trimming rules iteratively remove constituents until a desired sentence compression rate is reached ',Lin,'det_linguistically-motivated_These partmod_linguistically-motivated_trimming dobj_trimming_rules advmod_trimming_ advmod_constituents_iteratively dep_constituents_remove dep_linguistically-motivated_constituents mark_reached_until det_rate_a amod_rate_desired nn_rate_sentence nn_rate_compression nsubjpass_reached_rate auxpass_reached_is advcl_constituents_reached ' +o,3127,'The extension of dynamic SBNs with incrementally specified model structure -LRB- ie Incremental Sigmoid Belief Networks , used in this paper -RRB- was proposed and applied to constituent parsing in ',Titov,'det_extension_The nsubjpass_proposed_extension prep_extension_of amod_SBNs_dynamic pobj_of_SBNs prep_SBNs_with amod_structure_incrementally amod_structure_specified nn_structure_model pobj_with_structure dep_structure_ie nn_Networks_Incremental nn_Networks_Sigmoid nn_Networks_Belief dobj_ie_Networks partmod_Networks_used prep_used_in det_paper_this pobj_in_paper auxpass_proposed_was cc_proposed_and conj_proposed_applied prep_proposed_to pobj_to_constituent partmod_constituent_parsing prt_parsing_in ' +o,3128,'-LRB- In our experiments , we use maximum entropy classification -LRB- MaxEnt -RRB- to train this probability model -RRB- ',Berger,'prep_use_In poss_experiments_our pobj_In_experiments nsubj_use_we amod_classification_maximum nn_classification_entropy nsubj__classification appos_classification_MaxEnt ccomp_use_ aux_train_to xcomp__train det_model_this nn_model_probability dobj_train_model ' +o,3129,' show that treating U + as a source for a new feature function in a loglinear model for SMT allows us to maximally take advantage of unlabeled data by finding a weight for this feature using minimum error-rate training -LRB- MERT -RRB- ',Och,'nsubj_show_ complm_allows_that csubj_allows_treating nn_+_U dobj_treating_+ prep_+_as det_source_a pobj_as_source prep_source_for det_function_a amod_function_new nn_function_feature pobj_for_function prep_function_in det_model_a nn_model_loglinear pobj_in_model prep_model_for nn__SMT pobj_for_ ccomp_show_allows nsubj_take_us aux_take_to advmod_take_maximally xcomp_allows_take dobj_take_advantage prep_advantage_of amod_data_unlabeled pobj_of_data prep_take_by pcomp_by_finding det_weight_a dobj_finding_weight prep_weight_for det_feature_this pobj_for_feature partmod_feature_using amod_training_minimum amod_training_error-rate dobj_using_training abbrev_training_MERT ' +o,3130,'The second voting model , a maximum entropy model , was built as found that it yielded higher accuracy than nave Bayes in a subsequent comparison of WSD performance ',Klein,'det_model_The amod_model_second nn_model_voting nsubjpass_built_model det__a amod__maximum amod__entropy nn__model appos_model_ auxpass_built_was mark_found_as nsubj_found_ advcl_built_found complm_yielded_that nsubj_yielded_it ccomp_found_yielded amod_accuracy_higher dobj_yielded_accuracy prep_yielded_than amod_Bayes_nave pobj_than_Bayes prep_yielded_in det_comparison_a amod_comparison_subsequent pobj_in_comparison prep_comparison_of nn_performance_WSD pobj_of_performance ' +o,3131,'We can sum over all non-projective spanning trees by taking the determinant of the Kirchhoff matrix of the graph defined above , minus the row and column corresponding to the root node ',Smith,'nsubj_sum_We aux_sum_can prep_sum_over det_trees_all amod_trees_non-projective num_trees_spanning pobj_over_trees prep_sum_by pcomp_by_taking det_determinant_the dobj_taking_determinant prep_determinant_of det_matrix_the nn_matrix_Kirchhoff pobj_of_matrix prep_matrix_of det_graph_the pobj_of_graph partmod_graph_defined prep_defined_above cc_matrix_minus det_corresponding_the nn_corresponding_row cc_row_and conj_row_column conj_matrix_corresponding prep_corresponding_to det_node_the nn_node_root pobj_to_node ' +o,3132,'The weights are then averaged across all iterations of the perceptron , as in ',Collins,'det_weights_The nsubjpass_averaged_weights auxpass_averaged_are advmod_averaged_then prep_averaged_across det_iterations_all pobj_across_iterations prep_iterations_of det_perceptron_the pobj_of_perceptron dep_across_as dep_across_in ' +n,3133,'Experimental results indicate that our model outperforms coreference model by a large margin on the ACE data sets and compares favorably to a modified version of their model ',Haghighi,'amod_results_Experimental nsubj_indicate_results complm_outperforms_that poss_model_our nsubj_outperforms_model ccomp_indicate_outperforms amod_model_ nn_model_coreference dobj_outperforms_model prep_outperforms_by det_margin_a amod_margin_large pobj_by_margin prep_margin_on det_sets_the nn_sets_ACE nn_sets_data pobj_on_sets cc_outperforms_and conj_outperforms_compares advmod_compares_favorably dep_favorably_to det_version_a amod_version_modified pobj_to_version prep_version_of poss_model_their pobj_of_model ' +o,3134,'Note that it is straightforward to calculate these expected counts using a variant of the inside-outside algorithm applied to the dependency-parsing data structures for projective dependency structures , or the matrix-tree theorem for nonprojective dependency structures ',Smith,'complm_straightforward_that nsubj_straightforward_it cop_straightforward_is ccomp_Note_straightforward aux_calculate_to xcomp_straightforward_calculate det_counts_these amod_counts_expected dobj_calculate_counts partmod_counts_using det_variant_a dobj_using_variant prep_variant_of det__the amod__inside-outside nn__algorithm pobj_of_ partmod__applied prep_applied_to det__the amod__ amod__dependency-parsing nn__data nn__structures pobj_to_ prep__for amod_structures_projective amod_structures_dependency pobj_for_structures cc_counts_or det__the amod__matrix-tree nn__theorem conj_counts_ prep__for amod_structures_nonprojective amod_structures_dependency pobj_for_structures ' +o,3135,'We use minimum error rate training to tune the feature weights for the log-linear model ',Och,'nsubj_use_We amod__minimum nn__error nn__rate nn__training dobj_use_ aux_tune_to xcomp_use_tune det_weights_the nn_weights_feature dobj_tune_weights prep_tune_for det_model_the amod_model_log-linear pobj_for_model ' +o,3136,'For the first two tasks , all heuristics of the Pharaoh-Toolkit as well as the refined heuristic to combine both IBM4-alignments were tested and the best ones are shown in the tables ',Koehn,'prep_heuristics_For det_tasks_the amod_tasks_first num_tasks_two pobj_For_tasks nsubj_heuristics_all prep_heuristics_of det__the amod__Pharaoh-Toolkit pobj_of_ dep_well_as cc__well dep_well_as det__the amod__refined amod__heuristic conj__ aux_combine_to xcomp_heuristics_combine det_IBM4-alignments_both nsubjpass_tested_IBM4-alignments auxpass_tested_were dep_combine_tested cc_tested_and det_ones_the amod_ones_best nsubjpass_shown_ones auxpass_shown_are conj_tested_shown prep_shown_in det_tables_the pobj_in_tables ' +o,3137,'The decoder uses a binarized representation of the rules , which is obtained via a syncronous binarization procedure ',Zhang,'det_decoder_The nsubj_uses_decoder det_representation_a amod_representation_binarized dobj_uses_representation prep_representation_of det_rules_the pobj_of_rules nsubjpass_obtained_which auxpass_obtained_is rcmod_rules_obtained prep_obtained_via det_procedure_a amod_procedure_syncronous nn_procedure_binarization pobj_via_procedure ' +o,3138,'This kind of synchronizer stands in contrast to more ad-hoc approaches ',Brown,'det_kind_This dep__kind prep_kind_of nn_stands_synchronizer pobj_of_stands prep_stands_in dep_in_contrast dep_in_to amod_approaches_more amod_approaches_ad-hoc pobj_in_approaches ' +o,3139,'Sentiment classification is a well studied problem and in many domains users explicitly 1We use the term aspect to denote properties of an object that can be rated by a user as in Snyder and Barzilay ',Pang,'nn_classification_Sentiment nsubj__classification cop__is det__a advmod_studied_well amod__studied nn__problem cc__and conj__in amod_users_many nn_users_domains pobj_in_users advmod_use_explicitly nsubj_use_1We dep__use det_aspect_the nn_aspect_term nsubj_denote_aspect aux_denote_to xcomp_use_denote dobj_denote_properties prep_properties_of det_object_an pobj_of_object nsubjpass_rated_that aux_rated_can auxpass_rated_be rcmod_object_rated prep_rated_by det_user_a pobj_by_user dep_in_as prep_user_in nn__Snyder cc_Snyder_and conj_Snyder_Barzilay pobj_in_ ' +o,3140,'Every sentence was part-of-speech tagged using a maximum entropy tagger and parsed using a state-of-the-art wide coverage phrase structure parser ',Ratnaparkhi,'det_sentence_Every nsubjpass_tagged_sentence auxpass_tagged_was advmod_tagged_part-of-speech xcomp_tagged_using det__a amod__maximum amod__entropy nn__tagger dobj_using_ cc_tagged_and conj_tagged_parsed xcomp_parsed_using det__a amod__state-of-the-art amod__wide nn__coverage nn__phrase nn__structure nn__parser dobj_using_ ' +o,3141,'In each case the input to the network is a sequence of tag-word pairs5 5We used a publicly available tagger to provide the tags ',Ratnaparkhi,'prep_used_In det_case_each pobj_In_case det_input_the nsubj_sequence_input prep_input_to det_network_the pobj_to_network cop_sequence_is det_sequence_a rcmod_case_sequence prep_sequence_of amod_5We_tag-word amod_5We_pairs5 pobj_of_5We det_tagger_a advmod_available_publicly amod_tagger_available nsubj_used_tagger partmod_tagger_ aux_provide_to xcomp__provide det_tags_the dobj_provide_tags ' +o,3142,'The target set is built using the 88-89 Wall Street Journal Corpus (WSJ) tagged using the (Ratnaparkhi, 1996) tagger and the (Bangalore & Joshi, 1999) SuperTagger; the feedback sets are built using WSJ sentences con330 Algorithm 1 KE-train: (Karov & Edelman, 1998) algorithm adapted to literal/nonliteral classification Require: S: the set of sentences containing the target word Require: L: the set of literal seed sentences Require: N: the set of nonliteral seed sentences Require: W: the set of words/features, w s means w is in sentence s, s owner w means s contains w Require: epsilon1: threshold that determines the stopping condition 1: w-sim0(wx,wy) := 1 if wx = wy,0 otherwise 2: s-simI0(sx,sy) := 1, for all sx,sy S S where sx = sy, 0 otherwise 3: i := 0 4: while (true) do 5: s-simLi+1(sx,sy) := summationtextwxsx p(wx,sx)maxwysy w-simi(wx,wy), for all sx,sy S L 6: s-simNi+1(sx,sy) := summationtextwxsx p(wx,sx)maxwysy w-simi(wx,wy), for all sx,sy S N 7: for wx,wy W W do 8: w-simi+1(wx,wy) := braceleftBigg i = 0 summationtextsxownerwx p(wx,sx)maxsyownerwy s-simIi(sx,sy) else summationtextsxownerwx p(wx,sx)maxsyownerwys-simLi (sx,sy),s-simNi (sx,sy)} 9: end for 10: if wx,maxwyw-simi+1(wx,wy)w-simi(wx,wy)} epsilon1 then 11: break # algorithm converges in 1epsilon1 steps.',Ratnaparkhi,'' +n,3143,'This method was preferred against other related methods , like the one introduced in , since it embeds all the available semantic information existing in WordNet , even edges that cross POS , thus offering a richer semantic representation ',Mihalcea,'det_method_This nsubjpass_preferred_method auxpass_preferred_was prep_preferred_against amod_methods_other amod_methods_related pobj_against_methods prep_preferred_like det_one_the pobj_like_one partmod_one_introduced prep_introduced_in pobj_in_ mark_embeds_since nsubj_embeds_it advcl_preferred_embeds predet_information_all det_information_the amod_information_available dep_available_semantic nsubj_existing_information ccomp_embeds_existing prep_existing_in pobj_in_WordNet advmod_edges_even dep_existing_edges det_POS_that amod_POS_cross dobj_edges_POS advmod_existing_thus dep_existing_offering det_representation_a amod_representation_richer amod_representation_semantic dobj_offering_representation ' +o,3144,'Translation accuracy is measured in terms of the BLEU score , which is computed here for translations generated by using the tuple n-gram model alone , in the case of Table 2 , and by using the tuple n-gram model along with the additional four feature functions described in Section 32 , in the case of Table 3 ',Papineni,'nn_accuracy_Translation nsubjpass_measured_accuracy auxpass_measured_is prep_measured_in pobj_in_terms prep_terms_of det__the amod__BLEU nn__score pobj_of_ nsubjpass_computed_which auxpass_computed_is rcmod__computed advmod_computed_here prep_computed_for pobj_for_translations partmod_translations_generated prep_generated_by pcomp_by_using det_model_the amod_model_tuple amod_model_n-gram dobj_using_model advmod_using_alone prep_using_in det_case_the pobj_in_case prep_case_of pobj_of_Table num_Table_2 cc_in_and conj_in_by pcomp_by_using det_model_the amod_model_tuple amod_model_n-gram dobj_using_model prt_using_along prep_using_with det_functions_the amod_functions_additional num_functions_four nn_functions_feature pobj_with_functions partmod_functions_described prep_described_in pobj_in_Section num_Section_32 prep_described_in det_case_the pobj_in_case prep_case_of pobj_of_Table num_Table_3 ' +o,3145,'Such an approach contrasts with the log-linear HMM\\/Model -4 combination proposed by ',Och,'predet_approach_Such det_approach_an nsubj_contrasts_approach prep_contrasts_with det_combination_the amod_combination_log-linear nn_combination_HMM\\/Model num_combination_-4 pobj_with_combination partmod_combination_proposed prep_proposed_by ' +o,3146,'4 Dependency Parsing : Baseline 41 Learning Model and Features According to , all data-driven models for dependency parsing that have been proposed in recent years can be described as either graph-based or transition-based ',McDonald,'num_Parsing_4 nn_Parsing_Dependency nn_Model_Baseline num_Model_41 nn_Model_Learning nsubjpass_described_Model cc_Model_and conj_Model_Features prep_Model_According aux__to dep_models_ det_models_all amod_models_data-driven pobj_According_models prep_models_for amod_parsing_dependency pobj_for_parsing nsubjpass_proposed_that aux_proposed_have auxpass_proposed_been rcmod_models_proposed prep_proposed_in amod_years_recent pobj_in_years aux_described_can auxpass_described_be dep_Parsing_described prep_described_as preconj_graph-based_either pobj_as_graph-based cc_graph-based_or conj_graph-based_transition-based ' +o,3147,'They may rely only on this information -LRB- eg , -RRB- , or they may combine it with additional information as well -LRB- eg , -RRB- ',Bloom,'nsubj_rely_They aux_rely_may advmod_rely_only prep_rely_on det_information_this pobj_on_information appos_information_eg dep_eg_ cc_rely_or nsubj_combine_they aux_combine_may conj_rely_combine dobj_combine_it prep_combine_with amod_information_additional pobj_with_information advmod_well_as advmod_combine_well dep_well_eg dep_eg_ ' +o,3148,'The learning methods using in discriminative parsing are Perceptron and online large-margin learning -LRB- MIRA -RRB- ',Collins,'det_methods_The nn_methods_learning nsubj__methods partmod_methods_using prep_using_in amod_parsing_discriminative pobj_in_parsing cop__are nn__Perceptron cc__and amod_large-margin_online conj__large-margin partmod_large-margin_learning abbrev__MIRA dobj_learning_ ' +o,3149,'As an alternative , describes a forest-based reranking algorithm called cube growing , which also employs beam search , but focuses computation only where necessary in a top-down pass through a parse forest ',Huang,'prep_describes_As det_alternative_an pobj_As_alternative nsubj_describes_ det_algorithm_a amod_algorithm_forest-based nn_algorithm_reranking nsubj_called_algorithm ccomp_describes_called dobj_called_cube amod_cube_growing nsubj_employs_which advmod_employs_also rcmod_cube_employs nn_search_beam dobj_employs_search cc_employs_but conj_employs_focuses dobj_focuses_computation prep_focuses_only dep_necessary_where dep_only_necessary prep_called_in det_pass_a amod_pass_top-down pobj_in_pass prep_called_through det_forest_a amod_forest_parse pobj_through_forest ' +o,3150,' apply entropy regularization to dependency parsing ',Smith,'advmod_apply_ amod_regularization_entropy dobj_apply_regularization prep_apply_to amod_parsing_dependency pobj_to_parsing ' +o,3151,'In information retrieval , word similarity can be used to identify terms for pseudo-relevance feedback ',Brown,'prep_used_In nn_retrieval_information pobj_In_retrieval nn_similarity_word nsubjpass_used_similarity aux_used_can auxpass_used_be aux_identify_to purpcl_used_identify dobj_identify_terms prep_terms_for amod__pseudo-relevance nn__feedback pobj_for_ ' +o,3152,'One such relational reasoning task is the problem of compound noun interpretation , which has received a great deal of attention in recent years ',Turney,'num_task_One amod_task_such amod_task_relational nn_task_reasoning cop_problem_is det_problem_the rcmod_task_problem prep_problem_of amod_interpretation_compound nn_interpretation_noun pobj_of_interpretation nsubj_received_which aux_received_has rcmod_interpretation_received det_deal_a amod_deal_great dobj_received_deal prep_deal_of pobj_of_attention prep_received_in amod_years_recent pobj_in_years ' +o,3153,'A large corpus is vahmble as a source of such nouns ',Church,'det_corpus_A amod_corpus_large nsubj_vahmble_corpus cop_vahmble_is prep_vahmble_as det_source_a pobj_as_source prep_source_of amod_nouns_such pobj_of_nouns ' +o,3154,'Because our system uses a synchronous CFG , it could be thought of as an example of syntax-based statistical machine translation -LRB- MT -RRB- , joining a line of research that has been fruitful but has not previously produced systems that can compete with phrase-based systems in large-scale translation tasks such as the evaluations held by NIST ',Wu,'mark_uses_Because poss_system_our nsubj_uses_system advcl_thought_uses det_CFG_a amod_CFG_synchronous dobj_uses_CFG nsubjpass_thought_it aux_thought_could auxpass_thought_be dep_as_of prep_thought_as det_example_an pobj_as_example prep_example_of amod_translation_syntax-based amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_MT xcomp_thought_joining det_line_a dobj_joining_line prep_line_of nn__research pobj_of_ nsubj_fruitful_that aux_fruitful_has dep_fruitful_been rcmod_line_fruitful cc_fruitful_but aux_produced_has neg_produced_not advmod_produced_previously conj_fruitful_produced dobj_produced_systems nsubj_compete_that aux_compete_can rcmod_systems_compete prep_compete_with amod_systems_phrase-based pobj_with_systems prep_compete_in amod_tasks_large-scale nn_tasks_translation pobj_in_tasks dep_as_such prep_tasks_as det_evaluations_the pobj_as_evaluations partmod_evaluations_held prep_held_by pobj_by_NIST ' +o,3155,'-LRB- ~ -RRB- 1995 Association for Computational Linguistics Computational Linguistics Volume 21 , Number 2 and Mancini 1991 ; Meteer , Schwartz , and Weischedel 1991 ; Merialdo 1991 ; Pelillo , Moro , and Refice 1992 ; Weischedel et al 1993 ; Wothke et al 1993 -RRB- ',Marcus,'nn_Association_~ nn_Association_1995 prep_Association_for nn_Volume_Computational nn_Volume_Linguistics nn_Volume_Computational nn_Volume_Linguistics pobj_for_Volume num_Volume_21 appos_Volume_Number num_Number_2 cc_Number_and conj_Number_Mancini num_Mancini_1991 dep_Number_Meteer conj_Meteer_Schwartz cc_Meteer_and conj_Meteer_Weischedel num_Weischedel_1991 dep_Number_Merialdo dep_Merialdo_1991 dep_Number_Pelillo conj_Pelillo_Moro cc_Pelillo_and conj_Pelillo_Refice num_Refice_1992 dep_Number_Weischedel cc_Weischedel_et conj_Weischedel_al dep_Weischedel_1993 dep_Number_Wothke cc_Wothke_et conj_Wothke_al dep_Wothke_1993 ' +o,3156,' propose a MaxEnt-based reordering model for BTG while Setiawan et al ',Wu,'advmod_propose_ det_model_a amod_model_MaxEnt-based nn_model_reordering dobj_propose_model prep_model_for pobj_for_BTG amod_al_ nn_al_while nn_al_Setiawan nn_al_et nsubj_propose_al ' +o,3157,'One major focus is sentiment classification and opinion mining 2008 ',Pang,'num_focus_One amod_focus_major nsubj_classification_focus cop_classification_is nn_classification_sentiment cc_classification_and nn_mining_opinion conj_classification_mining partmod_classification_ dobj__2008 ' +o,3158,'It will also be relevant to apply advanced statistical models that can incorporate various useful information to this task , eg , the maximum entropy model ',Ratnaparkhi,'nsubj_relevant_It aux_relevant_will advmod_relevant_also cop_relevant_be aux_apply_to xcomp_relevant_apply amod_models_advanced amod_models_statistical dobj_apply_models nsubj_incorporate_that aux_incorporate_can rcmod_models_incorporate amod_information_various amod_information_useful dobj_incorporate_information prep_incorporate_to det_task_this pobj_to_task appos_task_eg det_model_the amod_model_maximum nn_model_entropy appos_eg_model ' +o,3159,'Some methods parse two flat strings at once using a bitext grammar ',Wu,'det_methods_Some nsubj_parse_methods num_strings_two amod_strings_flat dobj_parse_strings prep_parse_at advmod_using_once pcomp_at_using det_grammar_a amod_grammar_bitext dobj_using_grammar ' +o,3160,'Table 2 shows the unknown word tags for chunking , which are known as the IOB2 model ',Ramshaw,'nsubj_shows_Table num_Table_2 det_tags_the amod_tags_unknown nn_tags_word dobj_shows_tags prep_tags_for pobj_for_chunking nsubjpass_known_which auxpass_known_are rcmod_chunking_known prep_known_as det_model_the nn_model_IOB2 pobj_as_model ' +o,3161,' and compare with results reported by HK06 and CRR07 ',Haghighi,'cc__and conj__compare prep_compare_with pobj_with_results partmod_results_reported prep_reported_by nn__HK06 pobj_by_ cc__and nn__CRR07 conj__ ' +o,3162,'The frequency counts of dependency relationships are filtered with the loglikelihood ratio ',Dunning,'det_counts_The nn_counts_frequency nsubjpass_filtered_counts prep_counts_of amod_relationships_dependency pobj_of_relationships auxpass_filtered_are prep_filtered_with det_ratio_the amod_ratio_loglikelihood pobj_with_ratio ' +p,3163,'To some extent , this can probably be explained by the strong tradition of constituent analysis in Anglo-American linguistics , but this trend has been reinforced by the fact that the major treebank of American English , the Penn Treebank , is annotated primarily with constituent analysis ',Marcus,'prep_explained_To det_extent_some pobj_To_extent nsubjpass_explained_this aux_explained_can advmod_explained_probably auxpass_explained_be prep_explained_by det_tradition_the amod_tradition_strong pobj_by_tradition prep_tradition_of amod_analysis_constituent pobj_of_analysis prep_analysis_in nn_linguistics_Anglo-American pobj_in_linguistics cc_explained_but det_trend_this nsubjpass_reinforced_trend aux_reinforced_has auxpass_reinforced_been conj_explained_reinforced prep_reinforced_by det_fact_the pobj_by_fact complm_annotated_that det_treebank_the amod_treebank_major nsubjpass_annotated_treebank prep_treebank_of nn_English_American pobj_of_English det__the nn__Penn nn__Treebank appos_English_ auxpass_annotated_is ccomp_reinforced_annotated advmod_annotated_primarily prep_annotated_with amod_analysis_constituent pobj_with_analysis ' +o,3164,'We propose a probabilistic quasi-synchronous grammar , inspired by one proposed for machine translation , and parameterized by mixtures of a robust nonlexical syntax\\/alignment model with a -LRB- n optional -RRB- lexical-semantics-drivenlog-linear model ',Smith,'nsubj_propose_We det_grammar_a amod_grammar_probabilistic amod_grammar_quasi-synchronous dobj_propose_grammar dep_grammar_inspired prep_inspired_by pobj_by_one amod_one_proposed prep_one_for nn__machine nn__translation pobj_for_ cc_inspired_and conj_inspired_parameterized prep_parameterized_by pobj_by_mixtures prep_mixtures_of det_model_a amod_model_robust amod_model_nonlexical nn_model_syntax\\/alignment pobj_of_model prep_model_with det_model_a nn_optional_n appos_model_optional amod_model_lexical-semantics-drivenlog-linear pobj_with_model ' +o,3165,'Similarly , used Wikipedia , particularly the first sentence of each article , to create lists of entities ',Kazama,'advmod_used_Similarly nsubj_used_ dobj_used_Wikipedia advmod_Wikipedia_particularly det_sentence_the amod_sentence_first dep_Wikipedia_sentence prep_sentence_of det_article_each pobj_of_article aux_create_to xcomp_used_create dobj_create_lists prep_lists_of pobj_of_entities ' +o,3166,'Additionally , some research has explored cutting and pasting segments of text from the full document to generate a summary ',Jing,'advmod_explored_Additionally det_research_some nsubj_explored_research aux_explored_has ccomp_explored_cutting cc_cutting_and conj_cutting_pasting dobj_pasting_segments prep_segments_of pobj_of_text prep_pasting_from det_document_the amod_document_full pobj_from_document aux_generate_to xcomp_pasting_generate det_summary_a dobj_generate_summary ' +o,3167,'918 English For English we used the Wall Street Journal section of the Penn Treebank ',Marcus,'nn_English_918 dep_English_For pobj_For_English nsubj_used_we rcmod_English_used det_section_the nn_section_Wall nn_section_Street nn_section_Journal dobj_used_section prep_section_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,3168,' that draws on a stochastic tagger -LRB- see for details -RRB- as well as the SPECIALIST Lexicon5 , a large syntactic lexicon of both general and medical English that is distributed with the UMLS ',Cutting,'nsubj_draws_that dep__draws prep_draws_on det_tagger_a amod_tagger_stochastic pobj_on_tagger dep_tagger_see dobj_see_ prep__for pobj_for_details dep_well_as cc_tagger_well dep_well_as det_Lexicon5_the amod_Lexicon5_SPECIALIST conj_tagger_Lexicon5 det_lexicon_a amod_lexicon_large amod_lexicon_syntactic appos_Lexicon5_lexicon prep_lexicon_of pobj_of_both amod_both_general cc_general_and conj_general_medical dep_general_English nsubjpass_distributed_that auxpass_distributed_is rcmod_Lexicon5_distributed prep_distributed_with det_UMLS_the pobj_with_UMLS ' +o,3169,'8 Related Research Class-based LMs or factored LMs are very similar to our T+C scenario ',Brown,'number_Related_8 num__Related nn__Research nn__Class-based nn__LMs nsubj_similar_ cc__or amod__factored nn__LMs conj__ cop_similar_are advmod_similar_very prep_similar_to poss_scenario_our nn_scenario_T+C pobj_to_scenario ' +o,3170,'This second source of evidence is sometimes referred to as distributional similarity ',Hindle,'det_source_This amod_source_second nsubjpass_referred_source prep_source_of pobj_of_evidence auxpass_referred_is advmod_referred_sometimes prep_referred_to advmod_distributional_as amod_similarity_distributional pobj_to_similarity ' +o,3171,'Co-occurrence statistics is collected from either bilingual parallel and 334 non-parallel corpora , or monolingual corpora ',Yarowsky,'amod_statistics_Co-occurrence nsubjpass_collected_statistics auxpass_collected_is prep_collected_from det_parallel_either amod_parallel_bilingual pobj_from_parallel cc_parallel_and num__334 amod__non-parallel nn__corpora conj_parallel_ cc_parallel_or amod_corpora_monolingual conj_parallel_corpora ' +o,3172,' apply entropy regularization to dependency parsing ',Smith,'advmod_apply_ amod_regularization_entropy dobj_apply_regularization prep_apply_to amod_parsing_dependency pobj_to_parsing ' +o,3173,'When evaluated against the state-of-the-art, phrase-based decoder Pharaoh (Koehn, 2004), using the same experimental conditions translation table trained on the FBIS corpus (7.2M Chinese words and 9.2M English words of parallel text), trigram language model trained on 155M words of English newswire, interpolation weights a65 (Equation 2) trained using discriminative training (Och, 2003) (on the 2002 NIST MT evaluation set), probabilistic beam a90 set to 0.01, histogram beam a58 set to 10 and BLEU (Papineni et al. , 2002) as our metric, the WIDL-NGLM-Aa86 a129 algorithm produces translations that have a BLEU score of 0.2570, while Pharaoh translations have a BLEU score of 0.2635.',Och,'' +o,3174,'For scoring MT outputs , the proposed RSCM uses a score based on a translation model called IBM4 -LRB- TM-score -RRB- and a score based on a language model for the translation target language -LRB- LM-score -RRB- ',Brown,'prep_uses_For amod_outputs_scoring nn_outputs_MT pobj_For_outputs det_RSCM_the amod_RSCM_proposed appos_outputs_RSCM det_score_a dobj_uses_score prep_uses_based dep_based_on det_model_a nn_model_translation pobj_on_model amod__called nn__IBM4 nsubj_uses_ abbrev__TM-score cc__and det_score_a conj__score partmod_score_based prep_based_on det_model_a nn_model_language pobj_on_model prep_model_for det_language_the nn_language_translation nn_language_target pobj_for_language abbrev_language_LM-score ' +o,3175,' -RRB- , in which translation and language models are trainable separately too ',Koehn,'csubj_trainable_ dep_translation_in pobj_in_which dobj__translation cc_translation_and conj_translation_language advmod__models cop_trainable_are advmod_too_separately advmod_trainable_too ' +o,3176,'It is based on code and ideas from the system of , but also includes some ideas from GUITAR and other coreference systems ',Ponzetto,'nsubjpass_based_It auxpass_based_is prep_based_on pobj_on_code cc_code_and conj_code_ideas prep_based_from det_system_the pobj_from_system prep_system_of pobj_of_ cc_based_but advmod_includes_also conj_based_includes det_ideas_some dobj_includes_ideas prep_includes_from nn__GUITAR pobj_from_ cc__and amod__other amod__coreference nn__systems conj__ ' +o,3177,'However , union and rened alignments , which are many-to-many , are what are used to build competitive phrasal SMT systems , because intersection performs poorly , despite having been shown to have the best AER scores for the French\\/English corpus we are using ',Och,'advmod_are_However nn_alignments_union cc_union_and conj_union_rened nsubj_are_alignments nsubj_many-to-many_which cop_many-to-many_are rcmod_alignments_many-to-many nsubjpass_used_what auxpass_used_are ccomp_are_used aux_build_to xcomp_used_build amod_systems_competitive amod_systems_phrasal nn_systems_SMT dobj_build_systems mark_performs_because nsubj_performs_intersection advcl_are_performs advmod_performs_poorly prep_performs_despite aux_shown_having auxpass_shown_been pcomp_despite_shown aux_have_to xcomp_shown_have det_scores_the amod_scores_best nn_scores_AER dobj_have_scores prep_scores_for det_corpus_the nn_corpus_French\\/English pobj_for_corpus nsubj_using_we aux_using_are rcmod_scores_using ' +p,3178,'As a side product , we find empirical evidence to suggest that the effectiveness of rule lexicalization techniques and parent annotation techniques is due to the fact that both lead to a reduction in perplexity in the automata induced from training corpora ',Collins,'prep_find_As det_product_a nn_product_side pobj_As_product nsubj_find_we amod_evidence_empirical dobj_find_evidence aux_suggest_to xcomp_find_suggest complm_due_that det_effectiveness_the nsubj_due_effectiveness prep_effectiveness_of nn__rule nn__lexicalization nn__techniques pobj_of_ cc__and nn_techniques_parent nn_techniques_annotation conj__techniques advmod_due_ cop_due_is ccomp_suggest_due prep_due_to det_fact_the pobj_to_fact complm_lead_that nsubj_lead_both ccomp_due_lead prep_lead_to det_reduction_a pobj_to_reduction prep_reduction_in pobj_in_perplexity prep_lead_in det_automata_the pobj_in_automata partmod_automata_induced prep_induced_from nn_corpora_training pobj_from_corpora ' +o,3179,'The Penn Wall Street Journal treebank was used as training and test data ',Marcus,'det_Journal_The nn_Journal_Penn nn_Journal_Wall nn_Journal_Street nsubj_treebank_Journal nsubjpass_used_ auxpass_used_was ccomp_treebank_used prep_used_as nn_data_training cc_training_and conj_training_test pobj_as_data ' +o,3180,'Maximum Entropy models have been used to express the interactions among multiple feature variables -LRB- eg , -RRB- , but within this framework no systematic study of interactions has been proposed ',Berger,'nn_models_Maximum nn_models_Entropy nsubjpass_used_models aux_used_have auxpass_used_been aux_express_to xcomp_used_express det_interactions_the dobj_express_interactions prep_express_among amod_variables_multiple nn_variables_feature pobj_among_variables appos_variables_eg dep_eg_ cc_among_but conj_among_within det_framework_this pobj_within_framework det_study_no amod_study_systematic nsubjpass_proposed_study prep_study_of pobj_of_interactions aux_proposed_has auxpass_proposed_been dep_used_proposed ' +o,3181,'The final SMT system performance is evaluated on a uncased test set of 3071 sentences using the BLEU , NIST and METEOR scores ',Banerjee,'det_performance_The amod_performance_final nn_performance_SMT nn_performance_system nsubjpass_evaluated_performance auxpass_evaluated_is prep_evaluated_on det_set_a amod_set_uncased nn_set_test pobj_on_set prep_set_of num_sentences_3071 pobj_of_sentences partmod_sentences_using det__the amod__BLEU dobj_using_ nn__NIST conj__ cc__and amod_scores_METEOR amod_scores_ conj__scores ' +o,3182,'We also report on applying Factored Translation Models for English-to-Arabic translation ',Koehn,'nsubj_report_We advmod_report_also prep_report_on pcomp_on_applying nn__Factored nn__Translation nn__Models dobj_applying_ prep__for amod_translation_English-to-Arabic pobj_for_translation ' +o,3183,'We repeat Ramshaw and Marcus Transformation Based NP chunking algorithm by substituting supertags for POS tags in the dataset ',Ramshaw,'nsubj_repeat_We nn_NP_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nn_NP_Transformation nn_NP_Based nsubj_chunking_NP dep_repeat_chunking amod_algorithm_ dobj_chunking_algorithm prep_chunking_by pcomp_by_substituting dobj_substituting_supertags prep_substituting_for dep_tags_POS pobj_for_tags prep_tags_in det_dataset_the pobj_in_dataset ' +o,3184,'These techniques included unweighted FS morphology , conditional random fields , synchronous parsers , lexicalized parsers ,22 partially supervised training ` a la ,23 and grammar induction ',Wu,'det_techniques_These nsubj_included_techniques amod_morphology_unweighted nn_morphology_FS nsubj_lexicalized_morphology amod_fields_conditional amod_fields_random appos_morphology_fields amod__ amod__synchronous nn__parsers dep_fields_ ccomp_included_lexicalized nsubj_supervised_parsers num_,22_ dep_partially_,22 advmod_supervised_partially ccomp_lexicalized_supervised xcomp_supervised_training det__a nn__la dobj_training_ num__,23 cc__and nn__grammar nn__induction conj__ ' +o,3185,'We ran each estimator with the eight different combinations of values for the hyperparameters and prime listed below , which include the optimal values for the hyperparameters found by , and report results for the best combination for each estimator below 1 ',Johnson,'nsubj_ran_We det_estimator_each dobj_ran_estimator prep_ran_with det_combinations_the num_combinations_eight amod_combinations_different pobj_with_combinations prep_combinations_of pobj_of_values prep_values_for det_hyperparameters_the pobj_for_hyperparameters cc_combinations_and conj_combinations_prime partmod_prime_listed advmod_listed_below nsubj_include_which rcmod_prime_include det_values_the amod_values_optimal dobj_include_values prep_values_for det_hyperparameters_the pobj_for_hyperparameters dep_hyperparameters_found prep_found_by pobj_by_ cc_found_and conj_found_report dobj_report_results prep_report_for det_combination_the amod_combination_best pobj_for_combination prep_combination_for det_estimator_each pobj_for_estimator prep_report_below pobj_below_1 ' +o,3186,'3 Domain Adaptation Following , we present an application of structural correspondence learning -LRB- SCL -RRB- to non-projective dependency parsing ',Blitzer,'num_Adaptation_3 nn_Adaptation_Domain prep_present_Following pobj_Following_ nsubj_present_we dep_Adaptation_present det_application_an dobj_present_application prep_application_of amod_learning_structural nn_learning_correspondence pobj_of_learning abbrev_learning_SCL prep_present_to amod__non-projective amod__dependency nn__parsing pobj_to_ ' +p,3187,'The dif1The routinely used tool for automatic evaluation ROUGE was adopted exactly because it was demonstrated it is highly correlated with the manual DUC coverage scores ',Lin,'det_dif1The_The nsubj_used_dif1The advmod_used_routinely dobj_used_tool prep_used_for amod_evaluation_automatic pobj_for_evaluation nsubjpass_adopted_ROUGE auxpass_adopted_was dep_used_adopted advmod_adopted_exactly mark_demonstrated_because nsubjpass_demonstrated_it auxpass_demonstrated_was advcl_adopted_demonstrated nsubjpass_correlated_it auxpass_correlated_is advmod_correlated_highly ccomp_demonstrated_correlated prep_correlated_with det_scores_the amod_scores_manual nn_scores_DUC nn_scores_coverage pobj_with_scores ' +o,3188,'For the identification and labeling steps , we train a maximum entropy classifier over sections 02-21 of a version of the CCGbank corpus that has been augmented by projecting the Propbank semantic annotations ',Berger,'prep_train_For det_steps_the nn_steps_identification cc_identification_and conj_identification_labeling pobj_For_steps nsubj_train_we det__a amod__maximum amod__entropy nn__classifier dobj_train_ prep__over pobj_over_sections num_sections_02-21 prep_sections_of det_version_a pobj_of_version prep_version_of det__the nn__CCGbank nn__corpus pobj_of_ nsubjpass_augmented_that aux_augmented_has auxpass_augmented_been rcmod__augmented prep_augmented_by pcomp_by_projecting det__the nn__Propbank amod__semantic nn__annotations dobj_projecting_ ' +o,3189,'Evaluation 81 Effects of Unpublished Details In this section we present the results of effectively doing a clean-room implementation of Collins parsing model , that is , using only information available in , as shown in Table 4 ',Collins,'num_Effects_81 dobj_Evaluation_Effects prep_Effects_of nn_Details_Unpublished pobj_of_Details prep_present_In det_section_this pobj_In_section nsubj_present_we rcmod_Effects_present det_results_the dobj_present_results prep_results_of advmod_doing_effectively pcomp_of_doing det_implementation_a amod_implementation_clean-room dobj_doing_implementation prep_implementation_of nn_model_Collins amod_model_parsing pobj_of_model dep_Evaluation_that dep_that_is xcomp_Evaluation_using advmod_information_only nsubj_available_information xcomp_using_available prep_available_in pobj_in_ mark_shown_as advcl_using_shown prep_shown_in pobj_in_Table num_Table_4 ' +o,3190,'We contrast our work with , highlight some severe limitations of probability estimates computed from single derivations , and demonstrate that it is critical to account for many derivations for each sentence pair ',Galley,'nsubj_contrast_We poss_work_our dobj_contrast_work prep_contrast_with pobj_with_ conj_contrast_highlight det_limitations_some amod_limitations_severe dobj_highlight_limitations prep_limitations_of nn_estimates_probability pobj_of_estimates partmod_estimates_computed prep_computed_from amod_derivations_single pobj_from_derivations cc_contrast_and conj_contrast_demonstrate complm_critical_that nsubj_critical_it cop_critical_is ccomp_demonstrate_critical aux_account_to xcomp_critical_account prep_account_for amod_derivations_many pobj_for_derivations prep_derivations_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,3191,'The model scaling factors are optimized using minimum error rate training ',Och,'det_factors_The amod_factors_model amod_factors_scaling nsubjpass_optimized_factors auxpass_optimized_are xcomp_optimized_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +p,3192,'1 Introduction Phrase-based method and syntaxbased method represent the state-of-the-art technologies in statistical machine translation -LRB- SMT -RRB- ',Wu,'num__1 nn__Introduction amod__Phrase-based nn__method nsubj_represent_ cc__and nn__syntaxbased nn__method conj__ det_technologies_the amod_technologies_state-of-the-art dobj_represent_technologies prep_technologies_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +o,3193,'We compare TERp with BLEU , METEOR , and TER ',Papineni,'nsubj_compare_We dobj_compare_TERp prep_TERp_with nn__BLEU pobj_with_ nn__METEOR conj__ cc__and nn__TER conj__ ' +n,3194,'Both the global models use fairly small training sets , and there is no evidence that their techniques will scale to larger data sets ',Liang,'predet_models_Both det_models_the amod_models_global nsubj_use_models advmod_use_ advmod_small_fairly amod_sets_small nn_sets_training dobj_use_sets cc_use_and expl_is_there conj_use_is det_evidence_no nsubj_is_evidence complm_scale_that poss_techniques_their nsubj_scale_techniques aux_scale_will ccomp_is_scale prep_scale_to amod_sets_larger nn_sets_data pobj_to_sets ' +o,3195,' regarded MWE as connected collocations : a sequence of neighboring words whose exact meaning can not be derived from the meaning or connotation of its components , which means that MWEs also have low ST As some pioneers provide MWE identiflcation methods which are based on association metrics -LRB- AM -RRB- , such as likelihood ratio ',Dunning,'advmod_regarded_ advmod_regarded_MWE advmod_connected_as amod_collocations_connected dobj_regarded_collocations det_sequence_a dep_collocations_sequence prep_sequence_of amod_words_neighboring pobj_of_words poss_meaning_whose amod_meaning_exact nsubjpass_derived_meaning aux_derived_can neg_derived_not auxpass_derived_be rcmod_sequence_derived prep_derived_from det_meaning_the pobj_from_meaning cc_meaning_or conj_meaning_connotation prep_meaning_of poss_components_its pobj_of_components nsubj_means_which rcmod_collocations_means complm_have_that nsubj_have_MWEs advmod_have_also ccomp_means_have amod_ST_low dobj_have_ST mark_provide_As det_pioneers_some nsubj_provide_pioneers dep_regarded_provide nn_methods_MWE nn_methods_identiflcation dobj_provide_methods nsubjpass_based_which auxpass_based_are rcmod_methods_based prep_based_on nn_metrics_association pobj_on_metrics dep_metrics_AM dep_as_such prep_methods_as nn_ratio_likelihood pobj_as_ratio ' +p,3196,'7 Related Work Unannotated texts have been used successfully for a variety of NLP tasks , including named entity recognition , subjectivity classification , text classification , and word sense disambiguation ',Yarowsky,'num_texts_7 amod_texts_Related nn_texts_Work nn_texts_Unannotated nsubjpass_used_texts aux_used_have auxpass_used_been advmod_used_successfully prep_used_for det_variety_a pobj_for_variety prep_variety_of nn_tasks_NLP pobj_of_tasks prep_used_including amod__named nn__entity nn__recognition pobj_including_ amod__subjectivity nn__classification conj__ nn__text nn__classification conj__ cc__and nn_disambiguation_word nn_disambiguation_sense conj__disambiguation ' +o,3197,'More recently , the problem has been tackled using unsupervised -LRB- eg , -RRB- and supervised -LRB- eg , Evans -LRB- 2001 -RRB- , Ng and Cardie -LRB- 2002a -RRB- -RRB- approaches ',Bean,'advmod_recently_More advmod_tackled_recently det_problem_the nsubjpass_tackled_problem aux_tackled_has auxpass_tackled_been xcomp_tackled_using amod_approaches_unsupervised dep_unsupervised_eg dep_eg_ cc_unsupervised_and conj_unsupervised_supervised dep_supervised_eg appos_eg_Evans appos_Evans_2001 appos_Evans_Ng cc_Ng_and conj_Ng_Cardie appos_Ng_2002a dobj_using_approaches ' +p,3198,'2 Previous Work So far , Structural Correspondence Learning has been applied successfully to PoS tagging and Sentiment Analysis ',Blitzer,'num_Work_2 amod_Work_Previous dep_far_So advmod_applied_far amod_Learning_Structural nn_Learning_Correspondence nsubjpass_applied_Learning aux_applied_has auxpass_applied_been dep_Work_applied advmod_applied_successfully aux_PoS_to xcomp_applied_PoS nn_Analysis_tagging cc_tagging_and conj_tagging_Sentiment dobj_PoS_Analysis ' +o,3199,'a2 Maximum-entropy method The maximum-entropy method is useful with sparse data conditions and has been used by many researchers ',Berger,'amod_method_a2 amod_method_Maximum-entropy det_method_The amod_method_maximum-entropy nsubjpass_useful_method cop_useful_is dep_method_useful prep_useful_with amod_conditions_sparse nn_conditions_data pobj_with_conditions cc_useful_and aux_used_has auxpass_used_been conj_useful_used prep_used_by amod_researchers_many pobj_by_researchers ' +o,3200,'For this reason there is currently a great deal of interest in methods which incorporate syntactic information within statistical machine translation systems -LRB- eg , see -RRB- ',Och,'prep_is_For det_reason_this pobj_For_reason expl_is_there advmod_is_currently det_deal_a amod_deal_great nsubj_is_deal prep_deal_of pobj_of_interest prep_interest_in pobj_in_methods nsubj_incorporate_which rcmod_methods_incorporate amod_information_syntactic dobj_incorporate_information prep_incorporate_within amod_systems_statistical nn_systems_machine nn_systems_translation pobj_within_systems dep_deal_eg dep_eg_see dobj_eg_ ' +o,3201,'For that purpose , syntactical , statistical and hybrid syntaxicostatistical methodologies have been proposed ',Dunning,'prep_proposed_For det_purpose_that pobj_For_purpose amod__syntactical conj_purpose_ amod__statistical conj_purpose_ cc_purpose_and amod_methodologies_hybrid amod_methodologies_syntaxicostatistical conj_purpose_methodologies nsubjpass_proposed_ aux_proposed_have auxpass_proposed_been ' +o,3202,'But it is close to the paradigm described by Yarowsky and as it also employs self-training based on a relatively small seed data set which is incrementally enlarged with unlabelled samples ',Turney,'cc_close_But nsubj_close_it cop_close_is prep_close_to det_paradigm_the pobj_to_paradigm partmod_paradigm_described prep_described_by nn__Yarowsky pobj_by_ cc__and conj__ mark_employs_as nsubj_employs_it advmod_employs_also advcl_described_employs dobj_employs_self-training prep_employs_based dep_based_on det_set_a advmod_small_relatively amod_set_small nn_set_seed nn_set_data pobj_on_set nsubj_enlarged_which cop_enlarged_is advmod_enlarged_incrementally rcmod_set_enlarged prep_enlarged_with amod_samples_unlabelled pobj_with_samples ' +n,3203,'Our model improves the baseline provided by : -LRB- i -RRB- accuracy is increased by creating a lexicalised PCFG grammar and enriching conditioning context with parent f-structure features ; and -LRB- ii -RRB- coverage is increased by providing lexical smoothing and fuzzy matching techniques for rule smoothing ',Cahill,'poss_model_Our nsubj_improves_model det_baseline_the dobj_improves_baseline prep_improves_provided dep_provided_by pobj_by_ dep_increased_i nsubjpass_increased_accuracy auxpass_increased_is parataxis_improves_increased prep_increased_by pcomp_by_creating det_grammar_a amod_grammar_lexicalised nn_grammar_PCFG dobj_creating_grammar cc_creating_and conj_creating_enriching nn_context_conditioning dobj_enriching_context prep_enriching_with amod_features_parent amod_features_f-structure pobj_with_features cc_improves_and dep_increased_ii nsubjpass_increased_coverage auxpass_increased_is conj_improves_increased prep_increased_by pcomp_by_providing amod_techniques_lexical amod_techniques_smoothing cc_smoothing_and conj_smoothing_fuzzy amod_techniques_matching dobj_providing_techniques prep_techniques_for nn_smoothing_rule pobj_for_smoothing ' +o,3204,'The corpus consists of sections 15-18 and section 20 of the Penn Treebank , and is pre-divided into a 8936-sentence -LRB- 211727 tokens -RRB- training set and a 2012-sentence test set ',Marcus,'det_corpus_The nsubjpass_consists_corpus prep_consists_of pobj_of_sections num_sections_15-18 cc_sections_and conj_sections_section num_section_20 prep_section_of det__the nn__Penn nn__Treebank pobj_of_ cc_consists_and auxpass_pre-divided_is conj_consists_pre-divided prep_pre-divided_into det_set_a amod_set_8936-sentence num_tokens_211727 appos_set_tokens nn_set_training pobj_into_set cc_set_and det_set_a amod_set_2012-sentence amod_set_ nn_set_test conj_set_set ' +o,3205,'Here , we extract part-of-speech tags from the Collins parsers output for section 23 instead of reinventing a tagger ',Collins,'advmod_extract_Here nsubj_extract_we amod_tags_part-of-speech dobj_extract_tags prep_extract_from det__the nn__Collins nn__parsers nn__output pobj_from_ prep_extract_for pobj_for_section num_section_23 dep_of_instead prep_extract_of pcomp_of_reinventing det_tagger_a dobj_reinventing_tagger ' +o,3206,'46 Weakly-constrained algorithms In evaluation with ROUGE , summaries are truncated to a target length K Yih et al -LRB- 2007 -RRB- usedastackdecodingwithaslightmodication , which allows the last sentence in a summary to be truncated to a target length ',Lin,'num_algorithms_46 amod_algorithms_Weakly-constrained prep_truncated_In pobj_In_evaluation prep_evaluation_with nn__ROUGE pobj_with_ nsubj_truncated_summaries cop_truncated_are dep_algorithms_truncated prep_truncated_to det_usedastackdecodingwithaslightmodication_a nn_Yih_target nn_Yih_length nn_Yih_K nn_usedastackdecodingwithaslightmodication_Yih cc_Yih_et conj_Yih_al appos_Yih_2007 pobj_to_usedastackdecodingwithaslightmodication nsubj_allows_which rcmod_usedastackdecodingwithaslightmodication_allows det_sentence_the amod_sentence_last dobj_allows_sentence prep_sentence_in det_summary_a pobj_in_summary aux_truncated_to auxpass_truncated_be infmod_summary_truncated prep_truncated_to det_length_a nn_length_target pobj_to_length ' +o,3207,'For English there are many POS taggers , employing machine learning techniques like transformation-based error-driven learning , decision trees , markov model , maximum entropy methods etc There are also taggers which are hybrid using both stochastic and rule-based approaches , such as CLAWS ',Cutting,'prep_are_For pobj_For_English expl_are_there amod_taggers_many dep_taggers_POS nsubj_are_taggers xcomp_are_employing dobj_employing_machine partmod_machine_learning dobj_learning_techniques prep_learning_like amod__transformation-based amod__error-driven nn__learning pobj_like_ nn__decision nn__trees appos__ amod__markov nn__model appos__ amod_methods_maximum nn_methods_entropy appos__methods partmod_methods_ dobj__etc expl_are_There dep_are_are advmod_are_also nsubj_are_taggers nsubj_hybrid_which cop_hybrid_are rcmod_taggers_hybrid partmod_hybrid_using preconj_stochastic_both acomp_using_stochastic cc_stochastic_and amod_approaches_rule-based conj_stochastic_approaches dep_as_such conj_stochastic_as nn__CLAWS pobj_as_ ' +o,3208,' used noun-verb syntactic relations , and Hatzivassiloglou and McKeown -LRB- 1993 -RRB- used coordinated adjective-adjective modifier pairs ',Hindle,'nsubj_used_ amod_relations_noun-verb amod_relations_syntactic dobj_used_relations cc_used_and nsubj_used_Hatzivassiloglou cc_Hatzivassiloglou_and conj_Hatzivassiloglou_McKeown appos_Hatzivassiloglou_1993 conj_used_used dep_used_coordinated amod_pairs_adjective-adjective nn_pairs_modifier dobj_coordinated_pairs ' +p,3209,'We use a recently proposed dependency parser 1 which has demonstrated state-of-theart performance on a selection of languages from the 1The ISBN parser will be soon made downloadable from the authors web-page ',Titov,'nsubj_use_We det_parser_a advmod_proposed_recently amod_parser_proposed amod_parser_dependency nsubjpass_made_parser partmod_parser_ dobj__1 nsubj_demonstrated_which aux_demonstrated_has rcmod_1_demonstrated amod_performance_state-of-theart dobj_demonstrated_performance prep_demonstrated_on det_selection_a pobj_on_selection prep_selection_of pobj_of_languages prep_demonstrated_from det_parser_the amod_parser_1The nn_parser_ISBN pobj_from_parser aux_made_will auxpass_made_be advmod_made_soon ccomp_use_made dep_made_downloadable prep_made_from det_web-page_the nn_web-page_authors pobj_from_web-page ' +o,3210,'Furthermore , training corpora for information extraction are typically annotated with domain-specific tags , in contrast to general-purpose annotations such as part-of-speech tags or noun-phrase bracketing ',Marcus,'advmod_annotated_Furthermore nn_corpora_training nsubjpass_annotated_corpora prep_corpora_for nn_extraction_information pobj_for_extraction auxpass_annotated_are advmod_annotated_typically prep_annotated_with amod_tags_domain-specific pobj_with_tags prep_annotated_in dep_in_contrast dep_in_to amod_annotations_general-purpose pobj_in_annotations dep_as_such prep_annotations_as amod_tags_part-of-speech pobj_as_tags cc_tags_or amod_bracketing_noun-phrase conj_tags_bracketing ' +o,3211,'However , in experiments in unsupervised POS tag learning using HMM structured models , shows that VB is more effective than Gibbs sampling in approaching distributions that agree with the Zipfs law , which is prominent in natural languages ',Johnson,'advmod_shows_However prep_shows_in pobj_in_experiments prep_experiments_in amod_tag_unsupervised dep_tag_POS pobj_in_tag partmod_tag_learning xcomp_learning_using amod_models_HMM dep_HMM_structured dobj_using_models nsubj_shows_ complm_effective_that nsubj_effective_VB cop_effective_is advmod_effective_more ccomp_shows_effective prep_effective_than pobj_than_Gibbs xcomp_effective_sampling prep_sampling_in pcomp_in_approaching dobj_approaching_distributions nsubj_agree_that rcmod_distributions_agree prep_agree_with det_law_the nn_law_Zipfs pobj_with_law nsubj_prominent_which cop_prominent_is rcmod_law_prominent prep_prominent_in amod_languages_natural pobj_in_languages ' +o,3212,'1 Introduction Maximum Entropy -LRB- ME -RRB- modeling has received a lot of attention in language modeling and natural language processing for the past few years ',Berger,'num_Entropy_1 nn_Entropy_Introduction nn_Entropy_Maximum abbrev_Entropy_ME nsubj_received_modeling aux_received_has dep_Entropy_received det_lot_a dobj_received_lot prep_lot_of pobj_of_attention prep_received_in nn_modeling_language pobj_in_modeling cc_modeling_and amod_processing_natural nn_processing_language conj_modeling_processing prep_modeling_for det_years_the amod_years_past amod_years_few pobj_for_years ' +o,3213,'32 Automatic ROUGE Evaluation ROUGE measuresthen-grammatchbetween system generated summaries and human summaries ',Lin,'num_system_32 nn_system_Automatic nn_system_ROUGE nn_system_Evaluation nn_system_ROUGE nn_system_ amod_system_measuresthen-grammatchbetween nsubj_generated_system dobj_generated_summaries cc_summaries_and amod_summaries_human conj_summaries_summaries ' +o,3214,'5To test the reliability of the annotation scheme , we had a subset of the data annotated by two annotators and found a satisfactory - agreement of = 081 ',Carletta,'nsubj_test_5To det_reliability_the dobj_test_reliability prep_reliability_of det_scheme_the nn_scheme_annotation pobj_of_scheme nsubj_had_we dep_reliability_had det_subset_a dobj_had_subset prep_subset_of det_data_the pobj_of_data partmod_data_annotated prep_annotated_by num_annotators_two pobj_by_annotators cc_had_and conj_had_found det_satisfactory_a dobj_found_satisfactory nn__agreement dep_reliability_ prep__of dep_081_= pcomp_of_081 ' +o,3215,'We calculated the translation quality using Bleus modified n-gram precision metric for n-grams of up to length four ',Papineni,'nsubj_calculated_We det_quality_the nn_quality_translation dobj_calculated_quality partmod_quality_using amod__Bleus amod__modified amod__n-gram nn__precision amod__metric dobj_using_ prep_using_for pobj_for_n-grams prep_n-grams_of advmod_four_up dep_up_to pobj_to_length pobj_of_four ' +o,3216,'22 Global Linear Models We follow the framework of , recently applied to language modeling in Roark et al ',Collins,'num_Models_22 amod_Models_Global nn_Models_Linear nsubj_follow_We dep_Models_follow det_framework_the dobj_follow_framework prep_framework_of pobj_of_ advmod_applied_recently partmod_framework_applied prep_applied_to nn_modeling_language pobj_to_modeling prep_modeling_in pobj_in_Roark cc_Roark_et conj_Roark_al ' +o,3217,'In our experiments , we used the full parse output from Collins parser , in which every non-terminal node is already annotated with head information ',Collins,'prep_used_In poss_experiments_our pobj_In_experiments nsubj_used_we det_output_the amod_output_full nn_output_parse dobj_used_output prep_used_from nn__Collins nn__parser pobj_from_ rel_annotated_in pobj_in_which det_node_every amod_node_non-terminal nsubjpass_annotated_node auxpass_annotated_is advmod_annotated_already rcmod__annotated prep_annotated_with nn_information_head pobj_with_information ' +o,3218,'However , due to the computational issues with the voted perceptron , the averaged perceptron algorithm is used instead ',Collins,'advmod_used_However dep_to_due prep_used_to det_issues_the amod_issues_computational pobj_to_issues prep_issues_with det_perceptron_the amod_perceptron_voted pobj_with_perceptron det__the amod__averaged nn__perceptron nn__algorithm nsubjpass_used_ auxpass_used_is advmod_used_instead ' +o,3219,'3 The data 31 The supervised data For English , we use the same data division of Penn Treebank -LRB- PTB -RRB- parsed section as all of , , and do ; for details , see Table 1 ',Shen,'dep_do_3 det_data_The nsubj_do_data num_data_31 det_data_The amod_data_supervised nsubj_parsed_data prep_use_For pobj_For_English nsubj_use_we rcmod_data_use det_division_the amod_division_same nn_division_data dobj_use_division prep_division_of nn_Treebank_Penn pobj_of_Treebank abbrev_data_PTB rcmod_data_parsed nn__section dobj_parsed_ prep_parsed_as pobj_as_all prep_all_of pobj_of_ conj__ conj__ cc__and conj__ prep_see_for pobj_for_details parataxis_do_see dobj_see_Table num_Table_1 ' +p,3220,'Indeed , as for the voted perceptron of , we can get performance gains by reducing the support threshold for features to be included in the model ',Collins,'advmod_get_Indeed advmod_get_as dep_as_for det_perceptron_the amod_perceptron_voted pobj_for_perceptron prep_perceptron_of pobj_of_ nsubj_get_we aux_get_can nn_gains_performance dobj_get_gains prep_get_by pcomp_by_reducing det_threshold_the nn_threshold_support dobj_reducing_threshold prep_reducing_for pobj_for_features aux_included_to auxpass_included_be xcomp_reducing_included prep_included_in det_model_the pobj_in_model ' +o,3221,'The translation quality is evaluated by BLEU metric , as calculated by mteval-v11bpl 6 with case-sensitive matching of n-grams ',Papineni,'det_quality_The nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_is prep_evaluated_by amod__BLEU amod__metric pobj_by_ mark_calculated_as advcl_evaluated_calculated prep_calculated_by amod_6_mteval-v11bpl pobj_by_6 prep_calculated_with amod_matching_case-sensitive pobj_with_matching prep_matching_of pobj_of_n-grams ' +n,3222,'While the model of significantly outperforms the constrained model of , they both are well below the state-of-the-art in constituent parsing ',Matsuzaki,'mark_outperforms_While det_model_the nsubj_outperforms_model prep_model_of pobj_of_ advmod_outperforms_significantly advcl_parsing_outperforms det_model_the amod_model_constrained dobj_outperforms_model prep_model_of pobj_of_ nsubj_parsing_they advmod_parsing_both aux_parsing_are advmod_parsing_well dep_well_below det_state-of-the-art_the pobj_below_state-of-the-art prep_state-of-the-art_in pobj_in_constituent ' +o,3223,'52 Impact on translation quality As reported in Table 3 , small increases in METEOR , BLEU and NIST scores suggest that SMT output matches the references better after postprocessing or decoding with the suggested lemma translations ',Papineni,'num_Impact_52 dep_Impact_on nn_quality_translation pobj_on_quality mark_reported_As advcl_suggest_reported prep_reported_in pobj_in_Table num_Table_3 amod_increases_small nsubj_suggest_increases prep_increases_in nn__METEOR pobj_in_ nn__BLEU conj__ cc__and nn__NIST nn__scores conj__ rcmod_quality_suggest complm_matches_that nn_output_SMT nsubj_matches_output ccomp_suggest_matches det_references_the nsubj_better_references xcomp_matches_better prep_better_after pcomp_after_postprocessing cc_postprocessing_or conj_postprocessing_decoding prep_postprocessing_with det_translations_the amod_translations_suggested nn_translations_lemma pobj_with_translations ' +o,3224,'138 2 Rule Generation We start with phrase translations on the parallel training data using the techniques and implementation described in ',Koehn,'num_Generation_138 num_Generation_2 nn_Generation_Rule nsubj_start_We dep_Generation_start prep_start_with nn_translations_phrase pobj_with_translations prep_translations_on det_data_the amod_data_parallel nn_data_training pobj_on_data dep_Generation_using det_techniques_the dobj_using_techniques cc_using_and advmod_using_implementation conj_using_described prep_described_in ' +o,3225,'Instead of interpolating the two language models , we explicitly used them in the decoder and optimized their weights via minimumerror-rate -LRB- MER -RRB- training ',Och,'dep_of_Instead advcl_used_of pcomp_of_interpolating det_models_the num_models_two nn_models_language dobj_interpolating_models nsubj_used_we advmod_used_explicitly dobj_used_them prep_used_in det_decoder_the pobj_in_decoder cc_used_and conj_used_optimized poss_weights_their dobj_optimized_weights prep_optimized_via nn_training_minimumerror-rate nn_training_MER pobj_via_training ' +o,3226,'We use the adaptation of this algorithm to structure prediction , first proposed by ',Collins,'nsubj_use_We det_adaptation_the dobj_use_adaptation prep_adaptation_of det_algorithm_this pobj_of_algorithm prep_use_to nn_prediction_structure pobj_to_prediction advmod_proposed_first partmod_prediction_proposed prep_proposed_by ' +o,3227,'The flow using non-local features in two-stage architecture 24 Results We employ BIOE1 label scheme for the NER task because we found it performs better than IOB2 on Bakeoff 2006 NER MSRA and CityU corpora ',Levow,'det_flow_The nsubj_using_flow amod_features_non-local dobj_using_features prep_using_in amod_Results_two-stage nn_Results_architecture num_Results_24 pobj_in_Results nsubj_employ_We dep_using_employ nn_scheme_BIOE1 nn_scheme_label dobj_employ_scheme prep_scheme_for det_task_the nn_task_NER pobj_for_task mark_found_because nsubj_found_we advcl_employ_found nsubj_performs_it ccomp_found_performs dep_than_better quantmod_IOB2_than dobj_performs_IOB2 prep_IOB2_on nn_MSRA_Bakeoff num_MSRA_2006 nn_MSRA_ nn_MSRA_NER pobj_on_MSRA cc_MSRA_and nn_corpora_CityU conj_MSRA_corpora ' +o,3228,'321 Jensen-Shannon divergence is defined as D -LRB- q , r -RRB- = 12 parenleftbigg D parenleftbigg q q + r2 parenrightbigg + D parenleftbigg r q + r2 parenrightbiggparenrightbigg These experiments are a kind of poor mans version of the deterministic annealing clustering algorithm , which gradually increases the number of clusters during the clustering process ',Marcus,'num_divergence_321 amod_divergence_Jensen-Shannon nsubjpass_defined_divergence auxpass_defined_is mark_=_as nsubj_=_D dep_D_q appos_q_r advcl_defined_= num_q_12 amod_q_parenleftbigg nn_q_D amod_q_parenleftbigg nsubj_q_q ccomp_=_q cc_parenrightbigg_+ advmod_parenrightbigg_r2 acomp_q_parenrightbigg cc_parenrightbigg_+ advmod_parenleftbigg_D conj_parenrightbigg_parenleftbigg nn_q_r dep_parenleftbigg_q cc_parenrightbigg_+ advmod_parenrightbiggparenrightbigg_r2 conj_parenrightbigg_parenrightbiggparenrightbigg det_experiments_These nsubj_kind_experiments cop_kind_are det_kind_a ccomp_parenrightbiggparenrightbigg_kind prep_kind_of amod_version_poor nn_version_mans pobj_of_version prep_version_of det__the amod__deterministic amod__annealing nn__clustering nn__algorithm pobj_of_ nsubj_increases_which advmod_increases_gradually rcmod__increases det_number_the dobj_increases_number prep_number_of pobj_of_clusters prep_clusters_during det_process_the amod_process_clustering pobj_during_process ' +o,3229,'In fact, when the perceptron update rule of (Dekel et al. , 2004) which modifies the weights of every divergent node along the predicted and true paths is used in the ranking framework, it becomes virtually identical with the standard, flat, ranking perceptron of Collins (2002).5 In contrast, our approach shares the idea of (Cesa-Bianchi et al. , 2006a) that if a parent class has been predicted wrongly, then errors in the children should not be taken into account. We also view this as one of the key ideas of the incremental perceptron algorithm of (Collins and Roark, 2004), which searches through a complex decision space step-by-step and is immediately updated at the first wrong move.',Collins,'' +o,3230,'The annotation can be considered reliable with 95 \% agreement and a kappa of88 ',Carletta,'det_annotation_The nsubjpass_considered_annotation aux_considered_can auxpass_considered_be amod__reliable dobj_considered_ prep_considered_with dep_\%_95 amod_agreement_\% pobj_with_agreement cc_agreement_and det_of88_a dep__kappa amod_of88_ conj_agreement_of88 ' +o,3231,'Instead we report BLEU scores of the machine translation system using different combinations of wordand classbased models for translation tasks from English to Arabic and Arabic to English ',Papineni,'advmod_report_Instead nsubj_report_we amod_scores_BLEU dobj_report_scores nsubj_using_ prep__of det_system_the nn_system_machine nn_system_translation pobj_of_system dep_report_using amod_combinations_different dobj_using_combinations prep_combinations_of amod_models_wordand amod_models_classbased pobj_of_models prep_models_for nn_tasks_translation pobj_for_tasks prep_tasks_from pobj_from_English prep_using_to pobj_to_Arabic cc_Arabic_and conj_Arabic_Arabic prep_using_to pobj_to_English ' +o,3232,'Collocations have been widely used for tasks such as word sense disambiguation -LRB- WSD -RRB- , information extraction -LRB- IE -RRB- , and named-entity recognition ',Yarowsky,'nsubjpass_used_Collocations aux_used_have auxpass_used_been advmod_used_widely prep_used_for pobj_for_tasks dep_as_such prep_tasks_as nn_disambiguation_word nn_disambiguation_sense pobj_as_disambiguation abbrev_disambiguation_WSD amod_disambiguation_ nn_extraction_information measure__extraction abbrev_extraction_IE conj__ cc__and amod__named-entity nn__recognition conj__ ' +o,3233,'The POS tagger uses the same contextual predicates as ; the supertagger adds contextual predicates corresponding to POS tags and bigram combinations of POS tags ',Ratnaparkhi,'det_tagger_The dep_tagger_POS nsubj_uses_tagger det_predicates_the amod_predicates_same amod_predicates_contextual dobj_uses_predicates prep_predicates_as pobj_as_ det_supertagger_the nsubj_adds_supertagger parataxis_uses_adds amod_predicates_contextual dobj_adds_predicates partmod_predicates_corresponding aux_POS_to xcomp_corresponding_POS dobj_POS_tags cc_tags_and nn_combinations_bigram conj_tags_combinations prep_tags_of dep__POS nn__tags pobj_of_ ' +o,3234,'The principal training method is an adaptation of averaged perceptron learning as described by ',Collins,'det_method_The amod_method_principal nn_method_training nsubj_adaptation_method cop_adaptation_is det_adaptation_an prep_adaptation_of amod_perceptron_averaged pobj_of_perceptron partmod_perceptron_learning advmod_described_as acomp_learning_described prep_described_by ' +o,3235,'We optimize the model weights using a modified version of averaged perceptron learning as described by ',Collins,'nsubj_optimize_We det_weights_the nn_weights_model nsubj_using_weights dep_optimize_using det_version_a amod_version_modified dobj_using_version prep_version_of amod_perceptron_averaged pobj_of_perceptron partmod_perceptron_learning advmod_described_as acomp_learning_described prep_described_by ' +o,3236,'By habit , most systems for automatic role-semantic analysis have used Pennstyle constituents produced by or Charniaks -LRB- 2000 -RRB- parsers ',Collins,'prep_used_By pobj_By_habit amod_systems_most nsubj_used_systems prep_systems_for amod_analysis_automatic amod_analysis_role-semantic pobj_for_analysis aux_used_have nn__Pennstyle nn__constituents dobj_used_ partmod__produced prep_produced_by amod_parsers_ cc__or conj__Charniaks appos_parsers_2000 pobj_by_parsers ' +o,3237,'We used the Berkeley Parser 2 to train such grammars on sections 2-21 of the Penn Treebank ',Marcus,'nsubj_used_We det_Parser_the nn_Parser_Berkeley nsubj_train_Parser num_Parser_2 aux_train_to xcomp_used_train amod_grammars_such dobj_train_grammars prep_train_on pobj_on_sections num_sections_2-21 prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,3238,'In a standard phrase-based model is augmented with more than a million features whose weights are trained discriminatively by a variant of the perceptron algorithm ',Liang,'prep_augmented_In pobj_In_ det_model_a amod_model_standard amod_model_phrase-based nsubj_augmented_model cop_augmented_is prep_augmented_with dep_than_more quantmod_million_than quantmod_million_a num_features_million pobj_with_features poss_weights_whose nsubjpass_trained_weights auxpass_trained_are rcmod_features_trained advmod_trained_discriminatively prep_trained_by det_variant_a pobj_by_variant prep_variant_of det_algorithm_the nn_algorithm_perceptron pobj_of_algorithm ' +o,3239,'Despite the above differences , since the theorems of convergence and their proof are only dependent on the feature vectors , and not on the source of the feature definitions , the perceptron algorithm is applicable to the training of our CWS model ',Collins,'prep_applicable_Despite det_differences_the amod_differences_above pobj_Despite_differences mark_dependent_since det_theorems_the nsubj_dependent_theorems prep_theorems_of pobj_of_convergence cc_convergence_and poss__their nn__proof conj_convergence_ cop_dependent_are advmod_dependent_only dep_applicable_dependent prep_dependent_on det_vectors_the nn_vectors_feature pobj_on_vectors cc_on_and conj_on_not dep_on_on det_source_the pobj_on_source prep_source_of det_definitions_the nn_definitions_feature pobj_of_definitions det_algorithm_the nn_algorithm_perceptron nsubj_applicable_algorithm cop_applicable_is prep_applicable_to det_training_the pobj_to_training prep_training_of poss_model_our nn_model_CWS pobj_of_model ' +o,3240,'It is appreciated that multi-sense words appearing in the same document tend to be tagged with the same word sense if they belong to the same common domain in the semantic hierarchy ',Yarowsky,'nsubjpass_appreciated_It auxpass_appreciated_is complm_tend_that amod_words_multi-sense nsubj_tend_words partmod_words_appearing prep_appearing_in det_document_the amod_document_same pobj_in_document ccomp_appreciated_tend aux_tagged_to auxpass_tagged_be xcomp_tend_tagged prep_tagged_with det_sense_the amod_sense_same nn_sense_word pobj_with_sense mark_belong_if nsubj_belong_they advcl_tagged_belong prep_belong_to det_domain_the amod_domain_same amod_domain_common pobj_to_domain prep_domain_in det_hierarchy_the amod_hierarchy_semantic pobj_in_hierarchy ' +o,3241,' describes a method of sentiment classification using two human-selected seed words -LRB- the words poor and excellent -RRB- in conjunction with a very large text corpus ; the semantic orientation of phrases is computed as their association with the seed words -LRB- as measured by pointwise mutual information -RRB- ',Turney,'nsubj_describes_ det_method_a dobj_describes_method prep_method_of nn_classification_sentiment pobj_of_classification partmod_classification_using num_words_two amod_words_human-selected nn_words_seed dobj_using_words det_words_the dep_poor_words dep_words_poor cc_poor_and conj_poor_excellent prep_words_in pobj_in_conjunction prep_using_with det_corpus_a advmod_large_very amod_corpus_large nn_corpus_text pobj_with_corpus det_orientation_the amod_orientation_semantic nsubjpass_computed_orientation prep_orientation_of pobj_of_phrases auxpass_computed_is parataxis_describes_computed prep_computed_as poss_association_their pobj_as_association prep_computed_with det_words_the nn_words_seed pobj_with_words mark_measured_as dep_computed_measured prep_measured_by amod_information_pointwise amod_information_mutual pobj_by_information ' +o,3242,'24 GermanEnglish For GermanEnglish , we additionally incorporated rule-based reordering We parse the input using the Collins parser and apply a set of reordering rules to re-arrange the German sentence so that it corresponds more closely English word order ',Collins,'num_GermanEnglish_24 prep_incorporated_For pobj_For_GermanEnglish nsubj_incorporated_we advmod_incorporated_additionally dep_GermanEnglish_incorporated amod_reordering_rule-based dobj_incorporated_reordering nsubj_parse_We dep_GermanEnglish_parse det_input_the dobj_parse_input partmod_input_using det__the nn__Collins nn__parser dobj_using_ cc_parse_and conj_parse_apply det_set_a dobj_apply_set prep_set_of nn_rules_reordering pobj_of_rules aux_re-arrange_to xcomp_apply_re-arrange det_sentence_the amod_sentence_German dobj_re-arrange_sentence advmod_corresponds_so dep_corresponds_that nsubj_corresponds_it dep_re-arrange_corresponds advmod_closely_more advmod_corresponds_closely amod_order_English nn_order_word dobj_corresponds_order ' +o,3243,'There is potential of developing Sense Definition Model to identify and represent semantic and stylistic differentiation reflected in the MRD glosses pointed out in DiMarco , Hirst and Stede ',Dunning,'expl_is_There nsubj_is_potential prep_potential_of pcomp_of_developing nn_Model_Sense nn_Model_Definition nsubj_identify_Model aux_identify_to xcomp_developing_identify cc_identify_and conj_identify_represent amod_differentiation_semantic cc_semantic_and conj_semantic_stylistic dobj_identify_differentiation partmod_differentiation_reflected prep_reflected_in det_MRD_the pobj_in_MRD nsubj_pointed_glosses dep_is_pointed prt_pointed_out prep_pointed_in nn__DiMarco conj_DiMarco_Hirst cc_DiMarco_and conj_DiMarco_Stede pobj_in_ ' +o,3244,'On the same dataset as that of , our new supertagger achieves an accuracy of a2a4a3a6a5a8a7a10a9a12a11 Compared with the supertaggers with the same decoding complexity , our algorithm achieves an error reduction of a22a23a5a26a9a12a11 We repeat Ramshaw and Marcus Transformation Based NP chunking test by substituting supertags for POS tags in the dataset ',Ramshaw,'prep_achieves_On det_dataset_the amod_dataset_same pobj_On_dataset prep_dataset_as pobj_as_that prep_that_of pobj_of_ poss_supertagger_our amod_supertagger_new nsubj_achieves_supertagger parataxis_achieves_achieves det_accuracy_an dobj_achieves_accuracy prep_accuracy_of pobj_of_a2a4a3a6a5a8a7a10a9a12a11 prep_achieves_Compared dep_Compared_with det_supertaggers_the pobj_with_supertaggers prep_supertaggers_with det__the amod__same nn__decoding nn__complexity pobj_with_ poss_algorithm_our nsubj_achieves_algorithm det_reduction_an nn_reduction_error dobj_achieves_reduction prep_reduction_of pobj_of_a22a23a5a26a9a12a11 nsubj_repeat_We dep_achieves_repeat nn_NP_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nn_NP_Transformation nn_NP_Based nsubj_chunking_NP dep_repeat_chunking amod_test_ dobj_chunking_test prep_chunking_by pcomp_by_substituting dobj_substituting_supertags prep_substituting_for dep_tags_POS pobj_for_tags prep_tags_in det_dataset_the pobj_in_dataset ' +p,3245,'In addition to the widely used BLEU and NIST scores , we also evaluate translation quality with the recently proposed Meteor and four edit-distance style metrics , Word Error Rate -LRB- WER -RRB- , Positionindependent word Error Rate -LRB- PER -RRB- , CDER , which allows block reordering , and Translation Edit Rate -LRB- TER -RRB- ',Banerjee,'prep_evaluate_In dep_In_addition dep_In_to det__the advmod_used_widely amod__used nn__BLEU pobj_In_ cc__and amod_scores_NIST amod_scores_ conj__scores nsubj_evaluate_we advmod_evaluate_also nn_quality_translation dobj_evaluate_quality prep_evaluate_with det__the advmod_proposed_recently amod__proposed nn__Meteor pobj_with_ cc__and num_metrics_four amod_metrics_edit-distance nn_metrics_style conj__metrics nn_Rate_Word nn_Rate_Error appos_metrics_Rate abbrev_Rate_WER amod__Positionindependent nn__word nn__Error nn__Rate abbrev__PER conj_Rate_ conj_Rate_CDER nsubj_allows_which rcmod_CDER_allows xcomp_allows_block nn__reordering dobj_block_ cc_Rate_and nn__Translation nn__Edit nn__Rate abbrev__TER conj_Rate_ ' +o,3246,'We shall take HMM-based word alignment model as an example and follow the notation of ',Brown,'nsubj_take_We aux_take_shall amod__HMM-based nn__word nn__alignment nn__model dobj_take_ prep_take_as det_example_an pobj_as_example cc_take_and conj_take_follow det_notation_the dobj_follow_notation prep_notation_of ' +o,3247,'The baseline score using all phrase pairs was 5911 with a 95 \% confidence interval of -LRB- 5713 , 6109 -RRB- ',Papineni,'det_score_The amod_score_baseline nsubj__score partmod_score_using det_pairs_all nn_pairs_phrase dobj_using_pairs cop__was number__5911 prep__with det_interval_a dep_\%_95 amod_interval_\% nn_interval_confidence pobj_with_interval prep__of dep_of_5713 appos_5713_6109 ' +o,3248,' et al 1996 presented a way of computing conditional maximum entropy models directly by modifying equation 6 as follows -LRB- now instead of w we will explicitly use -LRB- x , y -RRB- -RRB- : i ~ Cx ~ -RRB- = ~ f ~ -LRB- ~ , y -RRB- \\* ~ -LRB- ~ , y -RRB- ~ ~ ~ -LRB- ~ , y -RRB- \\* ~ -LRB- ~ -RRB- \\* pCy I ~ -RRB- = p -LRB- xk -RRB- -LRB- 9 -RRB- x6X yEY xEX yEY where ~ -LRB- x , y -RRB- is an empirical probability of a joint configuration -LRB- w -RRB- of certain instantiated factor I variables with certain instantiated behavior variables ',Berger,'nn_al_ nn_al_et nsubj_presented_al num_al_1996 det_way_a dobj_presented_way prep_way_of pcomp_of_computing amod_models_conditional amod_models_maximum nn_models_entropy dobj_computing_models advmod_computing_directly prep_computing_by pcomp_by_modifying dobj_modifying_equation num_equation_6 mark_follows_as dep_equation_follows dep_equation_now dep_of_instead prep_now_of pobj_of_w nsubj_use_we aux_use_will advmod_use_explicitly rcmod_w_use dep_use_x appos_x_y dep_~_i nn_~_~ nn_~_Cx nsubj_~_~ dep_~_= parataxis_presented_~ nn_~_f dobj_~_~ appos_~_~ dep_~_y dep_~_\\* amod_~_~ appos_~_~ dep_~_y dobj_~_~ nn_~_~ dobj_~_~ appos_~_~ dep_~_y dep_~_\\* dobj_~_~ appos_~_~ dep_~_\\* nn_p_pCy nn_p_I nn_p_~ amod_p_= dobj_~_p appos_p_xk appos_p_9 nn_yEY_x6X nn_yEY_yEY nn_yEY_xEX dobj_~_yEY advmod_probability_where nsubj_probability_~ dep_~_x appos_x_y cop_probability_is det_probability_an amod_probability_empirical advcl_~_probability prep_probability_of det_configuration_a amod_configuration_joint pobj_of_configuration appos_configuration_w prep_configuration_of amod_variables_certain amod_variables_instantiated nn_variables_factor nn_variables_I pobj_of_variables prep_variables_with amod_variables_certain amod_variables_instantiated nn_variables_behavior pobj_with_variables ' +o,3249,'We also show that the domain adaptation work of , which is presented as an ad-hoc preprocessing step , is actually equivalent to our formal model ','Daume III','nsubj_show_We advmod_show_also complm_equivalent_that det_work_the nn_work_domain nn_work_adaptation nsubj_equivalent_work prep_work_of pobj_of_ nsubjpass_presented_which auxpass_presented_is rcmod_work_presented prep_presented_as det_step_an amod_step_ad-hoc nn_step_preprocessing pobj_as_step cop_equivalent_is advmod_equivalent_actually ccomp_show_equivalent prep_equivalent_to poss_model_our amod_model_formal pobj_to_model ' +o,3250,'Following the broad shift in the field from finite state transducers to grammar transducers , recent approaches to phrase-based alignment have used synchronous grammar formalisms permitting polynomial time inference ',Zhang,'prep_transducers_Following det_shift_the amod_shift_broad pobj_Following_shift prep_shift_in det_field_the pobj_in_field prep_field_from amod_state_finite pobj_from_state prep_transducers_to nn_transducers_grammar pobj_to_transducers nsubj_transducers_ amod_approaches_recent appos__approaches prep_approaches_to amod_alignment_phrase-based pobj_to_alignment aux_used_have rcmod_alignment_used amod_formalisms_synchronous nn_formalisms_grammar dobj_used_formalisms xcomp_used_permitting amod_inference_polynomial nn_inference_time dobj_permitting_inference ' +o,3251,'Within the NLP community , n-best list ranking has been looked at carefully in parsing , extractive summarization , and machine translation , to name a few ',Lin,'prep_looked_Within det_community_the nn_community_NLP pobj_Within_community amod_ranking_n-best nn_ranking_list nsubjpass_looked_ranking aux_looked_has auxpass_looked_been dep_carefully_at advmod_looked_carefully prep_looked_in pobj_in_parsing amod__extractive nn__summarization conj_parsing_ cc_parsing_and nn__machine nn__translation conj_parsing_ aux_name_to xcomp_looked_name det_few_a dobj_name_few ' +o,3252,'In the literature approaches to construction of taxonomies of concepts have been proposed ',Brown,'det_literature_the pobj_In_literature nsubjpass_proposed_approaches prep_approaches_to pobj_to_construction prep_construction_of pobj_of_taxonomies prep_taxonomies_of pobj_of_concepts aux_proposed_have auxpass_proposed_been rcmod_literature_proposed ' +o,3253,'-LRB- 1 -RRB- a I expected -LRB- nv the man who smoked NP -RRB- to eat ice-cream h I doubted -LRB- NP the man who liked to eat ice-cream NP -RRB- Current high-coverage parsers tend to use either custom , hand-generated lists of subcategorization frames , or published , handgenerated lists like the Ozford Advanced Learner \'s Dictionary of Contemporary English , Hornby and Covey ',Hindle,'dep_tend_1 det_I_a nsubj_tend_I partmod_I_expected parataxis_expected_nv det_man_the dobj_nv_man nsubj_smoked_who rcmod_man_smoked dobj_smoked_NP aux_eat_to xcomp_expected_eat amod_h_ice-cream dobj_eat_h nsubj_doubted_I rcmod_h_doubted parataxis_eat_NP det_man_the dobj_NP_man nsubj_liked_who rcmod_man_liked aux_eat_to xcomp_liked_eat amod_NP_ice-cream dobj_eat_NP amod_parsers_Current amod_parsers_high-coverage dobj_eat_parsers ccomp_handgenerated_tend aux_use_to xcomp_tend_use det_custom_either dobj_use_custom amod_lists_hand-generated nsubj__lists prep_lists_of nn_frames_subcategorization pobj_of_frames dep_tend_ cc__or conj__published dobj_handgenerated_lists prep_handgenerated_like det_Learner_the nn_Learner_Ozford nn_Learner_Advanced poss_Dictionary_Learner possessive_Learner_\'s pobj_like_Dictionary prep_Dictionary_of nn_English_Contemporary pobj_of_English conj_English_Hornby cc_English_and conj_English_Covey nn__ nsubj_handgenerated_ ' +o,3254,'The concept of mutual information , taken from information theory , was proposed as a measure of word association ',Brown,'det_concept_The nsubjpass_proposed_concept prep_concept_of amod_information_mutual pobj_of_information partmod_concept_taken prep_taken_from nn_theory_information pobj_from_theory auxpass_proposed_was prep_proposed_as det_measure_a pobj_as_measure prep_measure_of nn__word nn__association pobj_of_ ' +o,3255,'Here , we present experiments performed using two complex corpora , C1 and C2 , extracted from the Penn Treebank ',Marcus,'advmod_present_Here nsubj_present_we nsubj_performed_experiments ccomp_present_performed xcomp_performed_using num_corpora_two amod_corpora_complex nsubj_extracted_corpora appos_corpora_C1 cc_C1_and conj_C1_C2 ccomp_using_extracted prep_extracted_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3256,'FollowingtheworkofKooetal and and , it is possible to compute all expectations in O -LRB- n3 + L n2 -RRB- through matrix inversion ',Smith,'amod__FollowingtheworkofKooetal cc__and conj__ cc__and dep_possible_ dobj__ nsubj_possible_it cop_possible_is conj__possible aux_compute_to xcomp_possible_compute det_expectations_all dobj_compute_expectations prep_expectations_in pobj_in_O amod_n2_n3 nn_n2_+ nn_n2_L appos_O_n2 prep_compute_through nn_inversion_matrix pobj_through_inversion ' +o,3257,'To tune the decoder parameters , we conducted minimum error rate training with respect to the word BLEU score using 20K development sentence pairs ',Och,'aux_tune_To dep_conducted_tune det_parameters_the nn_parameters_decoder dobj_tune_parameters nsubj_conducted_we amod__minimum nn__error nn__rate nn__training dobj_conducted_ prep_conducted_with pobj_with_respect prep_conducted_to det__the nn__word nn__BLEU nn__score pobj_to_ partmod__using num_pairs_20K nn_pairs_development nn_pairs_sentence dobj_using_pairs ' +o,3258,'Agreement is sometimes measured as percentage of the cases on which the annotators agree , but more often expected agreement is taken into account in using the kappa statistic , which is given by : = po pe1 p e -LRB- 1 -RRB- where po is the observed proportion of agreement and pe is the proportion of agreement expected by chance ',Carletta,'nsubjpass_measured_Agreement auxpass_measured_is advmod_measured_sometimes mark_proportion_as nsubj_proportion_percentage prep_percentage_of det_cases_the pobj_of_cases rel_agree_on pobj_on_which det_annotators_the nsubj_agree_annotators rcmod_cases_agree cc_agree_but advmod_often_more advmod_expected_often conj_agree_expected nsubjpass_taken_agreement auxpass_taken_is ccomp_expected_taken prep_taken_into pobj_into_account prep_taken_in pcomp_in_using det__the nn__kappa nn__statistic dobj_using_ nsubjpass_given_which auxpass_given_is rcmod__given prep_given_by dep_agree_= iobj_=_po amod_e_pe1 nn_e_p dobj_=_e appos_e_1 advmod_proportion_where nsubj_proportion_po cop_proportion_is det_proportion_the amod_proportion_observed advcl_agree_proportion prep_proportion_of pobj_of_agreement cc_agreement_and conj_agreement_pe cop_proportion_is det_proportion_the advcl_measured_proportion prep_proportion_of pobj_of_agreement partmod_agreement_expected prep_expected_by pobj_by_chance ' +o,3259,'In recent years , HMMs have enjoyed great success in many tagging applications , most notably part-of-speech -LRB- POS -RRB- tagging and named entity recognition ',Marcus,'prep_enjoyed_In amod_years_recent pobj_In_years nsubj_enjoyed_HMMs aux_enjoyed_have amod_success_great dobj_enjoyed_success prep_enjoyed_in amod_applications_many amod_applications_tagging pobj_in_applications advmod_notably_most advmod__notably amod__part-of-speech appos__POS nn__tagging appos_applications_ cc_enjoyed_and conj_enjoyed_named nn__entity nn__recognition dobj_named_ ' +o,3260,'Inspired by the idea of graph based algorithms to collectively rank and select the best candidate , research efforts in the natural language community have applied graph-based approaches on keyword selection , text summarization , word sense disambiguation , sentiment analysis , and sentence retrieval for question answering ',Mihalcea,'prep_Inspired_by det_idea_the pobj_by_idea prep_idea_of pobj_of_graph partmod_graph_based dep_Inspired_algorithms aux_rank_to advmod_rank_collectively xcomp_algorithms_rank cc_rank_and conj_rank_select det_candidate_the amod_candidate_best nsubj_Inspired_candidate nn_efforts_research appos_candidate_efforts prep_efforts_in det_community_the amod_community_natural nn_community_language pobj_in_community aux_applied_have rcmod_community_applied amod_approaches_graph-based dobj_applied_approaches prep_applied_on amod__keyword nn__selection pobj_on_ nn__text nn__summarization conj__ nn__word nn__sense nn__disambiguation conj__ nn__sentiment nn__analysis conj__ cc__and nn_retrieval_sentence conj__retrieval prep_retrieval_for nn__question nn__answering pobj_for_ ' +p,3261,'Along this line , present convincing evidence that restricting phrasal translation to syntactic constituents yields poor translation performance the ability to translate nonconstituent phrases -LRB- such as there are , note that , and according to -RRB- turns out to be critical and pervasive ',Koehn,'prep_yields_Along det_evidence_this nn_evidence_line nn_evidence_ amod_evidence_present nn_evidence_convincing pobj_Along_evidence nsubj_restricting_that rcmod_evidence_restricting amod_translation_phrasal dobj_restricting_translation prep_restricting_to amod_constituents_syntactic pobj_to_constituents amod_performance_poor nn_performance_translation dobj_yields_performance det_ability_the nsubj_yields_ability aux_translate_to infmod_ability_translate amod_phrases_nonconstituent dobj_translate_phrases dep_as_such dep_phrases_as pobj_as_there dep_there_are appos_there_note amod_note_that cc_that_and conj_that_according dep_according_to rcmod_phrases_turns prt_turns_out aux_critical_to cop_critical_be xcomp_turns_critical cc_critical_and conj_critical_pervasive ' +o,3262,'The model can be seen as a bootstrapping learning process tbr disambiguation , where the information gained from one part -LRB- selectional preference -RRB- is used to improve tile other -LRB- disambiguation -RRB- and vice versa , reminiscent of the work by Riloff and Jones and ',Yarowsky,'det_model_The nsubjpass_seen_model aux_seen_can auxpass_seen_be dep_used_seen prep_seen_as det_bootstrapping_a pobj_as_bootstrapping partmod_bootstrapping_learning nn_disambiguation_process nn_disambiguation_tbr dobj_learning_disambiguation advmod_gained_where det_information_the nsubj_gained_information rcmod_disambiguation_gained prep_gained_from num_part_one pobj_from_part amod_preference_selectional appos_part_preference auxpass_used_is aux_improve_to xcomp_used_improve dobj_improve_tile amod_tile_other appos_tile_disambiguation cc_tile_and conj_tile_vice dep_reminiscent_versa amod_vice_reminiscent prep_vice_of det_work_the pobj_of_work prep_improve_by pobj_by_Riloff cc_Riloff_and conj_Riloff_Jones nn_and_ nsubjpass_used_and ' +o,3263,'Rather than explicit annotation , we could use latent annotations to split the POS tags , similarly to the introduction of latent annotations to PCFG grammars ',Matsuzaki,'dep_than_Rather advcl_use_than amod_annotation_explicit pobj_than_annotation nsubj_use_we aux_use_could amod_annotations_latent dobj_use_annotations aux_split_to xcomp_use_split det_tags_the dep_tags_POS dobj_split_tags advmod_split_similarly prep_split_to det_introduction_the pobj_to_introduction prep_introduction_of amod_annotations_latent pobj_of_annotations prep_annotations_to pcomp_to_PCFG dobj_PCFG_grammars ' +n,3264,'By doing so we must emphasize that , as described in the previous section , the BLEU score was not designed to deliver satisfactory results at the sentence level , and this also applies to the closely related NIST score ',Papineni,'prep_designed_By pcomp_By_doing mark_emphasize_so nsubj_emphasize_we aux_emphasize_must advcl_doing_emphasize dobj_emphasize_that advmod_described_as amod_that_described prep_emphasize_in det_section_the amod_section_previous pobj_in_section det_score_the amod_score_BLEU nsubjpass_designed_score auxpass_designed_was neg_designed_not aux_deliver_to xcomp_designed_deliver amod_results_satisfactory dobj_deliver_results prep_deliver_at det__the nn__sentence nn__level pobj_at_ cc_designed_and nsubj_applies_this advmod_applies_also conj_designed_applies prep_applies_to det_score_the advmod_related_closely amod_score_related nn_score_NIST pobj_to_score ' +p,3265,'One of the most notable examples is bootstrapping algorithm for word sense disambiguation ',Yarowsky,'nsubj_algorithm_One prep_One_of det_examples_the advmod_notable_most amod_examples_notable pobj_of_examples cop_algorithm_is amod_algorithm_ amod_algorithm_bootstrapping prep_algorithm_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +o,3266,'Similarly , the sense disambiguation problem is typically attacked by comparing the distribution of the neighbors of a word \'s occurrence to prototypical distributions associated with each of the word \'s senses ',Brown,'advmod_attacked_Similarly det_problem_the nn_problem_sense nn_problem_disambiguation nsubjpass_attacked_problem auxpass_attacked_is advmod_attacked_typically prep_attacked_by pcomp_by_comparing det_distribution_the dobj_comparing_distribution prep_distribution_of det_neighbors_the pobj_of_neighbors prep_neighbors_of det_word_a poss_occurrence_word possessive_word_\'s pobj_of_occurrence prep_comparing_to amod_distributions_prototypical pobj_to_distributions partmod_distributions_associated prep_associated_with pobj_with_each prep_each_of det_word_the poss__word possessive_word_\'s nn__senses pobj_of_ ' +o,3267,'31 Generation using PHARAOH PHARAOH is an SMT system that uses phrases as basic translation units ',Koehn,'num_Generation_31 nsubj_system_Generation partmod_Generation_using nn__PHARAOH nn__PHARAOH dobj_using_ cop_system_is det_system_an nn_system_SMT nsubj_uses_that rcmod_system_uses dobj_uses_phrases prep_phrases_as amod_units_basic nn_units_translation pobj_as_units ' +o,3268,'2 Related Work A number of researchers have described approaches that preprocess the source language input in SMT systems ',Berger,'num_Work_2 amod_Work_Related det_number_A nsubj_described_number prep_number_of nn__researchers pobj_of_ aux_described_have dep_Work_described dobj_described_approaches nsubj_preprocess_that rcmod_approaches_preprocess det_input_the nn_input_source nn_input_language dobj_preprocess_input prep_input_in nn_systems_SMT pobj_in_systems ' +o,3269,'3 Stochastic Inversion Transduction Grammars Stochastic Inversion Transduction Grammars -LRB- SITGs -RRB- can be viewed as a restricted subset of Stochastic Syntax-Directed Transduction Grammars ',Wu,'number_Stochastic_3 num_Grammars_Stochastic nn_Grammars_Inversion nn_Grammars_Transduction nn_Grammars_Grammars nn_Grammars_Stochastic nn_Grammars_Inversion nn_Grammars_Transduction abbrev_Grammars_SITGs nsubjpass_viewed_ aux_viewed_can auxpass_viewed_be dep_Grammars_viewed prep_viewed_as det_subset_a amod_subset_restricted pobj_as_subset prep_subset_of nn_Grammars_Stochastic nn_Grammars_Syntax-Directed nn_Grammars_Transduction pobj_of_Grammars ' +o,3270,' argue that precise alignment can improve transliteration effectiveness , experimenting on English-Chinese data and comparing IBM models with phonemebased alignments using direct probabilities ',Brown,'nsubj_argue_ complm_improve_that amod_alignment_precise nsubj_improve_alignment aux_improve_can ccomp_argue_improve nn_effectiveness_transliteration dobj_improve_effectiveness dep_improve_experimenting prep_experimenting_on amod_data_English-Chinese pobj_on_data cc_experimenting_and conj_experimenting_comparing nn_models_IBM nsubj_using_models dep_using_ prep__with amod_alignments_phonemebased pobj_with_alignments xcomp_comparing_using amod_probabilities_direct dobj_using_probabilities ' +o,3271,'The proposed synchronous grammar is able to cover the previous proposed grammar based on tree and tree sequence alignment ',Zhang,'det_grammar_The amod_grammar_proposed amod_grammar_synchronous nsubj_able_grammar cop_able_is aux_cover_to xcomp_able_cover det_grammar_the amod_grammar_previous amod_grammar_proposed dobj_cover_grammar prep_cover_based dep_based_on nn__tree pobj_on_ cc__and nn_alignment_tree nn_alignment_sequence nn_alignment_ conj__alignment ' +o,3272,'2 The Problem of Coverage in SMT Statistical machine translation made considerable advances in translation quality with the introduction of phrase-based translation ',Och,'det_Problem_The nsubj_made_Problem prep_Problem_of pobj_of_Coverage prep_Coverage_in nn_translation_SMT nn_translation_Statistical nn_translation_machine pobj_in_translation dep_2_made amod_advances_considerable dobj_made_advances prep_made_in nn_quality_translation pobj_in_quality prep_made_with det_introduction_the pobj_with_introduction prep_introduction_of amod_translation_phrase-based pobj_of_translation ' +o,3273,'A similar soft projection of dependencies was used in supervised machine translation by , who used a source sentences dependency paths to bias the generation of its translation ',Smith,'det_projection_A amod_projection_similar amod_projection_soft nsubjpass_used_projection prep_projection_of pobj_of_dependencies auxpass_used_was prep_used_in amod_translation_supervised nn_translation_machine pobj_in_translation prep_used_by pobj_by_ nsubj_used_who rcmod__used det_sentences_a nn_sentences_source nsubj_dependency_sentences ccomp_used_dependency nsubj_bias_paths aux_bias_to xcomp_dependency_bias det_generation_the dobj_bias_generation prep_generation_of poss_translation_its pobj_of_translation ' +o,3274,'Recent computational work either focuses on sentence subjectivity , concentrates just on explicit statements of evaluation , such as of films , or focuses on just one aspect of opinion , eg , on adjectives ',Turney,'amod_work_Recent amod_work_computational nsubj_focuses_work advmod_focuses_either prep_focuses_on nn__sentence nn__subjectivity pobj_on_ conj_focuses_concentrates advmod_concentrates_just prep_concentrates_on amod_statements_explicit pobj_on_statements prep_statements_of pobj_of_evaluation dep_as_such prep_statements_as dep_as_of nn__films pobj_of_ cc_focuses_or conj_focuses_focuses prep_focuses_on advmod_aspect_just num_aspect_one pobj_on_aspect prep_aspect_of nn__opinion dep__eg pobj_of_ prep_focuses_on pobj_on_adjectives ' +p,3275,'There has been considerable skepticism over whether WSD will actually improve performance of applications , but we are now starting to see improvement in performance due to WSD in cross-lingual information retrieval and machine translation and we hope that other applications such as question-answering , text simplication and summarisation might also benet as WSD methods improve ',Carpuat,'expl_been_There aux_been_has amod_skepticism_considerable dobj_been_skepticism prep_skepticism_over complm_improve_whether nsubj_improve_WSD aux_improve_will advmod_improve_actually pcomp_over_improve dobj_improve_performance prep_performance_of pobj_of_applications cc_been_but nsubj_starting_we aux_starting_are advmod_starting_now conj_been_starting aux_see_to xcomp_starting_see dobj_see_improvement prep_improvement_in pobj_in_performance dep_to_due prep_see_to pobj_to_WSD prep_WSD_in amod__cross-lingual nn__information nn__retrieval pobj_in_ cc__and nn__machine nn__translation conj__ cc_been_and nsubj_hope_we conj_been_hope complm_benet_that amod_applications_other nsubj_benet_applications dep_as_such prep_applications_as nn_simplication_question-answering conj_simplication_text pobj_as_simplication cc_simplication_and conj_simplication_summarisation aux_benet_might advmod_benet_also ccomp_hope_benet mark_improve_as nn_methods_WSD nsubj_improve_methods advcl_benet_improve ' +o,3276,'First , a parsing-based approach attempts to recover partial parses from the parse chart when the input can not be parsed in its entirety due to noise , in order to construct a -LRB- partial -RRB- semantic representation ',Brown,'advmod_attempts_First det_approach_a amod_approach_parsing-based nsubj_attempts_approach aux_recover_to xcomp_attempts_recover amod_parses_partial dobj_recover_parses prep_recover_from det_chart_the amod_chart_parse pobj_from_chart advmod_parsed_when det_input_the nsubjpass_parsed_input aux_parsed_can neg_parsed_not auxpass_parsed_be advcl_recover_parsed prep_parsed_in poss_entirety_its pobj_in_entirety amod_entirety_due prep_due_to pobj_to_noise mark_construct_in dep_construct_order aux_construct_to purpcl_attempts_construct amod__a dep_a_partial amod__semantic nn__representation dobj_construct_ ' +o,3277,'For example , in our previous work , we have used a statistical translation memory of phrases in conjunction with a statistical translation model ',Brown,'prep_used_For pobj_For_example prep_used_in poss__our amod__previous nn__work pobj_in_ nsubj_used_we aux_used_have det_memory_a amod_memory_statistical nn_memory_translation dobj_used_memory prep_memory_of pobj_of_phrases prep_phrases_in pobj_in_conjunction prep_used_with det_model_a amod_model_statistical nn_model_translation pobj_with_model ' +o,3278,' describe one application of MI to identify word collocations ; Kashioka et al ',Brown,'nsubj_describe_ num_application_one nsubj_identify_application prep_application_of pobj_of_MI aux_identify_to xcomp_describe_identify nn_collocations_word dobj_identify_collocations dep_collocations_Kashioka cc_Kashioka_et conj_Kashioka_al ' +o,3279,' and , the specific technique we used by means of a context language model is rather different ',He,'nsubj_different_ cc__and conj__ det_technique_the amod_technique_specific nsubj_different_technique nsubj_used_we rcmod_technique_used prep_used_by pobj_by_means prep_means_of det_model_a nn_model_context nn_model_language pobj_of_model cop_different_is advmod_different_rather ' +n,3280,' focus on alignment and do not present MT results , while May and Knight -LRB- 2007 -RRB- takesthesyntacticre-alignmentasaninputtoanEM algorithm where the unaligned target words are insertedintothetemplatesandminimumtemplatesare combinedintobiggertemplates ',DeNero,'nsubj_focus_ prep_focus_on pobj_on_alignment cc_focus_and aux_present_do neg_present_not conj_focus_present nn_results_MT dobj_present_results mark_combinedintobiggertemplates_while nn_algorithm_May cc_May_and conj_May_Knight appos_algorithm_2007 amod_algorithm_takesthesyntacticre-alignmentasaninputtoanEM nsubj_combinedintobiggertemplates_algorithm advmod_insertedintothetemplatesandminimumtemplatesare_where det_words_the amod_words_unaligned nn_words_target nsubj_insertedintothetemplatesandminimumtemplatesare_words cop_insertedintothetemplatesandminimumtemplatesare_are rcmod_algorithm_insertedintothetemplatesandminimumtemplatesare advcl_present_combinedintobiggertemplates acomp_combinedintobiggertemplates_ ' +o,3281,'1 Introduction Viewed at a very high level , statistical machine translationinvolvesfourphases : languageandtranslation model training , parameter tuning , decoding , and evaluation ',Koehn,'num_Viewed_1 nn_Viewed_Introduction prep_Viewed_at det_level_a advmod_high_very amod_level_high pobj_at_level amod_translationinvolvesfourphases_statistical nn_translationinvolvesfourphases_machine appos_level_translationinvolvesfourphases nn_training_languageandtranslation nn_training_model dep_Viewed_training nn_tuning_parameter conj_training_tuning conj_training_decoding cc_training_and conj_training_evaluation ' +o,3282,'Tillmann and Zhang , et al ',Liang,'cc_Tillmann_and nn__Zhang conj_Tillmann_ conj__ cc__et conj__al ' +o,3283,'html -RRB- provided by Lynette Hirschman ; syntactic structures in the style of the Penn TreeBank provided by Ann Taylor ; and an alternative annotation for the F0 aspects of prosody , known as Tilt and provided by its inventor , Paul Taylor ',Marcus,'dobj_html_provided prep_html_by nn_Hirschman_Lynette pobj_by_Hirschman amod_structures_syntactic dep_Hirschman_structures prep_structures_in det_style_the pobj_in_style prep_style_of det__the nn__Penn nn__TreeBank pobj_of_ partmod__provided prep_provided_by nn_Taylor_Ann pobj_by_Taylor cc_structures_and det_annotation_an amod_annotation_alternative conj_structures_annotation prep_annotation_for det_aspects_the nn_aspects_F0 pobj_for_aspects prep_aspects_of pobj_of_prosody dep_aspects_known prep_known_as nn__Tilt pobj_as_ cc_known_and conj_known_provided prep_provided_by poss_inventor_its pobj_by_inventor nn_Taylor_Paul nsubj_html_Taylor ' +o,3284,'DTM2, introduced in (Ittycheriah and Roukos, 2007), expresses the phrase-based translation task in a unified log-linear probabilistic framework consisting of three components: (i) a prior conditional distribution P0(.|S), (ii) a number of feature functions i() that capture the translation and language model effects, and (iii) the weights of the features i that are estimated under MaxEnt (Berger et al., 1996), as in (1): P(T|S) = P0(T,J|S)Z expsummationdisplay i ii(T,J,S) (1) Here J is the skip reordering factor for the phrase pair captured by i() and represents the jump from the previous source word, and Z is the per source sentence normalization term.',Berger,'' +o,3285,'The only difference is that we 5See also work on partial parsing as a task in its own right : inter alia ',Hindle,'det_difference_The amod_difference_only nsubj_is_difference complm_work_that dep_5See_we nsubj_work_5See advmod_work_also ccomp_is_work prep_work_on amod_parsing_partial pobj_on_parsing prep_work_as det_task_a pobj_as_task prep_task_in poss_right_its amod_right_own pobj_in_right amod_alia_ nn_alia_inter dobj_work_alia ' +o,3286,'The ve part-ofspeech -LRB- POS -RRB- patterns from were used for the extraction of indicators , all involving at least one adjective or adverb ',Turney,'det_part-ofspeech_The nn_part-ofspeech_ve appos_part-ofspeech_POS nsubjpass_used_patterns prep_patterns_from pobj_from_ auxpass_used_were dep_part-ofspeech_used prep_used_for det_extraction_the pobj_for_extraction prep_extraction_of pobj_of_indicators appos_indicators_all prep_all_involving quantmod_one_at dep_at_least num_adjective_one pobj_involving_adjective cc_adjective_or conj_adjective_adverb ' +o,3287,' adapted the perceptron learning algorithm to tagging tasks , via sentence-based global feedback ',Collins,'nsubj_adapted_ det_perceptron_the dobj_adapted_perceptron partmod_perceptron_learning dobj_learning_algorithm prep_learning_to amod_tasks_tagging pobj_to_tasks prep_tasks_via amod_feedback_sentence-based amod_feedback_global pobj_via_feedback ' +o,3288,'Finally , it would be nice to merge some of the approaches by and with the ideas of semi-supervised learning introduced here , since they seem orthogonal in at least some aspects -LRB- eg , to replace the rudimentary lookahead features with full bidirectionality -RRB- ',Shen,'advmod_nice_Finally nsubj_nice_it aux_nice_would cop_nice_be aux_merge_to xcomp_nice_merge dobj_merge_some prep_some_of det_approaches_the pobj_of_approaches prep_merge_by pobj_by_ cc__and conj__ prep_merge_with det_ideas_the pobj_with_ideas prep_ideas_of amod_learning_semi-supervised pobj_of_learning partmod_learning_introduced advmod_introduced_here mark_orthogonal_since nsubj_orthogonal_they cop_orthogonal_seem advcl_introduced_orthogonal prep_orthogonal_in quantmod_some_at dep_at_least num_aspects_some pobj_in_aspects dep_aspects_eg aux_replace_to xcomp_eg_replace det_features_the amod_features_rudimentary amod_features_lookahead dobj_replace_features prep_replace_with amod_bidirectionality_full pobj_with_bidirectionality ' +p,3289,'In addition , their system does not classify non-anaphoric pronouns , A third paper that has significantly influenced our work is that of ',Haghighi,'prep_classify_In pobj_In_addition poss_system_their nsubj_classify_system aux_classify_does neg_classify_not amod_pronouns_non-anaphoric dobj_classify_pronouns det_paper_A amod_paper_third appos_pronouns_paper nsubj_influenced_that aux_influenced_has advmod_influenced_significantly rcmod_paper_influenced poss_work_our nsubj_that_work cop_that_is ccomp_influenced_that prep_that_of ' +o,3290,' and the English parser developed by ',Collins,'cc__and det_parser_the amod_parser_English conj__parser partmod_parser_developed prep_developed_by nn__ pobj_by_ ' +o,3291,'We implemented these models within an maximum entropy framework ',Berger,'nsubj_implemented_We det_models_these dobj_implemented_models prep_implemented_within det_framework_an amod_framework_maximum nn_framework_entropy pobj_within_framework ' +o,3292,' -RRB- or Wikipedia , and the contextual role played by an NP -LRB- see Bean and Riloff -RRB- ',Ponzetto,'cc__or nn__Wikipedia dobj__ cc__and det_role_the amod_role_contextual conj__role advmod__played dep_played_by det_NP_an pobj_by_NP dep_NP_see nn__Bean cc_Bean_and conj_Bean_Riloff dobj_see_ ' +o,3293,'SCISSOR is implemented by augmenting head-driven parsing model II to incorporate the generation of semantic labels on internal nodes ',Collins,'nsubjpass_implemented_SCISSOR auxpass_implemented_is prep_implemented_by pcomp_by_augmenting num_model_ amod_model_head-driven nn_model_parsing nsubj_II_model xcomp_augmenting_II aux_incorporate_to xcomp_II_incorporate det_generation_the dobj_incorporate_generation prep_generation_of amod_labels_semantic pobj_of_labels prep_incorporate_on amod_nodes_internal pobj_on_nodes ' +o,3294,'We ran the trainer with its default settings -LRB- maximum phrase length 7 -RRB- , and then used Koehns implementation of minimumerror-rate training to tune the feature weights to maximize the systems BLEU score on our development set , yielding the values shown in Table 2 ',Och,'nsubj_ran_We det_trainer_the dobj_ran_trainer prep_ran_with poss_settings_its nn_settings_default pobj_with_settings amod_length_maximum nn_length_phrase dep_ran_length num_length_7 cc_ran_and advmod_used_then conj_ran_used nn_implementation_Koehns nsubj_tune_implementation prep_implementation_of amod__minimumerror-rate nn__training pobj_of_ aux_tune_to xcomp_used_tune det_weights_the nn_weights_feature nsubj_maximize_weights aux_maximize_to xcomp_tune_maximize det_score_the nn_score_systems nn_score_BLEU dobj_maximize_score prep_maximize_on poss_set_our nn_set_development pobj_on_set xcomp_used_yielding det_values_the dobj_yielding_values partmod_values_shown prep_shown_in pobj_in_Table num_Table_2 ' +o,3295,'= = = = -RRB- -LRB- -RRB- , -LRB- InverseM1 -RRB- -LRB- -RRB- , -LRB- DirectM1 -RRB- -LRB- -RRB- , -LRB- InverseMLE -RRB- -LRB- -RRB- , -LRB- DirectMLE -RRB- -LRB- -RRB- , , -LRB- -RRB- -LRB- -RRB- , , -LRB- -RRB- -LRB- \\* , -RRB- , -LRB- -RRB- , , -LRB- , \\* -RRB- -LRB- -RRB- , -LRB- -RRB- , , -LRB- Atreelets s t Atreelets t s Atreelets Atreelets tspATSf stpATSf c cATSf c cATSf We use word probability tables p -LRB- t s -RRB- and p -LRB- s t -RRB- estimated by IBM Model 1 ',Brown,'nn_=_= nn_=_= dobj_=_= appos_-RRB-_InverseM1 amod_InverseM1_DirectM1 appos_-RRB-_InverseMLE nn_-LRB-_DirectMLE dep_DirectM1_\\* dep_,_\\* nsubj_stpATSf_Atreelets dep_stpATSf_s nn_stpATSf_t nn_stpATSf_Atreelets nn_stpATSf_t nn_stpATSf_s nn_stpATSf_Atreelets nn_stpATSf_Atreelets nn_stpATSf_tspATSf rcmod_=_stpATSf dep_cATSf_c dep_use_cATSf dep_cATSf_c dep_use_cATSf nsubj_use_We rcmod_stpATSf_use nn_p_word nn_p_probability nn_p_tables dobj_use_p nn_s_t appos_p_s cc_use_and nsubj_estimated_p amod_t_s appos_p_t conj_use_estimated prep_estimated_by nn_Model_IBM pobj_by_Model nsubj_=_1 ' +o,3296,'240 2 Motivation Many approaches to identifying base noun phrases have been explored as part of chunking , but determining sub-NP structure is rarely addressed ',Ramshaw,'number_2_240 num_approaches_2 amod_approaches_Motivation amod_approaches_Many nsubjpass_addressed_approaches prep_approaches_to pcomp_to_identifying nn_phrases_base nn_phrases_noun nsubjpass_explored_phrases aux_explored_have auxpass_explored_been ccomp_identifying_explored prep_explored_as pobj_as_part prep_part_of amod__chunking pobj_of_ cc_identifying_but conj_identifying_determining amod_structure_sub-NP dobj_determining_structure auxpass_addressed_is advmod_addressed_rarely ' +o,3297,'For colnparison ~ we refer here to because this method and the proposed method have much in connnon ',Smadja,'prep_refer_For nn_~_colnparison pobj_For_~ nsubj_refer_we advmod_refer_here aux__to xcomp_refer_ mark_have_because det_method_this nsubj_have_method cc_method_and det_method_the amod_method_proposed conj_method_method advcl__have dobj_have_much prep_much_in pobj_in_connnon ' +o,3298,'Many researchers have focused the related problem of predicting sentiment and opinion in text , sometimes connected to extrinsic values like prediction markets ',Pang,'amod_researchers_Many nsubj_focused_researchers aux_focused_have det_problem_the amod_problem_related dobj_focused_problem prep_problem_of pcomp_of_predicting dobj_predicting_sentiment cc_sentiment_and conj_sentiment_opinion prep_predicting_in nn__text pobj_in_ advmod_problem_sometimes amod__connected prep_connected_to amod_values_extrinsic pobj_to_values prep_values_like nn_markets_prediction pobj_like_markets dep_problem_ ' +o,3299,'While word and phrasal paraphrases can be assimilated to the well-studied notion of synonymy , sentencelevel paraphrasingis moredifficult to grasp and can not be equated with word-for-word or phrase-by-phrase substitution since it might entail changes in the structure of the sentence ',Barzilay,'mark_assimilated_While nn_paraphrases_word cc_word_and conj_word_phrasal nsubjpass_assimilated_paraphrases aux_assimilated_can auxpass_assimilated_be advcl_moredifficult_assimilated prep_assimilated_to det_notion_the amod_notion_well-studied pobj_to_notion prep_notion_of pobj_of_synonymy amod_paraphrasingis_sentencelevel nsubjpass_moredifficult_paraphrasingis aux_grasp_to xcomp_moredifficult_grasp cc_moredifficult_and aux_equated_can neg_equated_not auxpass_equated_be conj_moredifficult_equated prep_equated_with amod_substitution_word-for-word cc_word-for-word_or conj_word-for-word_phrase-by-phrase pobj_with_substitution mark_entail_since nsubj_entail_it aux_entail_might advcl_equated_entail dobj_entail_changes prep_entail_in det_structure_the pobj_in_structure prep_structure_of det_sentence_the pobj_of_sentence ' +o,3300,'We used a feature set which included the current , next , and previous word ; the previous two tags ; various capitalization and other features of the word being tagged -LRB- the full feature set is described in -RRB- ',Collins,'nsubj_used_We det_set_a nn_set_feature dobj_used_set nsubj_included_which rcmod_set_included det_current_the dobj_included_current conj_current_next cc_current_and amod_word_previous conj_current_word det_tags_the amod_tags_previous num_tags_two dep_current_tags amod_capitalization_various dep_current_capitalization cc_current_and amod_features_other conj_current_features prep_included_of det_word_the pobj_of_word auxpass_tagged_being partmod_word_tagged det_set_the amod_set_full nn_set_feature nsubjpass_described_set auxpass_described_is dep_word_described prep_described_in pobj_in_ ' +o,3301,'Clark and Curran describe the supertagger , which uses log-linear models to define a distribution over the lexical category set for each local five-word context containing the target word ',Ratnaparkhi,'nn__Clark cc_Clark_and conj_Clark_Curran nsubj_describe_ det_supertagger_the dobj_describe_supertagger nsubj_uses_which rcmod_supertagger_uses amod_models_log-linear dobj_uses_models aux_define_to xcomp_uses_define det_distribution_a dobj_define_distribution prep_define_over det_category_the amod_category_lexical pobj_over_category partmod_category_set prep_set_for det_context_each amod_context_local amod_context_five-word pobj_for_context partmod_context_containing det_word_the nn_word_target dobj_containing_word ' +o,3302,'1 Introduction Distributional Similarity has been an active research area for more than a decade , , , , , , ',Hindle,'nn_Similarity_Introduction nn_Similarity_Distributional nsubj_been_Similarity aux_been_has dep_1_been det_area_an amod_area_active nn_area_research dobj_been_area prep_area_for dep_than_more quantmod_a_than num_decade_a pobj_for_decade amod_decade_ dep__ dep__ dep__ dep__ dep__ dep__ ' +o,3303,'Its applications range from sentence boundary disambiguation to part-of-speech tagging , parsing and machine translation ',Berger,'poss_applications_Its nsubj_range_applications prep_range_from nn__sentence nn__boundary nn__disambiguation pobj_from_ prep_range_to amod__part-of-speech nn__tagging pobj_to_ xcomp_range_parsing nn_translation_ cc__and conj__machine dobj_parsing_translation ' +n,3304,'But without the global normalization , the maximumlikelihood criterion motivated by the maximum entropy principle is no longer a feasible option as an optimization criterion ',Berger,'cc_option_But prep_option_without det_normalization_the amod_normalization_global pobj_without_normalization det_criterion_the amod_criterion_maximumlikelihood nsubj_option_criterion partmod_criterion_motivated prep_motivated_by det__the amod__maximum amod__entropy nn__principle pobj_by_ cop_option_is advmod_longer_no advmod_option_longer det_option_a amod_option_feasible prep_option_as det_criterion_an nn_criterion_optimization pobj_as_criterion ' +o,3305,'#Reference: If our player 2, 3, 7 or 5 has the ball and the ball is close to our goal line PHARAOH++: If player 3 has the ball is in 2 5 the ball is in the area near our goal line WASP1++: If players 2, 3, 7 and 5 has the ball and the ball is near our goal line Figure 4: Sample partial system output in the ROBOCUP domain ROBOCUP GEOQUERY BLEU NIST BLEU NIST PHARAOH 0.3247 5.0263 0.2070 3.1478 WASP1 0.4357 5.4486 0.4582 5.9900 PHARAOH++ 0.4336 5.9185 0.5354 6.3637 WASP1++ 0.6022 6.8976 0.5370 6.4808 Table 1: Results of automatic evaluation; bold type indicates the best performing system (or systems) for a given domain-metric pair (p < 0.05) 5.1 Automatic Evaluation Weperformed4runsof10-foldcrossvalidation,and measured the performance of the learned generators using the BLEU score (Papineni et al. , 2002) and the NIST score (Doddington, 2002).',Papineni,'' +o,3306,' prototype-driven approach requires just a few prototype examples for each POS tag , exploiting these labeled words to constrain the labels of their distributionally similar words when training a generative log-linear model for POS tagging ',Haghighi,'amod_approach_ amod_approach_prototype-driven nsubj_requires_approach quantmod_few_just quantmod_few_a num_examples_few nn_examples_prototype dobj_requires_examples prep_requires_for det_tag_each dep_tag_POS pobj_for_tag xcomp_requires_exploiting det_words_these amod_words_labeled dobj_exploiting_words aux_constrain_to xcomp_exploiting_constrain det_labels_the dobj_constrain_labels prep_labels_of poss_words_their advmod_similar_distributionally amod_words_similar pobj_of_words advmod_training_when advcl_constrain_training det_model_a amod_model_generative amod_model_log-linear dobj_training_model prep_training_for dep_tagging_POS pobj_for_tagging ' +o,3307,'Within this class would fall the Lexical Implication Rules -LRB- LIRs -RRB- of , the lexical rules of Copestake and Briscoe -LRB- 1991 -RRB- , the Generative Lexicon of Pustejovsky , and the ellipsis recovery procedUres of Viegas and Nirenburg ',Ostler,'prep_fall_Within det_class_this pobj_Within_class aux_fall_would det_Rules_the nn_Rules_Lexical nn_Rules_Implication dobj_fall_Rules abbrev_Rules_LIRs prep_Rules_of pobj_of_ det_rules_the amod_rules_lexical conj_Rules_rules prep_rules_of pobj_of_Copestake cc_Copestake_and conj_Copestake_Briscoe appos_Copestake_1991 det_Lexicon_the nn_Lexicon_Generative conj_Rules_Lexicon prep_Lexicon_of nn__Pustejovsky pobj_of_ cc_Rules_and det_ellipsis_the conj_Rules_ellipsis nn_procedUres_recovery nsubj_fall_procedUres prep_procedUres_of nn__Viegas cc_Viegas_and conj_Viegas_Nirenburg pobj_of_ ' +o,3308,'This model is similar in spirit to IBM model 1 ',Brown,'det_model_This nsubj_similar_model cop_similar_is prep_similar_in pobj_in_spirit prep_spirit_to nn_model_IBM pobj_to_model tmod_similar_1 ' +o,3309,'We are also interested in examining the approach within a standard phrase-based decoder such as Moses or a hierarchical phrase system ',Koehn,'nsubj_interested_We cop_interested_are advmod_interested_also prep_interested_in pcomp_in_examining det_approach_the dobj_examining_approach prep_examining_within det_decoder_a amod_decoder_standard amod_decoder_phrase-based pobj_within_decoder dep_as_such prep_decoder_as nn__Moses pobj_as_ cc__or det__a amod__hierarchical nn__phrase nn__system conj__ ' +o,3310,'Some papers based on learned an aided dictionary from a corpus to reduce the possibility of unknown words ',Smadja,'det__Some nn__papers nsubj_learned_ prep__based dep_based_on pobj_on_ det_dictionary_an amod_dictionary_aided dobj_learned_dictionary prep_learned_from det_corpus_a pobj_from_corpus aux_reduce_to xcomp_learned_reduce det_possibility_the dobj_reduce_possibility prep_possibility_of amod_words_unknown pobj_of_words ' +o,3311,'Distributional cluster (Brown et al. , 1992): cost, expense, risk, profitability, deferral, earmarks, capstone, cardinality, mintage, reseller Word \'cost\' (2 alternatives) 0.5426 cost, price, terms, damage: the amount of money paid for something 0.4574 monetary value, price, cost: the amount of money it would bring if sold Word \'expense\' (2 alternatives) 1.0000 expense, expenditure, outlay, outgo, spending, disbursal, disbursement 0.0000 expense: a detriment or sacrifice; \'at the expense of\' Word \'risk\' (2 alternatives) 0.6267 hazard, jeopardy, peril risk: subconeept of danger 0.3733 risk, peril danger: subeonceptofventure Word \'profitability\' (1 alternatives) 1.0000 profitableness, profitability: subconcept of advantage, benefit, usefulness Word \'deferral\' (3 alternatives) 0.6267 abeyance, deferral, recess: subconcept of inaction, inactivity, inactiveness 0.3733 postponement, deferment, deferral, moratorium: an agreed suspension of activity 0.3733 deferral: subconeeptofpause, wait Word \'earmarks\' (2 alternatives) 0.2898 earmark: identification mark on the ear of a domestic animal 0.7102 hallma.k, trademark, earmark: a distinguishing characteristic or attribute Word \'capstone\' (1 alternatives) 1.0000 capstone, coping stone, stretcher: used at top of wall Word \'eardinality\' Not in WordNet Word \'mintage\' (1 alternatives) 62 1.0000 coinage, mintage, specie, metal money: subconcept of cash Word \'reseller\' Not in WordNet This cluster was one presented by Brown et al. as a randomly-selected class, rather than one hand-picked for its coherence.',Brown,'' +o,3312,'The weights are trained using minimum error rate training with BLEU score as the objective function ',Och,'det_weights_The nsubjpass_trained_weights auxpass_trained_are xcomp_trained_using amod__minimum nn__error nn__rate nn__training dobj_using_ prep_using_with amod_score_BLEU pobj_with_score prep_score_as det_function_the amod_function_objective pobj_as_function ' +o,3313,' present a procedure to directly optimize the global scoring function used by a phrasebased decoder on the accuracy of the translations ',Tillmann,'nsubj_present_ det_procedure_a nsubj_optimize_procedure aux_optimize_to advmod_optimize_directly xcomp_present_optimize det_function_the amod_function_global amod_function_scoring dobj_optimize_function partmod_function_used prep_used_by det_decoder_a amod_decoder_phrasebased pobj_by_decoder prep_decoder_on det_accuracy_the pobj_on_accuracy prep_accuracy_of det_translations_the pobj_of_translations ' +o,3314,'Sentiment summarization has been well studied in the past decade ',Pang,'nn_summarization_Sentiment nsubjpass_studied_summarization aux_studied_has auxpass_studied_been advmod_studied_well prep_studied_in det_decade_the amod_decade_past pobj_in_decade ' +o,3315,'The Maximum Entropy model (Berger et al., 1996; Ratnaparkhi, 1997; Abney, 1997) is a conditional model that assigns a probability to every possible parse for a given sentence s. The model consists of a set of m feature functions fj() that describe properties of parses, together with their associated weights j. The denominator is a normalization term where Y (s) is the set of parses with yield s: p(|s;) = exp( summationtextm j=1 jfj())summationtext yY (s) exp( summationtextm j=1 jfj(y))) (1) The parameters (weights) j can be estimated efficiently by maximizing the regularized conditional likelihood of a training corpus (Johnson et al., 1999; van Noord and Malouf, 2005): = argmax logL() summationtextm j=1 2j 22 (2) where L() is the likelihood of the training data.',Berger,'' +o,3316,'Our conception of the task is inspired by Ramshaw and Marcus representation of text chunking as a tagging problem The information that can be used to train the system appears in columns 1 to 8 of Table 1 ',Ramshaw,'poss_conception_Our nsubjpass_inspired_conception prep_conception_of det_task_the pobj_of_task auxpass_inspired_is prep_inspired_by pobj_by_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nsubj_8_representation prep_representation_of pobj_of_text partmod_text_chunking prep_chunking_as det_problem_a amod_problem_tagging pobj_as_problem partmod_problem_ det_information_The dobj__information nsubjpass_used_that aux_used_can auxpass_used_be rcmod_information_used aux_train_to purpcl_used_train det_system_the dobj_train_system cop_8_appears prep_8_in pobj_in_columns number_8_1 dep_8_to dep_inspired_8 prep_8_of pobj_of_Table num_Table_1 ' +o,3317,'Sentiment classification at the document level investigates ways to classify each evaluative document -LRB- eg , product review -RRB- as positive or negative ',Turney,'nn_classification_Sentiment nsubj_investigates_classification prep_classification_at det_level_the nn_level_document pobj_at_level nsubj_classify_ways aux_classify_to xcomp_investigates_classify det_document_each amod_document_evaluative dobj_classify_document appos_document_eg nn_review_product dep_eg_review advmod_positive_as amod_document_positive cc_positive_or conj_positive_negative ' +o,3318,'If the alignments are not available , they can be automatically generated ; eg , using GIZA + + ',Och,'mark_available_If det_alignments_the nsubj_available_alignments cop_available_are neg_available_not advcl_generated_available nsubjpass_generated_they aux_generated_can auxpass_generated_be advmod_generated_automatically parataxis_generated_eg dep_eg_using nn_+_GIZA nn_+_+ dobj_using_+ ' +p,3319,'of the position infer marion of words at ltlat -LRB- ; hillg pairs of sellte\\/lCeS , which turned out useful ',Brown,'prep_infer_of det_position_the pobj_of_position nsubj_infer_marion prep_marion_of pobj_of_words prep_words_at pobj_at_ltlat amod_pairs_hillg dep_ltlat_pairs prep_pairs_of dep_of_sellte\\/lCeS nsubj_turned_which rcmod_ltlat_turned prt_turned_out dep_turned_useful ' +o,3320,'These include the bootstrapping approach and the context clustering approach ',Yarowsky,'nsubj_include_These det__the amod__bootstrapping nn__approach dobj_include_ cc__and det_context_the conj__context partmod_context_clustering nn__approach dobj_clustering_ ' +p,3321,'1 Introduction The most widely applied training procedure for statistical machine translation IBM model 4 unsupervised training followed by post-processing with symmetrization heuristics yields low quality word alignments ',Brown,'num_Introduction_1 det_most_The nsubj_applied_most advmod_applied_widely dep_Introduction_applied nn_procedure_training dobj_applied_procedure prep_applied_for amod_machine_statistical pobj_for_machine nn_training_translation nn_training_IBM nn_training_model num_training_4 amod_training_ nn_training_unsupervised nsubj_followed_training dep_Introduction_followed prep_followed_by pcomp_by_post-processing prep_post-processing_with nn_heuristics_symmetrization pobj_with_heuristics num_yields_ measure_low_yields amod_alignments_low nn_alignments_quality nn_alignments_word dobj_followed_alignments ' +o,3322,'All features encountered in the training data are ranked in the DL -LRB- best evidence first -RRB- according to the following loglikelihood ratio : Log Pr -LRB- reading i jfeature k -RRB- P j6 = i Pr -LRB- reading j jfeature k -RRB- We estimated probabilities via maximum likelihood , adopting a simple smoothing method : 01 is added to both the denominator and numerator ',Yarowsky,'det_features_All nsubj_encountered_features mark_ranked_in det_data_the nn_data_training nsubjpass_ranked_data auxpass_ranked_are advcl_encountered_ranked prep_ranked_in det_DL_the pobj_in_DL amod_evidence_best dep_first_evidence dep_DL_first prep_ranked_according dep_according_to det__the amod__following amod__loglikelihood nn__ratio pobj_to_ nn_Pr_Log dep__Pr dep_Pr_reading amod_k_i nn_k_jfeature dobj_reading_k nn_Pr_P nn_Pr_j6 nn_Pr_= nn_Pr_i dep_Pr_Pr nn_k_reading nn_k_j nn_k_jfeature appos_Pr_k nsubj_estimated_We rcmod_Pr_estimated dobj_estimated_probabilities prep_probabilities_via amod_likelihood_maximum pobj_via_likelihood xcomp_estimated_adopting det__a amod__simple amod__smoothing nn__method dobj_adopting_ nsubjpass_added_01 auxpass_added_is dep_encountered_added prep_added_to preconj_denominator_both det_denominator_the pobj_to_denominator cc_denominator_and conj_denominator_numerator ' +o,3323,' measure annotation quality in terms of precision and recall against manually constructed , gold-standard f-structures for 105 randomly selected trees from section 23 of the WSJ section of Penn-II ',Cahill,'nn_quality_ nn_quality_measure nn_quality_annotation dep_quality_in pobj_in_terms prep_terms_of pobj_of_precision cc_precision_and conj_precision_recall prep_terms_against advmod_constructed_manually amod_f-structures_constructed dep_constructed_gold-standard pobj_against_f-structures prep_f-structures_for num_trees_105 advmod_trees_randomly amod_trees_selected pobj_for_trees prep_trees_from pobj_from_section num_section_23 prep_section_of det_section_the nn_section_WSJ pobj_of_section prep_section_of pobj_of_Penn-II ' +o,3324,'As other researchers pursued efficient default unification , we also propose another definition of default unification , which we call lenient default unification ',Marcus,'mark_pursued_As amod_researchers_other nsubj_pursued_researchers advcl_propose_pursued amod__efficient nn__default nn__unification dobj_pursued_ nsubj_propose_we advmod_propose_also det_definition_another dobj_propose_definition prep_definition_of nn_unification_default pobj_of_unification dobj_call_which nsubj_call_we rcmod_unification_call amod_unification_lenient nn_unification_default dobj_call_unification ' +p,3325,'1 Introduction One of the major approaches to disambiguate word senses is supervised learning , , , , , , , ',Yarowsky,'num_Introduction_1 nsubjpass_supervised_Introduction num_Introduction_One prep_Introduction_of det_approaches_the amod_approaches_major pobj_of_approaches prep_approaches_to amod_senses_disambiguate nn_senses_word pobj_to_senses auxpass_supervised_is xcomp_supervised_learning nn__ appos__ appos__ appos__ appos__ appos__ dobj_learning_ partmod__ ' +o,3326,'We examine Structural Correspondence Learning -LRB- SCL -RRB- for this task , and compare it to several variants of Self-training ',Blitzer,'nsubj_examine_We nn__Structural nn__Correspondence nn__Learning abbrev__SCL dobj_examine_ prep__for det_task_this pobj_for_task cc_examine_and conj_examine_compare dobj_compare_it prep_compare_to amod_variants_several pobj_to_variants prep_variants_of nn__Self-training pobj_of_ ' +o,3327,'The WordNet : : Similarity package implements this distance measure and was used by the authors ',Pedersen,'det_WordNet_The nn__Similarity nn__package nsubjpass_implements_ dep_WordNet_implements det_measure_this nn_measure_distance dobj_implements_measure cc_implements_and auxpass_used_was conj_implements_used prep_used_by det_authors_the pobj_by_authors ' +o,3328,'The data consists of 2,544 main clauses from the Wall Street Journal Treebank corpus ',Marcus,'det_data_The nsubj_consists_data prep_consists_of num_clauses_2,544 amod_clauses_main pobj_of_clauses prep_clauses_from det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nn_corpus_Treebank pobj_from_corpus ' +p,3329,'ROUGE has been widely used for summarization evaluation ',Lin,'nn__ROUGE nsubjpass_used_ aux_used_has auxpass_used_been advmod_used_widely prep_used_for nn_evaluation_summarization pobj_for_evaluation ' +p,3330,'ROUGE version 155 was used for evaluation2 Among others , we focus on ROUGE-1 in the discussion of the result , because ROUGE-1 has proved to have strong correlation with human annotation ',Lin,'nn__ROUGE nn__version nn__155 nsubjpass_used_ auxpass_used_was prep_used_for pobj_for_evaluation2 prep_focus_Among pobj_Among_others nsubj_focus_we dep_used_focus prep_focus_on pobj_on_ROUGE-1 prep_ROUGE-1_in det_discussion_the pobj_in_discussion prep_discussion_of det_result_the pobj_of_result mark_proved_because nsubj_proved_ROUGE-1 aux_proved_has advcl_focus_proved aux_have_to xcomp_proved_have amod_correlation_strong dobj_have_correlation prep_correlation_with amod_annotation_human pobj_with_annotation ' +p,3331,'We report results using the well-known automatic evaluation metrics Bleu ',Papineni,'nsubj_report_We nsubj_using_results dep_report_using det_Bleu_the amod_Bleu_well-known amod_Bleu_automatic nn_Bleu_evaluation nn_Bleu_metrics dobj_using_Bleu ' +o,3332,'We train IBM Model-4 using GIZA + + toolkit in two translation directions and perform different word alignment combination ',Och,'nsubj_train_We nsubj_using_IBM dep_using_Model-4 xcomp_train_using nn__GIZA amod__+ cc_+_+ conj_+_toolkit dobj_using_ prep_using_in num_directions_two nn_directions_translation pobj_in_directions cc_train_and conj_train_perform amod_combination_different nn_combination_word nn_combination_alignment dobj_perform_combination ' +o,3333,' and Ros6 -LRB- 1995 -RRB- point out the importance of taking into account the expected chance agreement among judges when computing whether or not judges agree significantly ',Carletta,'cc__and conj__Ros6 dep__1995 conj__point prep_point_out det_importance_the pobj_out_importance prep_importance_of pcomp_of_taking prep_taking_into pobj_into_account det_agreement_the amod_agreement_expected nn_agreement_chance dobj_taking_agreement prep__among pobj_among_judges advmod_computing_when advcl__computing complm_agree_whether cc_agree_or conj_agree_not nsubj_agree_judges ccomp_computing_agree advmod_agree_significantly ' +o,3334,'For example , the words corruption and abuse are similar because both of them can be subjects of verbs like arouse , become , betray , cause , continue , cost , exist , force , go on , grow , have , increase , lead to , and persist , etc , and both of them can modify nouns like accusation , act , allegation , appearance , and case , etc Many methods have been proposed to compute distributional similarity between words , eg , , , and ',Hindle,'prep_similar_For pobj_For_example det_corruption_the nn_corruption_words nsubj_similar_corruption cc_corruption_and conj_corruption_abuse cop_similar_are mark_subjects_because nsubj_subjects_both prep_both_of pobj_of_them aux_subjects_can cop_subjects_be ccomp_similar_subjects prep_subjects_of pobj_of_verbs dep_verbs_like dep_like_arouse dep_arouse_become advmod_arouse_betray conj_arouse_cause conj_arouse_continue conj_arouse_cost conj_arouse_exist conj_arouse_force conj_arouse_go prt_go_on conj_arouse_grow conj_arouse_have conj_arouse_increase conj_arouse_lead prep_lead_to cc_arouse_and conj_arouse_persist dep_like_etc cc_like_and dep_of_both advmod_like_of pobj_of_them aux_modify_can conj_like_modify dobj_modify_nouns prep_modify_like pobj_like_accusation nn_case_act conj_case_allegation conj_case_appearance cc_case_and appos_accusation_case dep_accusation_etc amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been dep_similar_proposed aux_compute_to xcomp_proposed_compute amod_similarity_distributional dobj_compute_similarity prep_similarity_between pobj_between_words conj_words_eg conj_words_ conj_words_ conj_words_ cc_words_and conj_words_ ' +o,3335,'It consists of sections 15-18 of the Wall Street Journal part of the Penn Treebank II as training data -LRB- 211727 tokens -RRB- and section 20 as test data -LRB- 47377 tokens -RRB- ',Marcus,'nsubj_consists_It prep_consists_of pobj_of_sections num_sections_15-18 prep_sections_of det_part_the nn_part_Wall nn_part_Street nn_part_Journal pobj_of_part prep_part_of det__the nn__Penn nn__Treebank nn__II pobj_of_ prep_consists_as amod_data_training pobj_as_data num_tokens_211727 appos_data_tokens cc_data_and conj_data_section num_section_20 prep_section_as nn_data_test pobj_as_data num_tokens_47377 appos_data_tokens ' +p,3336,'Among recent top performing methods are Hidden Markov Models , maximum entropy approaches , and transformation-based learning ',Ratnaparkhi,'prep_are_Among amod_methods_recent amod_methods_top amod_methods_performing pobj_Among_methods nn__Hidden nn__Markov nn__Models nsubj_are_ amod_approaches_maximum nn_approaches_entropy appos__approaches dep_approaches_ cc__and conj__transformation-based xcomp_transformation-based_learning dobj_learning_ ' +o,3337,'sentence length : The longer the sentence is , the poorer the parser performs ',McDonald,'nn_length_sentence dep_longer_The dep_is_longer det_sentence_the nsubj_is_sentence dep_performs_is dep_poorer_the dep_performs_poorer det_parser_the nsubj_performs_parser dep_length_performs ' +p,3338,'Another interesting point is the relation to maximum entropy model , which is popular in the natural language processing community ',Berger,'det_point_Another amod_point_interesting nsubj_relation_point cop_relation_is det_relation_the prep_relation_to amod__maximum amod__entropy nn__model pobj_to_ nsubj_popular_which cop_popular_is rcmod__popular prep_popular_in det_community_the amod_community_natural nn_community_language nn_community_processing pobj_in_community ' +o,3339,'1 Introduction A recent development in data-driven parsing is the use of discriminative training methods ',Collins,'num_development_1 nn_development_Introduction nn_development_A amod_development_recent nsubj_use_development prep_development_in amod_parsing_data-driven pobj_in_parsing cop_use_is det_use_the prep_use_of amod_methods_discriminative nn_methods_training pobj_of_methods ' +o,3340,'This probability is computed using IBMs Model 1 : P -LRB- Q A -RRB- = productdisplay qQ P -LRB- q A -RRB- -LRB- 3 -RRB- P -LRB- q A -RRB- = -LRB- 1 -RRB- Pml -LRB- q A -RRB- + Pml -LRB- q C -RRB- -LRB- 4 -RRB- Pml -LRB- q A -RRB- = summationdisplay aA -LRB- T -LRB- q a -RRB- Pml -LRB- a A -RRB- -RRB- -LRB- 5 -RRB- where the probability that the question term q is generated from answer A , P -LRB- q A -RRB- , is smoothed using the prior probability that the term q is generated from the entire collection of answers C , Pml -LRB- q C -RRB- ',Brown,'det_probability_This nsubjpass_computed_probability auxpass_computed_is xcomp_computed_using nn__IBMs nn__Model num__1 dobj_using_ nsubj_P_P nn_A_Q appos_P_A dep_P_= advmod_P_productdisplay nn_P_qQ parataxis_computed_P nn_A_q appos_P_A appos_P_3 dobj_P_P nn_A_q appos_P_A nn_Pml_= appos_Pml_1 dobj_P_Pml nn_A_q appos_Pml_A nn_Pml_+ dobj_P_Pml nn_C_q appos_Pml_C appos_Pml_4 dobj_P_Pml nn_A_q appos_Pml_A amod_Pml_= nn_Pml_summationdisplay nn_Pml_aA nn_Pml_T nn_Pml_q nn_Pml_a dobj_P_Pml det_A_a appos_Pml_A appos_Pml_5 advmod_smoothed_where det_probability_the nsubjpass_smoothed_probability complm_generated_that det_q_the nn_q_question nn_q_term nsubjpass_generated_q auxpass_generated_is dep_probability_generated prep_generated_from nn_A_answer pobj_from_A appos_A_P nn_A_q appos_P_A auxpass_smoothed_is advcl_P_smoothed dep_probability_using det_probability_the amod_probability_prior xcomp_smoothed_probability complm_generated_that det_q_the nn_q_term nsubjpass_generated_q auxpass_generated_is ccomp_probability_generated prep_generated_from det_collection_the amod_collection_entire pobj_from_collection prep_collection_of nn_C_answers pobj_of_C appos_C_Pml nn_C_q appos_Pml_C ' +o,3341,'Lexical collocation functions , especially those determined statistically , have recently attracted considerable attention in computational linguistics mainly , though not exclusively , for use in disambiguation ',Marcus,'amod_functions_Lexical nn_functions_collocation nsubj_attracted_functions advmod_statistically_especially det_statistically_those amod_statistically_determined appos_functions_statistically aux_attracted_have advmod_attracted_recently amod_attention_considerable dobj_attracted_attention prep_attention_in amod_linguistics_computational pobj_in_linguistics partmod_linguistics_ advmod__mainly mark_exclusively_though neg_exclusively_not parataxis__exclusively prep_attracted_for pobj_for_use prep_use_in pobj_in_disambiguation ' +o,3342,'To optimize the system towards a maximal BLEU or NIST score , we use Minimum Error Rate -LRB- MER -RRB- Training as described in ',Och,'aux_optimize_To dep_use_optimize det_system_the dobj_optimize_system prep_system_towards det_BLEU_a amod_BLEU_maximal pobj_towards_BLEU cc_BLEU_or nn_score_NIST conj_BLEU_score nsubj_use_we nn_Rate_Minimum nn_Rate_Error nsubj_Training_Rate abbrev_Rate_MER dep_use_Training advmod_described_as acomp_Training_described dep_described_in ' +p,3343,'We evaluated the generator on the Penn Treebank , which is highly reliable corpus consisting of real-world texts ',Marcus,'nsubj_evaluated_We det_generator_the dobj_evaluated_generator prep_evaluated_on det__the nn__Penn nn__Treebank pobj_on_ nsubj_corpus_which cop_corpus_is advmod_reliable_highly amod_corpus_reliable rcmod__corpus partmod_corpus_consisting prep_consisting_of amod_texts_real-world pobj_of_texts ' +o,3344,'The word-based edit distance heuristic yields pairs that are relatively clean but offer relatively minor rewrites in generation , especially when compared to the MSA model of ',Barzilay,'dep_word-based_The dep_yields_word-based nn_yields_edit nn_yields_distance amod_yields_heuristic dep_pairs_yields nsubj_clean_that cop_clean_are advmod_clean_relatively rcmod_pairs_clean cc_clean_but conj_clean_offer advmod_minor_relatively amod_rewrites_minor dobj_offer_rewrites prep_offer_in pobj_in_generation advmod_when_especially dep_offer_when prep_pairs_compared dep_compared_to det_model_the nn_model_MSA pobj_to_model prep_model_of ' +o,3345,'We based our design on the IBM models 1 and 2 (Brown et al. , 1993), but taking into account that our model must generate correct derivations in a given grammar, not any seBEGIN some END eat (a) \'some a88 animalsa89 eat a88 animalsa89 \' BEGIN some END eat are dangerous (b) \'some a88 animalsa89 are dangerous\' BEGIN some END eat are dangerous (c) \'a88 animalsa89 are dangerous\' BEGIN snakes rats people some END eat are snakes rats people dangerous (d) Expansion of a88 animalsa89 Figure 3: Using a category a86 animalsa87 for \'snakes\', \'rats\' and \'people\' in the example of Figure 1.',Brown,'' +o,3346,'This generates tens of millions features , so we prune those features that occur fewer than 10 total times , as in ',Smith,'nsubj_generates_This dep_millions_tens quantmod_millions_of num_features_millions dobj_generates_features mark_prune_so nsubj_prune_we advcl_generates_prune det_features_those dobj_prune_features nsubj_occur_that rcmod_features_occur dep_than_fewer quantmod_10_than num_times_10 amod_times_total dobj_occur_times mark_in_as advcl_occur_in ' +o,3347,'This metric tests the hypothesis that the probability of phrase is the same whether phrase has been seen or not by calculating the likelihood of the observed data under a binomial distribution using probabilities derived using each hypothesis ',Dunning,'det_metric_This nsubj_hypothesis_metric dep_hypothesis_tests det_hypothesis_the complm_distribution_that det_probability_the nsubj_distribution_probability prep_probability_of pobj_of_phrase cop_distribution_is det_distribution_the amod_distribution_same dep_seen_whether nsubjpass_seen_phrase aux_seen_has auxpass_seen_been dep_distribution_seen cc_seen_or dep_by_not conj_seen_by pcomp_by_calculating det_likelihood_the dobj_calculating_likelihood prep_likelihood_of det_data_the amod_data_observed pobj_of_data prep_data_under det_binomial_a pobj_under_binomial ccomp_hypothesis_distribution partmod_distribution_using dobj_using_probabilities partmod_probabilities_derived xcomp_derived_using det_hypothesis_each dobj_using_hypothesis ' +p,3348,'Averaging parameters is a way to reduce overfitting for perceptron training , and is applied to all our experiments ',Collins,'nsubj_way_parameters cop_way_is det_way_a ccomp_Averaging_way aux_reduce_to infmod_way_reduce xcomp_reduce_overfitting prep_overfitting_for nn__perceptron nn__training pobj_for_ cc_Averaging_and auxpass_applied_is conj_Averaging_applied prep_applied_to predet_experiments_all poss_experiments_our pobj_to_experiments ' +o,3349,'Thus , some research has been focused on deriving different word-sense groupings to overcome the finegrained distinctions of WN , , , , and ',Navigli,'advmod_focused_Thus det_research_some nsubjpass_focused_research aux_focused_has auxpass_focused_been prep_focused_on pcomp_on_deriving amod_groupings_different amod_groupings_word-sense dobj_deriving_groupings aux_overcome_to xcomp_deriving_overcome det_distinctions_the amod_distinctions_finegrained dobj_overcome_distinctions prep_distinctions_of nn__WN pobj_of_ conj__ conj__ conj__ conj__ cc__and conj__ ' +o,3350,'Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ',Collins,'advmod_made_Thus prep_made_over det_years_the amod_years_past amod_years_few pobj_over_years advmod_made_along dep_along_with pobj_with_advances prep_advances_in det_use_the pobj_in_use prep_use_of pobj_of_learning cc_use_and amod_methods_statistical conj_use_methods prep_methods_for pobj_for_acquisition prep_acquisition_of amod__full nn__parsers pobj_of_ amod_progress_significant nsubjpass_made_progress aux_made_has auxpass_made_been prep_made_on det_use_the pobj_on_use prep_use_of amod_methods_statistical nn_methods_learning pobj_of_methods aux_recognize_to xcomp_made_recognize nn_patterns_shallow nn_patterns_parsing dobj_recognize_patterns amod_phrases_syntactic tmod_recognize_phrases cc_phrases_or conj_phrases_words nsubj_participate_that dep_phrases_participate prep_participate_in det__a amod__syntactic nn__relationship pobj_in_ ' +o,3351,'We use Minimal Error Rate Training to maximize BLEU on the complete development data ',Och,'nsubj_use_We nn__Minimal nn__Error nn__Rate nn__Training nsubj_maximize_ aux_maximize_to xcomp_use_maximize dobj_maximize_BLEU prep_maximize_on det_data_the amod_data_complete nn_data_development pobj_on_data ' +o,3352,'5http : \\/ \\/ clcsokayama-uacjp\\/rsc \\/ jacabit \\/ a4a6a5 which gathers the set of co-occurrence units a7 associated with the number of times that a7 and a2 occur together a8a6a9a10a9 a5 a11 In order to identify speci c words in the lexical context and to reduce word-frequency effects , we normalize context vectors using an association score such as Mutual Information or Log-likelihood ',Dunning,'nn_a4a6a5_\\/ nn_a4a6a5_\\/ amod_a4a6a5_clcsokayama-uacjp\\/rsc amod_a4a6a5_\\/ nn_a4a6a5_jacabit nn_a4a6a5_\\/ dep_5http_a4a6a5 nsubj_gathers_which rcmod_a4a6a5_gathers det_set_the dobj_gathers_set prep_set_of amod_a7_co-occurrence nn_a7_units pobj_of_a7 partmod_a7_associated prep_associated_with det_number_the pobj_with_number prep_number_of pobj_of_times complm_occur_that nsubj_occur_a7 cc_a7_and conj_a7_a2 ccomp_associated_occur prt_occur_together amod_a11_a8a6a9a10a9 amod_a11_a5 dobj_occur_a11 mark_identify_In dep_identify_order aux_identify_to dep_normalize_identify nn_words_speci nn_words_c dobj_identify_words prep_identify_in det_context_the amod_context_lexical pobj_in_context cc_identify_and aux_reduce_to conj_identify_reduce amod_effects_word-frequency dobj_reduce_effects nsubj_normalize_we rcmod_a11_normalize nn_vectors_context dobj_normalize_vectors partmod_vectors_using det_score_an nn_score_association dobj_using_score dep_as_such prep_score_as nn_Information_Mutual pobj_as_Information amod_Information_ cc__or conj__Log-likelihood ' +o,3353,'One of the simplest models that can be seen in the context of lexical triggers is the IBM model 1 which captures lexical dependencies between source and target words ',Brown,'nsubj__One prep_One_of det_models_the amod_models_simplest pobj_of_models nsubjpass_seen_that aux_seen_can auxpass_seen_be rcmod_models_seen prep_seen_in det_context_the pobj_in_context prep_context_of amod_triggers_lexical pobj_of_triggers cop__is det__the nn__IBM nn__model num__1 nsubj_captures_which rcmod__captures amod_dependencies_lexical dobj_captures_dependencies prep_dependencies_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,3354,'The corpus was automatically derived from the Penn Treebank II corpus , by means of the script chunklinkpl that we modified to fit our purposes ',Marcus,'det_corpus_The nsubjpass_derived_corpus auxpass_derived_was advmod_derived_automatically prep_derived_from det__the nn__Penn nn__Treebank nn__II nn__corpus pobj_from_ prep_derived_by nsubj__means prep_means_of det_chunklinkpl_the nn_chunklinkpl_script pobj_of_chunklinkpl pcomp_by_ complm_modified_that nsubj_modified_we ccomp__modified aux_fit_to xcomp_modified_fit poss_purposes_our dobj_fit_purposes ' +p,3355,'2 Maximum Entropy Models Maximum entropy -LRB- ME -RRB- models , also known as log-linear and exponential learning models , provideageneralpurposemachinelearningtechnique for classification and prediction which has been successfully applied to natural language processing including part of speech tagging , named entity recognition etc Maximum entropy models can integrate features from many heterogeneous information sources for classification ',Manning,'number_Maximum_2 num_models_Maximum nn_models_Entropy nn_models_Models nn_models_Maximum nn_models_entropy abbrev_models_ME nsubj_named_ advmod_known_also partmod__known prep_known_as amod_models_log-linear cc_log-linear_and conj_log-linear_exponential nn_models_learning pobj_as_models appos_models_provideageneralpurposemachinelearningtechnique prep_provideageneralpurposemachinelearningtechnique_for pobj_for_classification cc_classification_and conj_classification_prediction nsubjpass_applied_which aux_applied_has auxpass_applied_been advmod_applied_successfully rcmod_provideageneralpurposemachinelearningtechnique_applied prep_applied_to amod_processing_natural nn_processing_language pobj_to_processing prep_processing_including pobj_including_part prep_part_of nn_tagging_speech pobj_of_tagging dep_models_named nn_recognition_entity iobj_named_recognition nn_entropy_etc nn_entropy_Maximum dobj_named_entropy nsubj_integrate_models aux_integrate_can rcmod_entropy_integrate dobj_integrate_features prep_integrate_from amod_sources_many amod_sources_heterogeneous nn_sources_information pobj_from_sources prep_sources_for pobj_for_classification ' +o,3356,'There has been some previous work on accuracy-driven training techniques for SMT , such as MERT and the Simplex Armijo Downhill method , which tune the parameters in a linear combination of various phrase scores according to a held-out tuning set ',Och,'expl_been_There aux_been_has det_work_some amod_work_previous dobj_been_work prep_work_on amod_techniques_accuracy-driven nn_techniques_training pobj_on_techniques prep_techniques_for pobj_for_SMT dep_as_such prep_work_as nn__MERT pobj_as_ cc__and det__the nn__Simplex nn__Armijo nn__Downhill nn__method conj__ nsubj_tune_which rcmod_work_tune det_parameters_the dobj_tune_parameters prep_parameters_in det_combination_a amod_combination_linear pobj_in_combination prep_combination_of amod_scores_various nn_scores_phrase pobj_of_scores prep_tune_according dep_according_to det_set_a amod_set_held-out nn_set_tuning pobj_to_set ' +o,3357,' , , and Karol & Edelman -LRB- 1996 -RRB- where strong reliance on statistical techniques for the calculation of word and context similarity commands large source corpora ',Yarowsky,'nn__ cc__and conj__Karol cc_Karol_& conj_Karol_Edelman appos_Karol_1996 advmod_commands_where amod_reliance_strong nsubj_commands_reliance prep_reliance_on amod_techniques_statistical pobj_on_techniques prep_techniques_for det_calculation_the pobj_for_calculation prep_calculation_of nn_similarity_word cc_word_and conj_word_context pobj_of_similarity rcmod_Karol_commands amod_corpora_large nn_corpora_source dobj_commands_corpora ' +o,3358,'5 Synchronous DIG 51 Definition introduced synchronous binary trees and introduced synchronous tree adjoining grammars , both of which view the translation process as a synchronous derivation process of parallel trees ',Wu,'num__5 nn__Synchronous nn__DIG num__51 nn__Definition nsubj_introduced_ amod_trees_synchronous nn_trees_binary nsubj_introduced_trees cc_trees_and conj_trees_ ccomp_introduced_introduced amod_tree_synchronous dobj_introduced_tree amod_grammars_adjoining tmod_introduced_grammars dep_of_both nsubj_view_of pobj_of_which dep_grammars_view det_process_the nn_process_translation dobj_view_process prep_view_as det_process_a amod_process_synchronous nn_process_derivation pobj_as_process prep_process_of amod_trees_parallel pobj_of_trees ' +p,3359,'1 Introduction State-of-the-art Statistical Machine Translation -LRB- SMT -RRB- systems usually adopt a two-pass search strategy as shown in Figure 1 ',Och,'num_Translation_1 nn_Translation_Introduction nn_Translation_State-of-the-art nn_Translation_Statistical nn_Translation_Machine abbrev_Translation_SMT nsubj_adopt_systems advmod_adopt_usually dep_Translation_adopt det_strategy_a amod_strategy_two-pass nn_strategy_search dobj_adopt_strategy dep_as_ advmod_shown_as partmod_strategy_shown prep_shown_in pobj_in_Figure num_Figure_1 ' +o,3360,'This analysis depends on the SPECIALIST Lexicon and the Xerox part-of-speech tagger and provides simple noun phrases that are mapped to concepts in the UMLS Metathesaurus using MetaMap ',Cutting,'det_analysis_This nsubj_depends_analysis prep_depends_on det_Lexicon_the nn_Lexicon_SPECIALIST pobj_on_Lexicon cc_Lexicon_and det__the nn__Xerox amod__part-of-speech nn__tagger conj_Lexicon_ cc_depends_and conj_depends_provides amod_phrases_simple nn_phrases_noun dobj_provides_phrases nsubjpass_mapped_that auxpass_mapped_are rcmod_phrases_mapped prep_mapped_to pobj_to_concepts prep_concepts_in det_Metathesaurus_the nn_Metathesaurus_UMLS pobj_in_Metathesaurus xcomp_mapped_using nn__MetaMap dobj_using_ ' +o,3361,' presented results suggesting that the additional parameters required to ensure that a model is not deficient result in inferior performance , but we plan to study whether this is the case for our generative model in future work ',Och,'nsubj_presented_ dobj_presented_results xcomp_presented_suggesting complm_required_that det_parameters_the amod_parameters_additional nsubj_required_parameters ccomp_suggesting_required aux_ensure_to xcomp_required_ensure complm_result_that det_model_a nsubj_result_model cop_result_is neg_result_not amod_result_deficient ccomp_ensure_result prep_result_in amod_performance_inferior pobj_in_performance cc_result_but nsubj_plan_we conj_result_plan aux_study_to xcomp_plan_study complm_case_whether nsubj_case_this cop_case_is det_case_the ccomp_study_case prep_case_for poss_model_our amod_model_generative pobj_for_model prep_model_in amod_work_future pobj_in_work ' +o,3362,'High quality word alignments can yield more accurate phrase-pairs which improve quality of a phrase-based SMT system ',Fraser,'nn_alignments_High nn_alignments_quality nn_alignments_word nsubj_yield_alignments aux_yield_can advmod_accurate_more amod_phrase-pairs_accurate dobj_yield_phrase-pairs nsubj_improve_which rcmod_phrase-pairs_improve dobj_improve_quality prep_quality_of det_system_a amod_system_phrase-based nn_system_SMT pobj_of_system ' +o,3363,'The last row shows the results for the feature augmentation algorithm ','Daume III','det_row_The amod_row_last nsubj_shows_row det_results_the dobj_shows_results prep_results_for det_algorithm_the nn_algorithm_feature nn_algorithm_augmentation pobj_for_algorithm ' +p,3364,'51 The statistical parser The parsing model is the one proposed in Merlo and Musillo , which extends the syntactic parser of Henderson and with annotations which identify semantic role labels , and has competitive performance ',Titov,'det_parser_The amod_parser_statistical nsubj_one_parser det_model_The amod_model_parsing dep_parser_model cop_one_is det_one_the dep_51_one dep_one_proposed prep_proposed_in nn__Merlo cc_Merlo_and conj_Merlo_Musillo pobj_in_ nsubj_extends_which rcmod__extends det_parser_the amod_parser_syntactic dobj_extends_parser prep_parser_of nn__Henderson pobj_of_ cc_proposed_and conj_proposed_ prep__with pobj_with_annotations nsubj_identify_which rcmod_annotations_identify amod_labels_semantic nn_labels_role dobj_identify_labels cc_identify_and conj_identify_has amod_performance_competitive dobj_has_performance ' +o,3365,'Another WSD approach incorporating context-dependent phrasal translation lexicons is given in and has been evaluated on several translation tasks ',Carpuat,'det_approach_Another nn_approach_WSD nsubjpass_given_approach partmod_approach_incorporating amod_lexicons_context-dependent amod_lexicons_phrasal nn_lexicons_translation dobj_incorporating_lexicons auxpass_given_is prep_given_in pobj_in_ cc_given_and aux_evaluated_has auxpass_evaluated_been conj_given_evaluated prep_evaluated_on amod_tasks_several nn_tasks_translation pobj_on_tasks ' +o,3366,'In our research , 23 scores , namely BLEU with maximum n-gram lengths of 1 , 2 , 3 , and 4 , NIST with maximum n-gram lengths of 1 , 2 , 3 , 4 , and 5 , GTM with exponents of 10 , 20 , and 30 , METEOR -LRB- exact -RRB- , WER , PER , and ROUGE with n-gram lengths of 1 , 2 , 3 , and 4 and 4 variants -LRB- LCS , S , SU , W-12 -RRB- , were used to calculate each similarity S i Therefore , the value of m in Eq ',Papineni,'prep_used_In poss_research_our pobj_In_research num_scores_23 nsubjpass_used_scores advmod_scores_namely nn__BLEU dep_scores_ prep__with amod_lengths_maximum amod_lengths_n-gram pobj_with_lengths prep_lengths_of pobj_of_1 conj_scores_2 conj_2_3 cc_2_and conj_2_4 nn__NIST conj_scores_ prep__with amod_lengths_maximum amod_lengths_n-gram pobj_with_lengths prep_lengths_of pobj_of_1 conj_scores_2 conj_scores_3 conj_3_4 cc_3_and conj_3_5 nn__GTM conj_scores_ prep__with pobj_with_exponents prep_exponents_of pobj_of_10 conj_scores_20 cc_scores_and conj_scores_30 nn__METEOR dep__exact conj_scores_ nn__WER conj__ nn__PER appos__ cc__and nn__ROUGE conj__ prep__with amod_lengths_n-gram pobj_with_lengths prep_lengths_of pobj_of_1 conj_scores_2 conj_scores_3 cc_3_and num_variants_4 cc_4_and conj_4_4 conj_3_variants appos_variants_LCS nn_SU_S dep_LCS_SU dep_LCS_W-12 auxpass_used_were aux_calculate_to xcomp_used_calculate det_similarity_each dobj_calculate_similarity nsubj_i_S dep_used_i advmod_i_Therefore det_value_the dep_i_value prep_value_of pobj_of_m prep_m_in pobj_in_Eq ' +n,3367,'? ? word class : measures polarity using only adjectives , however in our approach we consider the noun , the verb , the adverb and the adjective content words ',Turney,'nn_class_word amod_measures_ dep_class_measures advmod_using_polarity partmod_measures_using amod_adjectives_only dobj_using_adjectives dep_in_however prep_class_in poss_approach_our pobj_in_approach nsubj_consider_we rcmod_approach_consider det_noun_the iobj_consider_noun det_verb_the conj_noun_verb det_adverb_the conj_noun_adverb cc_noun_and det_adjective_the conj_noun_adjective nn_words_content dobj_consider_words ' +o,3368,'It has been shown that human knowledge , in the form of a small amount of manually annotated parallel data to be used to seed or guide model training , can significantly improve word alignment F-measure and translation performance ',Fraser,'nsubjpass_shown_It aux_shown_has auxpass_shown_been complm_improve_that amod_knowledge_human nsubj_improve_knowledge prep_knowledge_in det_form_the pobj_in_form prep_form_of det_amount_a amod_amount_small pobj_of_amount prep_amount_of dep_of_manually partmod_form_annotated amod_data_parallel dobj_annotated_data aux_used_to auxpass_used_be xcomp_annotated_used prep_used_to nn_training_seed cc_seed_or conj_seed_guide nn_training_model pobj_to_training aux_improve_can advmod_improve_significantly ccomp_shown_improve nn_F-measure_word nn_F-measure_alignment dobj_improve_F-measure cc_F-measure_and nn_performance_translation conj_F-measure_performance ' +o,3369,'Intuitively, if we allow any Source words to be aligned to any Target words, the best alignment that we can come up with is the one in Figure 1.c. Sentence pair (S2, T2) offers strong evidence that b c in language S means the same thing as x in language T. On the basis of this evidence, we expect the system to also learn from sentence pair (S1, T1) that a in language S means the same thing as y in language T. Unfortunately, if one works with translation models that do not allow Target words to be aligned to more than one Source word as it is the case in the IBM models (Brown et al. , 1993) it is impossible to learn that the phrase b c in language S means the same thing as word x in language T. The IBM Model 4 (Brown et al. , 1993), for example, converges to the word alignments shown in Figure 1.b and learns the translation probabilities shown in Figure 1.a.2 Since in the IBM model one cannot link a Target word to more than a Source word, the training procedure 2To train the IBM-4 model, we used Giza (Al-Onaizan et al. , 1999).',Brown,'' +o,3370,'Our approach is data-driven : following the methodology in , we automatically convert the English PennII treebank and the Chinese Penn Treebank into f-structure banks ',Cahill,'poss_approach_Our nsubj_data-driven_approach cop_data-driven_is prep_convert_following det_methodology_the pobj_following_methodology prep_methodology_in pobj_in_ nsubj_convert_we advmod_convert_automatically parataxis_data-driven_convert det_treebank_the nn_treebank_English nn_treebank_PennII dobj_convert_treebank cc_treebank_and det__the amod__Chinese nn__Penn nn__Treebank conj_treebank_ prep_convert_into amod_banks_f-structure pobj_into_banks ' +o,3371,'In this paper we present a novel PCFG-based architecture for probabilistic generation based on wide-coverage , robust Lexical Functional Grammar -LRB- LFG -RRB- approximations automatically extracted from treebanks ',Cahill,'prep_present_In det_paper_this pobj_In_paper nsubj_present_we det_architecture_a amod_architecture_novel amod_architecture_PCFG-based nsubj_extracted_architecture prep_architecture_for amod_generation_probabilistic pobj_for_generation partmod_generation_based prep_based_on amod_approximations_wide-coverage dep_wide-coverage_robust nn_approximations_Lexical nn_approximations_Functional nn_approximations_Grammar abbrev_approximations_LFG pobj_on_approximations advmod_extracted_automatically ccomp_present_extracted prep_extracted_from pobj_from_treebanks ' +o,3372,'BLEU is a precision metric that assesses the quality of a translation in terms of the proportion of its word n-grams -LRB- n 4 has become standard -RRB- that it shares with several reference translations ',Papineni,'nn__BLEU nsubj_metric_ cop_metric_is det_metric_a nn_metric_precision nsubj_assesses_that rcmod_metric_assesses det_quality_the dobj_assesses_quality prep_quality_of det_translation_a pobj_of_translation prep_translation_in pobj_in_terms prep_terms_of det_proportion_the pobj_of_proportion prep_proportion_of poss_n-grams_its nn_n-grams_word pobj_of_n-grams nsubj_standard_n num_n_4 aux_standard_has cop_standard_become parataxis_assesses_standard complm_shares_that nsubj_shares_it ccomp_assesses_shares prep_shares_with amod_translations_several nn_translations_reference pobj_with_translations ' +o,3373,'These IBM models and more recent refinements as well as algorithms that bootstrap from these models like the HMM algorithm described in are unsupervised algorithms ',Berger,'det_models_These nn_models_IBM nsubj__models cc_models_and amod_refinements_more amod_refinements_recent conj_models_refinements dep_well_as cc__well dep_well_as conj__algorithms complm_unsupervised_that nsubjpass_unsupervised_bootstrap prep_bootstrap_from det_models_these pobj_from_models prep_models_like det_algorithm_the nn_algorithm_HMM pobj_like_algorithm partmod_algorithm_described prep_described_in pobj_in_ auxpass_unsupervised_are ccomp_algorithms_unsupervised dobj_unsupervised_algorithms ' +o,3374,'This situation is very similar to that involved in training HMM text taggers , where joint probabilities are computed that a particular word corresponds to a particular part-ofspeech , and the rest of the words in the sentence are also generated ',Cutting,'det_situation_This nsubj_similar_situation cop_similar_is advmod_similar_very prep_similar_to pobj_to_that partmod_that_involved prep_involved_in nn_taggers_training nn_taggers_HMM nn_taggers_text pobj_in_taggers advmod_computed_where amod_probabilities_joint nsubjpass_computed_probabilities auxpass_computed_are advcl_similar_computed complm_corresponds_that det_word_a amod_word_particular nsubj_corresponds_word ccomp_computed_corresponds prep_corresponds_to det_part-ofspeech_a amod_part-ofspeech_particular pobj_to_part-ofspeech cc_similar_and det_rest_the nsubjpass_generated_rest prep_rest_of det_words_the pobj_of_words prep_words_in det_sentence_the pobj_in_sentence auxpass_generated_are advmod_generated_also conj_similar_generated ' +n,3375,'While most parsing methods are currently supervised or semi-supervised , they depend on hand-annotated data which are difficult to come by and which exist only for a few languages ',McClosky,'mark_supervised_While amod_methods_most nn_methods_parsing nsubj_supervised_methods cop_supervised_are advmod_supervised_currently advcl_depend_supervised cc_supervised_or conj_supervised_semi-supervised advmod_supervised_ nsubj_depend_they prep_depend_on amod_data_hand-annotated pobj_on_data nsubj_difficult_which cop_difficult_are dep_depend_difficult aux_come_to xcomp_difficult_come prep_come_by cc_difficult_and nsubj_exist_which conj_difficult_exist advmod_exist_only prep_exist_for det_languages_a amod_languages_few pobj_for_languages ' +o,3376,'Almost all of the work in the area of automatically trained taggers has explored Markov-model based part of speech tagging ',Cutting,'advmod_based_Almost dep_Almost_all prep_all_of det_work_the pobj_of_work dep_Almost_in det_area_the pobj_in_area dep_Almost_of advmod_trained_automatically amod_taggers_trained pobj_of_taggers aux_based_has nn_Markov-model_explored nsubj_based_Markov-model dobj_based_part prep_part_of nn_tagging_speech pobj_of_tagging ' +n,3377,'To analyze our methods on IV and OOV words , we use a detailed evaluation metric than Bakeoff 2006 which includes Foov and Fiv ',Levow,'aux_analyze_To dep_use_analyze poss_methods_our dobj_analyze_methods prep_analyze_on nn_words_IV cc_IV_and conj_IV_OOV pobj_on_words nsubj_use_we det_metric_a amod_metric_detailed nn_metric_evaluation dobj_use_metric prep_metric_than nn__Bakeoff num__2006 pobj_than_ nsubj_includes_which rcmod_metric_includes dobj_includes_Foov cc_Foov_and conj_Foov_Fiv ' +o,3378,'Recently , methods for training binary classifiers to maximize the F 1 - score have been proposed for SVM and LRM ',Jansche,'advmod_classifiers_Recently nsubj_classifiers_methods prep_methods_for nn_binary_training pobj_for_binary aux_maximize_to xcomp_classifiers_maximize det_F_the dobj_maximize_F num_F_1 nsubjpass_proposed_score aux_proposed_have auxpass_proposed_been parataxis_classifiers_proposed prep_proposed_for nn__SVM pobj_for_ cc__and conj__LRM ' +o,3379,'PB , available at wwwcisupennedu\\/ace , is used along with the Penn TreeBank 2 -LRB- wwwcisupennedu \\/ treebank -RRB- ',Marcus,'nsubjpass_used_PB amod_PB_available prep_available_at pobj_at_wwwcisupennedu\\/ace auxpass_used_is prt_used_along prep_used_with det_TreeBank_the nn_TreeBank_Penn pobj_with_TreeBank num_TreeBank_2 amod_treebank_wwwcisupennedu amod_treebank_\\/ appos_TreeBank_treebank ' +o,3380,'Many traditional clustering techniques attempt to maximize the average mutual information of adjacent clusters = 21 , 2 12 2121 -RRB- -LRB- -RRB- -LRB- log -RRB- -LRB- -RRB- , -LRB- WW WP WWP WWPWWI , -LRB- 2 -RRB- where the same clusters are used for both predicted and conditional words ',Brown,'amod_techniques_Many amod_techniques_traditional nn_techniques_clustering nsubj__techniques dobj__attempt aux_information_to dep_information_maximize det_information_the amod_information_average amod_information_mutual xcomp__information prep_information_of amod_clusters_adjacent pobj_of_clusters dep_21_= rcmod_clusters_21 number_12_2 appos_clusters_12 dep_12_2121 appos_-LRB-_log nn_WWPWWI_WW nn_WWPWWI_WP nn_WWPWWI_WWP appos_clusters_WWPWWI appos_clusters_2 advmod_used_where det_clusters_the amod_clusters_same nsubjpass_used_clusters auxpass_used_are advcl_information_used prep_used_for preconj_predicted_both amod_words_predicted cc_predicted_and conj_predicted_conditional pobj_for_words ' +o,3381,'1 Introduction Over the past five years progress in machine translation , and to a lesser extent progress in natural language generation tasks such as summarization , has been driven by optimizing against n-grambased evaluation metrics such as Bleu ',Papineni,'num_Introduction_1 dep_Introduction_Over det_years_the amod_years_past num_years_five pobj_Over_years nsubjpass_driven_progress prep_progress_in nn_translation_machine pobj_in_translation cc_in_and conj_in_to det_progress_a amod_progress_lesser nn_progress_extent pobj_to_progress prep_progress_in amod_tasks_natural nn_tasks_language nn_tasks_generation pobj_in_tasks dep_as_such prep_tasks_as pobj_as_summarization aux_driven_has auxpass_driven_been rcmod_years_driven prep_driven_by pcomp_by_optimizing prep_optimizing_against amod_metrics_n-grambased nn_metrics_evaluation pobj_against_metrics dep_as_such prep_metrics_as pobj_as_Bleu ' +p,3382,'Study in collocation extraction using lexical statistics has gained some insights to the issues faced in collocation extraction ',Smadja,'nsubj_gained_Study prep_Study_in nn_extraction_collocation pobj_in_extraction partmod_extraction_using amod_statistics_lexical dobj_using_statistics aux_gained_has det_insights_some dobj_gained_insights prep_gained_to det_issues_the pobj_to_issues partmod_issues_faced prep_faced_in nn_extraction_collocation pobj_in_extraction ' +p,3383,'Effective training algorithm exists once the set of features a42 a57 a16 a1a33a8 a71a54a8 a71a100a85a68a5 a53 is selected ',Berger,'amod_algorithm_Effective nn_algorithm_training nsubj_exists_algorithm xcomp_exists_ mark_selected_once det_set_the nsubjpass_selected_set prep_set_of pobj_of_features amod_a53_a42 amod_a53_a57 amod_a53_a16 amod_a53_a1a33a8 amod_a53_a71a54a8 nn_a53_a71a100a85a68a5 dep_features_a53 auxpass_selected_is advcl__selected ' +p,3384,'There are several distance measures suitable for this purpose , such as the mutual information , the dice coefficient , the phi coefficient , the cosine measure and the confidence ',Church,'expl_are_There amod_measures_several nn_measures_distance nsubj_are_measures amod_measures_suitable prep_suitable_for det_purpose_this pobj_for_purpose dep_as_such prep_measures_as det__the amod__mutual nn__information pobj_as_ det__the nn__dice nn__coefficient conj__ det__the nn__phi nn__coefficient conj__ det__the amod__cosine nn__measure conj__ cc__and det__the nn__confidence conj__ ' +p,3385,'1 Introduction Co-occurrence statistics extracted from corpora lead to good performance on a wide range of tasks that involve the identification of the semantic relation between two words or concepts ',Turney,'num_statistics_1 nn_statistics_Introduction nn_statistics_Co-occurrence partmod_statistics_extracted prep_extracted_from nn_lead_corpora pobj_from_lead prep_extracted_to amod_performance_good pobj_to_performance prep_statistics_on det_range_a amod_range_wide pobj_on_range prep_range_of pobj_of_tasks nsubj_involve_that rcmod_tasks_involve det_identification_the dobj_involve_identification prep_identification_of det_relation_the amod_relation_semantic pobj_of_relation prep_relation_between num_words_two pobj_between_words cc_words_or conj_words_concepts ' +o,3386,'Previouswork , eg , has focusedonimprovingtheperformanceofPowells algorithm ',Moore,'nsubj_has_Previouswork nn__eg appos_Previouswork_ amod_algorithm_focusedonimprovingtheperformanceofPowells dobj_has_algorithm ' +o,3387,'1984 -RRB- , written discourse , and conversational data ',Marcus,'dep_discourse_1984 amod_discourse_written nsubj__discourse partmod_discourse_ cc_discourse_and amod_data_conversational conj_discourse_data ' +o,3388,'Itowever , Harris \' methodology implies also to simplify and transform each parse tree 2 , so as to obtain so-called ` elementary sentences \' exhibiting the main conceptual classes for the domain -LRB- Sager lIa ` or instance , Hindle needs a six million word corpus in order to extract noun similarities from predicate-argunlent structures ',Hindle,'nsubj_implies_Itowever poss_methodology_Harris appos_Itowever_methodology advmod_implies_also aux_simplify_to xcomp_implies_simplify cc_simplify_and conj_simplify_transform det_tree_each amod_tree_parse dobj_simplify_tree num_tree_2 advmod_as_so advmod_obtain_as aux_obtain_to conj_simplify_obtain amod_sentences_so-called amod_sentences_elementary dobj_obtain_sentences xcomp_obtain_exhibiting det_classes_the amod_classes_main amod_classes_conceptual dobj_exhibiting_classes prep_classes_for det_domain_the pobj_for_domain prep_exhibiting_Sager dep_Sager_lIa cc_lIa_or conj_lIa_instance nn__Hindle nsubj_needs_ dep_implies_needs det_corpus_a number_million_six num_corpus_million nn_corpus_word dobj_needs_corpus mark_extract_in dep_extract_order aux_extract_to purpcl_needs_extract nn_similarities_noun dobj_extract_similarities prep_extract_from amod_structures_predicate-argunlent pobj_from_structures ' +o,3389,'Most statistical parsing research , such as , has centered on training probabilistic context-free grammars using the Penn Treebank ',Collins,'advmod_statistical_Most amod_research_statistical amod_research_parsing nsubj_centered_research dep_as_such prep_research_as pobj_as_ aux_centered_has prep_centered_on pcomp_on_training amod_grammars_probabilistic amod_grammars_context-free dobj_training_grammars xcomp_training_using det_Treebank_the nn_Treebank_Penn dobj_using_Treebank ' +o,3390,'Our aim is not only to determine the utility of citation texts for survey creation , but also to examine the quality distinctions between this form of input and others such as abstracts and full textscomparing the results to human-generated surveys using both automatic and nugget-based pyramid evaluation ',Lin,'poss_aim_Our nsubj_is_aim neg_is_not advmod_is_only aux_determine_to ccomp_is_determine det_utility_the dobj_determine_utility prep_utility_of nn_texts_citation pobj_of_texts prep_determine_for nn_creation_survey pobj_for_creation dep_also_but cc_determine_also aux_examine_to conj_determine_examine det_distinctions_the nn_distinctions_quality dobj_examine_distinctions prep_distinctions_between det_form_this pobj_between_form prep_form_of pobj_of_input cc_input_and conj_input_others dep_as_such prep_distinctions_as pobj_as_abstracts cc_abstracts_and conj_abstracts_full xcomp_examine_textscomparing det_results_the dobj_textscomparing_results prep_textscomparing_to amod_surveys_human-generated pobj_to_surveys partmod_surveys_using det_evaluation_both amod_evaluation_automatic cc_automatic_and conj_automatic_nugget-based nn_evaluation_pyramid dobj_using_evaluation ' +o,3391,'For better probability estimation , the model was extended to work with -LRB- hidden -RRB- word classes ',Brown,'prep_extended_For amod_estimation_better nn_estimation_probability pobj_For_estimation det_model_the nsubjpass_extended_model auxpass_extended_was aux_work_to xcomp_extended_work prep_work_with dep_with_hidden nn_classes_word pobj_with_classes ' +o,3392,'First , it has been noted that in many natural language applications it is sufficient to use shallow parsing information ; information such as noun phrases -LRB- NPs -RRB- and other syntactic sequences have been found useful in many large-scale language processing applications including information extraction and text summarization ',Marcus,'advmod_noted_First nsubjpass_noted_it aux_noted_has auxpass_noted_been complm_sufficient_that prep_sufficient_in amod_applications_many amod_applications_natural nn_applications_language pobj_in_applications nsubj_sufficient_it cop_sufficient_is ccomp_noted_sufficient aux_use_to xcomp_sufficient_use nn_information_shallow amod_information_parsing dobj_use_information dep_information_information dep_as_such prep_information_as nn_phrases_noun pobj_as_phrases abbrev_phrases_NPs cc_sufficient_and amod_sequences_other amod_sequences_syntactic nsubjpass_found_sequences aux_found_have auxpass_found_been conj_sufficient_found acomp_found_useful prep_found_in amod_applications_many amod_applications_large-scale nn_applications_language nn_applications_processing pobj_in_applications prep_applications_including nn_extraction_information pobj_including_extraction cc_extraction_and nn__text nn__summarization conj_extraction_ ' +o,3393,'However , the best performing statistical approaches to lexical ambiguity resolution l ; lmmselves rely on complex infornmtion sources such as ` lemmas , inflected forms , parts of speech and arbitrary word classes If -RRB- local and distant collocations , trigram sequences , and predicate m ` gument association \' -LRB- , p 190 -RRB- or large context-windows up to 1000 neighboring words ',Yarowsky,'advmod_approaches_However det_statistical_the dep_performing_best amod_statistical_performing nsubj_approaches_statistical aux_lexical_to xcomp_approaches_lexical amod_resolution_ambiguity dobj_lexical_resolution tmod_lexical_l nsubj_rely_lmmselves parataxis_approaches_rely prep_rely_on amod_sources_complex nn_sources_infornmtion pobj_on_sources dep_as_such prep_sources_as pobj_as_lemmas amod_forms_inflected appos_lemmas_forms appos_lemmas_parts dep_If_of nn_classes_speech cc_speech_and conj_speech_arbitrary nn_classes_word pobj_of_classes rcmod_parts_If amod_collocations_local cc_local_and conj_local_distant dep_lemmas_collocations nn_sequences_trigram appos_collocations_sequences cc_lemmas_and amod_association_predicate nn_association_m nn_association_gument conj_lemmas_association appos_association_ dep__p num_p_190 cc_association_or amod_context-windows_large conj_association_context-windows advmod__up dep_up_to num_words_1000 amod_words_neighboring pobj_to_words partmod_context-windows_ ' +o,3394,'One such relational reasoning task is the problem of compound noun interpretation , which has received a great deal of attention in recent years ',Turney,'num_task_One amod_task_such amod_task_relational nn_task_reasoning cop_problem_is det_problem_the rcmod_task_problem prep_problem_of amod_interpretation_compound nn_interpretation_noun pobj_of_interpretation nsubj_received_which aux_received_has rcmod_interpretation_received det_deal_a amod_deal_great dobj_received_deal prep_deal_of pobj_of_attention prep_received_in amod_years_recent pobj_in_years ' +o,3395,'Table 1 shows the percentage of agreement in classifying words as compounds or non-compounds -LRB- Compound Classification Agreement , CCA -RRB- for each language and the Kappa score obtained from it , and the percentage of words for which also the decomposition provided was identical -LRB- Decompounding Agreement , DA -RRB- ',Carletta,'nsubj_shows_Table num_Table_1 det_percentage_the dobj_shows_percentage prep_percentage_of pobj_of_agreement prep_shows_in pcomp_in_classifying dobj_classifying_words prep_classifying_as pobj_as_compounds cc_compounds_or conj_compounds_non-compounds nn_Agreement_Compound nn_Agreement_Classification appos_compounds_Agreement dep_Agreement_CCA prep_compounds_for det_language_each pobj_for_language cc_shows_and det__the nn__Kappa nn__score nsubj_obtained_ conj_shows_obtained prep_obtained_from pobj_from_it cc_shows_and det_percentage_the nsubj_identical_percentage prep_percentage_of pobj_of_words rel_provided_for pobj_for_which advmod_provided_also det_decomposition_the nsubj_provided_decomposition rcmod_words_provided cop_identical_was conj_shows_identical nn_Agreement_Decompounding dep_identical_Agreement dep_Agreement_DA ' +o,3396,'Thus , conventional methods had to introduce some kinds of restrictions such as the limitation of the kind of chains or the length of chains to be extracted ',Smadja,'advmod_had_Thus amod_methods_conventional nsubj_had_methods aux_introduce_to xcomp_had_introduce det_kinds_some dobj_introduce_kinds prep_kinds_of pobj_of_restrictions dep_as_such prep_restrictions_as det_limitation_the pobj_as_limitation prep_limitation_of det_kind_the pobj_of_kind prep_kind_of pobj_of_chains cc_kind_or det_length_the conj_kind_length prep_length_of pobj_of_chains aux_extracted_to auxpass_extracted_be xcomp_had_extracted ' +o,3397,'CIT -RRB- ',Galley,'nn_-RRB-_CIT ' +o,3398,'23 Probabilistic models for generation with HPSG Some existing studies on probabilistic models for HPSG parsing adopted log-linear models ',Berger,'num_models_23 amod_models_Probabilistic prep_models_for pobj_for_generation dep_models_with pcomp_with_HPSG det_studies_Some amod_studies_existing dobj_HPSG_studies prep_HPSG_on amod_models_probabilistic pobj_on_models prep_HPSG_for pcomp_for_HPSG xcomp_HPSG_parsing advmod_adopted_ amod_models_adopted amod_models_log-linear dobj_parsing_models ' +o,3399,'Statistical machine translation views the translation process as a noisy-channel signal recovery process in which one tries to recover the input signal e , from the observed output signal f1 Early statistical machine translation systems used a purely word-based approach without taking into account any of the morphological or syntactic properties of the languages ',Brown,'nn_translation_machine nsubj_views_translation rcmod_Statistical_views det_process_the nn_process_translation dobj_views_process prep_views_as det_process_a amod_process_noisy-channel nn_process_signal nn_process_recovery pobj_as_process rel_tries_in pobj_in_which nsubj_tries_one rcmod_process_tries aux_recover_to xcomp_tries_recover det_signal_the nn_signal_input dobj_recover_signal dep_systems_e dep_output_from det_output_the amod_output_observed dep_signal_output dep_systems_signal num_signal_f1 advmod_systems_Early amod_systems_statistical nn_systems_machine nn_systems_translation dep_Statistical_systems partmod_systems_used det_approach_a advmod_word-based_purely amod_approach_word-based dobj_used_approach prep_used_without pcomp_without_taking prep_taking_into pobj_into_account dobj_taking_any prep_any_of det_properties_the amod_properties_morphological cc_morphological_or conj_morphological_syntactic pobj_of_properties prep_properties_of det_languages_the pobj_of_languages ' +o,3400,'The difficulty of this task is that the standard method for converting NER to a sequence tagging problem with BIOencoding , where each 1http : \\/ \\/ wwwnistgov\\/speech\\/tests \\/ ace \\/ indexhtm token is assigned a tag to indicate whether it is at the beginning -LRB- B -RRB- , inside -LRB- I -RRB- , or outside -LRB- O -RRB- of an entity , is not directly applicable when tokens belong to more than one entity ',Ramshaw,'det_difficulty_The nsubj_is_difficulty prep_difficulty_of det_task_this pobj_of_task complm_applicable_that det_method_the amod_method_standard nsubj_applicable_method prep_method_for pcomp_for_converting dobj_converting_NER prep_converting_to det_problem_a nn_problem_sequence amod_problem_tagging pobj_to_problem prep_problem_with nn__BIOencoding pobj_with_ advmod_wwwnistgov\\/speech\\/tests_where det_1http_each nsubj_wwwnistgov\\/speech\\/tests_1http nn_\\/_\\/ dep_1http_\\/ rcmod_method_wwwnistgov\\/speech\\/tests amod_token_\\/ nn_token_ace amod_token_\\/ nn_token_indexhtm nsubjpass_assigned_token auxpass_assigned_is ccomp_wwwnistgov\\/speech\\/tests_assigned det_tag_a dobj_assigned_tag aux_indicate_to infmod_tag_indicate complm_is_whether nsubj_is_it ccomp_indicate_is prep_is_at det_beginning_the pobj_at_beginning appos_tag_B conj_tag_inside dep_inside_I cc_tag_or conj_tag_outside appos_outside_O prep_outside_of det_entity_an pobj_of_entity cop_applicable_is neg_applicable_not advmod_applicable_directly ccomp_is_applicable advmod_belong_when nsubj_belong_tokens advcl_applicable_belong prep_belong_to dep_than_more quantmod_one_than num_entity_one pobj_to_entity ' +p,3401,'Maximum entropy models are a class of exponential models which require no unwarranted independence assumptions and have proven to be very successful in general for integrating information from disparate and possibly overlapping sources ',Berger,'amod__Maximum nn__entropy nn__models nsubj_class_ cop_class_are det_class_a prep_class_of amod_models_exponential pobj_of_models nsubj_require_which rcmod_class_require det_assumptions_no amod_assumptions_unwarranted nn_assumptions_independence dobj_require_assumptions cc_require_and aux_proven_have conj_require_proven aux_successful_to cop_successful_be advmod_successful_very xcomp_proven_successful prep_successful_in pobj_in_general prep_successful_for pcomp_for_integrating dobj_integrating_information prep_integrating_from pobj_from_disparate cc_disparate_and advmod_sources_possibly amod_sources_overlapping conj_disparate_sources ' +o,3402,'21418 examples of structures of the kind ` VB N1 PREP N2 \' were extracted from the Penn-TreeBank Wall Street Journal ',Marcus,'num_examples_21418 nsubjpass_extracted_examples prep_examples_of pobj_of_structures prep_structures_of det_kind_the pobj_of_kind nn_N2_VB nn_N2_N1 nn_N2_PREP dep_kind_N2 auxpass_extracted_were prep_extracted_from det_Journal_the nn_Journal_Penn-TreeBank nn_Journal_Wall nn_Journal_Street pobj_from_Journal ' +o,3403,'A tight integration of morphosyntactic information into the translation model was proposed by where lemma and morphological information are translated separately , and this information is combined on the output side to generate the translation ',Koehn,'det_integration_A amod_integration_tight nsubjpass_proposed_integration prep_integration_of amod_information_morphosyntactic pobj_of_information prep_information_into det_model_the nn_model_translation pobj_into_model auxpass_proposed_was prep_proposed_by pcomp_by_ advmod_translated_where nsubjpass_translated_lemma cc_lemma_and amod_information_morphological conj_lemma_information auxpass_translated_are advcl__translated advmod_translated_separately cc_proposed_and det_information_this nsubjpass_combined_information auxpass_combined_is conj_proposed_combined prep_combined_on det_side_the nn_side_output pobj_on_side aux_generate_to xcomp_combined_generate det_translation_the dobj_generate_translation ' +n,3404,'Surprisingly , although JESS-CM is a simpler version of the hybrid model in terms of model structure and parameter estimation procedure , JESS-CM provides F-scores of 9445 and 8803 for CoNLL00 and 03 data , respectively , which are 015 and 083 points higher than those reported in for the same configurations ',Suzuki,'advmod_provides_Surprisingly mark_version_although nsubj_version_JESS-CM cop_version_is det_version_a amod_version_simpler dep_provides_version prep_version_of det_model_the amod_model_hybrid pobj_of_model prep_model_in pobj_in_terms prep_terms_of nn_structure_model pobj_of_structure cc_structure_and nn_procedure_parameter nn_procedure_estimation conj_structure_procedure nsubj_provides_JESS-CM dobj_provides_F-scores prep_F-scores_of pobj_of_9445 cc_9445_and conj_9445_8803 prep_9445_for pobj_for_CoNLL00 cc_CoNLL00_and num_data_03 conj_CoNLL00_data advmod_9445_respectively nsubj_higher_which cop_higher_are num_points_015 cc_015_and conj_015_083 measure_higher_points rcmod_9445_higher mark_reported_than nsubj_reported_those ccomp_higher_reported prep_reported_in pobj_in_ prep_reported_for det_configurations_the amod_configurations_same pobj_for_configurations ' +p,3405,'Support Vector Machines -LRB- SVMs -RRB- and Maximum Entropy -LRB- ME -RRB- method are powerful learning methods that satisfy such requirements , and are applied successfully to other NLP tasks ',Berger,'nn__Support nn__Vector nn__Machines abbrev__SVMs nsubj_powerful_ cc__and nn_Entropy_Maximum nn__Entropy abbrev__ME nn__method conj__ cop_powerful_are dep_powerful_learning dobj_learning_methods nsubj_satisfy_that rcmod_methods_satisfy amod_requirements_such dobj_satisfy_requirements cc_learning_and auxpass_applied_are conj_learning_applied advmod_applied_successfully prep_applied_to amod__other nn__NLP nn__tasks pobj_to_ ' +o,3406,'34 Related work and issues for future research and van der Eijk -LRB- 1993 -RRB- describe term translation methods that use bilingual texts that were aligned at the sentence level ',Smadja,'num_work_34 amod_work_Related nsubj_describe_work cc_work_and conj_work_issues prep_issues_for amod__future nn__research pobj_for_ cc_work_and nn_Eijk_van nn_Eijk_der conj_work_Eijk appos_work_1993 nn_methods_term nn_methods_translation dobj_describe_methods nsubj_use_that rcmod_methods_use amod_texts_bilingual dobj_use_texts nsubjpass_aligned_that auxpass_aligned_were rcmod_texts_aligned prep_aligned_at det_level_the nn_level_sentence pobj_at_level ' +o,3407,'Note in passing that the ratio 104-108\\/997 \% compares very favourably with other systems ; cf 30\\/993 \% by POST and 104\\/976 \% or 109\\/986 \% by de Marcken ',Marcus,'prep_Note_in pobj_in_passing complm_compares_that det_ratio_the nsubj_compares_ratio num_\%_104-108\\/997 dep_ratio_\% ccomp_passing_compares advmod_favourably_very acomp_compares_favourably prep_favourably_with amod_systems_other pobj_with_systems parataxis_passing_cf num_\%_30\\/993 dobj_cf_\% prep_cf_by nn__POST pobj_by_ cc_passing_and number_\%_104\\/976 dep_\%_\% cc_\%_or conj_\%_109\\/986 conj_passing_\% prep_\%_by nn__de nn__Marcken pobj_by_ ' +o,3408,'Some statistical model to estimate the part of speech of unknown words from the case of the first letter and the prefix and suffix is proposed ',Ratnaparkhi,'det_model_Some amod_model_statistical nsubjpass_proposed_model aux_estimate_to infmod_model_estimate det_part_the dobj_estimate_part prep_part_of pobj_of_speech prep_speech_of amod_words_unknown pobj_of_words prep_estimate_from det_case_the pobj_from_case prep_case_of det_letter_the amod_letter_first pobj_of_letter cc_letter_and det_prefix_the conj_letter_prefix cc_prefix_and conj_prefix_suffix auxpass_proposed_is ' +o,3409,'This corpus of 29 million words was provided to us by Michael ',Collins,'det_corpus_This nsubjpass_provided_corpus prep_corpus_of number_million_29 num_words_million pobj_of_words auxpass_provided_was prep_provided_to pobj_to_us prep_provided_by pobj_by_Michael ' +o,3410,'Following our previous work , we extract features from a sequence representation and a parse tree representation of each relation instance ',Jiang,'prep_extract_Following poss__our amod__previous nn__work pobj_Following_ nsubj_extract_we dobj_extract_features prep_features_from det_representation_a nn_representation_sequence pobj_from_representation cc_features_and det_representation_a amod_representation_parse nn_representation_tree conj_features_representation prep_representation_of det_instance_each nn_instance_relation pobj_of_instance ' +p,3411,'42 Support Vector Machines We chose to adopt a tagging perspective for the Simple NP chunking task , in which each word is to be tagged as either B , I or O depending on wether it is in the Beginning , Inside , or Outside of the given chunk , an approach first taken by , and which has become the de-facto standard for this task ',Ramshaw,'num_Machines_42 amod_Machines_Support nn_Machines_Vector nsubj_chose_We dep_Machines_chose aux_adopt_to xcomp_chose_adopt det_perspective_a amod_perspective_tagging dobj_adopt_perspective prep_adopt_for det_task_the nn_task_Simple nn_task_NP amod_task_chunking pobj_for_task rel_is_in pobj_in_which det_word_each nsubj_is_word rcmod_task_is aux_tagged_to auxpass_tagged_be xcomp_is_tagged prep_tagged_as det_B_either pobj_as_B conj_B_I cc_B_or conj_B_O prep_tagged_depending dep_depending_on pobj_on_wether nsubj_is_it dep_Machines_is prep_is_in det_Outside_the nn_Outside_Beginning conj_Outside_Inside cc_Outside_or pobj_in_Outside prep_Outside_of det_chunk_the amod_chunk_given pobj_of_chunk det_approach_an nsubj_taken_approach advmod_taken_first dep_is_taken prep_taken_by pobj_by_ cc_taken_and nsubj_standard_which aux_standard_has cop_standard_become det_standard_the amod_standard_de-facto conj_taken_standard prep_standard_for det_task_this pobj_for_task ' +o,3412,'Second , we follow on taxonomy induction in incorporating transitive closure constraints in our probability calculations , as explained below ',Snow,'advmod_follow_Second nsubj_follow_we acomp_follow_ prep__on amod_induction_taxonomy pobj_on_induction prep_induction_in pcomp_in_incorporating amod_constraints_transitive nn_constraints_closure dobj_incorporating_constraints prep_incorporating_in poss_calculations_our nn_calculations_probability pobj_in_calculations mark_explained_as advcl_follow_explained advmod_explained_below ' +o,3413,'We then ranked the collected query pairs using loglikelihoodratio -LRB- LLR -RRB- , whichmeasures the dependence between q1 and q2 within the context of web queries ',Dunning,'nsubj_ranked_We advmod_ranked_then ccomp_whichmeasures_ranked det_pairs_the amod_pairs_collected amod_pairs_query dobj_ranked_pairs partmod_pairs_using amod__loglikelihoodratio abbrev__LLR dobj_using_ det_dependence_the dobj_whichmeasures_dependence prep_dependence_between pobj_between_q1 cc_q1_and conj_q1_q2 prep_whichmeasures_within det_context_the pobj_within_context prep_context_of pobj_of_web nn__queries nsubj_whichmeasures_ ' +o,3414,'Prominent among these properties is the semi-free Language Size LR LP Source English 40,000 874 \% 881 \% Chinese 3,484 690 \% 748 \% Czech 19,000 800 \% Table 1 : Results for the Collins model for various languages -LRB- dependency precision for Czech -RRB- wordorder , ie , German wordorder is fixed in some respects , but variable in others ',Collins,'nsubj_English_Prominent prep_Prominent_among det_properties_these pobj_among_properties cop_English_is det_English_the amod_English_semi-free nn_English_Language nn_English_Size nn_English_LR nn_English_LP nn_English_Source number_874_40,000 dep_\%_874 dep__\% num_\%_881 measure__\% amod__ amod__Chinese number_690_3,484 num_\%_690 measure__\% number_\%_748 num_\%_\% amod__ amod__Czech num__19,000 dep_\%_800 amod__\% nsubjpass_fixed_ dep__Table num_Table_1 dep__Results prep_Results_for det_model_the nn_model_Collins nn_model_ pobj_for_model prep_model_for amod_languages_various pobj_for_languages amod_precision_dependency dep__precision prep_precision_for pobj_for_Czech nn_wordorder_wordorder dep_wordorder_ie amod_wordorder_German dep__wordorder auxpass_fixed_is rcmod_English_fixed prep_fixed_in det_respects_some pobj_in_respects cc_English_but conj_English_variable prep_variable_in pobj_in_others ' +p,3415,'In order to overcome this , some unsupervised learning methods and minimally-supervised methods , eg , , have been proposed ',Yarowsky,'mark_overcome_In dep_overcome_order aux_overcome_to dep_proposed_overcome dobj_overcome_this det_methods_some amod_methods_unsupervised nn_methods_learning conj_this_methods cc_this_and amod_methods_minimally-supervised conj_this_methods nsubjpass_proposed_eg partmod_eg_ aux_proposed_have auxpass_proposed_been ' +o,3416,'An example set of tags can be found in the Penn Treebank project ',Marcus,'det_set_An nn_set_example nsubjpass_found_set prep_set_of pobj_of_tags aux_found_can auxpass_found_be prep_found_in det_project_the nn_project_Penn nn_project_Treebank pobj_in_project ' +o,3417,'There are many research directions, e.g., sentiment classification (classifying an opinion document as positive or negative) (e.g., Pang, Lee and Vaithyanathan, 2002; Turney, 2002), subjectivity classification (determining whether a sentence is subjective or objective, and its associated opinion) (Wiebe and Wilson, 2002; Yu and Hatzivassiloglou, 2003; Wilson et al, 2004; Kim and Hovy, 2004; Riloff and Wiebe, 2005), feature/topic-based sentiment analysis (assigning positive or negative sentiments to topics or product features) (Hu and Liu 2004; Popescu and Etzioni, 2005; Carenini et al., 2005; Ku et al., 2006; Kobayashi, Inui and Matsumoto, 2007; Titov and McDonald.',Pang,'' +o,3418,'It has been shown repeatedly -- eg , Briscoe and Carroll , Charniak , Collins , Inui et al ',Marcus,'nsubjpass_shown_It aux_shown_has auxpass_shown_been advmod_shown_repeatedly dep_shown_eg nn__Briscoe cc_Briscoe_and conj_Briscoe_Carroll appos_eg_ nn__Charniak appos_eg_ nn__Collins appos_eg_ conj__Inui cc__et conj__al ' +o,3419,'For example , work which failed to detect improvements in translation quality with the integration of word sense disambiguation , or work which attempted to integrate syntactic information but which failed to improve Bleu may deserve a second look with a more targeted manual evaluation ',Och,'prep_deserve_For pobj_For_example nsubj_deserve_work nsubj_failed_which rcmod_work_failed aux_detect_to xcomp_failed_detect dobj_detect_improvements prep_improvements_in nn_quality_translation pobj_in_quality prep_detect_with det_integration_the pobj_with_integration prep_integration_of nn__word nn__sense nn__disambiguation pobj_of_ cc_detect_or conj_detect_work nsubj_attempted_which dep_work_attempted aux_integrate_to xcomp_attempted_integrate amod_information_syntactic dobj_integrate_information cc_attempted_but nsubj_failed_which conj_attempted_failed aux_improve_to xcomp_failed_improve nn__Bleu dobj_improve_ aux_deserve_may det_look_a amod_look_second dobj_deserve_look prep_deserve_with det_evaluation_a advmod_targeted_more amod_evaluation_targeted amod_evaluation_manual pobj_with_evaluation ' +o,3420,'As the baseline standard , we took the ending-guessing rule set supplied with the Xerox tagger ',Cutting,'prep_took_As det_standard_the nn_standard_baseline pobj_As_standard nsubj_took_we det_set_the amod_set_ending-guessing nn_set_rule dobj_took_set partmod_set_supplied prep_supplied_with det_tagger_the nn_tagger_Xerox pobj_with_tagger ' +p,3421,'For the current work , the Log-likelihood coefficient has been employed , as it is reported to perform well among other scoring methods ',Dunning,'prep_employed_For det_work_the amod_work_current pobj_For_work det_coefficient_the amod_coefficient_Log-likelihood nsubjpass_employed_coefficient aux_employed_has auxpass_employed_been advmod_employed_ mark_reported_as nsubjpass_reported_it auxpass_reported_is advcl_employed_reported aux_perform_to xcomp_reported_perform advmod_perform_well prep_perform_among amod__other amod__scoring nn__methods pobj_among_ ' +o,3422,'The current release of PDTB20 contains the annotations of 1,808 Wall Street Journal articles -LRB- ~ 1 million words -RRB- from the Penn TreeBank II distribution and a total of 40,600 discourse connective tokens ',Marcus,'det_release_The amod_release_current nsubj_contains_release prep_release_of pobj_of_PDTB20 det_annotations_the dobj_contains_annotations prep_annotations_of num_articles_1,808 nn_articles_Wall nn_articles_Street nn_articles_Journal pobj_of_articles parataxis_contains_~ number_million_1 num_words_million dobj_~_words prep_contains_from det_distribution_the nn_distribution_Penn nn_distribution_TreeBank nn_distribution_ nn_distribution_II pobj_from_distribution cc_distribution_and det_total_a conj_distribution_total prep_total_of num__40,600 nn__discourse nn__connective nn__tokens pobj_of_ ' +o,3423,'This allows us to compute the conditional probability as follows : ag ~ -LRB- h f -RRB- P -LRB- \\/ Ih -RRB- 1L \' -LRB- 2 -RRB- Z -LRB- h -RRB- ct i ',Berger,'dep_This_allows nsubj_compute_us aux_compute_to xcomp_allows_compute det_probability_the amod_probability_conditional dobj_compute_probability mark_follows_as advcl_compute_follows acomp_follows_ nn_P_ag nn_P_~ nn_f_h appos_P_f dep_This_P nn_Ih_\\/ appos_P_Ih dep_P_1L appos_P_2 nn_i_Z appos_i_h nn_i_ct dep_P_i ' +o,3424,'We believe that other kinds of translationunit such as n-gram , factoredphrasaltranslation , or treelet can be used in this method ',Koehn,'nsubj_believe_We complm_used_that amod_kinds_other nsubjpass_used_kinds prep_kinds_of pobj_of_translationunit dep_as_such prep_translationunit_as amod__n-gram pobj_as_ nn__factoredphrasaltranslation conj__ cc__or amod__treelet conj__ aux_used_can auxpass_used_be ccomp_believe_used prep_used_in det_method_this pobj_in_method ' +o,3425,'Taken together with cube pruning , k-best tree extraction , and cube growing , these results provide evidence that lazy techniques may penetrate deeper yet into MT decoding and other NLP search problems ',Huang,'ccomp_provide_Taken advmod_Taken_together prep_Taken_with nn__cube nn__pruning pobj_with_ amod__k-best nn__tree nn__extraction appos__ cc__and nn__cube amod__growing conj__ det_results_these nsubj_provide_results dobj_provide_evidence complm_penetrate_that amod_techniques_lazy nsubj_penetrate_techniques aux_penetrate_may ccomp_provide_penetrate advmod_penetrate_deeper advmod_penetrate_yet prep_penetrate_into nn_decoding_MT pobj_into_decoding cc_decoding_and amod_problems_other nn_problems_NLP nn_problems_search conj_decoding_problems ' +o,3426,'A class of training criteria that provides a tighter connection between the decision rule and the final error metric is known as Minimum Error Rate Training -LRB- MERT -RRB- and has been suggested for SMT in ',Och,'det_class_A nsubjpass_known_class prep_class_of nn_criteria_training pobj_of_criteria nsubj_provides_that rcmod_criteria_provides det_connection_a amod_connection_tighter dobj_provides_connection prep_connection_between det_rule_the nn_rule_decision pobj_between_rule cc_rule_and det_metric_the amod_metric_final nn_metric_error conj_rule_metric auxpass_known_is prep_known_as nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_as_Training abbrev_Training_MERT cc_known_and aux_suggested_has auxpass_suggested_been conj_known_suggested prep_suggested_for pobj_for_SMT prep_suggested_in ' +o,3427,'4 Experiments and Results We use the standard corpus for this task , the Penn Treebank ',Marcus,'num_Experiments_4 cc_Experiments_and conj_Experiments_Results nsubj_use_We dep_Experiments_use det_corpus_the amod_corpus_standard dobj_use_corpus prep_corpus_for det_task_this pobj_for_task det_Treebank_the nn_Treebank_Penn appos_task_Treebank ' +p,3428,'This averaging effect has been shown to help overfitting ',Collins,'det_effect_This nn_effect_averaging nsubjpass_shown_effect aux_shown_has auxpass_shown_been aux_help_to xcomp_shown_help dobj_help_overfitting ' +o,3429,'Intuitively , if we are able to find good correspondences among features , then the augmented labeled source domain data should transfer better to a target domain -LRB- where no labeled data is available -RRB- ',Blitzer,'advmod_labeled_Intuitively mark_able_if nsubj_able_we cop_able_are dep_labeled_able aux_find_to xcomp_able_find amod_correspondences_good dobj_find_correspondences prep_find_among pobj_among_features advmod_augmented_then det_augmented_the nsubj_labeled_augmented nn_data_source nn_data_domain nsubj_transfer_data aux_transfer_should ccomp_labeled_transfer advmod_transfer_better prep_transfer_to det_domain_a nn_domain_target pobj_to_domain advmod_available_where det_data_no amod_data_labeled nsubj_available_data cop_available_is dep_domain_available ' +o,3430,'The methodology used (Brown et al. , 1993) is based on the definition of a function Pr(tI1|sJ1) that returns the probability that tI1 is a 835 source Transferir documentos explorados a otro directorio interaction-0 Move documents scanned to other directory interaction-1 Move s canned documents to other directory interaction-2 Move scanned documents to a nother directory interaction-3 Move scanned documents to another f older acceptance Move scanned documents to another folder Figure 1: Example of CAT system interactions to translate the Spanish source sentence into English.',Brown,'' +o,3431,'For tuning of decoder parameters , we conducted minimum error training with respect to the BLEU score using 916 development sentence pairs ',Och,'prep_conducted_For pcomp_For_tuning prep_tuning_of nn_parameters_decoder pobj_of_parameters nsubj_conducted_we amod__minimum nn__error nn__training dobj_conducted_ prep_conducted_with pobj_with_respect prep_conducted_to det_score_the amod_score_BLEU pobj_to_score partmod_score_using num_pairs_916 nn_pairs_development nn_pairs_sentence dobj_using_pairs ' +o,3432,'To this end , the translational correspondence is described within a translation rule , ie , -LRB- or a synchronous production -RRB- , rather than a translational phrase pair ; and the training data will be derivation forests , instead of the phrase-aligned bilingual corpus ',Galley,'dep_ie_To det_end_this pobj_To_end det_correspondence_the amod_correspondence_translational nsubjpass_described_correspondence auxpass_described_is dep_ie_described prep_described_within det_rule_a nn_rule_translation pobj_within_rule partmod_ie_ dep_-LRB-_or dep_-LRB-_a dep_a_synchronous dep_a_production dep_than_rather dep_ie_than det_pair_a amod_pair_translational nn_pair_phrase pobj_than_pair cc_ie_and det_data_the nn_data_training nsubj_forests_data aux_forests_will cop_forests_be nn_forests_derivation conj_ie_forests cc_forests_instead dep_instead_of det_corpus_the amod_corpus_phrase-aligned amod_corpus_bilingual conj_forests_corpus ' +p,3433,'Finally , to estimate the parameters i of the weighted linear model , we adopt the popular minimum error rate training procedure which directly optimizes translation quality as measured by the BLEU metric ',Och,'advmod_adopt_Finally aux_estimate_to dep_adopt_estimate dobj_estimate_the nn_the_parameters dep_the_i prep_the_of det_model_the amod_model_weighted amod_model_linear pobj_of_model nsubj_adopt_we det_rate_the amod_rate_popular amod_rate_minimum nn_rate_error dobj_adopt_rate nn__training nn__procedure nsubj_measured_ nsubj_optimizes_which advmod_optimizes_directly rcmod__optimizes nn_quality_translation dobj_optimizes_quality prep_optimizes_as dep_adopt_measured prep_measured_by det_metric_the amod_metric_BLEU pobj_by_metric ' +o,3434,'These methods are based on IBM statistical translation Model 2 , but take advantage of certain characteristics of the segments of text that can typically be extracted from translation memories ',Brown,'det_methods_These nsubjpass_based_methods auxpass_based_are prep_based_on nn_translation_IBM amod_translation_statistical pobj_on_translation nsubj__Model num_Model_2 dep_based_ cc__but conj__take dobj_take_advantage prep_advantage_of amod_characteristics_certain pobj_of_characteristics prep_characteristics_of det_segments_the pobj_of_segments prep_segments_of pobj_of_text nsubjpass_extracted_that aux_extracted_can advmod_extracted_typically auxpass_extracted_be rcmod_advantage_extracted prep_extracted_from nn_memories_translation pobj_from_memories ' +o,3435,'While reranking has benefited many tagging and parsing tasks including semantic role labeling , it has not yet been applied to semantic parsing ',Collins,'mark_benefited_While nsubj_benefited_reranking aux_benefited_has advcl_applied_benefited dobj_benefited_many amod_many_tagging cc_many_and amod_tasks_parsing conj_many_tasks partmod_many_ prep__including amod__semantic nn__role nn__labeling pobj_including_ nsubjpass_applied_it aux_applied_has neg_applied_not advmod_applied_yet auxpass_applied_been prep_applied_to amod_parsing_semantic pobj_to_parsing ' +o,3436,'1 Introduction The reranking approach is widely used in parsing as well as in other structured classification problems ',Koo,'num_approach_1 nn_approach_Introduction nn_approach_The amod_approach_reranking nsubjpass_used_approach auxpass_used_is advmod_used_widely prep_used_in amod__parsing pobj_in_ dep_well_as cc_in_well dep_well_as conj_in_in amod_problems_other amod_problems_structured nn_problems_classification pobj_in_problems ' +n,3437,'We presented some theoretical arguments for not limiting extraction to minimal rules , validated them on concrete examples , and presented experiments showing that contextually richer rules provide a 363 BLEU point increase over the minimal rules of ',Galley,'nsubj_presented_We det_arguments_some amod_arguments_theoretical dobj_presented_arguments prep_presented_for neg_limiting_not pcomp_for_limiting dobj_limiting_extraction prep_limiting_to amod_rules_minimal pobj_to_rules conj_presented_validated dobj_validated_them prep_validated_on amod_examples_concrete pobj_on_examples cc_presented_and conj_presented_presented dobj_presented_experiments xcomp_presented_showing complm_provide_that advmod_richer_contextually amod_rules_richer nsubj_provide_rules ccomp_showing_provide det_increase_a num_increase_363 nn_increase_BLEU nn_increase_point dobj_provide_increase prep_provide_over det_rules_the amod_rules_minimal pobj_over_rules prep_rules_of ' +o,3438,'Unsupervised systems are based on generative models trained with the EM algorithm ',Och,'amod__Unsupervised nn__systems nsubjpass_based_ auxpass_based_are prep_based_on amod_models_generative pobj_on_models partmod_models_trained prep_trained_with det_algorithm_the nn_algorithm_EM pobj_with_algorithm ' +o,3439,'In our Machine \% ` anslation system , transfer rules are generated automatically from parsed parallel text along the lines of ',Brown,'prep_generated_In poss_system_our num_\%_Machine amod_system_\% nn_system_anslation pobj_In_system nn_rules_transfer nsubjpass_generated_rules auxpass_generated_are advmod_generated_automatically prep_generated_from amod_text_parsed amod_text_parallel pobj_from_text prep_text_along det_lines_the pobj_along_lines prep_lines_of pobj_of_ ' +o,3440,'These alignment models stem from the source-channel approach to statistical machine translation ',Brown,'det_models_These nn_models_alignment nsubj_stem_models prep_stem_from det_approach_the amod_approach_source-channel pobj_from_approach prep_stem_to amod_translation_statistical nn_translation_machine pobj_to_translation ' +o,3441,'1 Motivation Question Answering has emerged as a key area in natural language processing -LRB- NLP -RRB- to apply question parsing , information extraction , summarization , and language generation techniques ',Fleischman,'num_Answering_1 nn_Answering_Motivation nn_Answering_Question nsubj_emerged_Answering aux_emerged_has prep_emerged_as det_area_a amod_area_key pobj_as_area prep_area_in amod_processing_natural nn_processing_language pobj_in_processing abbrev_processing_NLP aux_apply_to xcomp_emerged_apply nn_parsing_question dobj_apply_parsing nn_extraction_information conj_parsing_extraction conj_parsing_summarization cc_parsing_and nn_techniques_language nn_techniques_generation conj_parsing_techniques ' +o,3442,'We use the beam search technique of to search the space of all hypotheses ',Ratnaparkhi,'nsubj_use_We det_technique_the nn_technique_beam nn_technique_search nsubj_search_technique prep_technique_of pobj_of_ aux_search_to xcomp_use_search det_space_the dobj_search_space prep_space_of det_hypotheses_all pobj_of_hypotheses ' +o,3443,'33 Tree Transducer Grammars Syntactic machine translation uses tree transducer grammars to translate sentences ',Galley,'num__33 nn__Tree nn__Transducer nn__Grammars nn__Syntactic nn__machine nn__translation nsubj_uses_ nn_grammars_tree nn_grammars_transducer nsubj_translate_grammars aux_translate_to xcomp_uses_translate dobj_translate_sentences ' +o,3444,'Previous work has shown that data collected through the Mechanical Turk service is reliable and comparable in quality with trusted sources ',Snow,'amod_work_Previous nsubj_shown_work aux_shown_has complm_reliable_that nsubj_reliable_data partmod_data_collected prep_collected_through det_service_the nn_service_Mechanical nn_service_Turk pobj_through_service cop_reliable_is ccomp_shown_reliable cc_reliable_and conj_reliable_comparable prep_reliable_in pobj_in_quality prep_quality_with amod_sources_trusted pobj_with_sources ' +o,3445,'41 Data Preparation NP chunking results have been reported on two slightly different data sets : the original RM data set of , and the modi ed CoNLL-2000 version of Tjong Kim Sang and Buchholz -LRB- 2000 -RRB- ',Ramshaw,'num_results_41 nn_results_Data nn_results_Preparation nn_results_NP nn_results_chunking nsubjpass_reported_results aux_reported_have auxpass_reported_been prep_reported_on num_sets_two advmod_different_slightly amod_sets_different nn_sets_data pobj_on_sets det_data_the amod_data_original nn_data_RM nsubj_set_data parataxis_reported_set prep_set_of pobj_of_ cc_reported_and det_modi_the nsubj_ed_modi conj_reported_ed amod_version_CoNLL-2000 dobj_ed_version prep_version_of nn_Sang_Tjong nn_Sang_Kim pobj_of_Sang cc_Sang_and conj_Sang_Buchholz appos_Sang_2000 ' +o,3446,'To determine the tree head-word we used a set of rules similar to that described by and also used by , which we modified in the following way : The head of a prepositional phrase -LRB- PP-IN NP -RRB- was substituted by a function the name of which corresponds to the preposition , and its sole argument corresponds to the head of the noun phrase NP ',Berger,'aux_determine_To csubj_head-word_determine det_tree_the dobj_determine_tree nsubj_used_we ccomp_head-word_used det_set_a dobj_used_set prep_set_of pobj_of_rules amod_rules_similar prep_similar_to pobj_to_that dep_that_described prep_described_by nn__ pobj_by_ cc_described_and advmod_used_also conj_described_used prep_used_by pobj_by_ rel_modified_which nsubj_modified_we rcmod__modified prep_modified_in det_way_the amod_way_following pobj_in_way det_head_The nsubjpass_substituted_head prep_head_of det_phrase_a amod_phrase_prepositional pobj_of_phrase nn_NP_PP-IN abbrev_phrase_NP auxpass_substituted_was parataxis_head-word_substituted prep_substituted_by det_function_a pobj_by_function dep_name_the nsubj_corresponds_name prep_name_of pobj_of_which rcmod_function_corresponds prep_corresponds_to det_preposition_the pobj_to_preposition cc_head-word_and poss_argument_its amod_argument_sole nsubj_corresponds_argument conj_head-word_corresponds prep_corresponds_to det_head_the pobj_to_head prep_head_of det_NP_the nn_NP_noun nn_NP_phrase pobj_of_NP ' +o,3447,'The first work in SMT , done at IBM , developed a noisy-channel model , factoring the translation process into two portions : the translation model and the language model ',Brown,'det_work_The amod_work_first nsubj_developed_work prep_work_in pobj_in_SMT partmod_work_done prep_done_at nn__IBM pobj_at_ det_model_a amod_model_noisy-channel dobj_developed_model xcomp_developed_factoring det_process_the nn_process_translation dobj_factoring_process prep_factoring_into num_portions_two pobj_into_portions det_model_the nn_model_translation dep_portions_model cc_model_and det_model_the nn_model_language conj_model_model ' +o,3448,'Recent work shows that k-best maximum spanning tree -LRB- MST -RRB- parsing and reranking is also viable ',Hall,'amod_work_Recent nsubj_shows_work complm_viable_that amod_maximum_k-best nsubj_viable_maximum partmod_maximum_spanning dobj_spanning_tree abbrev_tree_MST partmod_tree_parsing cc_parsing_and conj_parsing_reranking cop_viable_is advmod_viable_also ccomp_shows_viable ' +o,3449,'(Pang & Lee, 2004; Aue & Gamon, 2005).',Pang,'' +n,3450,'Our study also shows that the simulated-annealing algorithm is more effective 1552 than the perceptron algorithm for feature weight tuning ',Collins,'poss_study_Our nsubj_shows_study advmod_shows_also complm_effective_that det__the amod__simulated-annealing nn__algorithm nsubj_effective_ cop_effective_is advmod_effective_more ccomp_shows_effective tmod_effective_1552 mark__than det_algorithm_the nn_algorithm_perceptron nsubj__algorithm ccomp_effective_ prep__for nn_tuning_feature nn_tuning_weight pobj_for_tuning ' +p,3451,'Indeed , researchers have shown that gigantic language models are key to state-ofthe-art performance , and the ability of phrase-based decoders to handle large-size , high-order language models with no consequence on asymptotic running time during decoding presents a compelling advantage over CKYdecoders , whosetimecomplexitygrowsprohibitively large with higher-order language models ',Brants,'advmod_shown_Indeed nsubj_shown_researchers aux_shown_have complm_key_that amod_models_gigantic nn_models_language nsubj_key_models cop_key_are ccomp_shown_key prep_key_to amod__state-ofthe-art nn__performance pobj_to_ cc_shown_and det_ability_the nsubj_presents_ability prep_ability_of amod_decoders_phrase-based pobj_of_decoders aux_handle_to infmod_ability_handle amod_models_large-size amod_models_high-order nn_models_language dobj_handle_models prep_handle_with det_consequence_no pobj_with_consequence prep_consequence_on amod_time_asymptotic amod_time_running pobj_on_time prep_handle_during pcomp_during_decoding conj_shown_presents det_advantage_a amod_advantage_compelling dobj_presents_advantage prep_presents_over pobj_over_CKYdecoders advmod_large_whosetimecomplexitygrowsprohibitively dep_presents_large prep_large_with amod_models_higher-order nn_models_language pobj_with_models ' +o,3452,'Our hierarchical training method yields significant improvement when compared to a similar nonhierarchical model which instead uses the standard 2Data and code used in this paper are available at http://peoplecsailmitedu/edc/emnlp07/ perceptron update of ',Collins,'poss_method_Our amod_method_hierarchical nn_method_training nsubj_yields_method amod_improvement_significant dobj_yields_improvement advmod_available_when prep_available_compared dep_compared_to det_model_a amod_model_similar amod_model_nonhierarchical pobj_to_model nsubj_uses_which advmod_uses_instead rcmod_model_uses det_2Data_the amod_2Data_standard nsubj_available_2Data cc_2Data_and conj_2Data_code partmod_2Data_used prep_used_in det_paper_this pobj_in_paper cop_available_are dep_improvement_available prep_available_at amod_update_http://peoplecsailmitedu/edc/emnlp07/ nn_update_perceptron pobj_at_update prep_update_of ' +o,3453,'It would be necessary to apply either semiautomatic or automatic methods such as those in to extend FrameNet coverage for final application to machine translation tasks ',Green,'nsubj_necessary_It aux_necessary_would cop_necessary_be aux_apply_to xcomp_necessary_apply det_methods_either amod_methods_semiautomatic cc_semiautomatic_or conj_semiautomatic_automatic dobj_apply_methods dep_as_such prep_methods_as pobj_as_those prep_those_in pobj_in_ aux_extend_to xcomp_apply_extend amod_coverage_FrameNet dobj_extend_coverage prep_coverage_for amod_application_final pobj_for_application prep_extend_to nn_tasks_machine nn_tasks_translation pobj_to_tasks ' +o,3454,'In the iNeast system , the identification of relevant terms is oriented towards multi-document summarization , and they use a likelihood ratio which favors terms which are representative of the set of documents as opposed to the full collection ',Dunning,'prep_oriented_In det__the amod__iNeast nn__system pobj_In_ det_identification_the nsubjpass_oriented_identification prep_identification_of amod_terms_relevant pobj_of_terms auxpass_oriented_is prep_oriented_towards amod_summarization_multi-document pobj_towards_summarization cc_oriented_and nsubj_use_they conj_oriented_use det__a nn__likelihood nn__ratio dobj_use_ nsubj_favors_which rcmod__favors dobj_favors_terms nsubj_representative_which cop_representative_are rcmod_terms_representative prep_representative_of det_set_the pobj_of_set prep_set_of pobj_of_documents advmod_opposed_as amod_documents_opposed prep_use_to det_collection_the amod_collection_full pobj_to_collection ' +o,3455,'For evaluation , we used the BLEU metrics , which calculates the geometric mean of n-gram precision for the MT outputs found in reference translations ',Papineni,'prep_used_For pobj_For_evaluation nsubj_used_we det_metrics_the amod_metrics_BLEU dobj_used_metrics nsubj_calculates_which rcmod_metrics_calculates det_mean_the amod_mean_geometric dobj_calculates_mean prep_mean_of amod_precision_n-gram pobj_of_precision prep_precision_for det_outputs_the nn_outputs_MT pobj_for_outputs partmod_outputs_found prep_found_in nn_translations_reference pobj_in_translations ' +o,3456,'In Section 3 we then describe the probabilistic taxonomy learning model introduced by ',Snow,'prep_describe_In pobj_In_Section num_Section_3 nsubj_describe_we advmod_describe_then det_taxonomy_the amod_taxonomy_probabilistic dobj_describe_taxonomy partmod_taxonomy_learning dobj_learning_model partmod_model_introduced prep_introduced_by ' +o,3457,' use a graph-based technique to identify and analyze only subjective parts of texts ',Pang,'nsubj_use_ det_technique_a amod_technique_graph-based dobj_use_technique aux_identify_to infmod_technique_identify cc_identify_and conj_identify_analyze advmod_parts_only amod_parts_subjective dobj_identify_parts prep_parts_of pobj_of_texts ' +o,3458,'Given the estimated 3 \% error rate of the WSJ tagging , they argue that the difference in performance is not sufficient to establish which of the two taggers is actually better ',Marcus,'prep_argue_Given det_rate_the amod_rate_estimated dep_\%_3 amod_rate_\% nn_rate_error dep_Given_rate prep_rate_of det__the nn__WSJ nn__tagging pobj_of_ nsubj_argue_they complm_sufficient_that det_difference_the nsubj_sufficient_difference prep_difference_in pobj_in_performance cop_sufficient_is neg_sufficient_not ccomp_argue_sufficient aux_establish_to xcomp_sufficient_establish nsubj_better_which prep_which_of det_taggers_the num_taggers_two pobj_of_taggers cop_better_is advmod_better_actually ccomp_establish_better ' +o,3459,'In this paper , translation quality is evaluated according to -LRB- 1 -RRB- the BLEU metrics which calculates the geometric mean of ngram precision by the system output with respect to reference translations , and -LRB- 2 -RRB- the METEOR metrics that calculates unigram overlaps between translations ',Papineni,'dep__In det_paper_this pobj_In_paper nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_is dep__evaluated prep_evaluated_according dep_according_to dep__1 det_metrics_the amod_metrics_BLEU nsubj__metrics nsubj_calculates_which rcmod_metrics_calculates det_mean_the amod_mean_geometric dobj_calculates_mean prep_mean_of nn_precision_ngram pobj_of_precision prep_calculates_by det_output_the nn_output_system pobj_by_output prep_output_with pobj_with_respect prep_calculates_to nn_translations_reference pobj_to_translations cc__and dep__2 det_metrics_the amod_metrics_METEOR nsubj__metrics nsubj_calculates_that rcmod_metrics_calculates nn_overlaps_unigram dobj_calculates_overlaps prep_overlaps_between pobj_between_translations conj__ ' +o,3460,'The WSJ corpus is based on the WSJ part of the PENN TREEBANK ; we used the first 10,000 sentences of section 2-21 as the pool set , and section 00 as evaluation set -LRB- 1,921 sentences -RRB- ',Marcus,'det_corpus_The nn_corpus_WSJ nsubjpass_based_corpus auxpass_based_is prep_based_on det_part_the nn_part_WSJ pobj_on_part prep_part_of det__the nn__PENN nn__TREEBANK pobj_of_ nsubj_used_we parataxis_based_used det_sentences_the amod_sentences_first num_sentences_10,000 dobj_used_sentences prep_sentences_of pobj_of_section num_section_2-21 prep_used_as det_set_the nn_set_pool pobj_as_set cc_based_and nsubj_set_section num_section_00 prep_section_as pobj_as_evaluation conj_based_set num_sentences_1,921 dep_set_sentences ' +o,3461,'This upper bound is consistent with the upper limit of 50 \% found by which takes into account stemming differences ','Daume III','det_bound_This amod_bound_upper nsubj_consistent_bound cop_consistent_is prep_consistent_with det_limit_the amod_limit_upper pobj_with_limit prep_limit_of num_\%_50 pobj_of_\% partmod_\%_found prep_found_by pobj_by_ nsubj_takes_which rcmod__takes prep_takes_into pobj_into_account partmod_account_stemming dobj_stemming_differences ' +o,3462,' grow the set of word links by appending neighboring points , while and Hey try to avoid both horizontal and vertical neighbors ',Och,'nsubj_grow_ det_set_the dobj_grow_set prep_set_of nn_links_word pobj_of_links prep_grow_by pcomp_by_appending amod_points_neighboring dobj_appending_points mark_try_while nn__ cc__and conj__Hey nsubj_try_ advcl_grow_try aux_avoid_to xcomp_try_avoid det_horizontal_both dobj_avoid_horizontal cc_horizontal_and amod_neighbors_vertical conj_horizontal_neighbors ' +o,3463,'52 Impact on translation quality As reported in Table 3 , small increases in METEOR , BLEU and NIST scores suggest that SMT output matches the references better after postprocessing or decoding with the suggested lemma translations ',Banerjee,'num_Impact_52 dep_Impact_on nn_quality_translation pobj_on_quality mark_reported_As advcl_suggest_reported prep_reported_in pobj_in_Table num_Table_3 amod_increases_small nsubj_suggest_increases prep_increases_in nn__METEOR pobj_in_ nn__BLEU conj__ cc__and nn__NIST nn__scores conj__ rcmod_quality_suggest complm_matches_that nn_output_SMT nsubj_matches_output ccomp_suggest_matches det_references_the nsubj_better_references xcomp_matches_better prep_better_after pcomp_after_postprocessing cc_postprocessing_or conj_postprocessing_decoding prep_postprocessing_with det_translations_the amod_translations_suggested nn_translations_lemma pobj_with_translations ' +o,3464,'Almost all of the work in the area of automatically trained taggers has explored Markov-model based part of speech tagging ',Marcus,'advmod_based_Almost dep_Almost_all prep_all_of det_work_the pobj_of_work dep_Almost_in det_area_the pobj_in_area dep_Almost_of advmod_trained_automatically amod_taggers_trained pobj_of_taggers aux_based_has nn_Markov-model_explored nsubj_based_Markov-model dobj_based_part prep_part_of nn__speech amod__tagging pobj_of_ ' +p,3465,'A novel approach was described in , which used an unsupervised training technique , extracting relations that were explicitly and unamibiguously signalled and automatically labeling those examples as the training set ',Marcu,'det_approach_A amod_approach_novel nsubjpass_described_approach auxpass_described_was prep_described_in pobj_in_ nsubj_used_which rcmod__used det_technique_an amod_technique_unsupervised nn_technique_training dobj_used_technique dep_used_extracting dobj_extracting_relations nsubj_were_that rcmod_relations_were dep_were_explicitly cc_explicitly_and advmod_signalled_unamibiguously conj_explicitly_signalled cc_extracting_and advmod_extracting_automatically conj_extracting_labeling det_examples_those dobj_labeling_examples prep_labeling_as det_set_the nn_set_training pobj_as_set ' +o,3466,'33 Accuracy Results describe a model for unknown words that uses four features , but treats the features ms independent ',Marcus,'num__33 nn__Accuracy nn__Results nsubj_describe_ det_model_a dobj_describe_model prep_model_for amod_words_unknown pobj_for_words nsubj_uses_that rcmod_model_uses num_features_four dobj_uses_features cc_uses_but conj_uses_treats det_ms_the nn_ms_features nsubj_independent_ms xcomp_treats_independent ' +o,3467,'In previous work , we have reported some preliminary success in aligning the English and Japanese versions of the AWK manual -LRB- Aho , Kernighan , Weinberger -RRB- , using charalign , a method that looks for character sequences that are the same in both the source and target ',Brown,'prep_reported_In amod__previous nn__work pobj_In_ nsubj_reported_we aux_reported_have det_success_some amod_success_preliminary dobj_reported_success prep_reported_in pcomp_in_aligning det_versions_the amod_versions_English cc_English_and conj_English_Japanese dobj_aligning_versions prep_versions_of det_manual_the nn_manual_AWK pobj_of_manual appos_manual_Aho dep_Aho_Kernighan nn__Weinberger dep_Aho_ dep_aligning_using nn__charalign dobj_using_ det_method_a appos__method nsubj_looks_that rcmod_method_looks prep_looks_for nn_sequences_character pobj_for_sequences nsubj_same_that cop_same_are det_same_the rcmod_sequences_same prep_same_in preconj_source_both det_source_the pobj_in_source cc_source_and conj_source_target ' +o,3468,' ` Noun Classification from Predicate-Argument Structures , \' Proceedings of the 28th Annual Meeting of the ACL , pp ',Hindle,'nn_Classification_Noun dobj__Classification prep_Classification_from nn_Structures_Predicate-Argument pobj_from_Structures dep_Classification_Proceedings prep_Proceedings_of det_Meeting_the amod_Meeting_28th amod_Meeting_Annual pobj_of_Meeting prep_Meeting_of det_ACL_the pobj_of_ACL appos_ACL_pp ' +o,3469,'Differences in behavior of WSD systems when applied to lexical-sample and all-words datasets have been observed on previous Senseval and Semeval competitions : supervised systems attain results on the high 80s and beat the most frequent baseline by a large margin for lexical-sample datasets , but results on the all-words datasets were much more modest , on the low 70s , and a few points above the most frequent baseline ','Daume III','prep_Differences_in pobj_in_behavior prep_Differences_of nn_systems_WSD pobj_of_systems advmod_observed_when csubjpass_observed_applied prep_applied_to amod_datasets_lexical-sample cc_lexical-sample_and conj_lexical-sample_all-words pobj_to_datasets aux_observed_have auxpass_observed_been dep_Differences_observed prep_observed_on amod_Senseval_previous pobj_on_Senseval cc_observed_and nn_competitions_Semeval nsubj__competitions conj_observed_ amod_systems_supervised nsubj_attain_systems ccomp__attain dobj_attain_results prep_attain_on det_80s_the amod_80s_high pobj_on_80s cc_attain_and conj_attain_beat det_baseline_the advmod_frequent_most amod_baseline_frequent dobj_beat_baseline prep_beat_by det_margin_a amod_margin_large pobj_by_margin prep_margin_for amod_datasets_lexical-sample pobj_for_datasets cc__but nsubj_modest_results prep_results_on det_datasets_the amod_datasets_all-words pobj_on_datasets cop_modest_were advmod_modest_much advmod_modest_more conj__modest prep_modest_on det_70s_the amod_70s_low pobj_on_70s cc_70s_and det_points_a amod_points_few conj_70s_points prep_points_above det_baseline_the advmod_frequent_most amod_baseline_frequent pobj_above_baseline ' +o,3470,'Slrs Parse Base is 176 ',Marcus,'nn__Slrs nn__Parse nn__Base nsubj_176_ cop_176_is ' +p,3471,'The bigram translation probability relies on word context , known to be helpful in translation , to improve the identification of target phrases ',Berger,'det_probability_The nn_probability_bigram nn_probability_translation nsubj_relies_probability prep_relies_on nn_context_word pobj_on_context partmod_context_known aux_helpful_to cop_helpful_be xcomp_known_helpful prep_helpful_in nn__translation pobj_in_ aux_improve_to xcomp_helpful_improve det_identification_the dobj_improve_identification prep_identification_of nn_phrases_target pobj_of_phrases ' +o,3472,'Movie-domainSubjectivityDataSet -LRB- Movie -RRB- : used a collection of labeled subjective and objective sentences in their work on review classification5 The data set contains 5000 subjective sentences , extracted from movie reviews collected from the Rotten Tomatoes web formed best ',Pang,'appos_Movie-domainSubjectivityDataSet_Movie nsubj_used_ dep_Movie-domainSubjectivityDataSet_used det_collection_a dobj_used_collection prep_collection_of pobj_of_labeled dep_labeled_subjective cc_labeled_and amod_sentences_objective conj_labeled_sentences prep_used_in poss_work_their pobj_in_work prep_work_on nn_classification5_review pobj_on_classification5 det_set_The nn_set_data nsubj_contains_set rcmod_classification5_contains dep_subjective_5000 amod_sentences_subjective dobj_contains_sentences partmod_sentences_extracted prep_extracted_from nn_reviews_movie pobj_from_reviews partmod_reviews_collected prep_collected_from det_web_the nn_web_Rotten nn_web_Tomatoes pobj_from_web partmod_web_formed advmod_formed_best ' +o,3473,'The features used in this study are : the length of t ; a single-parameter distortion penalty on phrase reordering in a , as described in ; phrase translation model probabilities ; and trigram language model probabilities logp -LRB- t -RRB- , using Kneser-Ney smoothing as implemented in the SRILM toolkit ',Koehn,'det_features_The nsubj_length_features partmod_features_used prep_used_in det_study_this pobj_in_study cop_length_are det_length_the prep_length_of pobj_of_t det_penalty_a amod_penalty_single-parameter nn_penalty_distortion dep_length_penalty prep_penalty_on nn_reordering_phrase pobj_on_reordering prep_reordering_in det_logp_a advmod_described_as amod_logp_described prep_described_in pobj_in_ nn_probabilities_phrase nn_probabilities_translation nn_probabilities_model dep__probabilities cc__and nn_probabilities_trigram nn_probabilities_language nn_probabilities_model conj__probabilities pobj_in_logp appos_logp_t partmod_reordering_using dobj_using_Kneser-Ney partmod_Kneser-Ney_smoothing mark_implemented_as advcl_smoothing_implemented prep_implemented_in det__the nn__SRILM nn__toolkit pobj_in_ ' +n,3474,'By segmenting words into morphemes , we can improve the performance of natural language systems including machine translation and information retrieval ',Brown,'prep_improve_By pcomp_By_segmenting dobj_segmenting_words prep_segmenting_into pobj_into_morphemes nsubj_improve_we aux_improve_can det_performance_the dobj_improve_performance prep_performance_of amod_systems_natural nn_systems_language pobj_of_systems prep_systems_including nn__machine nn__translation pobj_including_ cc__and nn__information nn__retrieval conj__ ' +p,3475,'It is the most widely reported metric in MT research , and has been shown to correlate well with human judgment ',Papineni,'nsubjpass_reported_It cop_reported_is det_reported_the advmod_reported_most advmod_reported_widely amod_reported_metric prep_reported_in nn_research_MT pobj_in_research cc_reported_and aux_shown_has auxpass_shown_been conj_reported_shown aux_correlate_to xcomp_shown_correlate advmod_correlate_well prep_correlate_with amod_judgment_human pobj_with_judgment ' +o,3476,' use the Learning as Search Optimization framework to take into account the non-locality behavior of the coreference features ','Daume III','nsubj_use_ det_framework_the amod_framework_Learning prep_Learning_as nn_Optimization_Search pobj_as_Optimization dobj_use_framework aux_take_to xcomp_use_take prep_take_into pobj_into_account det_behavior_the amod_behavior_non-locality dobj_take_behavior prep_behavior_of det_features_the amod_features_coreference pobj_of_features ' +p,3477,'This means that the 1 -RRB- roblem of recognizing named entities in those cases can be solved by incorporating techniques of base noun phrase chunking ',Ramshaw,'nsubj_means_This complm_solved_that dep_roblem_the dep_roblem_1 nsubjpass_solved_roblem prep_roblem_of pcomp_of_recognizing amod_entities_named dobj_recognizing_entities prep_recognizing_in det_cases_those pobj_in_cases aux_solved_can auxpass_solved_be ccomp_means_solved prep_solved_by pcomp_by_incorporating dobj_incorporating_techniques prep_techniques_of nn_chunking_base nn_chunking_noun nn_chunking_phrase pobj_of_chunking ' +o,3478,'An alternative representation for baseNPs has been put tbrward by ',Ramshaw,'det_representation_An amod_representation_alternative nsubjpass_put_representation prep_representation_for pobj_for_baseNPs aux_put_has auxpass_put_been ccomp_put_tbrward prep_tbrward_by ' +o,3479,'Let us now compare our results to those obtained using shallow parsing , as previously done by ',Grefenstette,'nsubj_compare_us advmod_compare_now ccomp_Let_compare poss_results_our dobj_compare_results prep_compare_to pobj_to_those partmod_those_obtained xcomp_obtained_using nn_parsing_shallow dobj_using_parsing mark_done_as advmod_done_previously advcl_compare_done prep_done_by ' +o,3480,' used bootstrapping to train decision list classifiers to disambiguate between two senses of a word , achieving impressive classification accuracy ',Yarowsky,'nsubj_used_ ccomp_used_bootstrapping aux_train_to xcomp_bootstrapping_train nn_classifiers_decision nn_classifiers_list nsubj_disambiguate_classifiers aux_disambiguate_to xcomp_train_disambiguate prep_disambiguate_between num_senses_two pobj_between_senses prep_senses_of det_word_a pobj_of_word dep_bootstrapping_achieving amod_accuracy_impressive nn_accuracy_classification dobj_achieving_accuracy ' +o,3481,'Until now , we have defined BestLossk , a to be the minimum of the loss given that the kth feature is updated an optimal amount : BestLossk , amin d LogLossUpda , k , d In this section we sketch a different approach , based on results from , which leads to an algorithm very similar to that for ExpLoss in Figures 3 and 4 ',Collins,'prep_defined_Until pobj_Until_now nsubj_defined_we aux_defined_have dobj_defined_BestLossk dep_BestLossk_a aux_minimum_to cop_minimum_be det_minimum_the dep_a_minimum prep_minimum_of det_loss_the pobj_of_loss partmod_loss_given complm_BestLossk_that det_feature_the nn_feature_kth nsubj_BestLossk_feature cop_BestLossk_is amod_BestLossk_updated det_amount_an amod_amount_optimal dep_BestLossk_amount ccomp_given_BestLossk nn_LogLossUpda_amin nn_LogLossUpda_d appos_BestLossk_LogLossUpda appos_BestLossk_k dobj_sketch_d prep_d_In det_section_this pobj_In_section nsubj_sketch_we rcmod_BestLossk_sketch det_approach_a amod_approach_different dobj_sketch_approach partmod_BestLossk_based prep_based_on pobj_on_results prep_based_from pobj_from_ nsubj_leads_which rcmod__leads prep_leads_to det_algorithm_an pobj_to_algorithm advmod_similar_very amod_algorithm_similar prep_similar_to pobj_to_that prep_that_for pobj_for_ExpLoss prep_leads_in pobj_in_Figures num_Figures_3 cc_Figures_and conj_Figures_4 ' +o,3482,'On the other hand , works done by have proposed methodologies to automatically acquire these patterns mostly based on supervised learning to leverage manual work ',Bollegala,'prep_proposed_On det_hand_the amod_hand_other pobj_On_hand nsubj_proposed_works partmod_works_done prep_done_by pobj_by_ aux_proposed_have dobj_proposed_methodologies aux_acquire_to advmod_acquire_automatically xcomp_proposed_acquire det_patterns_these dobj_acquire_patterns advmod_acquire_mostly prep_acquire_based dep_based_on amod_work_supervised dep_supervised_learning prep_learning_to pobj_to_leverage amod_work_manual pobj_on_work ' +o,3483,'212 Research on Syntax-Based SMT A number of researchers have proposed models where the translation process involves syntactic representations of the source and\\/or target languages ',Wu,'nn_Research_212 dep_number_Research prep_Research_on pobj_on_Syntax-Based dep_number_SMT det_number_A dep_have_number prep_number_of pobj_of_researchers dep_have_ amod_models_proposed dobj_have_models advmod_involves_where det_process_the nn_process_translation nsubj_involves_process advcl_have_involves amod_representations_syntactic dobj_involves_representations prep_representations_of det_source_the pobj_of_source cc_have_and\\/or conj_have_target dobj_target_languages ' +o,3484,'As described in Section 4 , we define the problem of term variation identifica1484 tion as a binary classification task , and build two types of classifiers according to the maximum entropy model and the MART algorithm , where all term similarity metrics are incorporated as features and are jointly optimized ',Berger,'mark_described_As advcl_define_described prep_described_in pobj_in_Section num_Section_4 nsubj_define_we det_problem_the nsubj_tion_problem prep_problem_of nn_identifica1484_term nn_identifica1484_variation pobj_of_identifica1484 ccomp_define_tion prep_tion_as det_task_a amod_task_binary nn_task_classification pobj_as_task cc_tion_and conj_tion_build num_types_two dobj_build_types prep_types_of pobj_of_classifiers prep_build_according dep_according_to det__the amod__maximum amod__entropy nn__model pobj_to_ cc__and det__the amod__MART nn__algorithm conj__ advmod_incorporated_where det_metrics_all nn_metrics_term nn_metrics_similarity nsubjpass_incorporated_metrics auxpass_incorporated_are advcl_define_incorporated prep_incorporated_as pobj_as_features cc_incorporated_and auxpass_optimized_are advmod_optimized_jointly conj_incorporated_optimized ' +p,3485,'These models have achieved state-of-the-art performance in transcript-based speech summarization ',Penn,'det_models_These nsubj_achieved_models aux_achieved_have amod_performance_state-of-the-art dobj_achieved_performance prep_achieved_in amod_summarization_transcript-based nn_summarization_speech pobj_in_summarization ' +p,3486,'1 Introduction Research in language processing has benefited greatly from the collection of large annotated corpora such as Penn PropBank and Penn Treebank ',Marcus,'num_Research_1 nn_Research_Introduction nsubj_benefited_Research prep_Research_in nn_processing_language pobj_in_processing aux_benefited_has advmod_benefited_greatly prep_benefited_from det_collection_the pobj_from_collection prep_collection_of amod_corpora_large amod_corpora_annotated pobj_of_corpora dep_as_such prep_corpora_as nn__Penn nn__PropBank pobj_as_ cc__and nn_Treebank_Penn conj__Treebank ' +o,3487,'The resulting intercoder reliability , measured with the Kappa statistic , is considered excellent -LRB- = 080 -RRB- ',Carletta,'det_reliability_The amod_reliability_resulting nn_reliability_intercoder nsubjpass_considered_reliability partmod_reliability_measured prep_measured_with det__the nn__Kappa nn__statistic pobj_with_ auxpass_considered_is dep_considered_excellent number_080_= dep_excellent_080 ' +o,3488,'For the classifier , we used the OpenNLP MaxEnt implementation -LRB- maxentsourceforgenet -RRB- of the maximum entropy classification algorithm ',Berger,'prep_used_For det_classifier_the pobj_For_classifier nsubj_used_we det_implementation_the nn_implementation_OpenNLP nn_implementation_MaxEnt dobj_used_implementation appos_implementation_maxentsourceforgenet prep_implementation_of det_algorithm_the amod_algorithm_maximum amod_algorithm_entropy nn_algorithm_classification pobj_of_algorithm ' +n,3489,'Statistical disambiguation such as for PP-attachment or for generative parsing greatly improve disambiguation , but as they model by imitation instead of by understanding , complete soundness has to remain elusive ',Collins,'aux_improve_Statistical nsubj_improve_disambiguation dep_as_such prep_disambiguation_as pobj_as_ prep__for pobj_for_PP-attachment cc_PP-attachment_or conj_PP-attachment_ prep__for amod_parsing_generative pobj_for_parsing advmod_improve_greatly nsubj_has_disambiguation cc_disambiguation_but mark_model_as nsubj_model_they conj_disambiguation_model prep_model_by pobj_by_imitation cc_by_instead dep_instead_of conj_by_by pobj_by_understanding amod_soundness_complete nsubj_has_soundness ccomp_improve_has aux_elusive_to cop_elusive_remain xcomp_has_elusive ' +o,3490,'In contrast , approaches to WSD attempt to take advantage of many different sources of information -LRB- eg see -RRB- ; it seems possible to obtain benefit from sources ranging from local collocational clues to membership in semantically or topically related word classes to consistency of word usages within a discourse ; and disambignation seems highly lexically sensitive , in effect requiring specialized disamhignators for each polysemous word ',Brown,'prep_attempt_In pobj_In_contrast nsubj_attempt_approaches prep_approaches_to pobj_to_WSD aux_take_to xcomp_attempt_take dobj_take_advantage prep_advantage_of amod_sources_many amod_sources_different pobj_of_sources prep_sources_of pobj_of_information nsubj_see_eg dep_sources_see acomp_see_ nsubj_possible_it cop_possible_seems parataxis_attempt_possible aux_obtain_to xcomp_possible_obtain dobj_obtain_benefit prep_obtain_from pobj_from_sources partmod_sources_ranging prep_ranging_from amod_clues_local amod_clues_collocational pobj_from_clues partmod_clues_ prep__to pobj_to_membership prep_membership_in pobj_in_semantically cc_membership_or advmod_related_topically amod_classes_related nn_classes_word conj_membership_classes partmod_classes_ prep__to pobj_to_consistency prep_consistency_of nn_usages_word pobj_of_usages prep__within det__a nn__discourse pobj_within_ cc_attempt_and nsubj_sensitive_disambignation cop_sensitive_seems advmod_sensitive_highly advmod_sensitive_lexically conj_attempt_sensitive prep_sensitive_in pobj_in_effect partmod_effect_requiring amod_disamhignators_specialized dobj_requiring_disamhignators prep_requiring_for det_word_each amod_word_polysemous pobj_for_word ' +o,3491,'l lhmsetsu ideni , illcation is a ln ` oblem similar to ohm , king in other l ; mguages ',Ramshaw,'prep__l amod_ideni_lhmsetsu pobj_l_ideni nsubj__illcation cop__is det__a amod__ln nn__oblem amod__similar prep_similar_to pobj_to_ohm appos_ohm_king prep__in amod_l_other pobj_in_l dep_l_mguages ' +o,3492,'For example , given that each semantic class exhibits a particular syntactic behavior , information on the semantic class should improve POStagging for adjective-noun and adjective-participle ambiguities , probably the most difficult distinctions both for humans and computers ',Marcus,'prep_improve_For pobj_For_example ccomp_improve_given complm_exhibits_that det_class_each amod_class_semantic nsubj_exhibits_class ccomp_given_exhibits det_behavior_a amod_behavior_particular amod_behavior_syntactic dobj_exhibits_behavior nsubj_improve_information prep_information_on det_class_the amod_class_semantic pobj_on_class aux_improve_should dobj_improve_POStagging prep_POStagging_for amod_ambiguities_adjective-noun cc_adjective-noun_and conj_adjective-noun_adjective-participle pobj_for_ambiguities advmod_ambiguities_probably det_distinctions_the advmod_difficult_most amod_distinctions_difficult dep_ambiguities_distinctions dep_for_both prep_improve_for pobj_for_humans cc_humans_and conj_humans_computers ' +o,3493,'so they conform to the Penn Treebank corpus annotation style , and then do experiments using models built with Treebank data ',Marcus,'mark_conform_so nsubj_conform_they prep_conform_to det_style_the nn_style_Penn nn_style_Treebank nn_style_corpus nn_style_ nn_style_annotation pobj_to_style cc_conform_and advmod_do_then conj_conform_do dobj_do_experiments partmod_experiments_using dobj_using_models partmod_models_built prep_built_with nn_data_Treebank pobj_with_data ' +o,3494,'ROUGE is a set of recall-based criteria that is mainly used for evaluating summarization tasks ',Lin,'nn__ROUGE nsubj_set_ cop_set_is det_set_a prep_set_of amod_criteria_recall-based pobj_of_criteria nsubjpass_used_that auxpass_used_is advmod_used_mainly rcmod_set_used prep_used_for pcomp_for_evaluating nn_tasks_summarization dobj_evaluating_tasks ' +p,3495,'G-Theory and Agreement Indices Two well-known measures for capturing the quality of manual annotations are agreement percentages and the kappa statistic ',Carletta,'nn_Indices_G-Theory cc_G-Theory_and conj_G-Theory_Agreement num_measures_Two amod_measures_well-known nsubj_percentages_measures prep_measures_for pcomp_for_capturing det_quality_the dobj_capturing_quality prep_quality_of amod_annotations_manual pobj_of_annotations cop_percentages_are nn_percentages_agreement dep_Indices_percentages cc_percentages_and det_statistic_the nn_statistic_kappa conj_percentages_statistic ' +o,3496,'Pivots are features occurring frequently and behaving similarly in both domains ',Blitzer,'nsubj_features_Pivots cop_features_are dep_features_occurring advmod_occurring_frequently cc_occurring_and conj_occurring_behaving advmod_behaving_similarly prep_behaving_in det_domains_both pobj_in_domains ' +o,3497,' introduced one of those similarity schemes , ? two-level SoftTFIDF ? ? ',Chen,'nsubj_introduced_ dobj_introduced_one prep_one_of det_schemes_those nn_schemes_similarity pobj_of_schemes amod_one_two-level dep_two-level_SoftTFIDF ' +o,3498,'1 Introduction : Defining SCMs The work presented here was done in the context of phrase-based MT ',Koehn,'num_Introduction_1 nn_SCMs_Defining nsubjpass_done_SCMs det_work_The nsubj_presented_work rcmod_SCMs_presented advmod_done_here auxpass_done_was dep_Introduction_done prep_done_in det_context_the pobj_in_context prep_context_of amod_MT_phrase-based pobj_of_MT ' +o,3499,'Four teams had approaches that relied -LRB- to varying degrees -RRB- on an IBM model of statistical machine translation ',Brown,'num_teams_Four nsubj_had_teams dobj_had_approaches nsubj_relied_that rcmod_approaches_relied dep_relied_to amod_degrees_varying pobj_to_degrees prep_relied_on det_model_an nn_model_IBM pobj_on_model prep_model_of amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,3500,'For English , we used the Penn Treebank in our experiments and the tool Penn2Malt7 to convert the data into dependency structures using a standard set of head rules ',Marcus,'prep_used_For pobj_For_English nsubj_used_we det__the nn__Penn nn__Treebank dobj_used_ prep_used_in poss_experiments_our pobj_in_experiments cc_experiments_and det_Penn2Malt7_the nn_Penn2Malt7_tool conj_experiments_Penn2Malt7 aux_convert_to xcomp_used_convert det_data_the dobj_convert_data prep_convert_into amod_structures_dependency pobj_into_structures partmod_structures_using det_set_a amod_set_standard dobj_using_set prep_set_of nn_rules_head pobj_of_rules advmod_using_ ' +o,3501,'Non-anaphoric definite descriptions have been detected using heuristics -LRB- eg , Vieira and Poesio -RRB- and unsupervised methods -LRB- eg , -RRB- ',Bean,'amod_descriptions_Non-anaphoric amod_descriptions_definite nsubjpass_detected_descriptions aux_detected_have auxpass_detected_been xcomp_detected_using dobj_using_heuristics dep_heuristics_eg conj_eg_Vieira cc_eg_and nn__Poesio conj_eg_ cc_heuristics_and amod_methods_unsupervised conj_heuristics_methods appos_methods_eg dep_eg_ ' +o,3502,'The model presented above is based on our previous work , which bears the same spirit of some other recent work on multitask learning ','Daume III','det_model_The nsubjpass_based_model partmod_model_presented prep_presented_above auxpass_based_is prep_based_on poss__our amod__previous nn__work pobj_on_ nsubj_bears_which rcmod__bears det_spirit_the amod_spirit_same dobj_bears_spirit prep_spirit_of det_work_some amod_work_other amod_work_recent pobj_of_work prep_work_on nn_learning_multitask pobj_on_learning ' +o,3503,'The L1 or L2 norm is commonly used in statistical natural language processing ',Gao,'det_norm_The nn_norm_L1 cc_L1_or conj_L1_L2 nsubjpass_used_norm auxpass_used_is advmod_used_commonly prep_used_in amod_processing_statistical amod_processing_natural nn_processing_language pobj_in_processing ' +o,3504,'Whereas language generation has benefited from syntax , the performance of statistical phrase-based machine translation when relying solely on syntactic phrases has been reported to be poor ',Koehn,'mark_benefited_Whereas nn_generation_language nsubj_benefited_generation aux_benefited_has advcl_reported_benefited prep_benefited_from nn__syntax pobj_from_ det_performance_the nsubjpass_reported_performance prep_performance_of amod_translation_statistical amod_translation_phrase-based nn_translation_machine pobj_of_translation advmod_relying_when dep_translation_relying advmod_relying_solely prep_relying_on amod_phrases_syntactic pobj_on_phrases aux_reported_has auxpass_reported_been aux_poor_to cop_poor_be xcomp_reported_poor ' +o,3505,'Besides continued research on improving MT techniques , one line of research is dedicated to better exploitation of existing methods for the combination of their respective advantages ',Rosti,'prep_dedicated_Besides amod_research_continued pobj_Besides_research prep_research_on amod_techniques_improving nn_techniques_MT pobj_on_techniques num_line_one nsubjpass_dedicated_line prep_line_of pobj_of_research auxpass_dedicated_is prep_dedicated_to amod_exploitation_better pobj_to_exploitation prep_exploitation_of amod_methods_existing pobj_of_methods prep_methods_for det_combination_the pobj_for_combination prep_combination_of poss_advantages_their amod_advantages_respective pobj_of_advantages ' +o,3506,'RECALL F-SCORE Brackets 8917 8750 8833 Dependencies 9640 9640 9640 Brackets , revised 9756 9803 9779 Dependencies , revised 9927 9927 9927 Table 1 : Agreement between annotators few weeks , and increased to about 1000 words per hour after gaining more experience ',Marcus,'nn_Brackets_RECALL nn_Brackets_F-SCORE num_Brackets_8917 number_8833_8750 num_Brackets_8833 amod_Brackets_Dependencies num_Brackets_9640 number_9640_9640 num_Brackets_9640 nsubj_revised_Brackets partmod_Brackets_revised num_Dependencies_9756 number_9779_9803 num_Dependencies_9779 dobj_revised_Dependencies dep_Brackets_revised number_9927_9927 dobj_revised_9927 num_Table_9927 dep_9927_Table num_Table_1 dep_9927_Agreement prep_Agreement_between amod_weeks_annotators amod_weeks_few pobj_between_weeks cc_revised_and conj_revised_increased prep_increased_to advmod_words_about num_words_1000 pobj_to_words prep_words_per pobj_per_hour prep_increased_after pcomp_after_gaining amod_experience_more dobj_gaining_experience ' +o,3507,'Following , we do not distinguish rare words ',Collins,'prep_distinguish_Following pobj_Following_ nsubj_distinguish_we aux_distinguish_do neg_distinguish_not amod_words_rare dobj_distinguish_words ' +o,3508,'4 The Dependency Labeler 41 Classifier We used a maximum entropy classifier to assign labels to the unlabeled dependencies produced by the Bayes Point Machine ',Berger,'det_Classifier_The nn_Classifier_Dependency nn_Classifier_Labeler num_Classifier_41 nsubj_produced_Classifier nsubj_used_We rcmod_Classifier_used det__a amod__maximum amod__entropy nn__classifier dobj_used_ aux_assign_to xcomp_used_assign dobj_assign_labels prep_assign_to det_dependencies_the amod_dependencies_unlabeled pobj_to_dependencies dep_4_produced prep_produced_by det_Machine_the nn_Machine_Bayes nn_Machine_Point pobj_by_Machine ' +o,3509,'Carletta mentions this problem , asking what the difference would be if the kappa statistic were computed across ` clause boundaries , transcribed word boundaries , and transcribed phoneme boundaries \' rather than the sentence boundaries she suggested ',Carletta,'nsubj_mentions_Carletta det_problem_this dobj_mentions_problem xcomp_mentions_asking dobj_be_what det_difference_the nsubj_be_difference aux_be_would ccomp_asking_be mark_computed_if det_statistic_the nn_statistic_kappa nsubjpass_computed_statistic auxpass_computed_were advcl_be_computed prep_computed_across nn_boundaries_clause pobj_across_boundaries amod_boundaries_transcribed nn_boundaries_word conj_boundaries_boundaries cc_boundaries_and amod_boundaries_transcribed nn_boundaries_phoneme poss__boundaries conj_boundaries_ dep_than_rather prep__than det_boundaries_the nn_boundaries_sentence pobj_than_boundaries nsubj_suggested_she dep_mentions_suggested ' +o,3510,'Previous uses of this model include language modeling , machine translation , prepositional phrase attachment , and word morphology ',Berger,'amod_uses_Previous nsubj_include_uses prep_uses_of det_model_this pobj_of_model nn__language nn__modeling dobj_include_ nn__machine nn__translation conj__ amod__prepositional nn__phrase nn__attachment conj__ cc__and nn__word nn__morphology conj__ ' +o,3511,'1999 -RRB- , OpenCCG and XLE , or created semi-automatically , or fully automatically extracted from annotated corpora , like the HPSG , LFG and CCG resources derived from the Penn-II Treebank -LRB- PTB -RRB- ',Cahill,'dep__1999 nn__OpenCCG nsubj__ cc__and conj__XLE dep__ cc__or conj__created amod__semi-automatically dobj__ cc__or advmod__fully advmod_extracted_automatically conj__extracted prep_extracted_from amod_corpora_annotated pobj_from_corpora prep_extracted_like det__the nn__HPSG nn__ appos__LFG pobj_like_ cc__and nn_resources_CCG nn_resources_ conj__resources partmod__derived prep_derived_from det_Treebank_the nn_Treebank_Penn-II pobj_from_Treebank abbrev__PTB ' +o,3512,'To optimize the parameters of the decoder , we performed minimum error rate training on IWSLT04 optimizing for the IBM-BLEU metric ',Papineni,'aux_optimize_To dep_performed_optimize det_parameters_the dobj_optimize_parameters prep_parameters_of det_decoder_the pobj_of_decoder nsubj_performed_we amod_training_minimum nn_training_error nn_training_rate dobj_performed_training prep_performed_on nn_optimizing_IWSLT04 pobj_on_optimizing prep_performed_for det_metric_the amod_metric_IBM-BLEU pobj_for_metric ' +o,3513,'2 Discriminative Reordering Model Basic reordering models in phrase-based systems use linear distance as the cost for phrase movements ',Koehn,'num_models_2 nn_models_Discriminative nn_models_Reordering nn_models_Model nn_models_Basic nn_models_reordering nsubj_use_models prep_models_in amod_systems_phrase-based pobj_in_systems amod_distance_linear dobj_use_distance prep_use_as det_cost_the pobj_as_cost prep_cost_for nn_movements_phrase pobj_for_movements ' +p,3514,'The technique of averaging was introduced in the context of perceptrons as an approximation to taking a vote among all the models traversed during training , and has been shown to work well in practice ',Collins,'det_technique_The nsubjpass_introduced_technique prep_technique_of pobj_of_averaging auxpass_introduced_was prep_introduced_in det_context_the pobj_in_context prep_context_of pobj_of_perceptrons prep_introduced_as det_approximation_an pobj_as_approximation prep_approximation_to pcomp_to_taking det_vote_a dobj_taking_vote prep_taking_among predet_models_all det_models_the pobj_among_models partmod_models_traversed prep_traversed_during pobj_during_training cc_introduced_and aux_shown_has auxpass_shown_been conj_introduced_shown aux_work_to xcomp_shown_work advmod_work_well prep_work_in pobj_in_practice ' +o,3515,'Translation performance is measured using the automatic BLEU metric , on one reference translation ',Papineni,'nn_performance_Translation nsubjpass_measured_performance auxpass_measured_is xcomp_measured_using det_metric_the amod_metric_automatic nn_metric_BLEU nn_metric_ dobj_using_metric prep_measured_on num_translation_one nn_translation_reference pobj_on_translation ' +o,3516,' also uses wide context , but incorporates the one-senseper-discourse and one-sense-per-collocation constraints , using an unsupervised learning technique ',Yarowsky,'nsubj_uses_ advmod_uses_also amod_context_wide dobj_uses_context cc_uses_but conj_uses_incorporates det_constraints_the amod_constraints_one-senseper-discourse cc_one-senseper-discourse_and conj_one-senseper-discourse_one-sense-per-collocation dobj_incorporates_constraints partmod_constraints_using det_technique_an amod_technique_unsupervised nn_technique_learning dobj_using_technique ' +o,3517,'When an S alignment exists , there will always also exist a P alignment such that P a65 S The English sentences were parsed using a state-of-the-art statistical parser trained on the University of Pennsylvania Treebank ',Marcus,'advmod_exists_When det_alignment_an nn_alignment_S nsubj_exists_alignment dep_alignment_exists expl_alignment_there aux_alignment_will advmod_alignment_always advmod_alignment_also dep_alignment_exist det_alignment_a nn_alignment_P prep_alignment_such det_S_that nn_S_P nn_S_a65 dep_such_S det_sentences_The amod_sentences_English nsubjpass_parsed_sentences auxpass_parsed_were dep_alignment_parsed xcomp_parsed_using det__a amod__state-of-the-art amod__statistical nn__parser dobj_using_ partmod__trained prep_trained_on det_University_the pobj_on_University prep_University_of nn_Treebank_Pennsylvania pobj_of_Treebank ' +n,3518,'It can be applied to complicated models such IBM Model-4 ',Brown,'nsubjpass_applied_It aux_applied_can auxpass_applied_be prep_applied_to amod_models_complicated pobj_to_models prep_applied_such nn_Model-4_IBM dep_such_Model-4 ' +o,3519,' and Taskar et al ',Collins,'cc__and conj__Taskar cc_Taskar_et conj_Taskar_al ' +o,3520,' and and show how to employ the matrix-tree theorem ',Smith,'nsubj_show_ cc__and conj__ cc__and nn__ conj__ advmod_employ_how aux_employ_to ccomp_show_employ det_theorem_the amod_theorem_matrix-tree dobj_employ_theorem ' +o,3521,'Determining the sense of an ambiguous word , using bootstrapping and texts from a different language was done by , Hearst -LRB- 1991 -RRB- , Diab -LRB- 2002 -RRB- , and Li and Li -LRB- 2004 -RRB- ',Yarowsky,'det_sense_the dobj_Determining_sense prep_sense_of det_word_an amod_word_ambiguous pobj_of_word xcomp_Determining_using dobj_using_bootstrapping cc_bootstrapping_and conj_bootstrapping_texts prep_using_from det_language_a amod_language_different pobj_from_language auxpass_done_was dep_Determining_done prep_done_by nsubj_Determining_ appos__Hearst appos_Hearst_1991 conj_Hearst_Diab appos_Diab_2002 cc_Hearst_and conj_Hearst_Li cc_Li_and conj_Li_Li appos_Li_2004 ' +o,3522,' proposed a method for extracting opinion holders , topics and opinion words , in which they use semantic role labeling as an intermediate step to label opinion holders and topics ',Kim,'advmod_proposed_ det_method_a dobj_proposed_method prep_proposed_for pcomp_for_extracting nn_holders_opinion dobj_extracting_holders conj_holders_topics cc_holders_and conj_holders_opinion nsubj_proposed_words rel_use_in pobj_in_which nsubj_use_they rcmod_words_use amod_labeling_semantic nn_labeling_role dobj_use_labeling prep_use_as det_step_an amod_step_intermediate pobj_as_step aux_label_to xcomp_use_label nn_holders_opinion dobj_label_holders cc_holders_and conj_holders_topics ' +o,3523,'Following previous work , we assume that the tag of a word is independent of the tags of all preceding words given the tags of the previous two words -LRB- ie , = 2 in the equation above -RRB- ',Ratnaparkhi,'prep_assume_Following amod__previous nn__work pobj_Following_ nsubj_assume_we complm_independent_that det_tag_the nsubj_independent_tag prep_tag_of det_word_a pobj_of_word cop_independent_is ccomp_assume_independent prep_independent_of det_tags_the pobj_of_tags prep_tags_of det_words_all amod_words_preceding pobj_of_words partmod_words_given det_tags_the dobj_given_tags prep_given_of det_words_the amod_words_previous num_words_two pobj_of_words dep_given_ie number_2_= measure_above_2 prep_2_in det_equation_the pobj_in_equation advmod_ie_above ' +o,3524,'We then examined the inter-annotator reliability of the annotation by calculating the score ',Carletta,'nsubj_examined_We advmod_examined_then det_reliability_the amod_reliability_inter-annotator dobj_examined_reliability prep_reliability_of det_annotation_the pobj_of_annotation prep_examined_by pcomp_by_calculating det_score_the dobj_calculating_score ' +o,3525,'Such methods stand in sharp contrast to partially supervised techniques that have recently been proposed to induce hidden grammatical representations that are finer-grained than those that can be read off the parsed sentences in treebanks ',Matsuzaki,'amod_methods_Such nsubj_stand_methods prep_stand_in amod_contrast_sharp pobj_in_contrast prep_stand_to advmod_supervised_partially amod_techniques_supervised pobj_to_techniques nsubjpass_proposed_that aux_proposed_have advmod_proposed_recently auxpass_proposed_been rcmod_techniques_proposed aux_induce_to xcomp_proposed_induce amod_representations_hidden amod_representations_grammatical dobj_induce_representations nsubj_finer-grained_that cop_finer-grained_are rcmod_representations_finer-grained prep_finer-grained_than pobj_than_those nsubjpass_read_that aux_read_can auxpass_read_be rcmod_those_read prt_read_off det_sentences_the amod_sentences_parsed dobj_read_sentences prep_read_in pobj_in_treebanks ' +o,3526,'The translation model is estimated via the EM algorithm or approximations that are bootstrapped from the previous model in the sequence as introduced in ',Brown,'det_model_The nn_model_translation nsubjpass_estimated_model auxpass_estimated_is prep_estimated_via det_algorithm_the nn_algorithm_EM pobj_via_algorithm cc_algorithm_or conj_algorithm_approximations nsubjpass_bootstrapped_that auxpass_bootstrapped_are rcmod_approximations_bootstrapped prep_bootstrapped_from det_model_the amod_model_previous pobj_from_model prep_model_in det_sequence_the pobj_in_sequence mark_introduced_as advcl_bootstrapped_introduced prt_introduced_in ' +o,3527,'This clustering was created automatically with the aid of a methodology described in ',Navigli,'det_clustering_This nsubjpass_created_clustering auxpass_created_was advmod_created_automatically prep_created_with det_aid_the pobj_with_aid prep_aid_of det_methodology_a pobj_of_methodology partmod_methodology_described prep_described_in ' +p,3528,'2 Confusion-network-based MT system combination The current state-of-the-art is confusion-networkbased MT system combination as described by 98 Rosti and colleagues ',Rosti,'num_combination_2 nn_combination_Confusion-network-based nn_combination_MT nn_combination_system det_state-of-the-art_The amod_state-of-the-art_current nsubj_combination_state-of-the-art cop_combination_is amod_combination_confusion-networkbased nn_combination_MT nn_combination_system dep_combination_combination prep_combination_as amod_Rosti_described prep_described_by pobj_by_98 pobj_as_Rosti cc_Rosti_and conj_Rosti_colleagues ' +o,3529,'A similar approach is used here , including a collapsed version of the Treebank POS tag set , with additions for specific words -LRB- eg personal pronouns and filled pause markers -RRB- , compound punctuation -LRB- eg multiple exclamation marks -RRB- , and a general emoticon tag , resulting in a total of 41 tags ',Marcus,'det_approach_A amod_approach_similar nsubjpass_used_approach auxpass_used_is advmod_used_here prep_used_including det_version_a amod_version_collapsed pobj_including_version prep_version_of det__the nn__Treebank dep__POS nn__tag nn__set pobj_of_ dep_additions_with appos__additions prep_additions_for amod_words_specific pobj_for_words dep_additions_eg amod_pronouns_personal dobj_eg_pronouns cc_eg_and conj_eg_filled nn_markers_pause dobj_filled_markers amod_punctuation_compound conj_additions_punctuation nn_marks_eg nn_marks_multiple nn_marks_exclamation appos_punctuation_marks cc_additions_and det_tag_a amod_tag_general nn_tag_emoticon conj_additions_tag partmod_tag_resulting prep_resulting_in det_total_a pobj_in_total prep_total_of num_tags_41 pobj_of_tags ' +o,3530,'It was first cast as a classification problem by , as a problem of NP chunking ',Ramshaw,'nsubjpass_cast_It auxpass_cast_was advmod_cast_first prep_cast_as det_problem_a nn_problem_classification pobj_as_problem prep_cast_by pobj_by_ prep_cast_as det_problem_a pobj_as_problem prep_problem_of pobj_of_NP amod_NP_chunking ' +o,3531,'Standard CI Model 1 training , initialised with a uniform translation table so that t -LRB- ejf -RRB- is constant for all source\\/target word pairs -LRB- f , e -RRB- , was run on untagged data for 10 iterations in each direction ',Marcus,'nn_Model_Standard nn_Model_CI num_training_1 nsubjpass_run_training partmod_training_initialised prep_initialised_with det_table_a amod_table_uniform nn_table_translation pobj_with_table advmod_constant_so dep_constant_that nsubj_constant_t appos_t_ejf cop_constant_is ccomp_initialised_constant prep_constant_for det_pairs_all amod_pairs_source\\/target nn_pairs_word pobj_for_pairs dep_pairs_f appos_f_e auxpass_run_was dep_Model_run prep_run_on amod_data_untagged pobj_on_data prep_data_for num_iterations_10 pobj_for_iterations prep_iterations_in det__each nn__direction pobj_in_ ' +o,3532,'For the efficiency of minimum-error-rate training , we built our development set -LRB- 580 sentences -RRB- using sentences not exceeding 50 characters from the NIST MT-02 evaluation test data ',Och,'prep_built_For det_efficiency_the pobj_For_efficiency prep_efficiency_of amod__minimum-error-rate nn__training pobj_of_ nsubj_built_we poss_set_our nn_set_development dobj_built_set number_sentences_580 dep_set_sentences partmod_set_using dobj_using_sentences neg_exceeding_not xcomp_using_exceeding num_characters_50 dobj_exceeding_characters prep_exceeding_from det_data_the nn_data_NIST nn_data_MT-02 nn_data_evaluation nn_data_test pobj_from_data ' +o,3533,'This differs from typical generative settings for IR and MT , where all conditioned events are disjoint by construction ',Brown,'nsubj_differs_This prep_differs_from amod_settings_typical nn_settings_generative pobj_from_settings prep_settings_for nn__IR cc_IR_and conj_IR_MT pobj_for_ advmod_disjoint_where det_events_all amod_events_conditioned nsubj_disjoint_events cop_disjoint_are advcl_differs_disjoint prep_disjoint_by pobj_by_construction ' +o,3534,'The parser induction algorithm used in all of the experiments in this paper was a distribution of Collins \'s model 2 parser ',Collins,'det_algorithm_The nn_algorithm_parser nn_algorithm_induction nsubj_distribution_algorithm partmod_algorithm_used prep_used_in pobj_in_all prep_all_of det_experiments_the pobj_of_experiments prep_experiments_in det_paper_this pobj_in_paper cop_distribution_was det_distribution_a prep_distribution_of poss_parser_Collins possessive_Collins_\'s nn_parser_model num_parser_2 pobj_of_parser ' +o,3535,'4 Experiments Our experiments were conducted on CoNLL-2007 shared task domain adaptation track using treebanks ',Marcus,'num_Experiments_4 poss_experiments_Our nsubjpass_conducted_experiments auxpass_conducted_were dep_Experiments_conducted prep_conducted_on amod__CoNLL-2007 amod__shared nn__task nn__domain nn__adaptation nn__track pobj_on_ partmod__using dobj_using_treebanks ' +o,3536,' state that a baseNP aims to identify essentially the initial portions of nonrecursive noun phrases up to the head , including determiners but not including postmodifying prepositional phrases or clauses However , work on baseNPs has essentially always proceeded via algorithmic extraction from fully parsed corpora such as the Penn Treebank ',Ramshaw,'nsubj_state_ complm_aims_that det_baseNP_a nsubj_aims_baseNP ccomp_state_aims aux_identify_to xcomp_aims_identify advmod_portions_essentially det_portions_the amod_portions_initial dep_identify_portions prep_portions_of amod_phrases_nonrecursive nn_phrases_noun pobj_of_phrases advmod_portions_up dep_up_to det_head_the pobj_to_head prep_head_including pobj_including_determiners cc_portions_but neg_including_not conj_portions_including xcomp_including_postmodifying amod_phrases_prepositional dobj_postmodifying_phrases cc_phrases_or conj_phrases_clauses advmod_proceeded_However nsubj_proceeded_work prep_work_on pobj_on_baseNPs aux_proceeded_has advmod_proceeded_essentially advmod_proceeded_always dep_state_proceeded prep_proceeded_via amod_extraction_algorithmic pobj_via_extraction prep_proceeded_from advmod_parsed_fully amod_corpora_parsed pobj_from_corpora dep_as_such prep_corpora_as det_Treebank_the nn_Treebank_Penn pobj_as_Treebank ' +o,3537,'We used the same 58 feature types as ',Ratnaparkhi,'nsubj_used_We det_types_the amod_types_same num_types_58 nn_types_feature dobj_used_types prep_used_as ' +o,3538,'context-free rules Charniak , Eisner context-free rules , headwords Charniak -LRB- 1997 -RRB- context-free rules , headwords , grandparent nodes context-free rules , headwords , grandparent nodes\\/rules , bigrams , two-level rules , two-level bigrams , nonheadwords Bod all fragments within parse trees Scope of Statistical Dependencies Model Figure 4 ',Collins,'amod__context-free nn__rules nn__Charniak nn__ nn__ dep_fragments_ nn_rules_Eisner nn_rules_ amod_rules_context-free appos__rules nn_Charniak_headwords appos__Charniak appos_Charniak_1997 amod_rules_context-free dep_Charniak_rules appos__headwords amod_nodes_grandparent appos__nodes number__ num_rules_ amod_rules_context-free dep_nodes_rules appos__headwords amod_nodes\\/rules_grandparent appos__nodes\\/rules appos__bigrams amod_rules_two-level appos__rules amod_bigrams_two-level appos__bigrams appos__nonheadwords nn__Bod dep_fragments_ det_fragments_all dep_Scope_fragments prep_fragments_within amod_trees_parse pobj_within_trees prep_Scope_of nn_Figure_Statistical nn_Figure_Dependencies nn_Figure_Model pobj_of_Figure dep_Scope_4 ' +o,3539,'I have made a preliminary analysis of the inventory of syntactic categories used in the tagging for labeling trees in the 18 Penn Treebank , comparing them to the categories used in CGEL ',Marcus,'nsubj_made_I aux_made_have det_analysis_a amod_analysis_preliminary dobj_made_analysis prep_analysis_of det_inventory_the pobj_of_inventory prep_inventory_of amod_categories_syntactic pobj_of_categories partmod_categories_used prep_used_in det_tagging_the pobj_in_tagging prep_tagging_for nn_trees_labeling pobj_for_trees prep_trees_in det__the num__18 nn__Penn nn__Treebank pobj_in_ xcomp_made_comparing dobj_comparing_them prep_comparing_to det_categories_the pobj_to_categories partmod_categories_used prep_used_in pobj_in_CGEL ' +o,3540,'For all performance metrics , we show the 70 \% confidence interval with respect to the MAP baseline computed using bootstrap resampling ',Och,'prep_show_For det_metrics_all nn_metrics_performance pobj_For_metrics nsubj_show_we det_interval_the dep_\%_70 amod_interval_\% nn_interval_confidence dobj_show_interval prep_interval_with pobj_with_respect prep_show_to det_baseline_the nn_baseline_MAP pobj_to_baseline partmod_baseline_computed xcomp_computed_using nn_resampling_bootstrap dobj_using_resampling ' +o,3541,',(Brown et al. , 1992)).',Brown,'' +n,3542,'By 17 0 10 20 30 40 50 60 70 80 90 100 10000 100000 1e+06 1e+07 Test Set Items with Translations (\%) Training Corpus Size (num words) unigrams bigrams trigrams 4-grams Figure 1: Percent of unique unigrams, bigrams, trigrams, and 4-grams from the Europarl Spanish test sentences for which translations were learned in increasingly large training corpora increasing the size of the basic unit of translation, phrase-based machine translation does away with many of the problems associated with the original word-based formulation of statistical machine translation (Brown et al. , 1993).',Brown,'' +o,3543,'They use a conditional model , based on Collins , which , as the authors acknowledge , has a number of theoretical deficiencies ; thus the results of Clark et al provide a useful baseline for the new models presented here ',Berger,'nsubj_use_They det_model_a amod_model_conditional dobj_use_model partmod_model_based prep_based_on nn__Collins pobj_on_ nsubj_has_which mark_acknowledge_as det_authors_the nsubj_acknowledge_authors parataxis_has_acknowledge rcmod_model_has det_number_a dobj_has_number prep_number_of amod_deficiencies_theoretical pobj_of_deficiencies advmod_provide_thus det_results_the nsubj_provide_results prep_results_of pobj_of_Clark cc_Clark_et conj_Clark_al parataxis_use_provide det_baseline_a amod_baseline_useful dobj_provide_baseline prep_baseline_for det_models_the amod_models_new pobj_for_models partmod_models_presented advmod_presented_here ' +o,3544,'We report BLEU scores on untokenized , recapitalized output ',Papineni,'nsubj_report_We amod_scores_BLEU nsubj__scores ccomp_report_ prep__on amod_output_untokenized amod_output_recapitalized pobj_on_output ' +o,3545,'This implementation is exactly the one proposed in , and we will denote it as MB-D hereafter ',Yarowsky,'det_implementation_This nsubj_one_implementation cop_one_is advmod_one_exactly det_one_the partmod_one_proposed prep_proposed_in pobj_in_ cc_one_and nsubj_denote_we aux_denote_will conj_one_denote dobj_denote_it prep_denote_as pobj_as_MB-D advmod_denote_hereafter ' +o,3546,'We have also used TPTs to encode n-gram count databases such as the Google 1T web n-gram database (Brants and Franz, 2006), but are not able to provide detailed results within the space limitations of this paper.4 5.1 Perplexity computation with 5-gram language models We compared the performance of TPT-encoded language models against three other language model implementations: the SRI language modeling toolkit (Stolcke, 2002), IRSTLM (Federico and Cettolo, 2007), and the language model implementation currently used in the Portage SMT system (Badr et al., 2007), which uses a pointer-based implementation but is able to perform fast LM filtering at load time.',Brants,'' +o,3547,'For example , Och reported that the quality of MT results was improved by using automatic MT evaluation measures for the parameter tuning of an MT system ',Och,'prep_reported_For pobj_For_example nsubj_reported_Och complm_improved_that det_quality_the nsubjpass_improved_quality prep_quality_of nn_results_MT pobj_of_results auxpass_improved_was ccomp_reported_improved prep_improved_by pcomp_by_using amod_measures_automatic nn_measures_MT nn_measures_evaluation dobj_using_measures prep_measures_for det_tuning_the nn_tuning_parameter pobj_for_tuning prep_tuning_of det_system_an nn_system_MT pobj_of_system ' +o,3548,'The user can select characters by their frequencies -LRB- ie - f and - g options -RRB- , the top or bottom N \% -LRB- ie - m and - n options -RRB- , their ranks -LRB- ie - r and - s options -RRB- and by their frequencies above two standard deviations phlS the mean -LRB- ie - z option -RRB- ',Smadja,'det_user_The nsubj_select_user aux_select_can dobj_select_characters prep_characters_by poss_frequencies_their pobj_by_frequencies dep_frequencies_ie dep_ie_f cc_f_and nn_options_g conj_f_options det_top_the num_\%_top cc_top_or amod_N_bottom conj_top_N appos_frequencies_\% dep_\%_ie dep_ie_m cc_m_and nn_options_n conj_m_options poss_ranks_their appos_\%_ranks dep_ranks_ie dep_ie_r cc_r_and nn_options_s conj_r_options cc_by_and conj_by_by poss_frequencies_their pobj_by_frequencies prep_frequencies_above num_phlS_two amod_phlS_standard nn_phlS_deviations pobj_above_phlS det__the amod__mean dep_by_ appos__ie amod_option_z dep_ie_option ' +o,3549,'In this paper , we propose an alignment algorithm between English and Korean conceptual units -LRB- or between English and Korean term constituents -RRB- in English-Korean technical term pairs based on IBM Model ',Brown,'prep_propose_In det_paper_this pobj_In_paper nsubj_propose_we det_algorithm_an amod_algorithm_alignment dobj_propose_algorithm prep_algorithm_between amod_units_English cc_English_and conj_English_Korean amod_units_conceptual pobj_between_units dep_-LRB-_or dep_or_between amod_constituents_English cc_English_and conj_English_Korean nn_constituents_term pobj_between_constituents prep_propose_in amod_pairs_English-Korean amod_pairs_technical nn_pairs_term pobj_in_pairs partmod_pairs_based prep_based_on nn_Model_IBM pobj_on_Model ' +o,3550,'222 ENGLISH TRAINING DATA For training in the English experiments , we used WSJ ',Marcus,'nn_DATA_222 nn_DATA_ENGLISH nn_DATA_TRAINING prep_used_For pcomp_For_training prep_training_in det_experiments_the amod_experiments_English pobj_in_experiments nsubj_used_we dep_DATA_used dobj_used_WSJ ' +o,3551,'The Decision List -LRB- DL -RRB- algorithm is described in ',Yarowsky,'det_algorithm_The amod_algorithm_Decision nn_algorithm_List abbrev_algorithm_DL nsubjpass_described_algorithm auxpass_described_is prep_described_in ' +o,3552,'Much work has been performed on learning to identify and classify polarity terms -LRB- ie , terms expressing a positive sentiment -LRB- eg , happy -RRB- or a negative sentiment -LRB- eg , terrible -RRB- -RRB- and exploiting them to do polarity classification -LRB- eg , Hatzivassiloglou and McKeown , , Kim and Hovy -LRB- 2004 -RRB- , Whitelaw et al ',Turney,'amod_work_Much nsubjpass_performed_work aux_performed_has auxpass_performed_been prep_performed_on pcomp_on_learning aux_identify_to xcomp_learning_identify cc_identify_and conj_identify_classify amod_terms_polarity dobj_identify_terms dep_terms_ie appos_ie_terms partmod_terms_expressing det_sentiment_a amod_sentiment_positive dobj_expressing_sentiment appos_sentiment_eg dep_eg_happy cc_sentiment_or det_sentiment_a amod_sentiment_negative conj_sentiment_sentiment dep_sentiment_eg dep_eg_terrible cc_learning_and conj_learning_exploiting nsubj_do_them aux_do_to xcomp_exploiting_do amod_classification_polarity dobj_do_classification dep_classification_eg dep_eg_Hatzivassiloglou cc_classification_and nn__McKeown conj_classification_ conj__ conj__Kim cc__and conj__Hovy appos__2004 appos__Whitelaw cc_Whitelaw_et conj_Whitelaw_al ' +o,3553,'It is potentially useful in other natural language processing tasks , such as the problem of estimating n-gram models or the problem of semantic tagging ',Brown,'nsubj_useful_It cop_useful_is advmod_useful_potentially prep_useful_in amod_tasks_other amod_tasks_natural nn_tasks_language nn_tasks_processing pobj_in_tasks dep_as_such prep_tasks_as det_problem_the pobj_as_problem prep_problem_of pcomp_of_estimating amod_models_n-gram dobj_estimating_models advmod_estimating_ cc_problem_or det_problem_the conj_problem_problem prep_problem_of amod__semantic amod__tagging pobj_of_ ' +o,3554,'However , the maximum entropy was found to yield higher accuracy than nave Bayes in a subsequent comparison by , who used a different subset of either Senseval-1 or Senseval-2 English lexical sample data ',Klein,'advmod_found_However det__the amod__maximum nn__entropy nsubjpass_found_ auxpass_found_was aux_yield_to xcomp_found_yield amod_accuracy_higher dobj_yield_accuracy prep_yield_than amod_Bayes_nave pobj_than_Bayes prep_Bayes_in det_comparison_a amod_comparison_subsequent pobj_in_comparison prep_yield_by pobj_by_ nsubj_used_who rcmod__used det_subset_a amod_subset_different dobj_used_subset prep_subset_of preconj_Senseval-1_either amod_data_Senseval-1 cc_Senseval-1_or conj_Senseval-1_Senseval-2 amod_data_English amod_data_lexical nn_data_sample pobj_of_data ' +o,3555,'Automatic approaches to creating a semantic orientation lexicon and , more generally , approaches for word-level sentiment annotation can be grouped into two kinds : -LRB- 1 -RRB- those that rely on manually created lexical resourcesmost of which use WordNet ; and -LRB- 2 -RRB- those that rely on text corpora ',Kanayama,'nsubj_approaches_Automatic prep_approaches_to pcomp_to_creating det_lexicon_a amod_lexicon_semantic nn_lexicon_orientation dobj_creating_lexicon cc_approaches_and dep_approaches_more dep_more_generally nsubjpass_grouped_approaches prep_approaches_for amod_annotation_word-level nn_annotation_sentiment pobj_for_annotation aux_grouped_can auxpass_grouped_be conj_approaches_grouped prep_grouped_into num_kinds_two pobj_into_kinds dep_created_1 nsubj_created_those nsubj_rely_that rcmod_those_rely prt_rely_on advmod_rely_manually parataxis_approaches_created amod_resourcesmost_lexical dobj_created_resourcesmost prep_resourcesmost_of nsubj_use_which pcomp_of_use nn__WordNet dobj_use_ cc_approaches_and dep__2 nsubj__those nsubj_rely_that rcmod_those_rely prep_rely_on nn_corpora_text pobj_on_corpora conj_approaches_ ' +o,3556,'Recent work has applied Bayesian non-parametric models to anaphora resolution , lexical acquisition and language modeling with good results ',Teh,'amod_work_Recent nsubj_applied_work aux_applied_has amod_models_Bayesian amod_models_non-parametric dobj_applied_models prep_applied_to nn__anaphora nn__resolution pobj_to_ amod__lexical nn__acquisition conj__ cc__and nn__language nn__modeling conj__ prep_applied_with amod_results_good pobj_with_results ' +p,3557,' successfully used this observation as an approximate annotation technique in an unsupervised WSD model ',Yarowsky,'nsubj_used_ advmod_used_successfully det_observation_this dobj_used_observation prep_used_as det_technique_an amod_technique_approximate nn_technique_annotation pobj_as_technique prep_technique_in det_model_an amod_model_unsupervised nn_model_WSD pobj_in_model ' +p,3558,'So far , SCL has been applied successfully in NLP for Part-of-Speech tagging and Sentiment Analysis ',Blitzer,'dep_far_So advmod_applied_far nsubjpass_applied_SCL aux_applied_has auxpass_applied_been advmod_applied_successfully prep_applied_in pobj_in_NLP prep_NLP_for amod_tagging_Part-of-Speech pobj_for_tagging cc_tagging_and nn_Analysis_Sentiment conj_tagging_Analysis ' +p,3559,'4 Experiments Phrase-based SMT systems have been shown to outperform word-based approaches ',Koehn,'num_Experiments_4 measure_Phrase-based_Experiments amod_systems_Phrase-based nn_systems_SMT nsubjpass_shown_systems aux_shown_have auxpass_shown_been aux_outperform_to xcomp_shown_outperform amod_approaches_word-based dobj_outperform_approaches ' +o,3560,': there is : want to : need not : in front of : as soon as : look at Figure 2 : Examples of entries from the manually developed dictionary 4 Experimental Setting 41 Evaluation The intrinsic quality of word alignment can be assessed using the Alignment Error Rate -LRB- AER -RRB- metric , that compares a systems alignment output to a set of gold-standard alignment ',Och,'expl_is_there dep_is_want prep_want_to dep_is_need neg_need_not prep_is_in pobj_in_front prep_front_of dep_of_as dep_of_soon dep_of_as dep_front_look prep_look_at pobj_at_Figure num_Figure_2 dep_front_Examples prep_Examples_of pobj_of_entries prep_entries_from det_manually_the pobj_from_manually partmod_front_developed nn_Experimental_dictionary num_Experimental_4 dobj_developed_Experimental partmod_Experimental_Setting num_Evaluation_41 dobj_Setting_Evaluation det_quality_The amod_quality_intrinsic nsubjpass_assessed_quality prep_quality_of nn_alignment_word pobj_of_alignment aux_assessed_can auxpass_assessed_be rcmod_Evaluation_assessed purpcl_assessed_using det__the nn_Rate_Alignment nn_Rate_Error nn__Rate abbrev_Rate_AER amod__metric dobj_using_ nsubj_compares_that rcmod__compares det_output_a nn_output_systems nn_output_alignment dobj_compares_output prep_compares_to det_set_a pobj_to_set prep_set_of amod_alignment_gold-standard pobj_of_alignment ' +o,3561,'If distributional similarity is conceived of as substitutability , as and Lee -LRB- 1999 -RRB- emphasize , then asymmetries arise when one word appears in a subset of the contexts in which the other appears ; for example , the adjectives that typically modify apple are a subset of those that modify fruit , sofruit substitutes for apple better than apple substitutes for fruit ',Weeds,'mark_conceived_If amod_similarity_distributional nsubjpass_conceived_similarity auxpass_conceived_is advcl_arise_conceived dep_as_of prep_conceived_as pobj_as_substitutability mark_emphasize_as nsubj_emphasize_ cc__and conj__Lee appos__1999 dep_substitutability_emphasize advmod_arise_then nsubj_arise_asymmetries advmod_appears_when num_word_one nsubj_appears_word advcl_arise_appears prep_appears_in det_subset_a pobj_in_subset prep_subset_of det_contexts_the pobj_of_contexts rel_appears_in pobj_in_which det_other_the nsubj_appears_other rcmod_contexts_appears prep_subset_for pobj_for_example det_adjectives_the nsubj_subset_adjectives nsubj_modify_that advmod_modify_typically rcmod_adjectives_modify dobj_modify_apple cop_subset_are det_subset_a parataxis_arise_subset prep_subset_of pobj_of_those nsubj_modify_that rcmod_those_modify dobj_modify_fruit amod_substitutes_sofruit appos_fruit_substitutes prep_substitutes_for pobj_for_apple advmod_modify_better prep_modify_than nn_substitutes_apple pobj_than_substitutes prep_substitutes_for pobj_for_fruit ' +o,3562,'So unlike some other studies , we used manually annotated alignments instead of automatically generated ones ',Zhang,'cc_used_So prep_used_unlike det__some amod__other nn__studies pobj_unlike_ nsubj_used_we advmod_used_manually amod_alignments_annotated dobj_used_alignments cc_used_instead dep_instead_of advmod_generated_automatically conj_used_generated dobj_generated_ones ' +o,3563,'The HWC metrics compare dependency and constituency trees for both reference and machine translations ',Liu,'det_metrics_The nn_metrics_HWC nsubj_compare_metrics nn_trees_dependency cc_dependency_and conj_dependency_constituency dobj_compare_trees prep_compare_for preconj_translations_both nn_translations_reference cc_reference_and conj_reference_machine pobj_for_translations ' +o,3564,'The inclusion of phrases longer than three words in translation resources has been avoided , as it has been shown not to have a strong impact on translation performance ',Koehn,'det_inclusion_The nsubjpass_avoided_inclusion prep_inclusion_of pobj_of_phrases advmod_phrases_longer dep_longer_than num_words_three pobj_than_words prep_words_in nn_resources_translation pobj_in_resources aux_avoided_has auxpass_avoided_been mark_shown_as nsubjpass_shown_it aux_shown_has auxpass_shown_been advcl_avoided_shown neg_have_not aux_have_to xcomp_shown_have det_impact_a amod_impact_strong dobj_have_impact prep_impact_on nn_performance_translation pobj_on_performance ' +o,3565,'2 Related work Our approach for emotion classification is based on the idea of and is similar to those of and ',Turney,'num_work_2 amod_work_Related poss_approach_Our nsubjpass_based_approach prep_approach_for nn_classification_emotion pobj_for_classification auxpass_based_is dep_work_based prep_based_on det_idea_the pobj_on_idea prep_idea_of pobj_of_ cc_based_and cop_similar_is conj_based_similar prep_similar_to pobj_to_those prep_those_of nn_and_ pobj_of_and ' +o,3566,'As expected , we see that MST does better than Malt for all categories except nouns and pronouns ',McDonald,'mark_expected_As advcl_see_expected nsubj_see_we complm_does_that nsubj_does_MST ccomp_see_does advmod_does_better prep_does_than pobj_than_Malt prep_Malt_for det_categories_all pobj_for_categories prep_categories_except pobj_except_nouns cc_nouns_and conj_nouns_pronouns ' +o,3567,'1 Introduction The probabilistic relation between verbs and their arguments plays an important role in modern statistical parsers and supertaggers , and in psychological theories of language processing ',Collins,'num_Introduction_1 det_relation_The amod_relation_probabilistic nsubj_plays_relation prep_relation_between pobj_between_verbs cc_verbs_and poss_arguments_their conj_verbs_arguments rcmod_Introduction_plays det_role_an amod_role_important dobj_plays_role prep_role_in amod_parsers_modern amod_parsers_statistical pobj_in_parsers cc_parsers_and amod__supertaggers conj_parsers_ cc_Introduction_and conj_Introduction_in amod_theories_psychological pobj_in_theories prep_theories_of nn__language nn__processing pobj_of_ ' +o,3568,'Moreover , the overall BLEU and METEOR scores , as well as numbers of exact string matches -LRB- as measured against to the original sentences in the CCGbank -RRB- are higher for the hypertagger-seeded realizer than for the preexisting realizer ',Papineni,'advmod_higher_Moreover det__the amod__overall amod__BLEU nsubj_higher_ cc__and amod_scores_METEOR amod_scores_ conj__scores dep_well_as cc__well dep_well_as conj__numbers prep_numbers_of amod_matches_exact nn_matches_string pobj_of_matches mark_measured_as dep__measured prep_measured_against dep_against_to det_sentences_the amod_sentences_original pobj_to_sentences prep_sentences_in det_CCGbank_the pobj_in_CCGbank cop_higher_are prep_higher_for det_realizer_the amod_realizer_hypertagger-seeded pobj_for_realizer dep_for_than prep_higher_for det_realizer_the amod_realizer_preexisting pobj_for_realizer ' +o,3569,'ald , 2008 -RRB- , and is also similar to the Pred baseline for domain adaptation in ','Daume III','prep_ald_2008 cc_ald_and cop_similar_is advmod_similar_also conj_ald_similar prep_similar_to det_baseline_the nn_baseline_Pred pobj_to_baseline prep_baseline_for nn_adaptation_domain pobj_for_adaptation prep_similar_in ' +o,3570,'21 Conditional Maximum Entropy Model The goal of CME is to find the most uniform conditional distribution of y given observation x , -LRB- -RRB- xyp , subject to constraints specified by a set of features -LRB- -RRB- yxf i , , where features typically take the value of either 0 or 1 ',Berger,'nsubj_is_21 nn_Model_Conditional nn_Model_Maximum nn_Model_Entropy dep_21_Model det_goal_The dep_21_goal prep_goal_of pobj_of_CME aux_find_to xcomp_is_find det_distribution_the advmod_uniform_most amod_distribution_uniform amod_distribution_conditional dobj_find_distribution prep_distribution_of pobj_of_y prep_find_given dep_given_observation dep_observation_x dep_observation_xyp amod_observation_subject prep_subject_to pobj_to_constraints partmod_observation_specified prep_specified_by det_set_a pobj_by_set prep_set_of pobj_of_features nn_-RRB-_yxf dep_-RRB-_i advmod_take_where nsubj_take_features advmod_take_typically rcmod_observation_take det_value_the dobj_take_value prep_value_of preconj_0_either pobj_of_0 cc_0_or conj_0_1 ' +p,3571,'21 The sparked considerable interest in bootstrapping with his successful method for word sense disambiguation ',Yarowsky,'det__The nsubj_sparked_ dep_21_sparked amod_interest_considerable dobj_sparked_interest prep_sparked_in pcomp_in_bootstrapping prep_bootstrapping_with poss_method_his amod_method_successful pobj_with_method prep_method_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +p,3572,'In statistical machine translation , IBM 1 ~ 5 models based on the source-chmmel model have been widely used and revised for many language donmins and applications ',Brown,'prep_used_In amod_translation_statistical nn_translation_machine pobj_In_translation nsubjpass_used_IBM num_IBM_1 partmod_IBM_~ num_models_5 dobj_~_models partmod_models_ prep__based dep_based_on det_model_the amod_model_source-chmmel pobj_on_model aux_used_have auxpass_used_been advmod_used_widely cc_used_and conj_used_revised prep_used_for amod_donmins_many nn_donmins_language pobj_for_donmins cc_donmins_and conj_donmins_applications ' +o,3573,'To tackle this problem , we defined 2The best results of -LRB- LR = 884 \% , LP = 891 \% and F = 888 \% -RRB- are achieved when the parser utilizes the information about the final punctuation and the look-ahead ',Collins,'aux_tackle_To dep_defined_tackle det_problem_this dobj_tackle_problem nsubj_defined_we amod_results_2The amod_results_best nsubjpass_achieved_results prep_results_of pobj_of_ dep_results_LR dep_\%_= num_\%_884 rcmod_LR_\% conj_LR_LP dep_\%_= num_\%_891 rcmod_LP_\% cc_LR_and conj_LR_F dep_\%_= num_\%_888 rcmod_F_\% auxpass_achieved_are ccomp_defined_achieved advmod_utilizes_when det_parser_the nsubj_utilizes_parser advcl_achieved_utilizes det_information_the dobj_utilizes_information prep_information_about det_punctuation_the amod_punctuation_final pobj_about_punctuation cc_punctuation_and det_look-ahead_the conj_punctuation_look-ahead ' +o,3574,'47 Fertility-Based Transducer In , three alignment models are described that include fertility models , these are IBM Models 3 , 4 , and 5 ',Brown,'num_Transducer_47 amod_Transducer_Fertility-Based prep_described_In pobj_In_ num_models_three amod_models_alignment nsubjpass_described_models auxpass_described_are ccomp_Models_described complm_include_that ccomp_described_include nn_models_fertility dobj_include_models nsubj_Models_these cop_Models_are nn_Models_IBM dep_Transducer_Models num_Models_3 num_Models_4 cc_Models_and conj_Models_5 ' +p,3575,'6 Discussion Noting that adding latent features to nonterminals in unlexicalized context-free parsing has been very successful , we were surprised not to see a 3Czech experiments were not done , since the number of features -LRB- more than 14 million -RRB- was too high to multiply out by clusters ',Matsuzaki,'nsubj_Noting_Discussion dep_6_Noting complm_successful_that csubj_successful_adding amod_features_latent dobj_adding_features prep_adding_to pobj_to_nonterminals prep_nonterminals_in amod_parsing_unlexicalized amod_parsing_context-free pobj_in_parsing aux_successful_has dep_successful_been advmod_successful_very ccomp_Noting_successful advmod_surprised_ nsubjpass_surprised_we auxpass_surprised_were dep_6_surprised neg_see_not aux_see_to xcomp_surprised_see det_experiments_a amod_experiments_3Czech nsubjpass_done_experiments auxpass_done_were neg_done_not ccomp_see_done mark_high_since det_number_the nsubj_high_number prep_number_of pobj_of_features dep_than_more quantmod_million_than number_million_14 dep_features_million cop_high_was advmod_high_too advcl_done_high aux_multiply_to xcomp_high_multiply prt_multiply_out prep_multiply_by pobj_by_clusters ' +o,3576,'Similarly to , the tree-to-string alignment templates discussed in this paper are actually transformation rules ',Galley,'advmod_rules_Similarly aux__to dep_rules_ det_templates_the amod_templates_tree-to-string nn_templates_alignment nsubj_rules_templates partmod_templates_discussed prep_discussed_in det_paper_this pobj_in_paper cop_rules_are advmod_rules_actually nn_rules_transformation ' +o,3577,'The hierarchical translation operations introduced in these methods call for extensions to the traditional beam decoder ',Koehn,'det_operations_The amod_operations_hierarchical nn_operations_translation nsubj_introduced_operations prep_introduced_in det_call_these nn_call_methods pobj_in_call prep_call_for pobj_for_extensions prep_introduced_to det_decoder_the amod_decoder_traditional nn_decoder_beam pobj_to_decoder ' +o,3578,'1http : \\/ \\/ wwwnistgov\\/speech\\/tests \\/ ace \\/ 49 Bootstrapping techniques have been used for such diverse NLP problems as : word sense disambiguation , named entity classification , IE pattern acquisition , document classification , fact extraction from the web and hyponymy relation extraction ',Yarowsky,'dep_wwwnistgov\\/speech\\/tests_1http nn_\\/_\\/ nsubj_wwwnistgov\\/speech\\/tests_\\/ amod_techniques_\\/ nn_techniques_ace nn_techniques_\\/ num_techniques_49 nn_techniques_Bootstrapping nsubjpass_used_techniques aux_used_have auxpass_used_been ccomp_wwwnistgov\\/speech\\/tests_used prep_used_for amod_problems_such amod_problems_diverse nn_problems_NLP pobj_for_problems prep_used_as nn__word nn__sense nn__disambiguation pobj_as_ amod__named nn__entity nn__classification nn__ dep__IE nn__pattern nn__acquisition nn__ dep__document nn__classification conj__ nn_extraction_fact conj__extraction prep_extraction_from det__the nn__web pobj_from_ cc__and amod__hyponymy nn__relation nn__extraction conj__ ' +o,3579,' used the one sense per collocation property as an essential ingredient for an unsupervised Word-SenseDisambiguationalgorithm ',Yarowsky,'nsubj_used_ det_sense_the num_sense_one dobj_used_sense prep_sense_per nn_property_collocation pobj_per_property prep_used_as det_ingredient_an amod_ingredient_essential pobj_as_ingredient prep_ingredient_for det_Word-SenseDisambiguationalgorithm_an amod_Word-SenseDisambiguationalgorithm_unsupervised pobj_for_Word-SenseDisambiguationalgorithm ' +o,3580,'In order to extract the linguistic features necessary for the model , all sentences were first automatically part-of-speech-tagged using a maximum entropy tagger and parsed using the Collins parser ',Collins,'mark_extract_In dep_extract_order aux_extract_to dep_part-of-speech-tagged_extract det_features_the amod_features_linguistic nsubj_necessary_features xcomp_extract_necessary prep_necessary_for det_model_the pobj_for_model det_sentences_all nsubjpass_part-of-speech-tagged_sentences auxpass_part-of-speech-tagged_were advmod_part-of-speech-tagged_first advmod_part-of-speech-tagged_automatically xcomp_part-of-speech-tagged_using det__a amod__maximum amod__entropy nn__tagger dobj_using_ cc_part-of-speech-tagged_and conj_part-of-speech-tagged_parsed xcomp_parsed_using det_parser_the nn_parser_Collins dobj_using_parser ' +p,3581,'1 Introduction During the last four years , various implementations and extentions to phrase-based statistical models have led to significant increases in machine translation accuracy ',Koehn,'dep_led_1 dep_During_Introduction prep_led_During det_years_the amod_years_last num_years_four pobj_During_years amod_implementations_various nsubj_led_implementations cc_implementations_and conj_implementations_extentions prep_implementations_to amod__phrase-based amod__statistical nn__models pobj_to_ aux_led_have prep_led_to amod_increases_significant pobj_to_increases prep_increases_in nn_accuracy_machine nn_accuracy_translation pobj_in_accuracy ' +o,3582,'2 Previous Work Other researchers have investigated the topic of automatic generation of abstracts , but the focus has been different , eg , sentence extraction , processing of structured templates , sentence compression , and generation of abstracts from multiple sources ',Collins,'num_researchers_2 amod_researchers_Previous nn_researchers_Work amod_researchers_Other nsubj_investigated_researchers aux_investigated_have det_topic_the dobj_investigated_topic prep_topic_of amod_generation_automatic pobj_of_generation prep_generation_of pobj_of_abstracts cc_investigated_but det_focus_the nsubj_been_focus aux_been_has conj_investigated_been amod__different dep__eg dep__sentence nn__extraction dobj_been_ appos__processing prep_processing_of amod__structured nn__templates nn__ dep__sentence nn__compression pobj_of_ cc_investigated_and nsubj__generation prep_generation_of pobj_of_abstracts prep_abstracts_from amod_sources_multiple pobj_from_sources conj_investigated_ ' +o,3583,'The learning algorithm for level-0 dependency is similar to the guided learning algorithm for labeling as described in ',Shen,'det_algorithm_The nn_algorithm_learning nsubj_similar_algorithm prep_algorithm_for amod_dependency_level-0 pobj_for_dependency cop_similar_is prep_similar_to det_guided_the pobj_to_guided partmod_guided_learning dobj_learning_algorithm prep_learning_for pcomp_for_labeling advmod_described_as acomp_labeling_described dep_described_in ' +o,3584,'Semantic -LRB- 1 -RRB- : The named entity -LRB- NE -RRB- tag of wi obtained using the Stanford CRF-based NE recognizer ',Finkel,'appos_Semantic_1 det_tag_The amod_tag_named nn_tag_entity abbrev_tag_NE nsubj_obtained_tag prep_tag_of pobj_of_wi dep_Semantic_obtained xcomp_obtained_using det_recognizer_the nn_recognizer_Stanford nn_recognizer_CRF-based nn_recognizer_NE dobj_using_recognizer ' +o,3585,'To make sense tagging more precise , it is advisable to place constraint on the translation counterpart c of w SWAT considers only those translations c that has been linked with w based the Competitive Linking Algorithm and logarithmic likelihood ratio ',Dunning,'aux_make_To dep_advisable_make dobj_make_sense partmod_sense_tagging advmod_precise_more acomp_tagging_precise nsubj_advisable_it cop_advisable_is aux_place_to xcomp_advisable_place dobj_place_constraint prep_place_on det_counterpart_the nn_counterpart_translation pobj_on_counterpart nsubj_considers_c prep_c_of nn_SWAT_w pobj_of_SWAT rcmod_counterpart_considers prep_considers_only det_translations_those dep_only_translations dep_translations_c nsubjpass_linked_that aux_linked_has auxpass_linked_been rcmod_translations_linked prep_linked_with pobj_with_w partmod_w_based det__the nn__Competitive nn__Linking nn__Algorithm dobj_based_ cc__and amod_ratio_logarithmic nn_ratio_likelihood conj__ratio ' +o,3586,'In summary , the strength of our approach is to exploit extremely precise structural clues , and to use 5 Semantic Orientation in ',Turney,'prep_is_In pobj_In_summary det_strength_the nsubj_is_strength prep_strength_of poss_approach_our pobj_of_approach aux_exploit_to ccomp_is_exploit advmod_precise_extremely amod_clues_precise amod_clues_structural dobj_exploit_clues cc_exploit_and aux_use_to conj_exploit_use num_Orientation_5 nn_Orientation_Semantic dobj_use_Orientation prep_Orientation_in ' +o,3587,'Various learning models have been studied such as Hidden Markov models -LRB- HMMs -RRB- , decision trees and maximum entropy models ',Berger,'amod_models_Various nn_models_learning nsubjpass_studied_models aux_studied_have auxpass_studied_been dep_as_such prep_studied_as nn_models_Hidden nn_models_Markov pobj_as_models abbrev_models_HMMs dep_models_ nn__decision nn__trees conj_models_ cc_models_and amod_models_maximum nn_models_entropy conj_models_models ' +o,3588,'Phrase-pairs are then extracted from the word alignments ',Koehn,'nsubjpass_extracted_Phrase-pairs auxpass_extracted_are advmod_extracted_then prep_extracted_from det_alignments_the nn_alignments_word pobj_from_alignments ' +o,3589,'Since this transform takes a probabilistic grammar as input , it can also easily accommodate horizontal and vertical Markovisation -LRB- annotating grammar symbols with parent and sibling categories -RRB- as described by and subsequently ',Collins,'mark_takes_Since det_transform_this nsubj_takes_transform advcl_accommodate_takes det_grammar_a amod_grammar_probabilistic dobj_takes_grammar prep_grammar_as pobj_as_input nsubj_accommodate_it aux_accommodate_can advmod_accommodate_also advmod_accommodate_easily nn_Markovisation_horizontal cc_horizontal_and conj_horizontal_vertical dobj_accommodate_Markovisation amod_symbols_annotating nn_symbols_grammar dep_Markovisation_symbols prep_symbols_with nn_categories_parent cc_parent_and conj_parent_sibling pobj_with_categories advmod_described_as amod_Markovisation_described prep_accommodate_by pobj_by_ cc__and conj__subsequently ' +o,3590,'Aware of this problem , Resnik and suggest creating the sense distance matrix based on results in experimental psychology such as Miller and Charles or Resnik ',Yarowsky,'nsubj_suggest_Aware prep_Aware_of det_problem_this pobj_of_problem conj_problem_Resnik cc_problem_and conj_problem_ xcomp_suggest_creating det_matrix_the nn_matrix_sense nn_matrix_distance dobj_creating_matrix prep_creating_based dep_based_on pobj_on_results prep_results_in amod_psychology_experimental pobj_in_psychology dep_as_such prep_psychology_as nn__Miller cc_Miller_and conj_Miller_Charles pobj_as_ cc__or nn__Resnik conj__ ' +o,3591,' observe that their predominant sense method is not performing as well for 3We use the Lesk -LRB- overlap -RRB- similarity as implemented by the WordNet : : similarity package ',Pedersen,'advmod_observe_ complm_performing_that poss_method_their amod_method_predominant nn_method_sense nsubj_performing_method aux_performing_is neg_performing_not ccomp_observe_performing advmod_well_as advmod_performing_well prep_performing_for amod_use_3We pobj_for_use det_similarity_the nn_similarity_Lesk appos_similarity_overlap nsubj_observe_similarity mark_implemented_as dep_similarity_implemented prep_implemented_by det_WordNet_the pobj_by_WordNet nn_package_similarity dep_similarity_package ' +o,3592,'Using thesaurus categories directly as a coarse sense division may seem to be a viable alternative ',Yarowsky,'nn_categories_thesaurus dobj_Using_categories advmod_Using_directly mark_seem_as det_division_a amod_division_coarse nn_division_sense nsubj_seem_division aux_seem_may advcl_Using_seem aux_alternative_to cop_alternative_be det_alternative_a amod_alternative_viable xcomp_seem_alternative ' +o,3593,'Similar to BLEU score , we also use the similar Brevity Penalty BP to penalize the short translations in computing RAcc ',Papineni,'ccomp_use_Similar prep_Similar_to amod_score_BLEU pobj_to_score nsubj_use_we advmod_use_also det__the amod__similar nn__Brevity nn__Penalty nn__BP nsubj_penalize_ aux_penalize_to xcomp_use_penalize det_translations_the amod_translations_short dobj_penalize_translations prep_translations_in amod_RAcc_computing pobj_in_RAcc ' +o,3594,'Collins et al proposed two algorithms for NER by modifying Yarowskys method and the framework suggested by ',Yarowsky,'nn__Collins cc_Collins_et conj_Collins_al nsubj_proposed_ num_algorithms_two dobj_proposed_algorithms prep_algorithms_for pobj_for_NER prep_proposed_by amod__modifying nn__Yarowskys nn__method pobj_by_ cc__and det_framework_the conj__framework partmod_framework_suggested prep_suggested_by pobj_by_ ' +p,3595,'RANDLM performs well and scaled to the full data with improvement -LRB- resulting in our best overall system -RRB- ',Talbot,'nn__RANDLM nsubj_performs_ advmod_performs_well cc_performs_and conj_performs_scaled prep_scaled_to det_data_the amod_data_full pobj_to_data prep_data_with pobj_with_improvement dep_data_resulting prep_resulting_in poss_system_our amod_system_best amod_system_overall pobj_in_system ' +o,3596,'As far as we know , language modeling always improves with additional training data , so we add data from the North American News Text Corpus -LRB- NANC -RRB- automatically parsed with the Charniak parser to train our language model on up to 20 million additional words ',McClosky,'advmod_far_As mark_know_as nsubj_know_we dep_far_know nn_modeling_language nsubj_improves_modeling advmod_improves_always parataxis_know_improves prep_improves_with amod_data_additional nn_data_training pobj_with_data mark_add_so nsubj_add_we advcl_know_add dobj_add_data prep_data_from det_Corpus_the nn_Corpus_North nn_Corpus_American nn_Corpus_News nn_Corpus_Text pobj_from_Corpus abbrev_data_NANC dep_data_ advmod_parsed_automatically partmod__parsed prep_parsed_with det__the nn__Charniak nn__parser pobj_with_ aux_train_to xcomp_parsed_train poss_model_our nn_model_language dobj_train_model prep_train_on quantmod_million_up dep_million_to number_million_20 num_words_million amod_words_additional pobj_on_words ' +o,3597,'The problem itself has started to get attention only recently ',Blitzer,'det_problem_The nsubj_started_itself aux_started_has dep_problem_started aux_get_to xcomp_started_get dobj_get_attention advmod_recently_only advmod_get_recently ' +o,3598,'42 Models with Prior Distributions Minimum discrimination information models are exponential models with a prior distribution q -LRB- y x -RRB- : p -LRB- y x -RRB- = q -LRB- y x -RRB- exp -LRB- summationtextF i = 1 ifi -LRB- x , y -RRB- -RRB- Z -LRB- x -RRB- -LRB- 14 -RRB- The central issue in performance prediction for MDI models is whether q -LRB- y x -RRB- needs to be accounted for ',Brown,'num_Models_42 nsubj_models_Models prep_Models_with amod__Prior nn__Distributions nn__Minimum nn__discrimination nn__information nn__models pobj_with_ cop_models_are amod_models_exponential prep_models_with det_q_a amod_q_prior nn_q_distribution pobj_with_q number_x_y dep_q_x dobj_models_p number_x_y dep_p_x dep_p_= nn_x_q nn_x_y iobj_=_x nn_Z_exp nn_i_summationtextF nsubj_=_i dep_Z_= num_ifi_1 dobj_=_ifi dep_ifi_x appos_x_y dobj_=_Z appos_Z_x appos_Z_14 det_issue_The amod_issue_central nsubj_is_issue prep_issue_in nn_prediction_performance pobj_in_prediction prep_prediction_for nn_models_MDI pobj_for_models dep_models_is complm_needs_whether nsubj_needs_q number_x_y dep_q_x ccomp_is_needs aux_accounted_to auxpass_accounted_be xcomp_needs_accounted prep_accounted_for ' +o,3599,'The most commonly used MT evaluation metric in recent years has been IBM ? s Bleu metric ',Papineni,'det_metric_The advmod_used_most advmod_used_commonly amod_metric_used nn_metric_MT nn_metric_evaluation nsubj_been_metric prep_metric_in amod_years_recent pobj_in_years aux_been_has dobj_been_IBM dep_metric_s advmod_metric_Bleu rcmod_IBM_metric ' +o,3600,' propose a new metric that extends n-gram matching to include synonyms and paraphrases ; and Lavie ? s METEOR metric can be used with additionalknowledgesuchasWordNetinordertosupport inexact lexical matches ',Banerjee,'advmod_propose_ det_metric_a amod_metric_new dobj_propose_metric nsubj_extends_that dep_propose_extends amod_matching_n-gram dobj_extends_matching aux_include_to infmod_matching_include dobj_include_synonyms cc_synonyms_and conj_synonyms_paraphrases cc_extends_and amod_s_Lavie dobj_used_s amod__METEOR amod__metric nsubjpass_used_ aux_used_can auxpass_used_be conj_extends_used prep_used_with amod_lexical_additionalknowledgesuchasWordNetinordertosupport nn_lexical_inexact pobj_with_lexical dep_lexical_matches ' +o,3601,'We follow the work of and choose the hypothesis that best agrees with other hypotheses on average as the backbone by applying Minimum Bayes Risk -LRB- MBR -RRB- decoding ',Rosti,'nsubj_follow_We det_work_the dobj_follow_work prep_work_of pobj_of_ cc_follow_and conj_follow_choose det_hypothesis_the dobj_choose_hypothesis complm_agrees_that nsubj_agrees_best ccomp_choose_agrees prep_agrees_with amod_hypotheses_other pobj_with_hypotheses prep_hypotheses_on pobj_on_average prep_average_as det_backbone_the pobj_as_backbone prep_agrees_by pcomp_by_applying nn_Risk_Minimum nn_Risk_Bayes nsubj_decoding_Risk abbrev_Risk_MBR dep_applying_decoding dobj_decoding_ ' +n,3602,'2 21 Word Alignment Adaptation Bi-directional Word Alignment In statistical translation models , only one-to-one and more-to-one word alignment links can be found ',Brown,'number_21_2 num_Alignment_21 nn_Alignment_Word nn_Alignment_Alignment nn_Alignment_Adaptation nn_Alignment_Bi-directional nn_Alignment_Word dep_Alignment_In amod_models_statistical nn_models_translation pobj_In_models nsubjpass_found_ advmod__only amod_links_one-to-one cc_one-to-one_and conj_one-to-one_more-to-one nn_links_word nn_links_alignment dep__links aux_found_can auxpass_found_be dep_Alignment_found ' +o,3603,'Starting from the list of 12 ambiguous words provided by which is shown in table 2 , we created a concordance for each word , with the lines in the concordances each relating to a context window of 20 words ',Yarowsky,'prep_created_Starting dep_Starting_from det_list_the pobj_from_list prep_list_of num_words_12 amod_words_ambiguous pobj_of_words partmod_words_provided prep_provided_by pobj_by_ nsubjpass_shown_which auxpass_shown_is rcmod__shown prep_shown_in pobj_in_table num_table_2 nsubj_created_we det_concordance_a dobj_created_concordance prep_concordance_for det_word_each pobj_for_word prep_created_with det_lines_the nsubj_relating_lines prep_lines_in det_concordances_the pobj_in_concordances dep_concordances_each pcomp_with_relating prep_relating_to det_window_a nn_window_context pobj_to_window prep_window_of num_words_20 pobj_of_words ' +p,3604,'Bleu is fast and easy to run , and it can be used as a target function in parameter optimization training procedures that are commonly used in state-of-the-art statistical MT systems ',Och,'nsubj_fast_Bleu cop_fast_is cc_fast_and conj_fast_easy aux_run_to xcomp_fast_run cc_fast_and nsubjpass_used_it aux_used_can auxpass_used_be conj_fast_used prep_used_as det_function_a nn_function_target pobj_as_function prep_function_in nn_procedures_parameter nn_procedures_optimization nn_procedures_training pobj_in_procedures nsubjpass_used_that auxpass_used_are advmod_used_commonly rcmod_function_used prep_used_in amod_systems_state-of-the-art amod_systems_statistical nn_systems_MT pobj_in_systems ' +o,3605,'It has been shown repeatedly -- eg , Briscoe and Carroll , Charniak , , Inui et al ',Collins,'nsubjpass_shown_It aux_shown_has auxpass_shown_been advmod_shown_repeatedly dep_shown_eg nn__Briscoe cc_Briscoe_and conj_Briscoe_Carroll appos_eg_ nn__Charniak appos_eg_ nn__ appos_eg_ conj__Inui cc__et conj__al ' +o,3606,'When efficient techniques have been proposed , they have been mostly evaluated on safe pairs of languages where the notion of word is rather clear ',Dunning,'advmod_proposed_When amod_techniques_efficient nsubjpass_proposed_techniques aux_proposed_have auxpass_proposed_been dep_evaluated_proposed dep_proposed_ nsubjpass_evaluated_they aux_evaluated_have auxpass_evaluated_been advmod_evaluated_mostly prep_evaluated_on amod_pairs_safe pobj_on_pairs prep_pairs_of pobj_of_languages advmod_clear_where det_notion_the nsubj_clear_notion prep_notion_of pobj_of_word cop_clear_is advmod_clear_rather rcmod_languages_clear ' +p,3607,'1 Introduction Over the past decade , there has been tremendous progress on learning parsing models from treebank data ',Collins,'num_Introduction_1 prep_been_Over det_decade_the amod_decade_past pobj_Over_decade expl_been_there aux_been_has dep_Introduction_been amod_progress_tremendous dobj_been_progress prep_progress_on pcomp_on_learning amod_models_parsing dobj_learning_models prep_learning_from nn_data_treebank pobj_from_data ' +o,3608,'Three K-means algorithms using different distributional similarity or dissimilarity measures : cosine , - skew divergence 4 , and Lins similarity ',Lin,'num_K-means_Three dep_K-means_algorithms xcomp_algorithms_using amod_similarity_different amod_similarity_distributional dobj_using_similarity cc_similarity_or nn_measures_dissimilarity conj_similarity_measures amod_similarity_cosine nn_divergence_skew dep_K-means_divergence number_4_ dep_divergence_4 cc_4_and conj_4_Lins dep_Lins_similarity ' +o,3609,'Given the following SCFG rule : VP VB NP JJR , VB NP will be JJR we can obtain a set of equivalent binary rules using the synchronous binarization method as follows : VP V1 JJR , V1 JJR V1 VB V2 , VB V2 V2 NP , NP will be This binarization is shown with the solid lines as binarization -LRB- a -RRB- in Figure 1 ',Zhang,'prep_JJR_Given det_rule_the amod_rule_following nn_rule_SCFG dep_Given_rule nn_JJR_VP nn_JJR_VB nn_JJR_NP dep_rule_JJR nn_NP_VB nsubj_JJR_NP aux_JJR_will cop_JJR_be nsubj_obtain_we aux_obtain_can ccomp_JJR_obtain det_set_a dobj_obtain_set prep_set_of amod_rules_equivalent nn_rules_binary pobj_of_rules xcomp_obtain_using det__the amod__synchronous nn__binarization nn__method dobj_using_ mark_follows_as advcl_JJR_follows nn_JJR_VP nn_JJR_V1 nsubj_binarization_JJR nn_V2_V1 nn_V2_JJR nn_V2_V1 nn_V2_VB appos_JJR_V2 nn_NP_VB nn_NP_V2 nn_NP_V2 appos_V2_NP advmod_binarization_NP aux_binarization_will cop_binarization_be det_binarization_This parataxis_JJR_binarization auxpass_shown_is rcmod_binarization_shown prep_shown_with det_lines_the amod_lines_solid pobj_with_lines prep_shown_as pobj_as_binarization dep_-LRB-_a prep_binarization_in pobj_in_Figure num_Figure_1 ' +o,3610,'Weights on the components were assigned using the method for max-BLEU training on the development set ',Och,'nsubjpass_assigned_Weights prep_Weights_on det_components_the pobj_on_components auxpass_assigned_were xcomp_assigned_using det_method_the amod_method_ dobj_using_method prep_using_for amod_training_max-BLEU pobj_for_training prep_training_on det_set_the nn_set_development pobj_on_set ' +o,3611,'Inter-annotator agreement was assessed mainly using f-score and percentage agreement as well as 11 Table 1: Annotation examples of superlative adjectives example sup span det num car mod comp set The third-largest thrift institution in Puerto Rico also [] 22 def sg no ord 37 The Agriculture Department reported that feedlots in the 13 biggest ranch states held [] 910 def pl yes no 1112 The failed takeover would have given UAL employees 75 \% voting control of the nation s second-largest airline [] 1717 pos sg no ord 1418 the kappa statistics (K), where applicable (Carletta, 1996).',Carletta,'' +o,3612,'We evaluate the summaries using the automatic evaluation tool ROUGE -LRB- described in Section 6 -RRB- and the ROUGE value works as the feedback to our learning loop ',Lin,'nsubj_evaluate_We det_summaries_the nsubj_using_summaries dep_evaluate_using det_tool_the amod_tool_automatic nn_tool_evaluation dobj_using_tool nn__ROUGE nsubj_works_ dep__described prep_described_in pobj_in_Section num_Section_6 cc__and det_value_the amod_value_ROUGE conj__value dep_evaluate_works prep_works_as det_feedback_the pobj_as_feedback prep_works_to poss_loop_our nn_loop_learning pobj_to_loop ' +o,3613,'The models were originally introduced in ; the current article 1 gives considerably more detail about the models and discusses them in greater depth ',Collins,'det_models_The nsubjpass_introduced_models auxpass_introduced_were advmod_introduced_originally prep_introduced_in pobj_in_ det_article_the amod_article_current nsubj_gives_article num_article_1 parataxis_introduced_gives advmod_more_considerably amod_detail_more dobj_gives_detail prep_gives_about det_models_the pobj_about_models cc_gives_and conj_gives_discusses dobj_discusses_them prep_discusses_in amod_depth_greater pobj_in_depth ' +o,3614,'We applied the union , intersection and refined symmetrization metrics to the final alignments output from training , as well as evaluating the two final alignments directly ',Och,'nsubj_applied_We det_metrics_the nn_intersection_union nn_metrics_intersection cc_intersection_and conj_intersection_refined nn_metrics_symmetrization dobj_applied_metrics dep_applied_ prep__to det_output_the amod_output_final nn_output_alignments pobj_to_output prep_output_from pobj_from_training dep_well_as cc__well dep_well_as conj__evaluating det_alignments_the num_alignments_two amod_alignments_final dobj_evaluating_alignments advmod_evaluating_directly ' +o,3615,'Feature weights were set with minimum error rate training on a development set using BLEU as the objective function ',Och,'amod_weights_Feature nsubjpass_set_weights auxpass_set_were prep_set_with amod__minimum nn__error nn__rate nn__training pobj_with_ prep__on det_development_a pobj_on_development partmod_development_set xcomp_set_using nn__BLEU dobj_using_ prep__as det_function_the amod_function_objective pobj_as_function ' +o,3616,'The judges had an acceptable 074 mean agreement for the assignment of the primary class , but a meaningless 021 for the secondary class -LRB- they did not even agree on which lemmata were polysemous -RRB- ',Carletta,'det_judges_The nsubj_had_judges det__an amod__acceptable tmod_acceptable_074 amod__mean nn__agreement dobj_had_ prep__for det_assignment_the pobj_for_assignment prep_assignment_of det_class_the amod_class_primary pobj_of_class cc_had_but det_021_a amod_021_meaningless conj_had_021 prep_021_for det_class_the amod_class_secondary pobj_for_class nsubj_agree_they aux_agree_did neg_agree_not advmod_agree_even dep_class_agree prep_agree_on dobj_polysemous_which nsubj_polysemous_lemmata cop_polysemous_were pcomp_on_polysemous ' +o,3617,'Pereira et al , Curran and Moens and use syntactic features in the vector definition ',Lin,'nn__Pereira cc_Pereira_et conj_Pereira_al nsubj_use_ conj__Curran cc__and nn__Moens conj__ cc__and conj__ amod_features_syntactic dobj_use_features prep_features_in det_definition_the nn_definition_vector pobj_in_definition ' +o,3618,'3 Perceptron Training The parsing problem is to find a mapping from a set of sentences x ? ? X to a set of parses y ? ? Y We assume that the mapping F is represented through a feature vector -LRB- x , y -RRB- ? ? Rd and a parameter vector ? ? Rd in the following way : F -LRB- x -RRB- = argmax y ? GEN -LRB- x -RRB- -LRB- x , y -RRB- -LRB- 1 -RRB- where GEN -LRB- x -RRB- denotes the set of possible parses for sentence x and -LRB- x , y -RRB- = summationtexti ii -LRB- x , y -RRB- is the inner product ',Collins,'num_Training_3 nn_Training_Perceptron det_problem_The amod_problem_parsing nsubj_is_problem rcmod_Training_is aux_find_to xcomp_is_find det_mapping_a dobj_find_mapping prep_find_from det_set_a pobj_from_set prep_set_of dep_x_sentences dep_of_x dep_?_X prep_X_to det_set_a pobj_to_set prep_set_of amod_y_parses pobj_of_y dep_Training_Y nsubj_assume_We rcmod_Y_assume advmod_represented_that det_F_the nn_F_mapping nsubjpass_represented_F auxpass_represented_is dep_assume_represented prep_represented_through det_vector_a nn_vector_feature pobj_through_vector dep_vector_x appos_x_y nsubj_parameter_Rd cc_Rd_and conj_Rd_a dep_represented_parameter dobj_parameter_vector dep_represented_Rd prep_Rd_in det__the amod__following nn__way pobj_in_ nn_x_F dep_Rd_x dep_y_= nn_y_argmax rcmod_x_y dobj_assume_GEN dep_GEN_x dep_GEN_x appos_x_y appos_GEN_1 advmod_denotes_where nsubj_denotes_GEN dep_GEN_x rcmod_GEN_denotes det_set_the dobj_denotes_set prep_set_of amod_parses_possible pobj_of_parses prep_parses_for nn_x_sentence pobj_for_x cc_denotes_and dep_denotes_x appos_x_y amod_ii_= nn_ii_summationtexti nsubj_product_ii dep_ii_x appos_x_y cop_product_is det_product_the amod_product_inner conj_denotes_product ' +o,3619,'Unknown words were not identified in as a useful predictor for the benefit of self-training ',McClosky,'amod_words_Unknown nsubjpass_identified_words auxpass_identified_were neg_identified_not prep_identified_in pobj_in_ prep_identified_as det_predictor_a amod_predictor_useful pobj_as_predictor prep_predictor_for det_benefit_the pobj_for_benefit prep_benefit_of pobj_of_self-training ' +o,3620,'Unsupervised approaches are attractive due to the the availability of large quantities of unlabeled text , and unsupervised morphological segmentation has been extensively studied for a number of languages ',Dasgupta,'amod_approaches_Unsupervised nsubj_attractive_approaches cop_attractive_are dep_attractive_due prep_attractive_to det_availability_the dep_availability_the pobj_to_availability prep_availability_of amod_quantities_large pobj_of_quantities prep_quantities_of amod_text_unlabeled pobj_of_text cc_attractive_and amod_segmentation_unsupervised amod_segmentation_morphological nsubjpass_studied_segmentation aux_studied_has auxpass_studied_been advmod_studied_extensively conj_attractive_studied prep_studied_for det_number_a pobj_for_number prep_number_of pobj_of_languages ' +p,3621,'On the Hansards data , the simple averaging technique described by yields a reasonable model ',Collins,'prep_described_On det_data_the nn_data_Hansards pobj_On_data det_technique_the amod_technique_simple nn_technique_averaging nsubj_described_technique prep_described_by amod_yields_ pobj_by_yields det_model_a amod_model_reasonable dep_yields_model ' +o,3622,'The weights 1 , , M are typically learned to directly minimize a standard evaluation criterion on development data -LRB- eg , the BLEU score ; -RRB- using numerical search ',Papineni,'det_M_The amod_M_weights tmod_weights_1 nsubjpass_learned_M auxpass_learned_are advmod_learned_typically aux_minimize_to advmod_minimize_directly xcomp_learned_minimize det_criterion_a amod_criterion_standard nn_criterion_evaluation dobj_minimize_criterion prep_criterion_on nn_data_development pobj_on_data appos_data_eg det__the num__BLEU nn__score dep_eg_ partmod_data_using amod__numerical nn__search dobj_using_ ' +o,3623,'A growing body of recent research has focused on the problems of identifying and generating paraphrases , eg , , Lin & Pantel -LRB- 2002 -RRB- , Shinyama et al , -LRB- 2002 -RRB- , Barzilay & Lee -LRB- 2003 -RRB- , and Pang et al ',Barzilay,'det_body_A amod_body_growing nsubj_focused_body prep_body_of amod_research_recent pobj_of_research aux_focused_has prep_focused_on det_problems_the pobj_on_problems prep_problems_of pcomp_of_identifying cc_identifying_and conj_identifying_generating dobj_identifying_paraphrases conj_paraphrases_eg conj_paraphrases_ conj__Lin cc__& conj__Pantel appos__2002 appos__Shinyama cc_Shinyama_et conj_Shinyama_al appos__2002 conj_paraphrases_Barzilay cc_Barzilay_& conj_Barzilay_Lee appos_Barzilay_2003 cc_paraphrases_and conj_paraphrases_Pang cc_Pang_et conj_Pang_al ' +o,3624,'So far , most previous work on domain adaptation for parsing has focused on data-driven systems , ie systems employing -LRB- constituent or dependency based -RRB- treebank grammars ',Berger,'dep_far_So advmod_focused_far amod_work_most amod_work_previous nsubj_focused_work prep_work_on nn_adaptation_domain pobj_on_adaptation prep_adaptation_for pobj_for_parsing aux_focused_has prep_focused_on amod__data-driven nn__systems pobj_on_ nn_systems_ie appos__systems xcomp_focused_employing dep_employing_constituent cc_constituent_or conj_constituent_dependency partmod_constituent_based nn_grammars_treebank nsubj__grammars ccomp_employing_ ' +o,3625,' 1996 -RRB- ',Ratnaparkhi,'dep__1996 ' +o,3626,'Several researchers also studied feature\\/topicbased sentiment analysis ',McDonald,'amod_researchers_Several nsubj_studied_researchers advmod_studied_also amod_analysis_feature\\/topicbased nn_analysis_sentiment dobj_studied_analysis ' +o,3627,'The boosting approach to ranking has been applied to named entity segmentation and natural language generation ',Collins,'det_approach_The amod_approach_boosting nsubjpass_applied_approach prep_approach_to pobj_to_ranking aux_applied_has auxpass_applied_been prep_applied_to amod__named nn__entity nn__segmentation pobj_to_ cc__and amod__natural nn__language nn__generation conj__ ' +o,3628,'1 minority report 2 box office 3 scooby doo 4 sixth sense 5 national guard 6 bourne identity 7 air national guard 8 united states 9 phantom menace 10 special effects 11 hotel room 12 comic book 13 blair witch project 14 short story 15 real life 16 jude law 17 iron giant 18 bin laden 19 black people 20 opening weekend 21 bad guy 22 country bears 23 mans man 24 long time 25 spoiler space 26 empire strikes back 27 top ten 28 politically correct 29 white people 30 tv show 31 bad guys 32 freddie prinze jr 33 monsters ball 34 good thing 35 evil minions 36 big screen 37 political correctness 38 martial arts 39 supreme court 40 beautiful mind Figure 7: Result of re-ranking output from the phrase extension module 6.4 Revisiting unigram informativeness An alternative approach to calculate informativeness from the foreground LM and the background LM is just to take the ratio of likelihood scores, a11 fga9a54a86 a15 a23 a11 bga9a54a86 a15 . This is a smoothed version of relative frequency ratio which is commonly used to find subject-specific terms (Damerau, 1993).',Dunning,'' +o,3629,'Finally , the parameters i of the log-linear model -LRB- 18 -RRB- are learned by minimumerror-rate training , which tries to set the parameters so as to maximize the BLEU score of a development set ',Och,'advmod_learned_Finally nsubjpass_learned_the nn_the_parameters dep_the_i prep_the_of det_model_the amod_model_log-linear pobj_of_model appos_model_18 auxpass_learned_are prep_learned_by amod__minimumerror-rate nn__training pobj_by_ nsubj_tries_which rcmod__tries aux_set_to xcomp_tries_set det_parameters_the dobj_set_parameters advmod_as_so advmod_set_as aux_maximize_to dep_as_maximize det__the amod__BLEU nn__score dobj_maximize_ prep__of det_set_a nn_set_development pobj_of_set ' +o,3630,'Given a manually compiled lexicon containing words and their relative frequencies Ps(fprimej), the best segmentationfJ1 is the one that maximizes the joint probability of all words in the sentence, with the assumption that words are independent of each other1: fJ1 = argmax fprimeJprime1 Pr(fprimeJprime1 |cK1 ) argmax fprimeJprime1 Jprimeproductdisplay j=1 Ps(fprimej), where the maximization is taken over Chinese word sequences whose character sequence is cK1 . 2.2 Translation system Once we have segmented the Chinese sentences into words, we train standard alignment models in both directions with GIZA++ (Och and Ney, 2002) using models of IBM-1 (Brown et al., 1993), HMM (Vogel et al., 1996) and IBM-4 (Brown et al., 1993).',Brown,'' +p,3631,'With the in-depth study of opinion mining , researchers committed their efforts for more accurate results : the research of sentiment summarization , domain transfer problem of the sentiment analysis and finegrained opinion mining are the main branches of the research of opinion mining ',Blitzer,'prep_committed_With det_study_the amod_study_in-depth pobj_With_study prep_study_of nn_mining_opinion pobj_of_mining nsubj_committed_researchers poss_efforts_their dobj_committed_efforts prep_committed_for advmod_accurate_more amod_results_accurate pobj_for_results det_research_the nsubj_branches_research prep_research_of nn__sentiment nn__summarization pobj_of_ nn_problem_domain nn_problem_transfer conj__problem prep_problem_of det__the nn__sentiment nn__analysis pobj_of_ cc__and amod__finegrained nn__opinion nn__mining conj__ cop_branches_are det_branches_the amod_branches_main parataxis_committed_branches prep_branches_of det_research_the pobj_of_research prep_research_of nn_mining_opinion pobj_of_mining ' +o,3632,'Second , the significance of the K-S distance in case of the null hypothesis -LRB- data sets are drawn from same distribution -RRB- can be calculated ',Dunning,'advmod_calculated_Second det_significance_the nsubjpass_calculated_significance prep_significance_of det_distance_the amod_distance_K-S pobj_of_distance prep_distance_in pobj_in_case prep_case_of det_hypothesis_the amod_hypothesis_null pobj_of_hypothesis nn_sets_data nsubjpass_drawn_sets auxpass_drawn_are dep_distance_drawn prep_drawn_from amod_distribution_same pobj_from_distribution aux_calculated_can auxpass_calculated_be dep_calculated_ ' +o,3633,' ask the user to suggest a few prototypes -LRB- examples -RRB- for each class and use those as features ',Haghighi,'advmod_ask_ det_user_the dobj_ask_user aux_suggest_to infmod_user_suggest det_prototypes_a amod_prototypes_few dobj_suggest_prototypes appos_prototypes_examples prep_prototypes_for det_class_each pobj_for_class cc_ask_and conj_ask_use dobj_use_those prep_those_as pobj_as_features ' +o,3634,'For these experiments , we have implemented an alignment package for IBM Model 4 using a hillclimbing search and Viterbi training as described in , and extended this to use new submodels ',Brown,'prep_implemented_For det_experiments_these pobj_For_experiments nsubj_implemented_we aux_implemented_have det_package_an amod_package_alignment dobj_implemented_package prep_implemented_for nn_Model_IBM pobj_for_Model num_Model_4 xcomp_implemented_using det_search_a amod_search_hillclimbing dobj_using_search cc_implemented_and amod_training_Viterbi dep_as_training advmod_described_as conj_implemented_described prep_described_in pobj_in_ cc_implemented_and conj_implemented_extended nsubj_use_this aux_use_to xcomp_extended_use amod_submodels_new dobj_use_submodels ' +o,3635,'The forest concept is also used in machine translation decoding , for example to characterize the search space of decoding with integrated language models ',Huang,'det_concept_The nn_concept_forest nsubjpass_used_concept auxpass_used_is advmod_used_also prep_used_in nn_translation_machine pobj_in_translation partmod_translation_decoding prep_decoding_for pobj_for_example aux_characterize_to xcomp_decoding_characterize det_space_the nn_space_search dobj_characterize_space prep_space_of pcomp_of_decoding prep_decoding_with amod_models_integrated nn_models_language pobj_with_models ' +o,3636,'POS tagging and phrase chunking in English were done using the trained systems provided with the fnTBL Toolkit ; both were trained from the annotated Penn Treebank corpus ',Marcus,'dep_tagging_POS nsubjpass_done_tagging cc_tagging_and nn_chunking_phrase conj_tagging_chunking prep_tagging_in pobj_in_English auxpass_done_were xcomp_done_using det_systems_the amod_systems_trained dobj_using_systems prep_using_provided dep_provided_with det__the amod__fnTBL nn__Toolkit pobj_with_ nsubjpass_trained_both auxpass_trained_were parataxis_done_trained prep_trained_from det_corpus_the amod_corpus_annotated nn_corpus_Penn nn_corpus_Treebank pobj_from_corpus ' +o,3637,'In the experiment , only the first 500 sentences were used to train the log-linear model weight vector , where minimum error rate -LRB- MER -RRB- training was used ',Och,'prep_used_In det_experiment_the pobj_In_experiment advmod_experiment_only det_sentences_the amod_sentences_first num_sentences_500 dep_experiment_sentences auxpass_used_were aux_train_to xcomp_used_train det_model_the amod_model_log-linear dobj_train_model nn_vector_weight nsubjpass_used_vector advmod_used_where amod_rate_minimum nn_rate_error nsubjpass_used_rate abbrev_rate_MER partmod_rate_training auxpass_used_was rcmod_vector_used ' +o,3638,'and Semantic Knowledge Sources for Coreference Resolution and Strube & Ponzetto -LRB- 2006 -RRB- aimed at showing that the encyclopedia that anyone can edit can be indeed used as a semantic resource for research in NLP ',Ponzetto,'nn_Knowledge_Semantic dep_and_Knowledge nsubj_aimed_Sources prep_Sources_for nn__Coreference nn__Resolution pobj_for_ cc__and conj__Strube cc_Strube_& conj_Strube_Ponzetto appos_Sources_2006 rcmod_Knowledge_aimed prep_aimed_at pcomp_at_showing complm_used_that det_encyclopedia_the nsubjpass_used_encyclopedia complm_edit_that nsubj_edit_anyone aux_edit_can dep_encyclopedia_edit aux_used_can auxpass_used_be advmod_used_indeed ccomp_showing_used prep_used_as det_resource_a amod_resource_semantic pobj_as_resource prep_resource_for pobj_for_research prep_research_in pobj_in_NLP ' +o,3639,'the remarks on the a3 a4 measure in -RRB- ',Dunning,'det_remarks_the prep_remarks_on det_measure_the amod_measure_a3 amod_measure_a4 pobj_on_measure prep_measure_in pobj_in_ ' +o,3640,'First , manyto-many word alignments are induced by running a one-to-many word alignment model , such as GIZA + + , in both directions and by combining the results based on a heuristic ',Och,'advmod_induced_First amod_alignments_manyto-many nn_alignments_word nsubjpass_induced_alignments auxpass_induced_are prep_induced_by pcomp_by_running det_model_a amod_model_one-to-many nn_model_word nn_model_alignment dobj_running_model dep_as_such prep_model_as pobj_as_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_running_in det_directions_both pobj_in_directions cc_by_and conj_by_by pcomp_by_combining det_results_the dobj_combining_results prep_combining_based dep_based_on det_heuristic_a pobj_on_heuristic ' +o,3641,'1 Introduction Word alignment was first proposed as an intermediate result of statistical machine translation ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubjpass_proposed_alignment auxpass_proposed_was advmod_proposed_first prep_proposed_as det_result_an amod_result_intermediate pobj_as_result prep_result_of amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,3642,'22 Creation of a Coarse-Grained Sense Inventory To tackle the granularity issue , we produced a coarser-grained version of the WordNet sense inventory3 based on the procedure described by ',Navigli,'num_Creation_22 nsubj_tackle_Creation prep_Creation_of det_Inventory_a amod_Inventory_Coarse-Grained nn_Inventory_Sense pobj_of_Inventory aux_tackle_To dep_produced_tackle det_issue_the amod_issue_granularity dobj_tackle_issue nsubj_produced_we det_version_a amod_version_coarser-grained dobj_produced_version prep_version_of det_inventory3_the nn_inventory3_WordNet nn_inventory3_sense pobj_of_inventory3 prep_produced_based dep_based_on det_procedure_the pobj_on_procedure partmod_procedure_described prep_described_by ' +o,3643,'To tune the decoder parameters , we conducted minimum error rate training with respect to the word BLEU score using 20K development sentence pairs ',Papineni,'aux_tune_To dep_conducted_tune det_parameters_the nn_parameters_decoder dobj_tune_parameters nsubj_conducted_we amod__minimum nn__error nn__rate nn__training dobj_conducted_ prep_conducted_with pobj_with_respect prep_conducted_to det__the nn__word nn__BLEU nn__score pobj_to_ partmod__using num_pairs_20K nn_pairs_development nn_pairs_sentence dobj_using_pairs ' +p,3644,'Finally , the translation model can be formalized as the following optimization problem argmax logPr -LRB- D ; -RRB- st mwsummationdisplay j = 1 Pr -LRB- wj ok -RRB- = 1 , k This optimization problem can be solved by the EM algorithm ',Brown,'advmod_formalized_Finally det_model_the nn_model_translation nsubjpass_formalized_model aux_formalized_can auxpass_formalized_be prep_formalized_as det_logPr_the amod_logPr_following amod_logPr_optimization nn_logPr_problem nn_logPr_argmax pobj_as_logPr dep_logPr_D nn_j_st nn_j_mwsummationdisplay nsubj_=_j rcmod_logPr_= num_Pr_1 dobj_=_Pr nn_ok_wj appos_Pr_ok dep_1_= ccomp_solved_1 dep_solved_k det_problem_This nn_problem_optimization nsubjpass_solved_problem aux_solved_can auxpass_solved_be rcmod_Pr_solved prep_solved_by det_algorithm_the nn_algorithm_EM pobj_by_algorithm ' +o,3645,'History-based models for predicting the next parser action 3 ',Collins,'amod_models_History-based nsubj__models prep_models_for pcomp_for_predicting det_action_the amod_action_next nn_action_parser dobj_predicting_action dobj__3 ' +o,3646,'Pharaoh also includes lexical weighting parameters that are derived from the alignments used to induce its phrase pairs ',Koehn,'nsubj_includes_Pharaoh advmod_includes_also amod_parameters_lexical nn_parameters_weighting dobj_includes_parameters nsubjpass_derived_that auxpass_derived_are rcmod_parameters_derived prep_derived_from det_alignments_the pobj_from_alignments partmod_alignments_used aux_induce_to xcomp_used_induce poss_pairs_its nn_pairs_phrase dobj_induce_pairs ' +o,3647,'32 Questions and Corpus To get a clear picture of the impact of using different information extraction methods for the offline construction of knowledge bases , similarly to , we focused only on questions about persons , taken from the TREC8 through TREC 2003 question sets ',Fleischman,'num_Questions_32 cc_Questions_and conj_Questions_Corpus aux_get_To dep_focused_get det_picture_a amod_picture_clear dobj_get_picture prep_picture_of det_impact_the pobj_of_impact prep_impact_of pcomp_of_using amod_methods_different nn_methods_information nn_methods_extraction dobj_using_methods prep_using_for det_construction_the amod_construction_offline pobj_for_construction prep_construction_of nn_bases_knowledge pobj_of_bases advmod__similarly aux__to xcomp_using_ nsubj_focused_we dep_Questions_focused advmod_focused_only prep_focused_on pobj_on_questions prep_questions_about pobj_about_persons partmod_persons_taken prep_taken_from det_TREC8_the pobj_from_TREC8 prep_taken_through nn_sets_TREC num_sets_2003 nn_sets_question pobj_through_sets ' +o,3648,'Moreover , in order to determine whether the performances of the predictive criteria are consistent across different learning models within the same domain , we have performed the study on two parsing models : one based on a context-free variant of tree-adjoining grammars , the Probabilistic Lexicalized Tree Insertion Grammar -LRB- PLTIG -RRB- formalism , and ',Collins,'advmod_performed_Moreover mark_determine_in dep_determine_order aux_determine_to dep_performed_determine complm_consistent_whether det_performances_the nsubj_consistent_performances prep_performances_of det_criteria_the amod_criteria_predictive pobj_of_criteria cop_consistent_are ccomp_determine_consistent prep_consistent_across acomp_learning_different pcomp_across_learning dobj_learning_models prep_learning_within det_domain_the amod_domain_same pobj_within_domain nsubj_performed_we aux_performed_have det_study_the dobj_performed_study prep_performed_on num_models_two amod_models_parsing pobj_on_models dep_models_one partmod_one_based prep_based_on det_variant_a amod_variant_context-free pobj_on_variant prep_variant_of amod__tree-adjoining nn__grammars pobj_of_ det_Grammar_the nn_Grammar_Probabilistic nn_Grammar_Lexicalized nn_Grammar_Tree nn_Grammar_Insertion appos__Grammar abbrev__PLTIG nn__formalism dep__ appos__and ' +o,3649,'1 Introduction Robust statistical syntactic parsers , made possible by new statistical techniques and by the availability of large , hand-annotated training corpora such as WSJ and Switchboard , have had a major impact on the field of natural language processing ',Marcus,'num_parsers_1 nn_parsers_Introduction nn_parsers_Robust amod_parsers_statistical amod_parsers_syntactic nsubj_had_parsers partmod_parsers_made acomp_made_possible prep_made_by amod__new amod__statistical nn__techniques pobj_by_ cc_by_and conj_by_by det_availability_the pobj_by_availability prep_availability_of amod_corpora_large amod_corpora_hand-annotated nn_corpora_training pobj_of_corpora dep_as_such prep_availability_as nn__WSJ pobj_as_ cc__and nn__Switchboard conj__ aux_had_have det_impact_a amod_impact_major dobj_had_impact prep_had_on det_field_the pobj_on_field prep_field_of amod_processing_natural nn_processing_language pobj_of_processing ' +o,3650,'data set ',Ramshaw,'nsubj_set_data acomp_set_ ' +o,3651,'Reported and direct speech are certainly important in discourse ; we do not believe , however , that they enter discourse relations of the type that RST attempts to capture ',Prasad,'amod_speech_Reported cc_Reported_and conj_Reported_direct nsubj_important_speech cop_important_are advmod_important_certainly prep_important_in nn__discourse pobj_in_ nsubj_believe_we aux_believe_do neg_believe_not parataxis_important_believe advmod_believe_however complm_enter_that nsubj_enter_they ccomp_believe_enter nn_relations_discourse dobj_enter_relations prep_relations_of det_type_the pobj_of_type complm_attempts_that nsubj_attempts_RST ccomp_enter_attempts aux_capture_to xcomp_attempts_capture ' +o,3652,'In this paper , we follow this line of research and try to solve the problem by extending Collins perceptron algorithm ',Collins,'prep_follow_In det_paper_this pobj_In_paper nsubj_follow_we det_line_this dobj_follow_line prep_line_of pobj_of_research cc_follow_and conj_follow_try aux_solve_to xcomp_try_solve det_problem_the dobj_solve_problem prep_solve_by pcomp_by_extending nn_algorithm_Collins nn_algorithm_perceptron dobj_extending_algorithm ' +o,3653,'A total of 216 collocations were extracted , shown in Appendix A We compared the collocations in Appendix A with the entries for the above 10 words in the NTC \'s English Idioms Dictionary -LRB- henceforth NTC-EID -RRB- , which contains approximately 6000 definitions of idioms ',Dunning,'det_total_A nsubj_extracted_total prep_total_of num_collocations_216 pobj_of_collocations aux_extracted_were dep_extracted_shown prep_shown_in nn_A_Appendix pobj_in_A dep_A_We prep_shown_compared det_collocations_the dep_in_collocations prep_collocations_in nn_A_Appendix pobj_in_A prep_A_with det_entries_the pobj_with_entries prep_entries_for det_words_the amod_words_above num_words_10 pobj_for_words dep_compared_in det_NTC_the poss__NTC possessive_NTC_\'s nn__English nn__Idioms nn__Dictionary nn_NTC-EID_henceforth abbrev__NTC-EID pobj_in_ nsubj_contains_which ccomp_extracted_contains quantmod_6000_approximately num_definitions_6000 dobj_contains_definitions prep_definitions_of pobj_of_idioms ' +o,3654,'Our chunks and functions are based on the annotations in the third release of the Penn Treebank ',Marcus,'poss_chunks_Our nsubjpass_based_chunks cc_chunks_and conj_chunks_functions auxpass_based_are prep_based_on det_annotations_the pobj_on_annotations prep_annotations_in det_release_the amod_release_third pobj_in_release prep_release_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,3655,'Others proposed distributional similarity measures between words ',Hindle,'amod_measures_proposed amod_measures_distributional nn_measures_similarity dep_Others_measures prep_measures_between pobj_between_words ' +o,3656,'Eisner , Charniak , , and many subsequent researchers1 annotated every node with lexical features passed up from its head child , in order to more precisely reflect the nodes inside contents ',Collins,'nn__Eisner nsubj_annotated_ nn__Charniak conj__ nn__ conj__ cc__and amod_researchers1_many amod_researchers1_subsequent conj__researchers1 det_node_every dobj_annotated_node prep_annotated_with amod_features_lexical pobj_with_features partmod_features_passed prt_passed_up prep_passed_from poss_child_its nn_child_head pobj_from_child mark_reflect_in dep_reflect_order aux_reflect_to advmod_reflect_more advmod_reflect_precisely purpcl_passed_reflect det_nodes_the dobj_reflect_nodes prep_nodes_inside pobj_inside_contents ' +o,3657,'Model 1 is the word-pair translation model used in simple machine translation and understanding models ',Brown,'nsubj_model_Model num_Model_1 cop_model_is det_model_the amod_model_word-pair nn_model_translation partmod_model_used prep_used_in amod_translation_simple nn_translation_machine pobj_in_translation cc_translation_and nn_models_understanding conj_translation_models ' +o,3658,'43 Scoring All-N Rules We observed that the likelihood of nouns mentioned in a definition to be referred by the concept title depends greatly on the syntactic path connecting them -LRB- which was exploited also in -RRB- ',Snow,'num_Rules_43 nn_Rules_Scoring nn_Rules_All-N nsubj_observed_We rcmod_Rules_observed complm_depends_that det_likelihood_the nsubj_depends_likelihood prep_likelihood_of pobj_of_nouns partmod_nouns_mentioned prep_mentioned_in det_definition_a pobj_in_definition aux_referred_to auxpass_referred_be infmod_definition_referred prep_referred_by det_title_the nn_title_concept pobj_by_title ccomp_observed_depends advmod_depends_greatly prep_depends_on det_path_the amod_path_syntactic pobj_on_path partmod_path_connecting dobj_connecting_them nsubjpass_exploited_which auxpass_exploited_was dep_Rules_exploited advmod_exploited_also prep_exploited_in pobj_in_ ' +o,3659,'Kupiec has proposed an estimation method for the N-gram language model using the Baum-Welch reestimation algorithm from an untagged corpus and et al ',Cutting,'nn__Kupiec nsubj_proposed_ aux_proposed_has det_method_an nn_method_estimation dobj_proposed_method prep_proposed_for det_model_the amod_model_N-gram nn_model_language pobj_for_model partmod_model_using det__the amod__Baum-Welch nn__reestimation nn__algorithm dobj_using_ prep_using_from det_corpus_an amod_corpus_untagged pobj_from_corpus cc_corpus_and conj_corpus_ cc__et conj__al ' +o,3660,'Empirical evaluations using two standard summarization metricsthe Pyramid method and ROUGE show that the best performing system is a CRF incorporating both order-2 Markov dependencies and skip-chain dependencies , which achieves 913 \% of human performance in Pyramid score , and outperforms our best-performing non-sequential model by 39 \% ',Lin,'amod_evaluations_Empirical nsubj_show_evaluations partmod_evaluations_using num__two amod__standard nn__summarization nn__metricsthe nn__Pyramid nn__method dobj_using_ cc__and nn__ROUGE conj__ complm_CRF_that det_system_the dep_performing_best amod_system_performing nsubj_CRF_system cop_CRF_is det_CRF_a ccomp_show_CRF partmod_CRF_incorporating det_dependencies_both amod_dependencies_order-2 nn_dependencies_Markov amod_dependencies_dependencies cc_dependencies_and conj_dependencies_skip-chain dobj_incorporating_dependencies nsubj_achieves_which rcmod_dependencies_achieves num_\%_913 dobj_achieves_\% prep_\%_of amod_performance_human pobj_of_performance prep_achieves_in nn_score_Pyramid pobj_in_score cc_achieves_and conj_achieves_outperforms poss_model_our amod_model_best-performing amod_model_non-sequential dobj_outperforms_model prep_outperforms_by num_\%_39 pobj_by_\% ' +o,3661,'The features we used are as follows : Direct and inverse IBM model ; 3 , 4-gram target language model ; 3 , 4 , 5-gram POS language model ; 96 Sentence length posterior probability ; N-gram posterior probabilities within the NBest list ; Minimum Bayes Risk probability ; Length ratio between source and target sentence ; The weights are optimized via MERT algorithm ',Ratnaparkhi,'det_features_The nsubj_Direct_features nsubj_used_we rcmod_features_used cop_Direct_are mark_follows_as dep_Direct_follows cc_Direct_and amod_model_inverse nn_model_IBM conj_Direct_model dep_Direct_3 amod_model_4-gram nn_model_target nn_model_language appos_3_model dep_Direct_3 appos_3_4 amod__5-gram dep__POS nn__language nn__model appos_3_ num_posterior_96 nn_posterior_Sentence nn_posterior_length dep_Direct_posterior nn__probability dep_posterior_ amod_probabilities_N-gram nn_probabilities_posterior dep_Direct_probabilities prep_probabilities_within det__the nn__NBest nn__list pobj_within_ nn_Bayes_Minimum dep_Direct_Bayes nn_probability_Risk dep_Bayes_probability amod_ratio_Length dep_Direct_ratio prep_ratio_between nn_sentence_source cc_source_and conj_source_target pobj_between_sentence det_weights_The nsubjpass_optimized_weights auxpass_optimized_are parataxis_Direct_optimized prep_optimized_via amod_algorithm_MERT pobj_via_algorithm ' +o,3662,'corpus , the Penn Treebank , the SUSANNE corpus , the Spoken English Corpus , the Oxford Psycholinguistic Database , and the ` Computer-Usable \' version of the Oxford Advanced Learner \'s Dictionary of Current English ',Marcus,'nn__corpus det__the nn__Penn nn__Treebank conj__ det__the amod__SUSANNE nn__corpus conj__ det__the nn__Spoken nn__English nn__Corpus conj__ det__the nn__Oxford nn__Psycholinguistic nn__Database conj__ cc__and det_version_the amod_version_Computer-Usable conj__version prep__of det_Learner_the nn_Learner_Oxford nn_Learner_Advanced poss_Dictionary_Learner possessive_Learner_\'s pobj_of_Dictionary prep_Dictionary_of nn__Current nn__English pobj_of_ ' +p,3663,'First , we compared our system output to human reference translations using Bleu , a widelyaccepted objective metric for evaluation of machine translations ',Papineni,'advmod_compared_First nsubj_compared_we poss_output_our nn_output_system dobj_compared_output prep_compared_to amod_translations_human nn_translations_reference pobj_to_translations xcomp_compared_using nn__Bleu dobj_using_ det_metric_a amod_metric_widelyaccepted nn_metric_objective appos__metric prep_metric_for pobj_for_evaluation prep_metric_of nn_translations_machine pobj_of_translations ' +o,3664,'Following , we call the first the source domain , and the second the target domain ',Blitzer,'prep_call_Following pobj_Following_ nsubj_call_we det_domain_the amod_domain_first det_domain_the nn_domain_source dobj_call_domain cc_domain_and det_domain_the amod_domain_second det_domain_the nn_domain_target conj_domain_domain ' +o,3665,'Generally , two edges can be re-combined if they satisfy the following two constraints : 1 -RRB- the LHS -LRB- left-hand side -RRB- nonterminals are identical and the sub-alignments are the same ; and 2 -RRB- the boundary words 1 on both sides of the partial translations are equal between the two edges ',Zhang,'advmod_re-combined_Generally num_edges_two nsubj_re-combined_edges aux_re-combined_can cop_re-combined_be mark_satisfy_if nsubj_satisfy_they advcl_re-combined_satisfy det_constraints_the amod_constraints_following num_constraints_two dobj_satisfy_constraints dep_identical_1 det_LHS_the nsubj_identical_LHS amod_side_left-hand appos_LHS_side amod_LHS_nonterminals cop_identical_are dep_constraints_identical cc_identical_and det_sub-alignments_the nsubj__sub-alignments cop__are det__the amod__same conj_identical_ cc_identical_and dep__2 det_words_the nn_words_boundary nsubj__words nsubj_equal_1 prep_1_on det_sides_both pobj_on_sides prep_sides_of det_translations_the amod_translations_partial pobj_of_translations cop_equal_are rcmod_words_equal prep_equal_between det_edges_the num_edges_two pobj_between_edges conj_identical_ ' +o,3666,'For the first set of experiments , we divide all inputs based on the mean value of the average system scores as in ',Nenkova,'prep_divide_For det_set_the amod_set_first pobj_For_set prep_set_of pobj_of_experiments nsubj_divide_we det_inputs_all dobj_divide_inputs partmod_inputs_based prep_based_on det_value_the amod_value_mean pobj_on_value prep_value_of det_scores_the amod_scores_average nn_scores_system pobj_of_scores prep_based_as dep_as_in ' +o,3667,'The first , Powells method , was advocated by when MERT was first introduced for statistical machine translation ',Och,'det_first_The nsubjpass_advocated_first nn_method_Powells appos_first_method auxpass_advocated_was prep_advocated_by pcomp_by_ advmod_introduced_when nsubjpass_introduced_MERT auxpass_introduced_was advmod_introduced_first advcl__introduced prep_introduced_for amod_translation_statistical nn_translation_machine pobj_for_translation ' +o,3668,'The true segmentation can now be compared with the N-best list in order to train an averaged perceptron algorithm ',Collins,'det_segmentation_The amod_segmentation_true nsubjpass_compared_segmentation aux_compared_can advmod_compared_now auxpass_compared_be prep_compared_with det_list_the amod_list_N-best pobj_with_list prep_list_in pobj_in_order aux_train_to purpcl_compared_train det_algorithm_an amod_algorithm_averaged nn_algorithm_perceptron dobj_train_algorithm ' +p,3669,'1 Introduction Phrase-based method and syntaxbased method represent the state-of-the-art technologies in statistical machine translation -LRB- SMT -RRB- ',Och,'num__1 nn__Introduction amod__Phrase-based nn__method nsubj_represent_ cc__and nn__syntaxbased nn__method conj__ det_technologies_the amod_technologies_state-of-the-art dobj_represent_technologies prep_technologies_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +o,3670,'It is available in several formats , and in this paper , we use the Penn Treebank format of NEGRA ',Marcus,'nsubj_available_It cop_available_is prep_available_in amod_formats_several pobj_in_formats cc_available_and prep_use_in det_paper_this pobj_in_paper nsubj_use_we conj_available_use det_format_the nn_format_Penn nn_format_Treebank nn_format_ dobj_use_format prep_format_of pobj_of_NEGRA ' +o,3671,'In previous research on splitting sentences , many methods have been based on word-sequence characteristics like N-gram ',Berger,'prep_based_In amod_research_previous pobj_In_research prep_research_on nn_sentences_splitting pobj_on_sentences amod_methods_many nsubjpass_based_methods aux_based_have auxpass_based_been prep_based_on amod_characteristics_word-sequence pobj_on_characteristics prep_characteristics_like pobj_like_N-gram ' +o,3672,'For example , sentence alignment of bilingual texts are performed just by measuring sentence lengths in words or in characters , or by statistically estimating word level correspondences ',Brown,'prep_performed_For pobj_For_example nn_alignment_sentence nsubjpass_performed_alignment prep_alignment_of amod_texts_bilingual pobj_of_texts auxpass_performed_are dep_by_just prep_performed_by pcomp_by_measuring nn_lengths_sentence dobj_measuring_lengths prep_measuring_in pobj_in_words cc_in_or conj_in_in nn__characters pobj_in_ cc_by_or conj_by_by amod__statistically dep_statistically_estimating nn_correspondences_word nn_correspondences_level dobj_estimating_correspondences pobj_by_ ' +o,3673,'2 Statistical Machine Translation We use a log-linear approach in which a foreign language sentence f is translated into another language , for example English , e , by seeking a maximum solution : e = argmax e wT h -LRB- f , e -RRB- -LRB- 1 -RRB- where h -LRB- f , e -RRB- is a large-dimension feature vector ',Och,'num_Translation_2 nn_Translation_Statistical nn_Translation_Machine nsubj_use_We dep_Translation_use det__a amod__log-linear nn__approach dobj_use_ rel_translated_in pobj_in_which det_f_a amod_f_foreign nn_f_language nn_f_sentence nsubjpass_translated_f auxpass_translated_is rcmod__translated prep_translated_into det_language_another pobj_into_language prep_translated_for pobj_for_example dep_e_English amod_example_e prep_translated_by pcomp_by_seeking det_solution_a amod_solution_maximum dobj_seeking_solution dep_translated_e dep_h_= amod_h_argmax amod_h_e nn_h_wT dep_e_h dep_h_f appos_f_e appos_h_1 advmod_vector_where nsubj_vector_h dep_h_f appos_f_e cop_vector_is det_vector_a amod_vector_large-dimension nn_vector_feature advcl_h_vector ' +o,3674,'The quality of the translation output is evaluated using BLEU ',Papineni,'det_quality_The nsubjpass_evaluated_quality prep_quality_of det_output_the nn_output_translation pobj_of_output auxpass_evaluated_is xcomp_evaluated_using dobj_using_BLEU ' +o,3675,'On one hand , as evidence , clusters of paraphrases can lead to better learning of text-totext rewriting rules compared to just pairs of paraphrases ',Barzilay,'prep_lead_On num_hand_one pobj_On_hand prep_lead_as amod_evidence_ pobj_as_evidence nsubj_lead_clusters prep_clusters_of pobj_of_paraphrases aux_lead_can prep_lead_to amod_learning_better pobj_to_learning prep_learning_of amod_rules_text-totext amod_rules_rewriting pobj_of_rules prep_rules_compared dep_compared_to advmod_pairs_just pobj_to_pairs prep_pairs_of pobj_of_paraphrases ' +p,3676,'According to our experience , the best performance is achieved when the union of the source-to-target and target-to-source alignment sets is used for tuple extraction -LRB- some experimental results regarding this issue are presented in Section 422 -RRB- ',Brown,'prep_achieved_According dep_According_to poss_experience_our pobj_to_experience det_performance_the amod_performance_best nsubjpass_achieved_performance auxpass_achieved_is advmod_sets_when det_union_the nsubj_sets_union prep_union_of det_alignment_the amod_alignment_source-to-target cc_source-to-target_and conj_source-to-target_target-to-source pobj_of_alignment advcl_achieved_sets nsubjpass_used_ auxpass_used_is ccomp_sets_used prep_used_for amod_extraction_tuple pobj_for_extraction det_results_some amod_results_experimental dep_extraction_results prep_results_regarding det_issue_this pobj_regarding_issue auxpass_presented_are rcmod_issue_presented prep_presented_in nn_422_Section pobj_in_422 ' +o,3677,'We then apply Brills rule-based tagger and BaseNP noun phrase chunker to extract noun phrases from these sentences ',Ramshaw,'nsubj_apply_We advmod_apply_then nn__Brills amod__rule-based nn__tagger dobj_apply_ cc__and nn__BaseNP nn__noun nn__phrase nn__chunker conj__ aux_extract_to xcomp_apply_extract nn_phrases_noun dobj_extract_phrases prep_extract_from det_sentences_these pobj_from_sentences ' +o,3678,'Except where noted , each system was trained on 27 million words of newswire data , aligned with GIZA + + and symmetrized with the grow-diag-final-and heuristic ',Koehn,'dep_trained_Except advmod_noted_where pcomp_Except_noted det_system_each nsubjpass_trained_system auxpass_trained_was prep_trained_on number_million_27 num_words_million pobj_on_words prep_words_of nn_data_newswire pobj_of_data conj_trained_aligned prep_aligned_with pobj_with_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_trained_and conj_trained_symmetrized prep_symmetrized_with det_heuristic_the amod_heuristic_grow-diag-final-and pobj_with_heuristic ' +o,3679,'When an S alignment exists , there will always also exist a P alignment such that P a65 S The English sentences were parsed using a state-of-the-art statistical parser trained on the University of Pennsylvania Treebank ',Brown,'advmod_exists_When det_alignment_an nn_alignment_S nsubj_exists_alignment dep_alignment_exists expl_alignment_there aux_alignment_will advmod_alignment_always advmod_alignment_also dep_alignment_exist det_alignment_a nn_alignment_P prep_alignment_such det_S_that nn_S_P nn_S_a65 dep_such_S det_sentences_The amod_sentences_English nsubjpass_parsed_sentences auxpass_parsed_were dep_alignment_parsed xcomp_parsed_using det__a amod__state-of-the-art amod__statistical nn__parser dobj_using_ partmod__trained prep_trained_on det_University_the pobj_on_University prep_University_of nn__Pennsylvania nn__Treebank pobj_of_ ' +o,3680,'For each word pair from the antonym set , we calculated the distributional distance between each of their senses using Mohammad and Hirsts method of concept distance along with the modified form of distributional measure -LRB- equation 2 -RRB- ',Lin,'prep_calculated_For det_pair_each nn_pair_word pobj_For_pair prep_pair_from det_set_the nn_set_antonym pobj_from_set nsubj_calculated_we det_distance_the amod_distance_distributional dobj_calculated_distance prep_distance_between pobj_between_each prep_each_of poss_senses_their pobj_of_senses partmod_senses_using nn_method_Mohammad cc_Mohammad_and conj_Mohammad_Hirsts nn_method_ dobj_using_method prep_method_of nn_distance_concept pobj_of_distance advmod_using_along dep_along_with det_form_the amod_form_modified pobj_with_form prep_form_of amod_measure_ amod_measure_distributional pobj_of_measure appos_measure_equation num_equation_2 ' +o,3681,'We use the same feature processing as , with the addition of context features in a window of3 ',Haghighi,'nsubj_use_We det_processing_the amod_processing_same nn_processing_feature dobj_use_processing prep_use_as pobj_as_ prep_use_with det_addition_the pobj_with_addition prep_addition_of nn_features_context pobj_of_features prep_features_in det_of3_a nn_of3_window pobj_in_of3 ' +o,3682,'1 Introduction The goal of this study has been to automatically extract a large set of hyponymy relations , which play a critical role in many NLP applications , such as Q&A systems ',Fleischman,'num_Introduction_1 det_goal_The nsubj_been_goal prep_goal_of det_study_this pobj_of_study aux_been_has dep_Introduction_been aux_extract_to advmod_extract_automatically xcomp_been_extract det_set_a amod_set_large dobj_extract_set prep_set_of amod_relations_hyponymy pobj_of_relations nsubj_play_which rcmod_set_play det_role_a amod_role_critical dobj_play_role prep_role_in amod_applications_many nn_applications_NLP pobj_in_applications dep_as_such prep_set_as nn_systems_Q&A pobj_as_systems ' +o,3683,'Distributional approaches , on the other hand , rely on text corpora , and model relatedness by comparing the contexts in which two words occur , assuming that related words occur in similar context -LRB- eg , , Lin -LRB- 1998 -RRB- , Mohammad and Hirst -LRB- 2006 -RRB- -RRB- ',Hindle,'amod_approaches_Distributional nsubj_rely_approaches prep_approaches_on det_hand_the amod_hand_other pobj_on_hand prep_rely_on nn_corpora_text pobj_on_corpora cc_rely_and nsubj_relatedness_model conj_rely_relatedness prep_relatedness_by pcomp_by_comparing det_contexts_the dobj_comparing_contexts rel_occur_in pobj_in_which num_words_two nsubj_occur_words rcmod_contexts_occur dep_comparing_assuming complm_occur_that amod_words_related nsubj_occur_words ccomp_assuming_occur prep_occur_in amod_context_similar pobj_in_context dep_context_eg nn_Lin_ appos_eg_Lin appos_Lin_1998 appos_Lin_Mohammad cc_Mohammad_and conj_Mohammad_Hirst appos_Mohammad_2006 ' +o,3684,'Related Work The recent availability of large amounts of bilingual data has attracted interest in several areas , including sentence alignment , word alignment , alignment of groups of words , and statistical translation ',Dagan,'nn_Work_Related det_availability_The amod_availability_recent nsubj_attracted_availability prep_availability_of amod_amounts_large pobj_of_amounts prep_amounts_of amod_data_bilingual pobj_of_data aux_attracted_has dep_Work_attracted dobj_attracted_interest prep_attracted_in amod_areas_several pobj_in_areas prep_attracted_including nn__sentence nn__alignment pobj_including_ nn__word nn__alignment conj__ conj__alignment prep_alignment_of pobj_of_groups prep_groups_of nn__words pobj_of_ cc__and amod__statistical nn__translation conj__ ' +p,3685,'Nonparametricmodels may be appropriate ',Teh,'nn__Nonparametricmodels nsubj_appropriate_ aux_appropriate_may cop_appropriate_be ' +o,3686,'The most commonly used MT evaluation metric in recent years has been IBMs Bleu metric ',Papineni,'det_metric_The advmod_used_most advmod_used_commonly amod_metric_used nn_metric_MT nn_metric_evaluation nsubj_been_metric prep_metric_in amod_years_recent pobj_in_years aux_been_has nn_Bleu_IBMs dobj_been_Bleu amod_Bleu_metric ' +o,3687,'On the other hand , extracted hyponymy relations , which are independent of the NE categories , from Wikipedia and utilized it as a gazetteer ',Kazama,'prep_extracted_On det_hand_the amod_hand_other pobj_On_hand nsubj_extracted_ amod_relations_hyponymy dobj_extracted_relations nsubjpass_independent_which auxpass_independent_are rcmod_relations_independent prep_independent_of det_categories_the nn_categories_NE pobj_of_categories prep_independent_from pobj_from_Wikipedia cc_independent_and conj_independent_utilized dobj_utilized_it prep_utilized_as det_gazetteer_a pobj_as_gazetteer ' +o,3688,'For instance , word alignment models are often trained using the GIZA + + toolkit ; error minimizing training criteria such as the Minimum Error Rate Training are employed in order to learn feature function weights for log-linear models ; and translation candidates are produced using phrase-based decoders in combination with n-gram language models ',Brants,'prep_trained_For pobj_For_instance nn_models_word nn_models_alignment nsubjpass_trained_models auxpass_trained_are advmod_trained_often xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__toolkit dobj_using_ nsubjpass_employed_error partmod_error_minimizing nn_criteria_training dobj_minimizing_criteria dep_as_such prep_criteria_as det__the nn__Minimum nn__Error nn__Rate nn__Training pobj_as_ auxpass_employed_are parataxis_trained_employed prep_employed_in pobj_in_order aux_learn_to xcomp_employed_learn nn_weights_feature nn_weights_function dobj_learn_weights prep_learn_for amod_models_log-linear pobj_for_models cc_trained_and nn_candidates_translation nsubjpass_produced_candidates auxpass_produced_are conj_trained_produced xcomp_produced_using amod_decoders_phrase-based dobj_using_decoders advmod_using_ prep_using_in pobj_in_combination prep_combination_with amod_models_n-gram nn_models_language pobj_with_models ' +p,3689,'6 Related Work The popular IBM models for statistical machine translation are described in ',Brown,'num_Work_6 amod_Work_Related det_models_The amod_models_popular nn_models_IBM nsubjpass_described_models prep_models_for amod_translation_statistical nn_translation_machine pobj_for_translation auxpass_described_are dep_Work_described prep_described_in ' +o,3690,'machine translation but also in other applications such as word sense disanabiguation and bilingnal lexicography ',Brown,'nn_translation_machine nsubj__translation dep_in_but dep_in_also dep__in amod_applications_other pobj_in_applications dep_as_such prep_applications_as nn_sense_word pobj_as_sense dep__disanabiguation dobj__ cc__and amod__bilingnal nn__lexicography conj__ ' +o,3691,'More recently , & Marcus -LRB- In press -RRB- apply transformation-based learning to the problem ',Ramshaw,'advmod_recently_More advmod_apply_recently nsubj_apply_ cc__& conj__Marcus dep__In pobj_In_press amod__transformation-based nn__learning dobj_apply_ prep_apply_to det_problem_the pobj_to_problem ' +o,3692,'In comparison , achieved 48 Table 1 : A summary of the experimental results on four polysemous words ',Yarowsky,'prep_achieved_In pobj_In_comparison nsubj_achieved_ num_Table_48 dobj_achieved_Table num_Table_1 det_summary_A dep_Table_summary prep_summary_of det_results_the amod_results_experimental pobj_of_results prep_achieved_on num_words_four amod_words_polysemous pobj_on_words ' +o,3693,'3 Synchronous Binarization Optimization by Cost Reduction As discussed in Section 1 , binarizing an SCFG in a fixed -LRB- left-heavy -RRB- way may lead to a large number of competing edges and consequently high risk of making search errors ',Zhang,'num_Optimization_3 nn_Optimization_Synchronous nn_Optimization_Binarization prep_Optimization_by nn_Reduction_Cost pobj_by_Reduction mark_lead_As csubj_lead_discussed prep_discussed_in pobj_in_Section num_Section_1 xcomp_discussed_binarizing det_SCFG_an dobj_binarizing_SCFG prep_SCFG_in det_way_a amod_way_fixed appos_way_left-heavy pobj_in_way advmod_lead_ aux_lead_may dep_Optimization_lead prep_lead_to det_number_a amod_number_large pobj_to_number prep_number_of amod_edges_competing pobj_of_edges dep_Optimization_and dep_Optimization_consequently amod_risk_high dep_consequently_risk prep_risk_of pcomp_of_making nn_errors_search dobj_making_errors ' +o,3694,'However , CHECK moves are almost always about some information which the speaker has been told - a description that models the backward looking functionality of a dialogue act ',Carletta,'advmod_description_However nn_moves_CHECK nsubj_description_moves cop_description_are advmod_always_almost advmod_description_always prep_description_about det_information_some pobj_about_information dobj_told_which det_speaker_the nsubjpass_told_speaker aux_told_has auxpass_told_been ccomp_description_told dep_told_ det_description_a nsubj_models_that rcmod_description_models det_functionality_the advmod_looking_backward amod_functionality_looking dobj_models_functionality prep_functionality_of det_act_a nn_act_dialogue pobj_of_act ' +o,3695,'No artificial glue-rules or rule span limits were employed7 The parameters of the translation system were trained to maximize BLEU on the MT02 test set ',Och,'det_glue-rules_No amod_glue-rules_artificial nsubj_employed7_glue-rules cc_glue-rules_or nn_limits_rule nn_limits_span conj_glue-rules_limits cop_employed7_were det_parameters_The nsubjpass_trained_parameters prep_parameters_of det_system_the nn_system_translation pobj_of_system auxpass_trained_were rcmod_employed7_trained aux_maximize_to xcomp_trained_maximize dobj_maximize_BLEU prep_maximize_on det_set_the nn_set_MT02 nn_set_test pobj_on_set ' +o,3696,'23 Forest minimum error training To tune the feature weights of our system , we used a variant of the minimum error training algorithm that computes the error statistics from the target sentences from the translation search space -LRB- represented by a packed forest -RRB- that are exactly those that are minimally discriminable by changing the feature weights along a single vector in the dimensions of the feature space ',Och,'num_error_23 nn_error_Forest nn_error_minimum dep_used_training aux_tune_To xcomp_training_tune det_weights_the nn_weights_feature dobj_tune_weights prep_weights_of poss_system_our pobj_of_system nsubj_used_we dep_error_used det_variant_a dobj_used_variant prep_variant_of det__the amod__minimum nn__error nn__training nn__algorithm pobj_of_ nsubj_computes_that rcmod_variant_computes det_statistics_the nn_statistics_error dobj_computes_statistics prep_computes_from det_sentences_the nn_sentences_target pobj_from_sentences prep_sentences_from det_space_the nn_space_translation nn_space_search pobj_from_space dep_space_represented prep_represented_by det_forest_a amod_forest_packed pobj_by_forest nsubj_those_that cop_those_are advmod_those_exactly rcmod_space_those nsubj_discriminable_that cop_discriminable_are advmod_discriminable_minimally rcmod_those_discriminable prep_discriminable_by pcomp_by_changing det_weights_the nn_weights_feature dobj_changing_weights prep_changing_along det_vector_a amod_vector_single pobj_along_vector prep_vector_in det_dimensions_the pobj_in_dimensions prep_dimensions_of det__the nn__feature nn__space pobj_of_ ' +o,3697,'The production weights are estimated either by heuristic counting or using the EM algorithm ',Koehn,'det_weights_The nn_weights_production nsubjpass_estimated_weights auxpass_estimated_are advmod_estimated_either prep_estimated_by amod__heuristic amod__counting pobj_by_ cc_estimated_or conj_estimated_using det_algorithm_the nn_algorithm_EM dobj_using_algorithm ' +o,3698,'Consequently , considerable effort has gone into devising and improving automatic word alignment algorithms , and into evaluating their performance ',Fraser,'advmod_gone_Consequently amod_effort_considerable nsubj_gone_effort aux_gone_has prep_gone_into pcomp_into_devising cc_devising_and conj_devising_improving amod_algorithms_automatic nn_algorithms_word nn_algorithms_alignment dobj_devising_algorithms cc_into_and conj_into_into pcomp_into_evaluating poss_performance_their dobj_evaluating_performance ' +o,3699,'Baron and Hirst extracted collocations with Xtract and classified the collocations using the orientations of the words in the neighboring sentences ',Smadja,'nn__Baron cc_Baron_and conj_Baron_Hirst nsubj_extracted_ dobj_extracted_collocations prep_extracted_with nn__Xtract pobj_with_ cc_extracted_and conj_extracted_classified det_collocations_the dobj_classified_collocations xcomp_classified_using det_orientations_the dobj_using_orientations prep_orientations_of det_words_the pobj_of_words prep_words_in det_sentences_the amod_sentences_neighboring pobj_in_sentences ' +o,3700,' proposed a symmetrical measure : Par Lin -LRB- s t -RRB- = summationtext fF s F t -LRB- w -LRB- s , f -RRB- + w -LRB- t , f -RRB- -RRB- summationtext fF s w -LRB- s , f -RRB- + summationtext fF t w -LRB- t , f -RRB- , where F s and F t denote sets of features with positive weights for words s and t , respectively ',Lin,'advmod_proposed_ det_measure_a amod_measure_symmetrical dobj_proposed_measure nn_Lin_Par dep_measure_Lin amod_t_s appos_Lin_t rcmod_Lin_= amod_t_summationtext nn_t_fF nn_t_s nn_t_F dobj_=_t dep_t_w dep_w_s appos_s_f amod_w_+ dobj_w_w dep_w_t appos_t_f nn_fF_summationtext nsubj_w_fF dep_w_s rcmod_t_w dep_w_s appos_s_f cc_w_+ conj_w_summationtext nn_w_fF nn_w_t nsubj_proposed_w dep_w_t appos_t_f advmod_t_where nn_s_F nsubj_t_s cc_s_and conj_s_F rcmod_w_t amod_sets_denote dobj_t_sets prep_sets_of pobj_of_features prep_features_with amod_weights_positive pobj_with_weights prep_t_for nn_s_words pobj_for_s cc_s_and conj_s_t advmod_t_respectively ' +o,3701,'159 21 Baseline System The baseline system is a phrase-based SMT system , built almost entirely using freely available components ',Koehn,'number_21_159 num_System_21 nn_System_Baseline det_system_The nn_system_baseline nsubj__system cop__is det__a amod__phrase-based nn__SMT nn__system dep_System_ partmod__built advmod_entirely_almost advmod_built_entirely xcomp_built_using advmod_available_freely amod_components_available dobj_using_components ' +o,3702,'Many reordering constraints have been used for word reorderings , such as ITG constraints , IBM constraints and local constraints ',Berger,'amod_constraints_Many nn_constraints_reordering nsubjpass_used_constraints aux_used_have auxpass_used_been prep_used_for nn_reorderings_word pobj_for_reorderings dep_as_such prep_reorderings_as nn__ITG nn__constraints pobj_as_ nn__IBM nn__constraints conj__ cc__and amod__local nn__constraints conj__ ' +o,3703,'For transfer-learning baseline , we implement traditional SCL model -LRB- T-SCL -RRB- ',Blitzer,'prep_implement_For amod_baseline_transfer-learning pobj_For_baseline nsubj_implement_we amod_model_traditional nn_model_SCL dobj_implement_model appos_model_T-SCL ' +o,3704,'We rescore the ASR N-best lists with the standard HMM and IBM MT models ',Brown,'nsubj_rescore_We det_N-best_the nn_N-best_ASR nsubj_lists_N-best ccomp_rescore_lists prep_lists_with det__the amod__standard nn__HMM pobj_with_ cc__and nn_models_IBM nn_models_ nn_models_MT conj__models ' +p,3705,'We use five sentiment classification datasets , including the widely-used movie review dataset -LRB- MOV -RRB- as well as four datasets containing reviews of four different types of products from Amazon -LRB- books -LRB- BOO -RRB- , DVDs -LRB- DVD -RRB- , electronics -LRB- ELE -RRB- , and kitchen appliances -LRB- KIT -RRB- -RRB- ',Pang,'nsubj_use_We num_datasets_five nn_datasets_sentiment nn_datasets_classification dobj_use_datasets prep_datasets_including det_dataset_the amod_dataset_widely-used nn_dataset_movie nn_dataset_review pobj_including_dataset abbrev_dataset_MOV dep_dataset_ dep_well_as cc_dataset_well dep_well_as num_datasets_four conj_dataset_datasets partmod_datasets_containing dobj_containing_reviews prep_reviews_of num_types_four amod_types_different pobj_of_types prep_types_of pobj_of_products prep_containing_from pobj_from_Amazon dep_Amazon_books abbrev_books_BOO conj_books_DVDs abbrev_DVDs_DVD conj_books_electronics abbrev_electronics_ELE cc_books_and nn_appliances_kitchen conj_books_appliances abbrev_appliances_KIT partmod_Amazon_ ' +o,3706,'While choosing an optimum window size for an application is often subject to trial and error , there are some generally recognized trade-offs between small versus large windows , such as the impact of data-sparseness , and the nature of the associations retrieved Measures based on distance between words in the text ',Rapp,'mark_subject_While csubj_subject_choosing det_size_an amod_size_optimum nn_size_window dobj_choosing_size prep_choosing_for det_application_an pobj_for_application cop_subject_is advmod_subject_often advcl_are_subject prep_subject_to pobj_to_trial cc_trial_and conj_trial_error expl_are_there det_trade-offs_some advmod_recognized_generally amod_trade-offs_recognized nsubj_are_trade-offs prep_trade-offs_between pobj_between_small prep_small_versus amod_windows_large pobj_versus_windows dep_as_such prep_small_as det_impact_the pobj_as_impact prep_impact_of pobj_of_data-sparseness cc_trade-offs_and det_nature_the conj_trade-offs_nature prep_nature_of det_associations_the pobj_of_associations partmod_associations_retrieved amod_Measures_ dobj_retrieved_Measures prep_retrieved_based dep_based_on pobj_on_distance prep_distance_between pobj_between_words prep_words_in det_text_the pobj_in_text ' +o,3707,' , and the third type is a mixture of the first and second type , employing n-gram and grammarbased features , eg ',Cahill,'cc__and det_type_the amod_type_third conj__type cop_mixture_is det_mixture_a rcmod_type_mixture prep_mixture_of det_type_the amod_type_first cc_first_and conj_first_second pobj_of_type partmod_mixture_employing amod_features_n-gram cc_n-gram_and conj_n-gram_grammarbased dobj_employing_features appos_features_eg ' +o,3708,'The task originally emerged as an intermediate result of training the IBM translation models ',Brown,'det_task_The nsubj_emerged_task advmod_emerged_originally prep_emerged_as det_result_an amod_result_intermediate pobj_as_result prep_result_of pcomp_of_training det_models_the nn_models_IBM nn_models_translation dobj_training_models ' +o,3709,' also states that in the behavioral sciences , K -RRB- 8 signals good replicability , and 67 -LRB- K -LRB- 8 allows tentative conclusions to be drawn ',Carletta,'nsubj_states_ advmod_states_also dobj_states_that prep_that_in det_sciences_the amod_sciences_behavioral dep_in_sciences appos_sciences_K num_signals_8 measure_good_signals amod_replicability_good pobj_in_replicability cc_replicability_and conj_replicability_67 dep_67_K number_allows_8 dep_K_allows amod_conclusions_tentative dep_allows_conclusions aux_drawn_to auxpass_drawn_be infmod_that_drawn ' +o,3710,'Furthermore , these systems have tackled the problem at different levels of granularity , from the document level , sentence level , phrase level , as well as the speaker level in debates ',Thomas,'advmod_tackled_Furthermore det_systems_these nsubj_tackled_systems aux_tackled_have det_problem_the dobj_tackled_problem prep_problem_at amod_levels_different pobj_at_levels prep_levels_of pobj_of_granularity prep_levels_from det__the nn__document nn__level pobj_from_ nn__sentence nn__level conj_problem_ nn__phrase nn__level appos__ dep_well_as cc_problem_well dep_well_as det_level_the nn_level_speaker conj_problem_level prep_level_in pobj_in_debates ' +o,3711,'Aspect-based sentiment analysis summarizes sentiments with diverse attributes , so that customers may have to look more closely into analyzed sentiments ',Titov,'amod_analysis_Aspect-based nn_analysis_sentiment nsubj_summarizes_analysis dobj_summarizes_sentiments prep_sentiments_with amod_attributes_diverse pobj_with_attributes advmod_have_so dep_have_that nsubj_have_customers aux_have_may dep_summarizes_have aux_look_to xcomp_have_look advmod_closely_more advmod_look_closely prep_look_into amod_sentiments_analyzed pobj_into_sentiments ' +o,3712,'The final model V uses the weight vector w = summationtextk j = 1 -LRB- cjwj -RRB- Tn ',Collins,'det_V_The amod_V_final nn_V_model nsubj_uses_V det_w_the nn_w_weight nn_w_vector nsubj_Tn_w dep_Tn_= nn_Tn_summationtextk nn_Tn_j nn_Tn_= num_Tn_1 nn_Tn_cjwj ccomp_uses_Tn ' +o,3713,'Appendix B gives a sketch of one such approach , which is based on results from ',Collins,'nn_B_Appendix nsubj_gives_B det_sketch_a iobj_gives_sketch prep_sketch_of pobj_of_one amod_approach_such dobj_gives_approach nsubjpass_based_which auxpass_based_is rcmod_approach_based prep_based_on pobj_on_results prep_based_from ' +o,3714,'As the third test set we selected all tokens of the Brown corpus part of the Penn Treebank , a selected portion of the original one-million word Brown corpus , a collection of samples of American English in many different genres , from sources printed in 1961 ; we refer to this test set as BROWN ',Marcus,'prep_selected_As det_set_the amod_set_third nn_set_test pobj_As_set nsubj_selected_we dep_tokens_all dobj_selected_tokens prep_tokens_of det_part_the nn_part_Brown nn_part_corpus pobj_of_part prep_part_of det__the nn__Penn nn__Treebank pobj_of_ det_portion_a amod_portion_selected appos_tokens_portion prep_portion_of det__the amod__original amod__one-million nn__word nn__Brown nn__corpus pobj_of_ det_collection_a appos_tokens_collection prep_collection_of pobj_of_samples prep_samples_of nn_English_American pobj_of_English prep_English_in amod_genres_many amod_genres_different pobj_in_genres prep_selected_from pobj_from_sources partmod_sources_printed prep_printed_in pobj_in_1961 nsubj_refer_we parataxis_selected_refer prep_refer_to det_test_this pobj_to_test partmod_test_set prep_set_as pobj_as_BROWN ' +o,3715,'2 Architecture of the system The goal of statistical machine translation (SMT) is to produce a target sentence e from a source sentence f. It is today common practice to use phrases as translation units (Koehn et al., 2003; Och and Ney, 2003) and a log linear framework in order to introduce several models explaining the translation process: e = argmaxp(e|f) = argmaxe {exp(summationdisplay i ihi(e,f))} (1) The feature functions hi are the system models and the i weights are typically optimized to maximize a scoring function on a development set (Och and Ney, 2002).',Koehn,'' +o,3716,'Some studies have been done for acquiring collocation translations using parallel corpora ',Smadja,'det_studies_Some nsubjpass_done_studies aux_done_have auxpass_done_been prep_done_for pcomp_for_acquiring nn_translations_collocation dobj_acquiring_translations xcomp_acquiring_using amod_corpora_parallel dobj_using_corpora ' +o,3717,'In this paper , translation quality is evaluated according to -LRB- 1 -RRB- the BLEU metrics which calculates the geometric mean of ngram precision by the system output with respect to reference translations , and -LRB- 2 -RRB- the METEOR metrics that calculates unigram overlaps between translations ',Banerjee,'dep__In det_paper_this pobj_In_paper nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_is dep__evaluated prep_evaluated_according dep_according_to dep__1 det_metrics_the amod_metrics_BLEU nsubj__metrics nsubj_calculates_which rcmod_metrics_calculates det_mean_the amod_mean_geometric dobj_calculates_mean prep_mean_of nn_precision_ngram pobj_of_precision prep_calculates_by det_output_the nn_output_system pobj_by_output prep_output_with pobj_with_respect prep_calculates_to nn_translations_reference pobj_to_translations cc__and dep_overlaps_2 det_metrics_the amod_metrics_METEOR nsubj_overlaps_metrics nsubj_calculates_that rcmod_metrics_calculates dobj_calculates_unigram conj__overlaps prep_overlaps_between pobj_between_translations ' +o,3718,'We used the Penn Treebank WSJ corpus to perform the empirical evaluation of the considered approaches ',Marcus,'nsubj_used_We det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_WSJ nsubj__corpus ccomp_used_ aux_perform_to xcomp__perform det_evaluation_the amod_evaluation_empirical dobj_perform_evaluation prep_evaluation_of det_approaches_the amod_approaches_considered pobj_of_approaches ' +o,3719,'We compare our methods with both the averaged perceptron and conditional random fields using identical predicate sets ',Collins,'nsubj_compare_We poss_methods_our dobj_compare_methods prep_compare_with preconj__both det__the amod__averaged nn__perceptron nsubj_using_ cc__and amod_fields_conditional amod_fields_random conj__fields advmod_using_ pcomp_with_using amod_sets_identical nn_sets_predicate dobj_using_sets ' +o,3720,'We report case-insensitive scores on version 06 of METEOR with all modules enabled , version 104 of IBM-style BLEU , and version 5 of TER ',Papineni,'nsubj_report_We amod_scores_case-insensitive dobj_report_scores prep_scores_on pobj_on_version num_version_06 prep_version_of nn__METEOR pobj_of_ prep__with det_modules_all pobj_with_modules partmod_modules_enabled conj_scores_version num_version_104 prep_version_of nn__IBM-style nn__BLEU pobj_of_ cc_scores_and conj_scores_version num_version_5 prep_version_of nn__TER pobj_of_ ' +o,3721,'For more information on these models , please refer to ',Brown,'prep_please_For amod_information_more pobj_For_information prep_information_on det_models_these pobj_on_models dep_please_refer prep_refer_to ' +o,3722,'For example , developed a system to identify inflammatory texts and developed methods for classifying reviews as positive or negative ',Pang,'prep_developed_For pobj_For_example nsubj_developed_ det_system_a dobj_developed_system aux_identify_to infmod_system_identify amod_texts_inflammatory dobj_identify_texts cc_identify_and conj_identify_ amod_methods_developed dobj__methods prep__for pcomp_for_classifying nsubj_positive_reviews advmod_positive_as xcomp_classifying_positive cc_positive_or conj_positive_negative ' +o,3723,'For Japanese sentences , instead of using full parse trees , existing sentence compression methods trim dependency trees by the discriminative model through the use of simple linear combined features ',Nomoto,'prep_trim_For amod_sentences_Japanese pobj_For_sentences dep_of_instead prep_trim_of pcomp_of_using amod_trees_full nn_trees_parse dobj_using_trees amod_methods_existing nn_methods_sentence nn_methods_compression nsubj_trim_methods amod_trees_dependency dobj_trim_trees prep_trim_by det__the amod__discriminative nn__model pobj_by_ prep_trim_through det_use_the pobj_through_use prep_use_of amod__simple amod__linear amod__combined nn__features pobj_of_ ' +p,3724,' shows that baseNP recognition -LRB- Fz = I = 920 -RRB- is easier than finding both NP and VP chunks -LRB- Fz = 1 = 881 -RRB- and that increasing the size of the training data increases the performance on the test set ',Ramshaw,'nsubj_shows_ dep_easier_that nn_recognition_baseNP nsubj_easier_recognition dep_920_Fz punct_920_= nsubj_920_I dep_920_= dep_recognition_920 cop_easier_is dep_shows_easier prep_easier_than pcomp_than_finding preconj_chunks_both nn_chunks_NP cc_NP_and conj_NP_VP dobj_finding_chunks nsubj_881_Fz dep_881_= num_881_1 amod_881_= parataxis_finding_881 cc_easier_and dep_increases_that csubj_increases_increasing det_size_the dobj_increasing_size prep_size_of det_data_the nn_data_training pobj_of_data conj_easier_increases det_performance_the dobj_increases_performance prep_performance_on det_set_the nn_set_test pobj_on_set ' +o,3725,'Maximum Entropy models implement the intuition that the best model is the one that is consistent with the set of constraints imposed by the evidence but otherwise is as uniform as possible ',Berger,'nn_models_Entropy nsubj_Maximum_models dep_intuition_implement det_intuition_the dep_Maximum_intuition complm_one_that det_model_the amod_model_best nsubj_one_model cop_one_is det_one_the ccomp_intuition_one nsubj_consistent_that cop_consistent_is rcmod_one_consistent prep_consistent_with det_set_the pobj_with_set prep_set_of pobj_of_constraints partmod_constraints_imposed prep_imposed_by det_evidence_the pobj_by_evidence cc_consistent_but advmod_consistent_otherwise cop_uniform_is advmod_uniform_as conj_consistent_uniform prep_uniform_as pobj_as_possible ' +o,3726,'Although there is a modest cost associated with annotating data , we show that a reduction of 40 \% relative in alignment error -LRB- AER -RRB- is possible over the GIZA + + aligner ',Och,'mark_is_Although expl_is_there advcl_show_is det_cost_a amod_cost_modest nsubj_is_cost partmod_cost_associated prep_associated_with amod_data_annotating pobj_with_data nsubj_show_we complm_possible_that det_reduction_a nsubj_possible_reduction prep_reduction_of num_\%_40 pobj_of_\% dep_in_relative prep_\%_in amod_error_alignment pobj_in_error abbrev_error_AER cop_possible_is ccomp_show_possible prep_possible_over det_aligner_the amod_aligner_GIZA cc_GIZA_+ conj_GIZA_+ pobj_over_aligner ' +o,3727,'1113 : Recursive DP equations for summing over t and a alignments are treated as a hidden variable to be marginalized out10 Optimization problems of this form are by now widely known in NLP , and have recently been used for machinetranslationaswell ',Koo,'nn_DP_Recursive nsubj_equations_DP dep_1113_equations prep_equations_for pcomp_for_summing prep_summing_over pobj_over_t cc_equations_and det_alignments_a nsubjpass_treated_alignments auxpass_treated_are conj_equations_treated prep_treated_as det_variable_a amod_variable_hidden pobj_as_variable aux_marginalized_to auxpass_marginalized_be infmod_variable_marginalized amod_problems_out10 amod_problems_Optimization nsubjpass_known_problems prep_problems_of det_form_this pobj_of_form auxpass_known_are prep_known_by pobj_by_now advmod_known_widely ccomp_marginalized_known prep_known_in nn__NLP pobj_in_ cc_treated_and aux_used_have advmod_used_recently auxpass_used_been conj_treated_used prep_used_for amod__machinetranslationaswell pobj_for_ ' +o,3728,'BLEU Score : BLEU is an automatic metric designed by IBM , which uses several references ',Papineni,'amod_Score_BLEU nsubj_metric_BLEU cop_metric_is det_metric_an amod_metric_automatic dep_Score_metric partmod_metric_designed prep_designed_by pobj_by_IBM nsubj_uses_which rcmod_IBM_uses amod_references_several dobj_uses_references ' +o,3729,'There have been many approaches to compute the similarity between words based on their distribution in a corpus ',Hindle,'expl_been_There aux_been_have amod_approaches_many dobj_been_approaches aux_compute_to infmod_approaches_compute det_similarity_the dobj_compute_similarity prep_similarity_between pobj_between_words prep_compute_based dep_based_on poss_distribution_their pobj_on_distribution prep_distribution_in det_corpus_a pobj_in_corpus ' +o,3730,'For example , since the Collins parser depends on a prior part-of-speech tagger , we included the time for POS tagging in our Collins measurements ',Ratnaparkhi,'prep_included_For pobj_For_example mark_depends_since det_parser_the nn_parser_Collins nsubj_depends_parser dep_included_depends prep_depends_on det__a amod__prior amod__part-of-speech nn__tagger pobj_on_ nsubj_included_we det_time_the dobj_included_time prep_included_for pobj_for_POS xcomp_included_tagging prep_tagging_in poss_measurements_our nn_measurements_Collins pobj_in_measurements ' +o,3731,'In all experiments , word alignment was obtained using the grow-diag-final heuristic for symmetrizing GIZA + + alignments ',Och,'prep_obtained_In det_experiments_all pobj_In_experiments nn_alignment_word nsubjpass_obtained_alignment auxpass_obtained_was xcomp_obtained_using det_heuristic_the amod_heuristic_grow-diag-final dobj_using_heuristic prep_using_for pcomp_for_symmetrizing amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ amod_alignments_ dobj_symmetrizing_alignments ' +o,3732,'In this paper , we implement the SDB model in a state-of-the-art phrase-based system which adapts a binary bracketing transduction grammar -LRB- BTG -RRB- to phrase translation and reordering , described in ',Wu,'prep_implement_In det_paper_this pobj_In_paper nsubj_implement_we det_model_the nn_model_SDB dobj_implement_model prep_implement_in det_system_a amod_system_state-of-the-art amod_system_phrase-based pobj_in_system nsubj_adapts_which rcmod_system_adapts det_grammar_a advmod_bracketing_binary amod_grammar_bracketing nn_grammar_transduction nsubj__grammar abbrev_grammar_BTG dep_implement_ prep__to nn_translation_phrase pobj_to_translation cc_translation_and conj_translation_reordering dep__described prep_described_in pobj_in_ ' +o,3733,'Many studies focus on rare words ; butterflies are more interesting than moths ',Dunning,'amod_studies_Many nsubj_focus_studies prep_focus_on amod__rare nn__words pobj_on_ nsubj_interesting_butterflies cop_interesting_are advmod_interesting_more parataxis_focus_interesting prep_interesting_than pobj_than_moths ' +o,3734,'In this paper we use a non-projective dependency tree CRF ',Smith,'prep_use_In det_paper_this pobj_In_paper nsubj_use_we det_CRF_a amod_CRF_non-projective amod_CRF_dependency nn_CRF_tree dobj_use_CRF ' +o,3735,'For a full description of the algorithm , see ',Collins,'prep_see_For det_description_a amod_description_full pobj_For_description prep_description_of det_algorithm_the pobj_of_algorithm ' +o,3736,'This direction has been forming the mainstream of research on opinion-sensitive text processing ',Turney,'det_direction_This nsubj_forming_direction aux_forming_has aux_forming_been det_mainstream_the dobj_forming_mainstream prep_mainstream_of pobj_of_research prep_forming_on amod_processing_opinion-sensitive nn_processing_text pobj_on_processing ' +o,3737,'The term global feature vector is used by to distinguish between feature count vectors for whole sequences and the local feature vectors in ME tagging models , which are Boolean valued vectors containing the indicator features for one element in the sequence ',Collins,'det_term_The amod_vector_global nn_vector_feature nsubjpass_used_vector auxpass_used_is dep_term_used prep_used_by pobj_by_ aux_distinguish_to xcomp_used_distinguish prep_distinguish_between nn_vectors_feature nn_vectors_count pobj_between_vectors prep_vectors_for amod_sequences_whole pobj_for_sequences cc_sequences_and det_vectors_the amod_vectors_local nn_vectors_feature conj_sequences_vectors prep_distinguish_in nn_models_ME nn_models_tagging pobj_in_models nsubj_Boolean_which cop_Boolean_are rcmod_models_Boolean partmod_Boolean_valued nsubj_containing_vectors dep_term_containing det_features_the nn_features_indicator dobj_containing_features prep_containing_for num_element_one pobj_for_element prep_element_in det_sequence_the pobj_in_sequence ' +n,3738,'The method was intended as a replacement for sentence-based methods -LRB- eg , -RRB- , which are very sensitive to noise ',Brown,'det_method_The nsubjpass_intended_method auxpass_intended_was prep_intended_as det_replacement_a pobj_as_replacement prep_replacement_for amod_methods_sentence-based pobj_for_methods appos_methods_eg dep_eg_ nsubj_sensitive_which cop_sensitive_are advmod_sensitive_very rcmod_methods_sensitive prep_sensitive_to pobj_to_noise ' +o,3739,'edu Abstract This paper reports on our experience hand tagging the senses of 25 of the most frequent verbs in 12,925 sentences of the Wall Street Journal Treebank corpus ',Marcus,'nn_Abstract_edu det_paper_This nsubj_reports_paper dep_Abstract_reports prep_reports_on poss_hand_our nn_hand_experience pobj_on_hand partmod_hand_tagging det_senses_the dobj_tagging_senses prep_senses_of pobj_of_25 prep_senses_of det_verbs_the advmod_verbs_most amod_verbs_frequent pobj_of_verbs prep_verbs_in num_sentences_12,925 pobj_in_sentences prep_sentences_of det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nn_corpus_Treebank pobj_of_corpus ' +o,3740,'Additionally , we present results of the tagger on the NEGRA corpus and the Penn Treebank ',Marcus,'advmod_present_Additionally nsubj_present_we dobj_present_results prep_results_of det_tagger_the pobj_of_tagger prep_tagger_on det__the nn__NEGRA nn__corpus pobj_on_ cc__and det_Treebank_the nn_Treebank_Penn conj__Treebank ' +o,3741,'In this paper we will compare and evaluate several aspects of these techniques , focusing on Minimum Error Rate -LRB- MER -RRB- training and Minimum Bayes Risk -LRB- MBR -RRB- decision rules , within a novel training environment that isolates the impact of each component of these methods ',Och,'prep_compare_In det_paper_this pobj_In_paper nsubj_compare_we aux_compare_will cc_compare_and conj_compare_evaluate amod_aspects_several dobj_compare_aspects prep_aspects_of det_techniques_these pobj_of_techniques partmod_aspects_focusing prep_focusing_on nn_Rate_Minimum nn_Rate_Error nn__Rate abbrev__MER nn__training pobj_on_ cc__and nn_Risk_Minimum nn_Risk_Bayes conj__Risk abbrev_Risk_MBR nn_rules_decision dep_Risk_rules prep_focusing_within det_environment_a amod_environment_novel nn_environment_training pobj_within_environment nsubj_isolates_that rcmod_environment_isolates det_impact_the dobj_isolates_impact prep_impact_of det_component_each pobj_of_component prep_component_of det_methods_these pobj_of_methods ' +o,3742,'This situation is very similar to the training process of translation models in statistical machine translation , where parallel corpus is used to find the mappings between words from different languages by exploiting their co-occurrence patterns ',Brown,'det_situation_This nsubj_similar_situation cop_similar_is advmod_similar_very prep_similar_to det_process_the nn_process_training pobj_to_process prep_process_of nn_models_translation pobj_of_models prep_similar_in amod__statistical nn__machine nn__translation pobj_in_ advmod_used_where amod_corpus_parallel nsubjpass_used_corpus auxpass_used_is advcl_similar_used aux_find_to xcomp_used_find det_mappings_the dobj_find_mappings prep_mappings_between pobj_between_words prep_words_from amod_languages_different pobj_from_languages prep_find_by pcomp_by_exploiting poss_patterns_their amod_patterns_co-occurrence dobj_exploiting_patterns ' +o,3743,'We also implemented an averaged perceptron system -LRB- another online learning algorithm -RRB- for comparison ',Collins,'nsubj_implemented_We advmod_implemented_also det__an amod__averaged nn__perceptron nn__system dobj_implemented_ det_algorithm_another amod_algorithm_online nn_algorithm_learning appos__algorithm prep_implemented_for pobj_for_comparison ' +o,3744,'A variety of methods have been applied , ranging from simple frequency , modified frequency measures such as c-values and standard statistical significance tests such as the t-test , the chi-squared test , and loglikelihood , and information-based methods , eg pointwise mutual information ',Church,'det_variety_A nsubjpass_applied_variety prep_variety_of pobj_of_methods aux_applied_have auxpass_applied_been xcomp_applied_ranging prep_ranging_from amod__simple nn__frequency pobj_from_ amod_measures_modified nn_measures_frequency appos__measures dep_as_such prep_measures_as amod__c-values pobj_as_ cc__and amod_tests_standard amod_tests_statistical nn_tests_significance conj__tests dep_as_such prep__as det_t-test_the pobj_as_t-test det_test_the amod_test_chi-squared conj_t-test_test cc_t-test_and amod__loglikelihood conj_t-test_ cc__and amod_methods_information-based conj__methods amod_information_eg amod_information_pointwise amod_information_mutual appos_methods_information ' +p,3745,'Movies Reviews : This is a popular dataset in sentiment analysis literature ',Pang,'nn_Reviews_Movies dep_dataset_Reviews nsubj_dataset_This cop_dataset_is det_dataset_a amod_dataset_popular prep_dataset_in nn_literature_sentiment nn_literature_analysis pobj_in_literature ' +p,3746,'36 Parameter Estimation To estimate parameters k -LRB- 1 k K -RRB- , lm , and um , we adopt the approach of minimum error rate training -LRB- MERT -RRB- that is popular in SMT ',Och,'num_Estimation_36 nn_Estimation_Parameter aux_estimate_To dep_adopt_estimate nn_k_parameters dobj_estimate_k num_K_1 nn_K_k appos_k_K conj_k_lm cc_k_and conj_k_um nsubj_adopt_we dep_Estimation_adopt det_approach_the dobj_adopt_approach prep_approach_of amod_training_minimum nn_training_error nn_training_rate pobj_of_training abbrev_approach_MERT nsubj_popular_that cop_popular_is rcmod_approach_popular prep_popular_in pobj_in_SMT ' +o,3747,'We show that our semi-supervised approach yields improvements for fixed datasets by performing parsing experiments on the Penn Treebank and Prague Dependency Treebank -LRB- see Sections 41 and 43 -RRB- ',Marcus,'nsubj_show_We complm_yields_that poss_approach_our amod_approach_semi-supervised nsubj_yields_approach ccomp_show_yields dobj_yields_improvements prep_improvements_for amod_datasets_fixed pobj_for_datasets prep_datasets_by pcomp_by_performing amod_experiments_parsing dobj_performing_experiments prep_performing_on det__the nn__Penn nn__Treebank pobj_on_ cc__and nn__Prague nn__Dependency nn__Treebank conj__ parataxis_yields_see dobj_see_Sections num_Sections_41 cc_Sections_and conj_Sections_43 ' +o,3748,'Deterministic Annealing : In this system , instead of using the regular MERT whose training objective is to minimize the onebest error , we use the deterministic annealing training procedure described in Smith and Eisner , whose objective is to minimize the expected error -LRB- together with the entropy regularization technique -RRB- ',Och,'nn_Annealing_Deterministic prep_use_In det_system_this pobj_In_system dep_of_instead prep_system_of pcomp_of_using det__the amod__regular nn__MERT dobj_using_ poss_objective_whose nn_objective_training nsubj_is_objective rcmod__is aux_minimize_to xcomp_is_minimize det_error_the amod_error_onebest dobj_minimize_error nsubj_use_we dep_Annealing_use det_procedure_the amod_procedure_deterministic amod_procedure_annealing nn_procedure_training dobj_use_procedure partmod_procedure_described prep_described_in nn__Smith cc_Smith_and conj_Smith_Eisner pobj_in_ poss_objective_whose nsubj_is_objective rcmod__is aux_minimize_to xcomp_is_minimize det_error_the amod_error_expected dobj_minimize_error dep_with_together dep_minimize_with det_technique_the amod_technique_entropy nn_technique_regularization pobj_with_technique ' +o,3749,'Since many concepts are expressed by idiomatic multiword expressions instead of single words , and different languages may realize the same concept using different numbers of words , word alignment based methods , which are highly dependent on the probability information at the lexical level , are not well suited for this type of translation ',Wu,'mark_expressed_Since amod_concepts_many nsubjpass_expressed_concepts auxpass_expressed_are advcl_suited_expressed prep_expressed_by amod_expressions_idiomatic amod_expressions_multiword pobj_by_expressions dep_of_instead prep_expressions_of amod_words_single pobj_of_words cc_expressed_and amod_languages_different nsubj_realize_languages aux_realize_may conj_expressed_realize det_concept_the amod_concept_same dobj_realize_concept partmod_concept_using amod_numbers_different dobj_using_numbers prep_numbers_of nn__words pobj_of_ nn_alignment_word nsubjpass_suited_alignment partmod_alignment_based dobj_based_methods nsubj_dependent_which cop_dependent_are advmod_dependent_highly rcmod_alignment_dependent prep_dependent_on det_information_the nn_information_probability pobj_on_information prep_information_at det_level_the amod_level_lexical pobj_at_level auxpass_suited_are neg_suited_not advmod_suited_well prep_suited_for det_type_this pobj_for_type prep_type_of pobj_of_translation ' +p,3750,'1 Introduction In recent years , Bracketing Transduction Grammar -LRB- BTG -RRB- proposed by has been widely used in statistical machine translation -LRB- SMT -RRB- ',Wu,'dep_used_1 dep_In_Introduction prep_used_In amod_years_recent pobj_In_years nn_Grammar_Bracketing nn_Grammar_Transduction nsubjpass_used_Grammar abbrev_Grammar_BTG partmod_Grammar_proposed prep_proposed_by pobj_by_ aux_used_has auxpass_used_been advmod_used_widely prep_used_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +o,3751,'In the following , ROUGE-SN denotes ROUGE-S with maximum skip distance N ROUGE-SU This measure is an extension of ROUGE-S ; it adds a unigram as a counting unit ',Lin,'prep__In det_following_the pobj_In_following nn_ROUGE-S_ROUGE-SN nn_ROUGE-S_denotes nsubj__ROUGE-S prep_ROUGE-S_with amod_ROUGE-SU_maximum amod_ROUGE-SU_skip nn_ROUGE-SU_distance nn_ROUGE-SU_N pobj_with_ROUGE-SU det_measure_This nsubj_extension_measure cop_extension_is det_extension_an ccomp__extension prep_extension_of pobj_of_ROUGE-S nsubj_adds_it parataxis__adds det_unigram_a dobj_adds_unigram prep_unigram_as det_unit_a amod_unit_counting pobj_as_unit ' +p,3752,'While significant time savings have already been reported on the basis of automatic pre-tagging -LRB- eg , for POS and parse tree taggings in the Penn TreeBank , or named entity taggings for the Genia corpus -RRB- , this kind of pre-processing does not reduce the number of text tokens actually to be considered ',Marcus,'mark_reported_While amod_savings_significant nn_savings_time nsubjpass_reported_savings aux_reported_have advmod_reported_already auxpass_reported_been advcl_reduce_reported prep_reported_on det_basis_the pobj_on_basis prep_basis_of amod_pre-tagging_automatic pobj_of_pre-tagging dep_basis_eg prep_eg_for amod_taggings_POS cc_POS_and conj_POS_parse nn_taggings_tree pobj_for_taggings prep_taggings_in det__the nn__Penn nn__TreeBank pobj_in_ cc_eg_or conj_eg_named nn_taggings_entity dobj_named_taggings prep_named_for det__the nn__Genia nn__corpus pobj_for_ det_kind_this nsubj_reduce_kind prep_kind_of pobj_of_pre-processing aux_reduce_does neg_reduce_not det_number_the dobj_reduce_number prep_number_of nn_tokens_text pobj_of_tokens advmod_reduce_actually aux_considered_to auxpass_considered_be xcomp_reduce_considered ' +o,3753,'The corpus lines retained are part-of-speech tagged ',Cutting,'det_lines_The nn_lines_corpus nsubj_retained_lines cop_tagged_are amod_tagged_part-of-speech ccomp_retained_tagged ' +o,3754,' used a quasisynchronous grammar to discover the correspondence between words implied by the correspondence between the trees ',Smith,'nsubj_used_ det_grammar_a amod_grammar_quasisynchronous dobj_used_grammar aux_discover_to xcomp_used_discover det_correspondence_the dobj_discover_correspondence prep_correspondence_between pobj_between_words partmod_words_implied prep_implied_by det_correspondence_the pobj_by_correspondence prep_correspondence_between det_trees_the pobj_between_trees ' +o,3755,' , which is based on that of ',Och,'dep_based_ nsubjpass_based_which auxpass_based_is prep_based_on pobj_on_that prep_that_of ' +o,3756,'BLEU is one of the methods for automatic evaluation of translation quality ',Papineni,'nn__BLEU nsubj_one_ cop_one_is prep_one_of det_methods_the pobj_of_methods prep_methods_for amod_evaluation_automatic pobj_for_evaluation prep_evaluation_of nn_quality_translation pobj_of_quality ' +o,3757,',2004 -RRB- appliedextractiontechniquessimilarto Xtractsystem ; Japanese : -LRB- Ikeharaetal ',Smadja,'dobj_,2004_appliedextractiontechniquessimilarto prep_,2004_Xtractsystem pobj_Xtractsystem_ rcmod__Japanese prep_Japanese_Ikeharaetal ' +o,3758,'A number of researches which utilized distributional similarity have been conducted , including and many others ',Hindle,'det_number_A nsubjpass_conducted_number prep_number_of pobj_of_researches nsubj_utilized_which rcmod_researches_utilized amod_similarity_distributional dobj_utilized_similarity aux_conducted_have auxpass_conducted_been prep_conducted_including amod_others_ cc__and conj__many pobj_including_others ' +o,3759,'Alignment spaces can emerge from generative stories , from syntactic notions , or they can be imposed to create competition between links ',Wu,'amod_spaces_Alignment nsubj_emerge_spaces aux_emerge_can prep_emerge_from amod__generative nn__stories pobj_from_ prep_emerge_from amod__syntactic nn__notions pobj_from_ cc_emerge_or nsubjpass_imposed_they aux_imposed_can auxpass_imposed_be conj_emerge_imposed aux_create_to purpcl_imposed_create dobj_create_competition prep_competition_between nn__links pobj_between_ ' +o,3760,'Given the parallel corpus , we tagged the English words with a publicly available maximum entropy tagger , and we used an implementation of the IBM translation model to align the words ',Ratnaparkhi,'prep_tagged_Given det_corpus_the amod_corpus_parallel dep_Given_corpus nsubj_tagged_we det_words_the amod_words_English dobj_tagged_words prep_words_with det__a advmod_available_publicly amod__available amod__maximum nn__entropy nn__tagger pobj_with_ cc_tagged_and nsubj_used_we conj_tagged_used det_implementation_an dobj_used_implementation prep_implementation_of det__the nn__IBM nn__translation nn__model pobj_of_ aux_align_to xcomp_used_align det_words_the dobj_align_words ' +o,3761,'53 Related works and discussion Our two-step model essentially belongs to the same category as the works of and ',Jing,'num_works_53 nn_works_Related cc_works_and conj_works_discussion poss_model_Our amod_model_two-step nsubj_belongs_model advmod_belongs_essentially dep_works_belongs prep_belongs_to det_category_the amod_category_same pobj_to_category prep_category_as det_works_the pobj_as_works prep_works_of nn_and_ pobj_of_and ' +o,3762,'It seems nevertheless that all 2Church and Hanks , use statistics in their algorithms to extract collocations from texts ',Smadja,'nsubj_seems_It advmod_seems_nevertheless complm_use_that det_2Church_all nsubj_use_2Church cc_2Church_and conj_2Church_Hanks num_Hanks_ num_Hanks_ ccomp_seems_use dobj_use_statistics prep_statistics_in poss_algorithms_their pobj_in_algorithms aux_extract_to infmod_algorithms_extract dobj_extract_collocations prep_extract_from pobj_from_texts ' +o,3763,'Before parsing , POS tags are assigned to the input sentence using our reimplementation of the POStagger from ',Collins,'prep_assigned_Before pobj_Before_parsing dep_tags_POS nsubjpass_assigned_tags auxpass_assigned_are prep_assigned_to det_sentence_the nn_sentence_input pobj_to_sentence xcomp_assigned_using poss_reimplementation_our dobj_using_reimplementation prep_reimplementation_of det_POStagger_the pobj_of_POStagger prep_using_from ' +o,3764,'1 Introduction For statistical machine translation -LRB- SMT -RRB- , phrasebased methods and syntax-based methods outperform word-based methods ',Och,'num_Introduction_1 dep_Introduction_For amod_translation_statistical nn_translation_machine pobj_For_translation abbrev_translation_SMT amod__phrasebased nn__methods conj_translation_ cc_translation_and amod_methods_syntax-based conj_translation_methods partmod_methods_ ccomp__outperform amod_methods_word-based iobj_outperform_methods dobj_outperform_ ' +o,3765,'The lexical acquisition phase uses the GIZA + + word-alignment tool , an implementation of IBM Model 5 to construct an alignment of MRs with NL strings ',Brown,'det_phase_The amod_phase_lexical nn_phase_acquisition nsubj_uses_phase det_tool_the amod_tool_GIZA cc_GIZA_+ conj_GIZA_+ amod_tool_word-alignment dobj_uses_tool det__an nn__implementation appos_tool_ prep__of nn__IBM nn__Model num__5 pobj_of_ aux_construct_to xcomp_uses_construct det_alignment_an dobj_construct_alignment prep_alignment_of pobj_of_MRs prep_construct_with nn_strings_NL pobj_with_strings ' +o,3766,'In the absence of an annotated corpus , dependencies can be derived by other means , eg part413 of-speech probabilities can be approximated from a raw corpus as in , word-sense dependencies can be derived as definition-based similarities , etc Label dependencies are set as weights on the arcs drawn between corresponding labels ',Cutting,'prep_approximated_In det_absence_the pobj_In_absence prep_absence_of det_corpus_an amod_corpus_annotated pobj_of_corpus nsubjpass_derived_dependencies aux_derived_can auxpass_derived_be parataxis_approximated_derived prep_derived_by amod_means_other pobj_by_means amod_probabilities_eg amod_probabilities_part413 amod_probabilities_of-speech nsubjpass_approximated_probabilities aux_approximated_can auxpass_approximated_be prep_approximated_from det_corpus_a amod_corpus_raw pobj_from_corpus mark_derived_as prep_derived_in pobj_in_ amod_dependencies_word-sense nsubjpass_derived_dependencies aux_derived_can auxpass_derived_be advcl_set_derived prep_derived_as amod_similarities_definition-based pobj_as_similarities nn_dependencies_etc nn_dependencies_Label nsubjpass_set_dependencies auxpass_set_are dep_approximated_set prep_set_as pobj_as_weights prep_weights_on det_arcs_the pobj_on_arcs partmod_arcs_drawn prep_drawn_between amod_labels_corresponding pobj_between_labels ' +o,3767,'The description of the minimum cut framework in Section 41 was inspired by ',Pang,'det_description_The nsubjpass_inspired_description prep_description_of det_framework_the amod_framework_minimum nn_framework_cut pobj_of_framework prep_framework_in pobj_in_Section num_Section_41 auxpass_inspired_was prep_inspired_by ' +o,3768,'As we noted in Section 5 , we are able to significantly outperform basic structural correspondence learning ',Blitzer,'mark_noted_As nsubj_noted_we advcl_able_noted prep_noted_in pobj_in_Section num_Section_5 nsubj_able_we cop_able_are aux_outperform_to advmod_outperform_significantly xcomp_able_outperform amod_correspondence_basic amod_correspondence_structural dobj_outperform_correspondence partmod_correspondence_learning ' +o,3769,' demonstrates that manual mappings can be created for a small number of words with relative ease , but for a very large number of words the e ort involved in mapping would approach presented involves no be considerable ',Mihalcea,'nsubj_demonstrates_ dep_created_that amod_mappings_manual nsubjpass_created_mappings aux_created_can auxpass_created_be dep_demonstrates_created prep_created_for det_number_a amod_number_small pobj_for_number prep_number_of pobj_of_words prep_words_with amod_ease_relative pobj_with_ease cc_created_but mark_involves_for det_number_a advmod_large_very amod_number_large nsubj_involves_number prep_number_of pobj_of_words det_ort_the amod_ort_e nsubj_approach_ort partmod_ort_involved prep_involved_in pobj_in_mapping aux_approach_would rcmod_words_approach dep_approach_presented conj_created_involves dep_considerable_no cop_considerable_be ccomp_involves_considerable ' +o,3770,'The output of a contextfree parser , such as that of or Charniak -LRB- 2000 -RRB- , can be transformed into a sequence of shallow constituents for comparison with the output of a shallow parser ',Collins,'det_output_The nsubjpass_transformed_output prep_output_of det_parser_a amod_parser_contextfree pobj_of_parser dep_as_such prep_output_as pobj_as_that prep_that_of pobj_of_ cc__or conj__Charniak appos__2000 aux_transformed_can auxpass_transformed_be prep_transformed_into det_sequence_a pobj_into_sequence prep_sequence_of nn_constituents_shallow pobj_of_constituents prep_constituents_for pobj_for_comparison prep_transformed_with det_output_the pobj_with_output prep_output_of det_parser_a nn_parser_shallow pobj_of_parser ' +p,3771,'Veale used WordNet to answer 374 multiple-choice SAT analogy questions , achieving an accuracy of 43 \% , but the best corpus-based approach attains an accuracy of 56 \% ',Turney,'nn__Veale nsubj_used_ nsubj_answer_WordNet aux_answer_to xcomp_used_answer num_questions_374 amod_questions_multiple-choice nn_questions_SAT nn_questions_analogy dobj_answer_questions xcomp_used_achieving det_accuracy_an dobj_achieving_accuracy prep_accuracy_of num_\%_43 pobj_of_\% cc_used_but det_approach_the advmod_approach_best amod_approach_corpus-based nsubj_attains_approach conj_used_attains det_accuracy_an dobj_attains_accuracy prep_accuracy_of num_\%_56 pobj_of_\% ' +n,3772,'This approach addresses the problematic aspects of both pure knowledge-based generation -LRB- where incomplete knowledge is inevitable -RRB- and pure statistical bag generation -LRB- where the statistical system has no linguistic guidance -RRB- ',Brown,'det_approach_This nsubj_addresses_approach det_aspects_the amod_aspects_problematic dobj_addresses_aspects prep_aspects_of det_generation_both amod_generation_pure amod_generation_knowledge-based pobj_of_generation advmod_inevitable_where amod_knowledge_incomplete nsubj_inevitable_knowledge cop_inevitable_is dep_generation_inevitable cc_generation_and amod__pure amod__statistical nn__bag nn__generation conj_generation_ advmod_has_where det_system_the amod_system_statistical nsubj_has_system dep__has det_guidance_no amod_guidance_linguistic dobj_has_guidance ' +o,3773,'The sequence Ws is thought as a noisy version of WT and the best guess I -RRB- d ~ is then computed as ^ W ~ = argmax P -LRB- WTWs -RRB- wT = argmax P -LRB- WslWT -RRB- P -LRB- WT -RRB- -LRB- 1 -RRB- wT In they propose a method for maximizing P -LRB- WTIWs -RRB- by estimating P -LRB- WT -RRB- and P -LRB- WsIWT -RRB- and solving the problem in equation 1 ',Brown,'det_Ws_The nn_Ws_sequence nsubjpass_thought_Ws auxpass_thought_is mark_computed_as det_version_a amod_version_noisy nsubj_computed_version prep_version_of pobj_of_WT cc_WT_and det_I_the amod_I_best nn_I_guess conj_WT_I advmod_~_d amod_WT_~ cop_computed_is advmod_computed_then advcl_thought_computed mark_=_as nn_~_^ nn_~_W nsubj_=_~ advcl_computed_= nn_P_argmax dobj_=_P abbrev_P_WTWs partmod_P_wT dep_P_= nn_P_argmax ccomp_wT_P appos_P_WslWT nsubj_wT_P abbrev_P_WT appos_P_1 dep_thought_wT prep_propose_In pobj_In_ nsubj_propose_they ccomp_wT_propose det_method_a dobj_propose_method prep_propose_for pcomp_for_maximizing dobj_maximizing_P abbrev_P_WTIWs prep_maximizing_by pcomp_by_estimating dobj_estimating_P abbrev_P_WT cc_P_and conj_P_P appos_P_WsIWT cc_estimating_and conj_estimating_solving det_problem_the dobj_solving_problem prep_problem_in pobj_in_equation num_equation_1 ' +o,3774,'21 The Standard Machine Learning Approach We use maximum entropy -LRB- MaxEnt -RRB- classification in conjunction with the 33 features described in Ng -LRB- 2007 -RRB- to acquire a model , PC , for determining the probability that two mentions , mi and mj , are coreferent ',Berger,'det_Learning_The nn_Learning_Standard nn_Learning_Machine nsubj_Approach_Learning dep_21_Approach nsubj_use_We ccomp_Approach_use amod__maximum amod__entropy appos__MaxEnt nn__classification dobj_use_ prep__in pobj_in_conjunction prep_use_with det_features_the num_features_33 pobj_with_features partmod_features_described prep_described_in pobj_in_Ng appos_Ng_2007 aux_acquire_to infmod_Ng_acquire det_model_a dobj_acquire_model appos_model_PC prep_acquire_for dep_probability_determining det_probability_the pcomp_for_probability complm_coreferent_that num_mentions_two nsubj_coreferent_mentions appos_mentions_mi cc_mi_and conj_mi_mj cop_coreferent_are ccomp_probability_coreferent ' +o,3775,'The resulting Kappa statistics over the annotated data yields a0a2a1 a3a5a4a7a6 , which seems to indicate that human annotators can reliably distinguish between coherent samples -LRB- as in Example -LRB- 1a -RRB- -RRB- and incoherent ones -LRB- as in Example -LRB- 1b -RRB- -RRB- ',Carletta,'det_statistics_The amod_statistics_resulting nn_statistics_Kappa nsubj__statistics prep__over det_a3a5a4a7a6_the amod_a3a5a4a7a6_annotated nn_a3a5a4a7a6_data nn_a3a5a4a7a6_yields nn_a3a5a4a7a6_a0a2a1 pobj_over_a3a5a4a7a6 nsubj_seems_which rcmod_a3a5a4a7a6_seems aux_indicate_to xcomp_seems_indicate complm_distinguish_that amod_annotators_human nsubj_distinguish_annotators aux_distinguish_can advmod_distinguish_reliably ccomp_indicate_distinguish prep_distinguish_between amod_samples_coherent pobj_between_samples dep_samples_as dep_as_in pobj_in_Example appos_Example_1a cc_samples_and amod_ones_incoherent conj_samples_ones dep_samples_as dep_as_in pobj_in_Example appos_Example_1b ' +o,3776,'There has been an increased interest recently in employing Bayesian modeling for probabilistic grammars in different settings , ranging from putting priors over grammar probabilities to putting non-parametric priors over derivations to learning the set of states in a grammar ',Liang,'expl_been_There aux_been_has det_interest_an amod_interest_increased dobj_been_interest advmod_been_recently prep_been_in pcomp_in_employing nn_modeling_Bayesian dobj_employing_modeling prep_modeling_for amod_grammars_probabilistic pobj_for_grammars prep_grammars_in amod_settings_different pobj_in_settings dep_employing_ranging prep_ranging_from pcomp_from_putting nsubj__priors prep_priors_over nn_probabilities_grammar pobj_over_probabilities xcomp_putting_ prep__to pcomp_to_putting amod_priors_non-parametric dobj_putting_priors prep_putting_over pobj_over_derivations partmod_derivations_ prep__to pcomp_to_learning det_set_the dobj_learning_set prep_set_of pobj_of_states prep_states_in det_grammar_a pobj_in_grammar ' +n,3777,'In addition , the clustering methods used , such as HMMs and Browns algorithm , seem unable to adequately capture the semantics of MNs since they are based only on the information of adjacent words ',Brown,'prep_unable_In pobj_In_addition det_methods_the amod_methods_clustering nsubj_unable_methods partmod_methods_used dep_as_such prep_methods_as nn__HMMs cc_HMMs_and conj_HMMs_Browns nn__algorithm pobj_as_ cop_unable_seem aux_capture_to advmod_capture_adequately xcomp_unable_capture det_semantics_the dobj_capture_semantics prep_semantics_of pobj_of_MNs mark_based_since nsubjpass_based_they auxpass_based_are advcl_unable_based advmod_based_only prep_based_on det_information_the pobj_on_information prep_information_of amod_words_adjacent pobj_of_words ' +p,3778,'The IOB1 format , introduced in , consistently -LRB- : ame out as the best format ',Ramshaw,'det_format_The nn_format_IOB1 partmod_format_introduced prep_introduced_in pobj_in_ dep_introduced_consistently dep_out_ame dep_as_out prep_consistently_as det_format_the amod_format_best pobj_as_format ' +o,3779,'When the data has distinct sub-structures , models that exploit hidden state variables are advantageous in learning ',Matsuzaki,'advmod_has_When det_data_the nsubj_has_data amod_sub-structures_distinct dobj_has_sub-structures appos_sub-structures_models nsubj_exploit_that rcmod_models_exploit amod_variables_hidden nn_variables_state nsubj_advantageous_variables cop_advantageous_are ccomp_exploit_advantageous prep_advantageous_in pcomp_in_learning ' +o,3780,'To model p -LRB- fJle ~ ; 8 , T -RRB- we assume the existence of an alignment a J We assume that every word fj is produced by the word e ~ j at position aj in the training corpus with the probability P -LRB- f ~ le , ~ i -RRB- : J p -LRB- f lc \' -RRB- = 1 -RRB- p -LRB- L Icon -RRB- j = l -LRB- 7 -RRB- The word alignment a J is trained automatically using statistical translation models as described in ',Brown,'aux_model_To dep_assume_model dobj_model_p amod_~_fJle dep_p_~ dep_~_8 appos_8_T nsubj_assume_we det_existence_the nsubj_=_existence prep_existence_of det_alignment_an pobj_of_alignment det_J_a dep_alignment_J nsubj_assume_We rcmod_J_assume complm_produced_that det_fj_every nn_fj_word nsubjpass_produced_fj auxpass_produced_is ccomp_assume_produced prep_produced_by det_word_the pobj_by_word dep_j_e amod_j_~ dep_J_j prep_j_at nn_aj_position pobj_at_aj prep_j_in det_corpus_the nn_corpus_training pobj_in_corpus prep_corpus_with det_P_the nn_P_probability pobj_with_P nsubj_~_f dep_j_~ dobj_~_le xcomp_~_~ dobj_~_i dep_j_J dep_J_p nn_lc_f appos_p_lc ccomp_assume_= dep_l_1 nsubj_l_p nn_Icon_L appos_p_Icon partmod_p_j ccomp_j_= dep_assume_l dep_l_7 det_alignment_The nn_alignment_word dobj_l_alignment det_J_a nsubjpass_trained_J auxpass_trained_is rcmod_alignment_trained advmod_using_automatically xcomp_trained_using amod_models_statistical nn_models_translation dobj_using_models mark_described_as advcl_trained_described prt_described_in ' +o,3781,'54 IBM-3 Word Alignment Models Since the true distribution over alignments is not known , we used the IBM-3 statistical translation model to approximate This model is specified through four components : Fertility probabilities for words ; Fertility probabilities for NULL ; Word Translation probabilities ; and Distortion probabilities ',Brown,'num_Models_54 nn_Models_IBM-3 nn_Models_Word nn_Models_Alignment mark_known_Since det_distribution_the amod_distribution_true nsubjpass_known_distribution prep_distribution_over pobj_over_alignments auxpass_known_is neg_known_not advcl_used_known nsubj_used_we dep_Models_used det_model_the nn_model_IBM-3 amod_model_statistical nn_model_translation dobj_used_model partmod_model_ aux_approximate_to xcomp__approximate det_model_This nsubjpass_specified_model auxpass_specified_is dep_Models_specified prep_specified_through num_components_four pobj_through_components nn_probabilities_Fertility dep_components_probabilities prep_probabilities_for pobj_for_words nn_probabilities_Fertility dep_components_probabilities prep_probabilities_for pobj_for_NULL nn_probabilities_Word nn_probabilities_Translation dep_components_probabilities cc_components_and amod_probabilities_Distortion conj_components_probabilities ' +o,3782,'The models are based on a maximum entropy framework ',Ratnaparkhi,'det_models_The nsubjpass_based_models auxpass_based_are prep_based_on det_framework_a amod_framework_maximum nn_framework_entropy pobj_on_framework ' +o,3783,'A number of researchers have explored learning words and phrases with prior positive or negative polarity -LRB- another term is semantic orientation -RRB- -LRB- eg , -RRB- ',Turney,'det_number_A nsubj_explored_number prep_number_of pobj_of_researchers aux_explored_have xcomp_explored_learning dobj_learning_words cc_words_and conj_words_phrases prep_words_with amod_polarity_prior amod_polarity_positive cc_positive_or conj_positive_negative pobj_with_polarity det_term_another nsubj_orientation_term cop_orientation_is amod_orientation_semantic dep_polarity_orientation appos_words_eg dep_eg_ ' +o,3784,'In the meantime , synchronous parsing methods efficiently process the same bitext phrases while building their bilingual constituents , but continue to be employed primarily for word-to-word analysis ',Wu,'prep_process_In det_meantime_the pobj_In_meantime amod_methods_synchronous nn_methods_parsing nsubj_process_methods advmod_process_efficiently det_phrases_the amod_phrases_same nn_phrases_bitext dobj_process_phrases prep_process_while pcomp_while_building poss_constituents_their amod_constituents_bilingual dobj_building_constituents cc_process_but conj_process_continue aux_employed_to auxpass_employed_be xcomp_continue_employed advmod_employed_primarily prep_employed_for amod_analysis_word-to-word pobj_for_analysis ' +o,3785,'What , therefore , has to be explored are various similarity metrics , defining similarity in a concrete way and evaluate the results against human annotations ',Papineni,'nsubj_has_What advmod_has_therefore aux_explored_to cop_explored_be xcomp_has_explored cop_metrics_are amod_metrics_various nn_metrics_similarity ccomp_explored_metrics amod_similarity_defining appos_metrics_similarity prep_similarity_in det_way_a amod_way_concrete pobj_in_way cc_explored_and conj_explored_evaluate det_results_the dobj_evaluate_results prep_evaluate_against amod_annotations_human pobj_against_annotations ' +o,3786,'One other work that investigates the use of a limited lexicon is , which develops a prototype-drive approach to propagate the categorical property using distributional similarity features ; using only three exemplars of each tag , they achieve a tagging accuracy of 805 \% using a somewhat larger dataset but also the full Penn tagset , which is much larger ',Haghighi,'num_work_One amod_work_other nsubj__work nsubj_investigates_that ccomp_work_investigates det_use_the dobj_investigates_use prep_use_of det_lexicon_a amod_lexicon_limited pobj_of_lexicon cop__is nsubj_develops_which rcmod__develops det_approach_a amod_approach_prototype-drive dobj_develops_approach aux_propagate_to infmod_approach_propagate det_property_the amod_property_categorical dobj_propagate_property partmod_property_using amod_features_distributional nn_features_similarity dobj_using_features dep_achieve_using quantmod_three_only num_exemplars_three dobj_using_exemplars prep_exemplars_of det_tag_each pobj_of_tag nsubj_achieve_they parataxis__achieve det_accuracy_a amod_accuracy_tagging dobj_achieve_accuracy prep_accuracy_of num_\%_805 pobj_of_\% partmod_\%_using det_dataset_a advmod_larger_somewhat amod_dataset_larger dobj_using_dataset dep_also_but cc_accuracy_also det_tagset_the amod_tagset_full nn_tagset_Penn conj_accuracy_tagset nsubj_larger_which cop_larger_is advmod_larger_much rcmod_tagset_larger ' +o,3787,'Although the above statement was made about translation problems faced by human translators , recent research suggests that it also applies to problems in machine translation ',Brown,'mark_made_Although det_statement_the amod_statement_above nsubjpass_made_statement auxpass_made_was advcl_suggests_made prep_made_about nn_problems_translation pobj_about_problems partmod_problems_faced prep_faced_by amod_translators_human pobj_by_translators amod__recent nn__research nsubj_suggests_ complm_applies_that nsubj_applies_it advmod_applies_also ccomp_suggests_applies prep_applies_to pobj_to_problems prep_problems_in nn_translation_machine pobj_in_translation ' +p,3788,'They were based on mutual information , conditional probabilities , or on some standard statistical tests , such as the chi-square test or the loglikelihood ratio ',Dunning,'nsubjpass_based_They auxpass_based_were prep_based_on amod__mutual nn__information pobj_on_ amod__conditional nn__probabilities appos__ cc__or conj__on det_tests_some amod_tests_standard amod_tests_statistical pobj_on_tests dep_as_such prep_on_as det_test_the amod_test_chi-square pobj_as_test cc_test_or det_ratio_the amod_ratio_loglikelihood conj_test_ratio ' +p,3789,'State-of-theart machine learning techniques including Support Vector Machines , AdaBoost and Maximum Entropy Models provide high performance classifiers if one has abundant correctly labeled examples ',Berger,'amod_machine_State-of-theart nsubj_provide_machine partmod_machine_learning dobj_learning_techniques prep_techniques_including nn__Support nn__Vector nn__Machines pobj_including_ nn__AdaBoost conj__ cc__and nn__Maximum nn__Entropy nn__Models conj__ amod_classifiers_high nn_classifiers_performance dobj_provide_classifiers mark_labeled_if nsubj_labeled_one aux_labeled_has advmod_labeled_abundant advmod_labeled_correctly advcl_provide_labeled dobj_labeled_examples ' +o,3790,'Originally introduced as a byproduct of training statistical translation models in , word alignment has become the first step in training most statistical translation systems , and alignments are useful to a host of other tasks ',Brown,'advmod_introduced_Originally ccomp_step_introduced prep_introduced_as det_byproduct_a pobj_as_byproduct prep_byproduct_of pcomp_of_training amod_models_statistical nn_models_translation dobj_training_models prep_training_in pobj_in_ nn_alignment_word nsubj_step_alignment aux_step_has cop_step_become det_step_the amod_step_first prep_step_in pcomp_in_training advmod_statistical_most amod_systems_statistical nn_systems_translation dobj_training_systems cc_step_and nsubj_useful_alignments cop_useful_are conj_step_useful prep_useful_to det_host_a pobj_to_host prep_host_of amod_tasks_other pobj_of_tasks ' +o,3791,'IBM constraints , the lexical word reordering model , and inversion transduction grammar -LRB- ITG -RRB- constraints belong to this type of approach ',Berger,'nn_constraints_IBM nsubj__constraints det_model_the amod_model_lexical nn_model_word nn_model_reordering nsubj__model ccomp__ cc__and nn_grammar_inversion nn_grammar_transduction nn_constraints_grammar abbrev_constraints_ITG nsubj_belong_constraints advmod_belong_ conj__belong prep_belong_to det_type_this pobj_to_type prep_type_of pobj_of_approach ' +o,3792,'Finally , achieve an SF of 9590 \% and a TF of 9134 \% by 10-fold cross validation using CTB data ',Zhang,'advmod_achieve_Finally nsubj_achieve_ det_SF_an dobj_achieve_SF prep_SF_of num_\%_9590 pobj_of_\% cc_SF_and det_TF_a conj_SF_TF prep_TF_of num_\%_9134 pobj_of_\% prep_achieve_by amod_validation_10-fold amod_validation_cross pobj_by_validation partmod_validation_using nn_data_CTB dobj_using_data ' +o,3793,'First , for each verb occurrence subjects and objects were extracted from a parsed corpus ',Collins,'dep_extracted_First mark_extracted_for det_subjects_each amod_subjects_verb nn_subjects_occurrence nsubjpass_extracted_subjects cc_subjects_and conj_subjects_objects auxpass_extracted_were prep_extracted_from det_corpus_a amod_corpus_parsed pobj_from_corpus ' +o,3794,'Like the work of and Mani et al ',Jing,'det_work_the dep_Like_work prep_work_of pobj_of_ cc__and conj__Mani cc_Mani_et conj_Mani_al ' +o,3795,'The huge increase in computational and storage cost of including longer phrases does not provide a signi cant improvement in quality as the probability of reappearance of larger phrases decreases ',Koehn,'det_increase_The amod_increase_huge nsubj_provide_increase prep_increase_in nn_cost_computational cc_computational_and conj_computational_storage pobj_in_cost prep_cost_of pcomp_of_including amod_phrases_longer dobj_including_phrases aux_provide_does neg_provide_not det_improvement_a nn_improvement_signi nn_improvement_cant dobj_provide_improvement prep_provide_in nn__quality pobj_in_ mark_decreases_as det_probability_the nsubj_decreases_probability prep_probability_of pobj_of_reappearance prep_reappearance_of amod_phrases_larger pobj_of_phrases advcl_provide_decreases ' +o,3796,'In their presentation of the factored SMT models , describe experiments for translating from English to German , Spanish and Czech , using morphology tags added on the morphologically rich side , along with POS tags ',Koehn,'prep_describe_In poss_presentation_their pobj_In_presentation prep_presentation_of det_models_the amod_models_factored nn_models_SMT pobj_of_models nsubj_describe_ dobj_describe_experiments prep_describe_for pcomp_for_translating prep_translating_from pobj_from_English prep_translating_to pobj_to_German conj_German_Spanish cc_German_and conj_German_Czech xcomp_describe_using amod_tags_morphology dobj_using_tags partmod_tags_added prep_added_on det_side_the advmod_rich_morphologically amod_side_rich pobj_on_side advmod_side_along dep_along_with dep_tags_POS pobj_with_tags ' +o,3797,'1 Introduction Word alignmentdetection of corresponding words between two sentences that are translations of each otheris usually an intermediate step of statistical machine translation -LRB- MT -RRB- , but also has been shown useful for other applications such as construction of bilingual lexicons , word-sense disambiguation , projection of resources , and crosslanguage information retrieval ',Koehn,'num_alignmentdetection_1 nn_alignmentdetection_Introduction nn_alignmentdetection_Word prep_alignmentdetection_of amod_words_corresponding pobj_of_words prep_words_between num_sentences_two pobj_between_sentences nsubj_translations_that cop_translations_are rcmod_sentences_translations prep_translations_of det_otheris_each pobj_of_otheris advmod_translations_usually det_step_an amod_step_intermediate nsubjpass__step prep_step_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_MT dep_alignmentdetection_ cc__but advmod_shown_also aux_shown_has auxpass_shown_been conj__shown acomp_shown_useful prep_shown_for amod_applications_other pobj_for_applications dep_as_such prep_applications_as pobj_as_construction prep_construction_of amod_lexicons_bilingual pobj_of_lexicons amod_disambiguation_word-sense conj_lexicons_disambiguation conj_lexicons_projection prep_projection_of pobj_of_resources cc_lexicons_and amod_retrieval_crosslanguage nn_retrieval_information conj_lexicons_retrieval ' +o,3798,'Our goal is to come up with a mechanism that , given an input string , identifies the phrases in this string , this is a fundamental task with applications in natural language ',Ramshaw,'poss_goal_Our nsubj_is_goal aux_come_to xcomp_is_come prt_come_up prep_come_with det_mechanism_a pobj_with_mechanism nsubj_identifies_that prep_identifies_given det_string_an nn_string_input dep_given_string rcmod_mechanism_identifies det_phrases_the dobj_identifies_phrases prep_task_in det_string_this pobj_in_string nsubj_task_this cop_task_is det_task_a amod_task_fundamental dep_is_task prep_task_with pobj_with_applications prep_applications_in amod_language_natural pobj_in_language ' +o,3799,' note that the bootstrapping algorithm works well but its performance can deteriorate rapidly when non-coreferring data enter as candidate heuristics ',Bean,'nsubj_note_ complm_works_that det_algorithm_the amod_algorithm_bootstrapping nsubj_works_algorithm ccomp_note_works advmod_works_well cc_note_but poss_performance_its nsubj_deteriorate_performance aux_deteriorate_can conj_note_deteriorate advmod_deteriorate_rapidly advmod_enter_when amod_data_non-coreferring nsubj_enter_data advcl_deteriorate_enter prep_enter_as nn_heuristics_candidate pobj_as_heuristics ' +o,3800,'Letter successor variety -LRB- LSV -RRB- models use the hypothesis that there is less certainty when predicting the next character at morpheme boundaries ',Dasgupta,'nn_variety_Letter nn_variety_successor nn_models_variety abbrev_models_LSV nsubj__models dep_hypothesis_use det_hypothesis_the xcomp__hypothesis complm_is_that expl_is_there ccomp_hypothesis_is advmod_certainty_less acomp_is_certainty advmod_predicting_when advcl_is_predicting det_character_the amod_character_next dobj_predicting_character prep_predicting_at amod_boundaries_morpheme pobj_at_boundaries ' +o,3801,'This can be seen as a simplified version of ',Rosti,'nsubjpass_seen_This aux_seen_can auxpass_seen_be prep_seen_as det_version_a amod_version_simplified pobj_as_version prep_version_of ' +o,3802,'The MSLR parser performs syntactic analysis of the sentence ',Marcus,'det__The amod__MSLR nn__parser nsubj_performs_ amod_analysis_syntactic dobj_performs_analysis prep_analysis_of det_sentence_the pobj_of_sentence ' +o,3803,' The syntactic annotation task consists of marking constituent boundaries , inserting empty categories -LRB- traces of movement , PRO , pro -RRB- , showing the relationships between constituents -LRB- argument\\/adjunct structures -RRB- , and specifying a particular subset of adverbial roles ',Marcus,'det_task_The amod_task_syntactic nn_task_annotation nsubj_consists_task ccomp__consists prep_consists_of pcomp_of_marking amod_boundaries_constituent dobj_marking_boundaries conj_marking_inserting amod_categories_empty dobj_inserting_categories dep_categories_traces prep_traces_of pobj_of_movement appos_traces_PRO dep_traces_pro conj_marking_showing det_relationships_the dobj_showing_relationships prep_relationships_between pobj_between_constituents nn_structures_argument\\/adjunct appos_constituents_structures cc_marking_and conj_marking_specifying det_subset_a amod_subset_particular dobj_specifying_subset prep_subset_of amod_roles_adverbial pobj_of_roles ' +o,3804,'766 System Beam Error \% 5 337 1 290 289 Guided Learning , feature B 3 285 all 285 284 276 Guided Learning , feature E 1 273 Guided Learning , feature E 3 267 Table 4 : Comparison with the previous works According to the experiments shown above , we build our best system by using feature set E with beam width B = 3 ',Berger,'num_Beam_766 nn_Beam_System num_\%_Error nsubj__\% dep_Beam_ number_337_5 dobj__337 nn_Learning_ num_Learning_1 num_Learning_290 num_Learning_ num_Learning_289 nn_Learning_Guided dep_337_Learning nn_B_feature appos_337_B number_285_3 num__285 dep_B_ det_Learning_all num_Learning_285 num_Learning_ num_Learning_284 num_Learning_ num_Learning_276 nn_Learning_Guided dep__Learning nn_E_feature appos_337_E number_273_1 num_Learning_273 nn_Learning_Guided dep_E_Learning nn_E_feature appos_337_E number_267_3 dep_E_267 dep_267_Table num_Table_4 dep_337_Comparison prep_Comparison_with det_works_the amod_works_previous pobj_with_works prep_build_According dep_According_to det_experiments_the pobj_to_experiments partmod_experiments_shown prep_shown_above nsubj_build_we dep_Beam_build poss_system_our amod_system_best dobj_build_system prep_build_by pcomp_by_using nn_E_feature nn_E_set dobj_using_E prep_E_with nn_B_beam amod_B_width pobj_with_B dep_3_= dep_B_3 ' +o,3805,'We will briefly review the perceptron algorithm , and its convergence properties see for a full description ',Collins,'nsubj_review_We aux_review_will advmod_review_briefly det_algorithm_the nn_algorithm_perceptron dobj_review_algorithm cc_review_and poss_properties_its amod_properties_convergence nsubj_see_properties conj_review_see dobj_see_ prep__for det_description_a amod_description_full pobj_for_description ' +o,3806,'1 Introduction Word alignment was first proposed as an intermediate result of statistical machine translation ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubjpass_proposed_alignment auxpass_proposed_was advmod_proposed_first prep_proposed_as det_result_an amod_result_intermediate pobj_as_result prep_result_of amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,3807,'The tree is produced by a state-of-the-art dependency parser trained on the Wall Street Journal Penn Treebank ',Brown,'det_tree_The nsubjpass_produced_tree auxpass_produced_is prep_produced_by det__a amod__state-of-the-art amod__dependency nn__parser pobj_by_ partmod__trained prep_trained_on det__the nn__Wall nn__Street nn__Journal nn__Penn nn__Treebank pobj_on_ ' +o,3808,'For our out-of-domain training condition , the parser was trained on sections 2-21 of the Wall Street Journal -LRB- WSJ -RRB- corpus ',Marcus,'prep_trained_For poss_condition_our amod_condition_out-of-domain nn_condition_training pobj_For_condition det_parser_the nsubjpass_trained_parser auxpass_trained_was prep_trained_on pobj_on_sections num_sections_2-21 prep_sections_of det_corpus_the nn_Journal_Wall nn_Journal_Street nn_corpus_Journal abbrev_Journal_WSJ pobj_of_corpus ' +o,3809,'Close to the problem studied here is Jing and McKeowns cut-and-paste method founded on EndresNiggemeyers observations ',Jing,'nsubj_method_Close prep_Close_to det_problem_the pobj_to_problem partmod_problem_studied advmod_studied_here cop_method_is nn_method_Jing cc_Jing_and conj_Jing_McKeowns nn_method_ amod_method_cut-and-paste partmod_method_founded prep_founded_on nn_observations_EndresNiggemeyers pobj_on_observations ' +o,3810,'We analyzed a set of articles and identified six major operations that can be used for editing the extracted sentences , including removing extraneous phrases from an extracted sentence , combining a reduced sentence with other sentences , syntactic transformation , substituting phrases in an extracted sentence with their paraphrases , substituting phrases with more general or specific descriptions , and reordering the extracted sentences ',Jing,'nsubj_analyzed_We det_set_a dobj_analyzed_set prep_set_of pobj_of_articles cc_analyzed_and conj_analyzed_identified num_operations_six amod_operations_major dobj_identified_operations nsubjpass_used_that aux_used_can auxpass_used_be rcmod_operations_used prep_used_for pcomp_for_editing det_sentences_the amod_sentences_extracted dobj_editing_sentences conj_editing_including xcomp_including_removing amod_phrases_extraneous dobj_removing_phrases prep_removing_from det_sentence_an amod_sentence_extracted pobj_from_sentence conj_editing_combining det_sentence_a amod_sentence_reduced dobj_combining_sentence prep_combining_with amod_sentences_other pobj_with_sentences amod_transformation_syntactic appos_sentences_transformation conj_editing_substituting dobj_substituting_phrases prep_substituting_in det_sentence_an amod_sentence_extracted pobj_in_sentence prep_sentence_with poss_paraphrases_their pobj_with_paraphrases conj_editing_substituting dobj_substituting_phrases prep_substituting_with advmod_general_more amod_descriptions_general cc_general_or conj_general_specific pobj_with_descriptions cc_editing_and conj_editing_reordering det_sentences_the amod_sentences_extracted dobj_reordering_sentences ' +o,3811,'Using techniques described in Church and , Church and Hanks -LRB- 1990 -RRB- , and Hindle and Rooth -LRB- 1991 -RRB- , Figure 4 shows some examples of the most frequent V-O pairs from the AP corpus ',Hindle,'dep_shows_Using dobj_Using_techniques partmod_techniques_described prep_described_in pobj_in_Church cc_Church_and conj_Church_ conj_techniques_Church cc_Church_and conj_Church_Hanks appos_Church_1990 cc_techniques_and conj_techniques_Hindle cc_Hindle_and conj_Hindle_Rooth appos_Hindle_1991 nsubj_shows_Figure num_Figure_4 det_examples_some dobj_shows_examples prep_examples_of det_pairs_the advmod_frequent_most amod_pairs_frequent amod_pairs_V-O pobj_of_pairs prep_pairs_from det_corpus_the nn_corpus_AP pobj_from_corpus ' +o,3812,'We used the procedure described in , with the only modification being the multiplication of the loglikelihood values with a triangular function that depends on the logarithm of a words frequency ',Rapp,'nsubj_used_We det_procedure_the nsubj_described_procedure ccomp_used_described prep_described_in pobj_in_ prep_described_with det_modification_the amod_modification_only pobj_with_modification cop_multiplication_being det_multiplication_the partmod_modification_multiplication prep_multiplication_of det_values_the amod_values_loglikelihood pobj_of_values prep_values_with det_function_a nn_function_triangular pobj_with_function nsubj_depends_that rcmod_multiplication_depends prep_depends_on det_logarithm_the pobj_on_logarithm prep_logarithm_of det_frequency_a nn_frequency_words pobj_of_frequency ' +o,3813,'3 Schone & Jurafsky \'s results indicate similar results for log-likelihood & T-score , and strong parallelism among information-theoretic measures such as ChiSquared , Selectional Association , Symmetric Conditional Probability and the Z-Score ',Smadja,'nn_Schone_3 nn_results_Schone cc_Schone_& conj_Schone_Jurafsky possessive_Jurafsky_\'s nsubj_indicate_results amod_results_similar dobj_indicate_results prep_results_for pobj_for_log-likelihood cc_log-likelihood_& conj_log-likelihood_T-score cc_results_and amod_parallelism_strong conj_results_parallelism prep_parallelism_among amod_measures_information-theoretic pobj_among_measures dep_as_such prep_measures_as pobj_as_ChiSquared nn__Selectional nn__Association conj_ChiSquared_ nn__Symmetric nn__Conditional nn__Probability conj_ChiSquared_ cc_ChiSquared_and det_Z-Score_the conj_ChiSquared_Z-Score ' +o,3814,'(Barzilay and McKeown, 2001; Shinyama et al. , 2002; Barzilay and Lee, 2003).',Barzilay,'' +o,3815,'32 Results In line with previous work , we first compare Naive Bayes and Logistic regression on the two NLP tasks ',Klein,'num_Results_32 prep_compare_In pobj_In_line prep_line_with amod__previous nn__work pobj_with_ nsubj_compare_we advmod_compare_first dep_Results_compare nn_Bayes_Naive dobj_compare_Bayes cc_Bayes_and nn_regression_Logistic conj_Bayes_regression prep_Bayes_on det_tasks_the num_tasks_two nn_tasks_NLP pobj_on_tasks ' +o,3816,'Note that it is straightforward to calculate these expected counts using a variant of the inside-outside algorithm applied to the dependency-parsing data structures for projective dependency structures , or the matrix-tree theorem for nonprojective dependency structures ',Koo,'complm_straightforward_that nsubj_straightforward_it cop_straightforward_is ccomp_Note_straightforward aux_calculate_to xcomp_straightforward_calculate det_counts_these amod_counts_expected dobj_calculate_counts partmod_counts_using det_variant_a dobj_using_variant prep_variant_of det__the amod__inside-outside nn__algorithm pobj_of_ partmod__applied prep_applied_to det__the amod__ amod__dependency-parsing nn__data nn__structures pobj_to_ prep__for amod_structures_projective amod_structures_dependency pobj_for_structures cc_counts_or det__the amod__matrix-tree nn__theorem conj_counts_ prep__for amod_structures_nonprojective amod_structures_dependency pobj_for_structures ' +o,3817,'5 Related work The methodology which is closest to our framework is ORANGE , which evaluates a similarity metric using the average ranks obtained by reference items within a baseline set ',Lin,'num_work_5 amod_work_Related det_methodology_The nsubj__methodology nsubj_closest_which cop_closest_is rcmod_methodology_closest prep_closest_to poss_framework_our pobj_to_framework cop__is nn__ORANGE dep_work_ nsubj_evaluates_which rcmod__evaluates det_similarity_a nsubj_using_similarity dep_using_metric xcomp_evaluates_using det_ranks_the amod_ranks_average dobj_using_ranks partmod_ranks_obtained prep_obtained_by nn_items_reference pobj_by_items prep_obtained_within det_set_a nn_set_baseline pobj_within_set ' +o,3818,'However , these unsupervised methodologies show a major drawback by extracting quasi-exact2 or even exact match pairs of sentences as they rely on classical string similarity measures such as the Edit Distance in the case of and word N-gram overlap for ',Barzilay,'advmod_show_However det_methodologies_these amod_methodologies_unsupervised nsubj_show_methodologies det_drawback_a amod_drawback_major dobj_show_drawback prep_show_by pcomp_by_extracting amod_pairs_quasi-exact2 cc_quasi-exact2_or advmod_exact_even conj_quasi-exact2_exact nn_pairs_match dobj_extracting_pairs prep_pairs_of pobj_of_sentences mark_rely_as nsubj_rely_they advcl_extracting_rely prep_rely_on amod_measures_classical nn_measures_string nn_measures_similarity pobj_on_measures dep_as_such prep_measures_as det_Distance_the nn_Distance_Edit pobj_as_Distance prep_Distance_in det_case_the pobj_in_case prep_case_of pobj_of_ cc__and conj__word nsubj_overlap_N-gram dep_show_overlap prep_overlap_for ' +o,3819,'These constraints tie words in such a way that the space of alignments can not be enumerated as in IBM models 1 and 2 ',Brown,'det_constraints_These nsubj_tie_constraints dobj_tie_words prep_tie_in predet_way_such det_way_a pobj_in_way complm_enumerated_that det_space_the nsubj_enumerated_space prep_space_of pobj_of_alignments aux_enumerated_can neg_enumerated_not cop_enumerated_be ccomp_tie_enumerated advmod_enumerated_as dep_as_in nn_models_IBM pobj_in_models tmod_enumerated_1 dep_1_and dep_1_2 ' +o,3820,'Our approach differs in important ways from the use of hidden Markov models -LRB- HMMs -RRB- for classbased language modeling ',Brown,'poss_approach_Our nsubj_differs_approach prep_differs_in amod_ways_important pobj_in_ways prep_differs_from det_use_the pobj_from_use prep_use_of amod_models_hidden nn_models_Markov pobj_of_models abbrev_models_HMMs prep_models_for amod__classbased nn__language nn__modeling pobj_for_ ' +o,3821,'4 Sub Translation Combining For sub translation combining , we mainly use the best-first expansion idea from cube pruning to combine subtranslations and generate the whole k-best translations ',Huang,'num_Combining_4 nn_Combining_Sub nn_Combining_Translation prep_use_For nn_translation_sub pobj_For_translation partmod_translation_combining nsubj_use_we advmod_use_mainly dep_Combining_use det_idea_the amod_idea_best-first nn_idea_expansion dobj_use_idea prep_idea_from nn__cube nn__pruning pobj_from_ aux_combine_to xcomp_use_combine dobj_combine_subtranslations cc_combine_and conj_combine_generate det_translations_the amod_translations_whole amod_translations_k-best dobj_generate_translations ' +o,3822,'1 Introduction Recent approaches to statistical machine translation -LRB- SMT -RRB- piggyback on the central concepts of phrasebased SMT and at the same time attempt to improve some of its shortcomings by incorporating syntactic knowledge in the translation process ',Koehn,'num_approaches_1 nn_approaches_Introduction amod_approaches_Recent nsubj_attempt_approaches prep_approaches_to amod_piggyback_statistical nn_piggyback_machine nn_piggyback_translation abbrev_piggyback_SMT pobj_to_piggyback prep_piggyback_on det_concepts_the amod_concepts_central pobj_on_concepts prep_concepts_of amod__phrasebased nn__SMT pobj_of_ cc_to_and conj_to_at det_time_the amod_time_same pobj_at_time aux_improve_to xcomp_attempt_improve dobj_improve_some prep_some_of poss_shortcomings_its pobj_of_shortcomings prep_improve_by pcomp_by_incorporating amod_knowledge_syntactic dobj_incorporating_knowledge prep_incorporating_in det_process_the nn_process_translation pobj_in_process ' +o,3823,'Both calculate the precision of a translation by comparing it to a reference translation and incorporating a length penalty -RRB- ',Papineni,'nsubj_calculate_Both det_precision_the dobj_calculate_precision prep_precision_of det_translation_a pobj_of_translation prep_calculate_by pcomp_by_comparing dobj_comparing_it prep_comparing_to det_translation_a nn_translation_reference pobj_to_translation cc_comparing_and conj_comparing_incorporating det__a nn__length nn__penalty dobj_incorporating_ ' +o,3824,'Version of the System P R F Baseline 50.8 100 67.4 Discourse-new detection only 69 72 70 Hand-coded DT: partial 62 85 71.7 Hand-coded DT: total 77 77 77 ID3 75 75 75 Table 1: Overall results by Vieira and Poesio 2.2 Bean and Riloff Bean and Riloff (1999) developed a system for identifying discourse-new DDs1 that incorporates, in addition to syntax-based heuristics aimed at recognizing predicative and established DDs using postmodification heuristics similar to those used by Vieira and Poesio, additional techniques for mining from corpora unfamiliar DDs including proper names, larger situation, and semantically functional.',Bean,'' +o,3825,'Phrase-based MT systems are straightforward to train from parallel corpora and , like the original IBM models , benefit from standard language models built on large monolingual , target-language corpora ',Brants,'amod_systems_Phrase-based nn_systems_MT nsubj_straightforward_systems cop_straightforward_are aux_train_to xcomp_straightforward_train prep_train_from amod__parallel nn__corpora pobj_from_ cc_train_and dep_train_like det_models_the amod_models_original nn_models_IBM pobj_like_models dep_like_ conj_train_benefit prep_benefit_from amod_models_standard nn_models_language pobj_from_models partmod_models_built prep_built_on amod_monolingual_large pobj_on_monolingual amod_corpora_target-language appos_monolingual_corpora ' +p,3826,'21 The BLEU Metric The metric most often used with MERT is BLEU , where the score of a candidate c against a reference translation r is : BLEU = BP -LRB- len -LRB- c -RRB- , len -LRB- r -RRB- -RRB- exp -LRB- 4summationdisplay n = 1 1 4 logpn -RRB- , where pn is the n-gram precision2 and BP is a brevity penalty meant to penalize short outputs , to discourage improving precision at the expense of recall ',Papineni,'det_BLEU_The nsubj_Metric_BLEU dep_21_Metric det_most_The amod_most_metric nsubj__most advmod_used_often partmod_most_used prep_used_with pobj_with_MERT cop__is nn__BLEU ccomp_Metric_ advmod_is_where det_score_the nsubj_is_score prep_score_of det_candidate_a pobj_of_candidate dep_score_c prep_score_against det_r_a nn_r_reference nn_r_translation pobj_against_r rcmod__is parataxis_is_BLEU nn_exp_= nn_exp_BP dep_exp_len dep_-LRB-_c nn_r_len appos_len_r dobj_BLEU_exp amod_n_4summationdisplay nsubj_=_n dep_exp_= num_1_1 iobj_=_1 num_logpn_4 dobj_=_logpn advmod_precision2_where nsubj_precision2_pn cop_precision2_is det_precision2_the amod_precision2_n-gram rcmod_exp_precision2 cc_Metric_and nsubj_penalty_BP cop_penalty_is det_penalty_a amod_penalty_brevity conj_Metric_penalty partmod_penalty_meant aux_penalize_to xcomp_meant_penalize amod_outputs_short dobj_penalize_outputs aux_discourage_to xcomp_meant_discourage xcomp_discourage_improving dobj_improving_precision prep_improving_at det_expense_the pobj_at_expense prep_expense_of pobj_of_recall ' +o,3827,'Set Test Set ENGLISH-WSJ Sections Section 22 Section 23 2-21 ENGLISH-BROWN see 10 \% of 10 \% of the ENGLISH-WSJ the data6 the data6 FRENCH7 Sentences Sentences Sentences 1-18 ,609 18,610-19 ,609 19,609-20 ,610 GERMAN Sentences Sentences Sentences 1-18 ,602 18,603-19 ,602 19,603-20 ,602 Table 1 : Corpora and standard experimental setups ',Marcus,'nn_Set_Set nn_Set_Test partmod_Set_ENGLISH-WSJ nn_ENGLISH-BROWN_Sections nn_ENGLISH-BROWN_Section num_ENGLISH-BROWN_22 nn_ENGLISH-BROWN_Section num_ENGLISH-BROWN_23 nn_ENGLISH-BROWN_ num_ENGLISH-BROWN_2-21 nsubj_see_ENGLISH-BROWN ccomp_ENGLISH-WSJ_see num_\%_10 dobj_see_\% prep_\%_of num_\%_10 pobj_of_\% prep_\%_of det_ENGLISH-WSJ_the amod_ENGLISH-WSJ_ pobj_of_ENGLISH-WSJ det_data6_the dep_Sentences_data6 det_Sentences_the amod_Sentences_data6 nn_Sentences_FRENCH7 dep_data6_Sentences num_Sentences_Sentences dep_Sentences_Sentences nn_Sentences_ num_Sentences_1-18 num_Sentences_,609 num_Sentences_18,610-19 num_Sentences_,609 num_Sentences_19,609-20 num_Sentences_,610 nn_Sentences_GERMAN dep_Set_Sentences number_,602_Sentences cc_,602_Sentences conj_,602_ number_,602_1-18 number_,602_,602 number_,602_18,603-19 number_,602_,602 number_,602_19,603-20 dep_Sentences_,602 cc_,602_Table conj_,602_1 dep_,602_Corpora cc_Corpora_and amod_setups_standard amod_setups_experimental conj_Corpora_setups ' +o,3828,'Similarly to MERT , Tillmann and Zhang estimate the parameters of a weight vector on a linear combination of -LRB- binary -RRB- features using a global objective function correlated with BLEU ',Papineni,'advmod_estimate_Similarly dep_Similarly_to pobj_to_MERT conj_MERT_Tillmann cc_MERT_and conj_MERT_Zhang det_parameters_the dobj_estimate_parameters prep_parameters_of det_vector_a nn_vector_weight pobj_of_vector prep_vector_on det_combination_a amod_combination_linear pobj_on_combination prep_combination_of appos_features_binary pobj_of_features partmod_parameters_using det_function_a amod_function_global nn_function_objective dobj_using_function partmod_function_correlated prep_correlated_with pobj_with_BLEU ' +o,3829,'Lacking an automatic method , recent WSD works still resort to human intervention to identify and group closely related senses in an MRD ',Yarowsky,'dep_resort_Lacking det_method_an amod_method_automatic dobj_Lacking_method amod_works_recent nn_works_WSD nsubj_resort_works advmod_resort_ advmod_resort_still prep_resort_to amod_intervention_human pobj_to_intervention aux_identify_to xcomp_resort_identify cc_resort_and conj_resort_group advmod_related_closely amod_senses_related dep_in_senses prep_group_in det_MRD_an pobj_in_MRD ' +o,3830,'It is easy to see that the main difference between the PA algorithms and the Perceptron algorithm -LRB- PC -RRB- as well as the MIRA algorithm is in line 9 ',Collins,'nsubj_easy_It cop_easy_is aux_see_to xcomp_easy_see complm_is_that det_difference_the amod_difference_main nsubj_is_difference prep_difference_between det__the nn_algorithms_PA nn__algorithms cc_algorithms_and det_algorithm_the amod_algorithm_Perceptron conj_algorithms_algorithm nn__PC pobj_between_ dep_well_as cc_difference_well dep_well_as det__the nn__MIRA nn__algorithm conj_difference_ ccomp_see_is prep_is_in pobj_in_line num_line_9 ' +o,3831,'While we can only compare class models with word models on the largest training set , for this training set model M outperforms the baseline Katzsmoothed word trigram model by 19 \% absolute6 4 Domain Adaptation In this section , we introduce another heuristic for improving exponential models and show how this heuristic can be used to motivate a regularized version of minimum discrimination information -LRB- MDI -RRB- models ',Brown,'mark_compare_While nsubj_compare_we aux_compare_can advmod_compare_only advcl_introduce_compare nn_models_class dobj_compare_models prep_compare_with nn_models_word pobj_with_models prep_compare_on det_training_the amod_training_largest pobj_on_training partmod_training_set mark_set_for det_training_this nsubj_set_training advcl_compare_set nn_M_model nsubj_outperforms_M ccomp_set_outperforms det_model_the amod_model_baseline nn_model_Katzsmoothed nn_model_word nn_model_trigram nsubj_absolute6_model prep_model_by num_\%_19 pobj_by_\% xcomp_outperforms_absolute6 tmod_absolute6_4 nn_Adaptation_Domain dep_absolute6_Adaptation dep_absolute6_In det_section_this pobj_In_section nsubj_introduce_we det_heuristic_another dobj_introduce_heuristic prep_introduce_for pcomp_for_improving amod_models_exponential dobj_improving_models cc_introduce_and conj_introduce_show advmod_used_how det_heuristic_this nsubjpass_used_heuristic aux_used_can auxpass_used_be ccomp_show_used aux_motivate_to purpcl_used_motivate det_version_a amod_version_regularized dobj_motivate_version prep_version_of amod_information_minimum nn_information_discrimination pobj_of_information abbrev_information_MDI nsubj__models dep_introduce_ ' +o,3832,'We use a statistical POS tagging system built on Arabic Treebank data with MaxEnt framework ',Ratnaparkhi,'nsubj_use_We det_system_a amod_system_statistical dep_system_POS amod_system_tagging dobj_use_system partmod_system_built prep_built_on nn_data_Arabic nn_data_Treebank pobj_on_data prep_built_with nn_framework_MaxEnt pobj_with_framework ' +o,3833,'Though our motivation is similar to that of , we chose to build an independent component for inflection prediction in isolation rather than folding morphological information into the main translation model ',Koehn,'mark_similar_Though poss_motivation_our nsubj_similar_motivation cop_similar_is advcl_chose_similar prep_similar_to pobj_to_that prep_that_of pobj_of_ nsubj_chose_we aux_build_to xcomp_chose_build det_component_an amod_component_independent dobj_build_component prep_component_for nn_prediction_inflection pobj_for_prediction prep_prediction_in pobj_in_isolation cc_component_rather dep_rather_than amod_information_folding amod_information_morphological conj_component_information prep_build_into det_model_the amod_model_main nn_model_translation pobj_into_model ' +o,3834,'Each dataset consisted of a collection of flat rules such as Sput ! NP put NP PP extracted from the Penn Treebank ',Marcus,'det_dataset_Each nsubj_put_dataset partmod_dataset_consisted prep_consisted_of det_collection_a pobj_of_collection prep_collection_of amod_rules_flat pobj_of_rules dep_as_such prep_rules_as nn_NP_Sput pobj_as_NP nn_PP_NP dobj_put_PP partmod_PP_extracted prep_extracted_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3835,'However , since we are interested in the word counts that correlate to w , we adopt the concept of the translation model proposed by ',Brown,'advmod_adopt_However mark_interested_since nsubj_interested_we cop_interested_are dep_adopt_interested prep_interested_in det_counts_the nn_counts_word pobj_in_counts nsubj_correlate_that rcmod_counts_correlate aux_w_to xcomp_correlate_w nsubj_adopt_we det_concept_the dobj_adopt_concept prep_concept_of det_model_the nn_model_translation pobj_of_model partmod_model_proposed prep_proposed_by ' +o,3836,'By no means an exhaustive list , the most commonly cited ranking and scoring algorithms are HITS and PageRank , which rank hyperlinked documents using the concepts of hubs and authorities ',Lin,'advmod_no_By advmod_means_no det_list_an amod_list_exhaustive nsubj_means_list dep_commonly_the dep_commonly_most nsubj_cited_commonly rcmod_list_cited nn_algorithms_ranking cc_ranking_and conj_ranking_scoring nsubj__algorithms cop__are nn__HITS ccomp_cited_ cc__and nn__PageRank conj__ nsubj_rank_which rcmod_list_rank amod_documents_hyperlinked dobj_rank_documents partmod_documents_using det_concepts_the dobj_using_concepts prep_concepts_of pobj_of_hubs cc_hubs_and conj_hubs_authorities ' +o,3837,'BLEU : Automatic evaluation by BLEU score ',Papineni,'nn_evaluation_Automatic dep_BLEU_evaluation prep_evaluation_by amod_score_BLEU pobj_by_score ' +o,3838,'The modified version of the Roark parser , trained on the Brown Corpus section of the Penn Treebank , was used to parse the different narratives and produce the word by word measures ',Marcus,'det_version_The amod_version_modified nsubjpass_used_version prep_version_of det_parser_the nn_parser_Roark pobj_of_parser partmod_version_trained prep_trained_on det_section_the nn_section_Brown nn_section_Corpus pobj_on_section prep_section_of det__the nn__Penn nn__Treebank pobj_of_ auxpass_used_was aux_parse_to xcomp_used_parse det_narratives_the amod_narratives_different dobj_parse_narratives cc_parse_and conj_parse_produce det_word_the dobj_produce_word prep_produce_by nn_measures_word pobj_by_measures ' +o,3839,'With respect to already available POS tagsets , the scheme allows corresponding extensions of the supertype POSTag to , eg , PennPOSTag -LRB- for the Penn Tag Set -RRB- or GeniaPOSTag -LRB- for the GENIA Tag Set -RRB- ',Marcus,'prep_allows_With pobj_With_respect prep_respect_to advmod_available_already amod_tagsets_available dep_tagsets_POS pobj_to_tagsets det_scheme_the nsubj_allows_scheme amod_extensions_corresponding dobj_allows_extensions prep_extensions_of det_supertype_the pobj_of_supertype partmod_supertype_POSTag prep_POSTag_to conj_supertype_eg appos_eg_PennPOSTag dep_PennPOSTag_for det__the nn__Penn nn__Tag nn__Set pobj_for_ cc_supertype_or conj_supertype_GeniaPOSTag dep_GeniaPOSTag_for det__the nn__GENIA nn__Tag nn__Set pobj_for_ ' +o,3840,'of Words Person names 803 1749 Organization names 312 867 Location names 345 614 The BLEU score with a single reference translation was deployed for evaluation ',Papineni,'prep_312_of nn_names_Words nn_names_Person pobj_of_names number_1749_803 num_names_1749 nn_names_Organization nsubj_312_names num_names_867 nn_names_Location iobj_312_names number_614_345 dobj_312_614 det__The amod__BLEU nn__score nsubjpass_deployed_ prep__with det_translation_a amod_translation_single nn_translation_reference pobj_with_translation auxpass_deployed_was dep_312_deployed prep_deployed_for pobj_for_evaluation ' +o,3841,'7 For the most frequent 184 expressions , on the average , the agreement rate between two human annotators is 093 and the Kappa value is 073 , which means allowing tentative conclusions to be drawn ',Carletta,'prep_093_For det_expressions_the advmod_expressions_most amod_expressions_frequent nn_expressions_184 pobj_For_expressions prep_093_on det_average_the pobj_on_average det_rate_the nn_rate_agreement nsubj_093_rate prep_rate_between num_annotators_two amod_annotators_human pobj_between_annotators cop_093_is dep_7_093 cc_093_and det_value_the nn_value_Kappa nsubj_073_value cop_073_is conj_093_073 nsubj_means_which rcmod_073_means xcomp_means_allowing amod_conclusions_tentative dobj_allowing_conclusions aux_drawn_to auxpass_drawn_be xcomp_allowing_drawn ' +o,3842,'Except where noted , each system was trained on 27 million words of newswire data , aligned with GIZA + + and symmetrized with the grow-diag-final-and heuristic ',Och,'dep_trained_Except advmod_noted_where pcomp_Except_noted det_system_each nsubjpass_trained_system auxpass_trained_was prep_trained_on number_million_27 num_words_million pobj_on_words prep_words_of nn_data_newswire pobj_of_data conj_trained_aligned prep_aligned_with pobj_with_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_trained_and conj_trained_symmetrized prep_symmetrized_with det__the amod__grow-diag-final-and amod__heuristic pobj_with_ ' +p,3843,'Lexicalization can increase parsing performance dramatically for English , and the lexicalized model proposed by Collins -LRB- 1997 -RRB- has been successfully applied to Czech and Chinese ',Collins,'nsubj_increase_Lexicalization aux_increase_can xcomp_increase_parsing dobj_parsing_performance advmod_parsing_dramatically prep_parsing_for nn__English pobj_for_ cc_increase_and det_model_the amod_model_lexicalized nsubjpass_applied_model partmod_model_proposed prep_proposed_by pobj_by_Collins appos_model_1997 aux_applied_has auxpass_applied_been advmod_applied_successfully conj_increase_applied prep_applied_to amod__Czech pobj_to_ cc__and amod__Chinese conj__ ' +o,3844,'It uses a log-linear model to define a distribution over the lexical category set for each word and the previous two categories and the forward backward algorithm efficiently sums over all histories to give a distibution for each word ',Ratnaparkhi,'nsubj_uses_It det_model_a amod_model_log-linear dobj_uses_model aux_define_to infmod_model_define det_distribution_a dobj_define_distribution prep_define_over det_category_the amod_category_lexical pobj_over_category partmod_category_set prep_set_for det_word_each pobj_for_word cc_word_and det_categories_the amod_categories_previous num_categories_two conj_word_categories nsubj_sums_ cc__and det_algorithm_the advmod_algorithm_forward amod_algorithm_backward conj__algorithm advmod_sums_efficiently dep_uses_sums prep_sums_over det_histories_all pobj_over_histories aux_give_to xcomp_sums_give det_distibution_a dobj_give_distibution prep_distibution_for det_word_each pobj_for_word ' +p,3845,'Some NLG researchers are impressed by the success of the BLEU evaluation metric in Machine Translation -LRB- MT -RRB- , which has transformed the MT field by allowing researchers to quickly and cheaply evaluate the impact of new ideas , algorithms , and data sets ',Papineni,'det_researchers_Some nn_researchers_NLG nsubjpass_impressed_researchers auxpass_impressed_are prep_impressed_by det_success_the pobj_by_success prep_success_of det_evaluation_the amod_evaluation_BLEU pobj_of_evaluation amod__metric nsubj_algorithms_ prep__in nn_Translation_Machine pobj_in_Translation abbrev__MT nsubj_transformed_which aux_transformed_has rcmod__transformed det_field_the nn_field_MT dobj_transformed_field prep_transformed_by pcomp_by_allowing dobj_allowing_researchers aux_evaluate_to advmod_evaluate_quickly cc_quickly_and conj_quickly_cheaply xcomp_allowing_evaluate det_impact_the dobj_evaluate_impact prep_impact_of amod_ideas_new pobj_of_ideas dep_impressed_algorithms cc_algorithms_and nsubj_sets_data conj_algorithms_sets ' +o,3846,'There has been a large number of studies in tagging and morphological disambiguation using various techniques such as statistical techniques , eg , , constraint-based techniques and transformation-based techniques ',Cutting,'expl_been_There aux_been_has det_number_a amod_number_large dobj_been_number prep_number_of pobj_of_studies prep_studies_in amod_disambiguation_tagging cc_tagging_and conj_tagging_morphological pobj_in_disambiguation partmod_disambiguation_using amod_techniques_various dobj_using_techniques dep_as_such prep_techniques_as amod__statistical nn__techniques dep__eg dep__ amod__constraint-based nn__techniques pobj_as_ cc__and amod__transformation-based nn__techniques conj__ ' +o,3847,'Because their joint distributions have such closed-form expressions , the parameters can be estimated directly from the training data without the need for an iterative fitting procedure -LRB- as is required , for example , to estimate the parameters of maximum entropy models ; -RRB- ',Berger,'mark_have_Because poss_distributions_their amod_distributions_joint nsubj_have_distributions advcl_estimated_have amod_expressions_such amod_expressions_closed-form dobj_have_expressions det_parameters_the nsubjpass_estimated_parameters aux_estimated_can auxpass_estimated_be advmod_estimated_directly prep_estimated_from det_data_the nn_data_training pobj_from_data prep_estimated_without det_need_the pobj_without_need prep_need_for det_procedure_an amod_procedure_iterative amod_procedure_fitting pobj_for_procedure mark_required_as auxpass_required_is dep_procedure_required prep_required_for pobj_for_example aux_estimate_to xcomp_required_estimate det_parameters_the dobj_estimate_parameters prep_parameters_of amod_models_maximum nn_models_entropy pobj_of_models dep_parameters_ ' +p,3848,'In addition , the averaged parameters technology is used to alleviate overfitting and achieve stable performance ',Collins,'prep_used_In pobj_In_addition det__the amod__averaged nn__parameters nn__technology nsubjpass_used_ auxpass_used_is aux_alleviate_to xcomp_used_alleviate dobj_alleviate_overfitting cc_alleviate_and conj_alleviate_achieve amod_performance_stable dobj_achieve_performance ' +o,3849,'For example , smoothing methods have played a central role in probabilistic approaches , and yet they are not being used in current large margin training algorithms ',Collins,'prep_played_For pobj_For_example amod_methods_smoothing nsubj_played_methods aux_played_have det_role_a amod_role_central dobj_played_role prep_role_in amod__probabilistic nn__approaches pobj_in_ cc_played_and advmod_used_yet nsubjpass_used_they aux_used_are neg_used_not auxpass_used_being conj_played_used prep_used_in amod_algorithms_current amod_algorithms_large nn_algorithms_margin nn_algorithms_training pobj_in_algorithms ' +o,3850,'Running words 1,864 14,437 Vocabulary size 569 1,081 Table 2 : ChineseEnglish corpus statistics using Phramer , a 3-gram language model with Kneser-Ney smoothing trained with SRILM on the English side of the training data and Pharaoh with default settings to decode ',Och,'csubj_size_Running nsubj_Vocabulary_words number_14,437_1,864 measure_Vocabulary_14,437 xcomp_Running_Vocabulary number_1,081_569 num_Table_1,081 dobj_size_Table num_Table_2 nn_statistics_ChineseEnglish nn_statistics_corpus nsubj__statistics parataxis_size_ xcomp__using nn__Phramer dobj_using_ det_model_a amod_model_3-gram nn_model_language appos__model prep_model_with amod_smoothing_Kneser-Ney pobj_with_smoothing partmod_smoothing_trained prep_trained_with nn__SRILM pobj_with_ prep__on det_side_the amod_side_English pobj_on_side prep_side_of det_data_the nn_data_training pobj_of_data cc_side_and nn__Pharaoh conj_side_ prep__with nn_settings_default pobj_with_settings aux_decode_to xcomp_trained_decode ' +o,3851,'This task is quite common in corpus linguistics and provides the starting point to many other algorithms , eg , for computing statistics such as pointwise mutual information , for unsupervised sense clustering , and more generally , a large body of work in lexical semantics based on distributional profiles , dating back to Firth and Harris ',Church,'det_task_This nsubj_common_task cop_common_is advmod_common_quite prep_common_in nn_linguistics_corpus pobj_in_linguistics cc_common_and conj_common_provides det_point_the amod_point_starting dobj_provides_point prep_provides_to amod_algorithms_many amod_algorithms_other pobj_to_algorithms appos_algorithms_eg dep_to_for amod_statistics_computing pobj_for_statistics dep_as_such prep_statistics_as amod__pointwise amod__mutual nn__information pobj_as_ dep_to_for amod_sense_unsupervised pobj_for_sense partmod_sense_clustering acomp_clustering_ cc__and conj__more advmod_clustering_generally det_body_a amod_body_large nsubj_based_body prep_body_of pobj_of_work prep_work_in amod_semantics_lexical pobj_in_semantics ccomp_clustering_based prep_based_on amod_profiles_distributional pobj_on_profiles dep_back_dating dep_to_back dep_to_to nn__Firth pobj_to_ cc__and nn__Harris conj__ ' +o,3852,'The score for a given candidate a9 is given by a modified IBM Model 1 probability (Brown et al. , 1993) as follows: a2a4a3a6a9a21a10a13a12a15a7a14a2 a15 a24a26a17a16 a2a4a3a6a9a19a18 a14a15a10a12 a7 (4) a2 a15 a20 a24a16a22a21a24a23a26a25a1a27a28a27a28a27 a20 a24a16a30a29a1a23a26a25 a31 a32 a33 a23a35a34a37a36 a3a38a12 a33 a10a12a9 a16a8a39 a7 (5) where a40 is the length of a9, a41 is the length of a12, a15 is a scaling factor based on the number of matches of a9 found, and a14 a33 is the index of the English word aligned with a12 a33 according to alignment a14 . The probability a36 a3a6a9 a16a8a39 a10a12 a33 a7 is a linear combination of the transliteration and translation score, where the translation score is a uniform probability over all dictionary entries for a12 a33 . The scored matches form the list of translation candidates.',Brown,'' +o,3853,'The heuristic estimator employs word-alignment -LRB- Giza + + -RRB- and a few thumb rules for defining phrase pairs , and then extracts a multi-set of phrase pairs and estimates their conditional probabilities based on the counts in the multi-set ',Och,'det_estimator_The amod_estimator_heuristic nsubj_employs_estimator amod__word-alignment nn_+_Giza nn_+_+ appos__+ dobj_employs_ cc__and det_rules_a amod_rules_few nn_rules_thumb conj__rules prep_rules_for amod_pairs_defining nn_pairs_phrase pobj_for_pairs cc_pairs_and advmod_pairs_then conj_pairs_extracts det_multi-set_a dep_extracts_multi-set prep_multi-set_of nn_pairs_phrase pobj_of_pairs cc_employs_and conj_employs_estimates poss_probabilities_their amod_probabilities_conditional dobj_estimates_probabilities prep_estimates_based dep_based_on det_counts_the pobj_on_counts prep_counts_in det_multi-set_the pobj_in_multi-set ' +o,3854,'We say that wv and nq are semantically related if w ~ i and nq are semantically related and -LRB- wp , nq -RRB- and -LRB- w ~ i , nq -RRB- are semantically similar ',Dagan,'nsubj_say_We complm_related_that nsubj_related_wv cc_wv_and conj_wv_nq cop_related_are advmod_related_semantically ccomp_say_related mark_related_if nn_i_w nn_i_~ nsubjpass_related_i cc_i_and conj_i_nq auxpass_related_are advmod_related_semantically dep_related_related cc_related_and dep_related_wp appos_wp_nq nsubj_similar_and nn_nq_w amod_nq_~ nn_nq_i appos_and_nq cop_similar_are advmod_similar_semantically conj_related_similar ' +o,3855,'These rules are learned using a word alignment model , which finds an optimal mapping from words to MR predicates given a set of training sentences and their correct MRs Word alignment models have been widely used for lexical acquisition in SMT ',Koehn,'det_rules_These nsubjpass_learned_rules auxpass_learned_are xcomp_learned_using det_model_a nn_model_word nn_model_alignment dobj_using_model nsubj_finds_which rcmod_model_finds det_mapping_an amod_mapping_optimal dobj_finds_mapping prep_finds_from pobj_from_words prep_words_to nn_predicates_MR pobj_to_predicates partmod_predicates_given det_set_a dobj_given_set prep_set_of nn_sentences_training pobj_of_sentences cc_sentences_and poss_alignment_their amod_alignment_correct nn_alignment_MRs nn_alignment_Word conj_sentences_alignment nsubjpass_used_models aux_used_have auxpass_used_been advmod_used_widely dep_learned_used prep_used_for amod_acquisition_lexical pobj_for_acquisition prep_acquisition_in pobj_in_SMT ' +o,3856,'In contrast to the opinion extracts produced by , our summaries are not text extracts , but rather explicitly identify and 337 characterize the relations between opinions and their sources ',Pang,'prep_extracts_In pobj_In_contrast prep_contrast_to det_extracts_the nn_extracts_opinion pobj_to_extracts partmod_extracts_produced prep_produced_by pobj_by_ poss_summaries_our nsubj_extracts_summaries cop_extracts_are neg_extracts_not nn_extracts_text dep_rather_but cc_extracts_rather advmod_identify_explicitly conj_extracts_identify cc_extracts_and advmod_characterize_337 conj_extracts_characterize det_relations_the dobj_characterize_relations prep_relations_between pobj_between_opinions cc_opinions_and poss_sources_their conj_opinions_sources ' +o,3857,' , , , , ',Pang,'nn__ appos__ appos__ ' +o,3858,'The hallucination process is motivated by the use of NULL alignments into Markov alignment models as done by ',Och,'det_process_The nn_process_hallucination nsubjpass_motivated_process auxpass_motivated_is prep_motivated_by det_use_the pobj_by_use prep_use_of amod_alignments_NULL pobj_of_alignments prep_motivated_into nn_models_Markov nn_models_alignment pobj_into_models mark_done_as advcl_motivated_done prep_done_by ' +o,3859,'Some of the data comes from the parsed files 2-21 of the Wall Street Journal Penn Treebank corpus , and additional parsed text was obtained by parsing the 1987 Wall Street Journal text using the parser described in Charniak et al ',Marcus,'nsubj_comes_Some prep_Some_of det_data_the pobj_of_data prep_comes_from det_2-21_the amod_2-21_parsed nn_2-21_files pobj_from_2-21 prep_2-21_of det__the nn__Wall nn__Street nn__Journal nn__Penn nn__Treebank nn__corpus pobj_of_ cc_comes_and amod_text_additional amod_text_parsed nsubjpass_obtained_text auxpass_obtained_was conj_comes_obtained prep_obtained_by pcomp_by_parsing det_text_the num_text_1987 nn_text_Wall nn_text_Street nn_text_Journal dobj_parsing_text partmod_text_using det_parser_the dobj_using_parser partmod_parser_described prep_described_in pobj_in_Charniak cc_Charniak_et conj_Charniak_al ' +o,3860,'and Gildea , 2007 ; et al , 2006 ; Gildea , Satta , and , 2006 -RRB- ',Zhang,'nn_Gildea_and appos_Gildea_2007 dep_Gildea_ cc__et conj__al appos__2006 dep_Gildea_Gildea conj_Gildea_Satta cc_Gildea_and nn_2006_ conj_Gildea_2006 ' +o,3861,'Work in modeled the limited information available at phrase-boundaries ',Zhang,'nsubj_modeled_Work prep_Work_in pobj_in_ det_information_the amod_information_limited nsubj_available_information xcomp_modeled_available prep_available_at pobj_at_phrase-boundaries ' +o,3862,'Incremental Sigmoid Belief Networks differ from simple dynamic SBNs in that they allow the model structure to depend on the output variable values ',Titov,'nn__Incremental nn__Sigmoid nn__Belief nn__Networks nsubj_differ_ prep_differ_from amod_SBNs_simple amod_SBNs_dynamic pobj_from_SBNs rel_allow_in pobj_in_that nsubj_allow_they rcmod_SBNs_allow det_structure_the nn_structure_model nsubj_depend_structure aux_depend_to xcomp_allow_depend prep_depend_on det_values_the nn_values_output nn_values_variable pobj_on_values ' +o,3863,'Second , the automatic approach , in which the model is automatically obtained from corpora -LRB- either raw or annotated -RRB- 1 , and consists of n-grams , rules or neural nets ',Cutting,'advmod_rules_Second det_approach_the amod_approach_automatic nsubj_rules_approach rel_obtained_in pobj_in_which det_model_the nsubjpass_obtained_model auxpass_obtained_is advmod_obtained_automatically rcmod_approach_obtained prep_obtained_from pobj_from_corpora preconj_raw_either dep_corpora_raw cc_raw_or conj_raw_annotated dep_corpora_1 cc_obtained_and conj_obtained_consists prep_consists_of amod__n-grams pobj_of_ amod__ cc__or conj__neural nn__nets dobj_rules_ ' +o,3864,'This approach is usually referred to as the noisy source-channel approach in statistical machine translation ',Brown,'det_approach_This nsubjpass_referred_approach auxpass_referred_is advmod_referred_usually prep_referred_to advmod_approach_as det_approach_the advmod_source-channel_noisy amod_approach_source-channel pobj_to_approach prep_approach_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,3865,' use a BLEU oracle decoder for discriminative training of a local reordering model ',Tillmann,'amod_use_ det_decoder_a amod_decoder_BLEU nn_decoder_oracle dep_use_decoder prep_decoder_for amod_training_discriminative pobj_for_training prep_training_of det_model_a amod_model_local nn_model_reordering pobj_of_model ' +o,3866,'Templates for local features are similar to the ones employed by for POS-tagging -LRB- Table 3 -RRB- , though as our input already includes POStags , we can make use of part-of-speech information as well ',Ratnaparkhi,'nsubj_similar_Templates prep_Templates_for amod_features_local pobj_for_features cop_similar_are prep_similar_to det_ones_the pobj_to_ones partmod_ones_employed prep_employed_by pobj_by_ prep__for pobj_for_POS-tagging number_3_Table dep_POS-tagging_3 mark_make_though mark_includes_as poss_input_our nsubj_includes_input advmod_includes_already advcl_make_includes dobj_includes_POStags nsubj_make_we aux_make_can advcl_employed_make dobj_make_use prep_use_of amod_information_part-of-speech pobj_of_information advmod_well_as advmod_make_well ' +o,3867,'From 9 Combined metric BY BP B4AC BE B7BDB5C8CABPB4AC BE C8 B7 CAB5 , from , AC BPBD ',Carletta,'number_9_ pobj_From_9 prep_9_Combined amod_CAB5_metric nn_CAB5_BY nn_CAB5_BP nn_CAB5_B4AC nn_CAB5_BE nn_CAB5_B7BDB5C8CABPB4AC nn_CAB5_BE nn_CAB5_C8 nn_CAB5_B7 dobj_BPBD_CAB5 dep_CAB5_from pobj_from_ nsubj_BPBD_AC dep_Combined_BPBD ' +o,3868,'Generally , WSD methods use the context of a word for its sense disambiguation , and the context information can come from either annotated\\/unannotated text or other knowledge resources , such as WordNet , SemCor , Open Mind Word Expert , eXtended WordNet , Wikipedia , parallel corpora ',Mihalcea,'advmod_use_Generally nn_methods_WSD nsubj_use_methods det_context_the dobj_use_context prep_context_of det_word_a pobj_of_word prep_word_for poss_disambiguation_its nn_disambiguation_sense pobj_for_disambiguation cc_use_and det_information_the nn_information_context nsubj_come_information aux_come_can conj_use_come prep_come_from det_text_either amod_text_annotated\\/unannotated pobj_from_text cc_text_or amod_resources_other nn_resources_knowledge conj_text_resources dep_as_such prep_text_as nn__WordNet pobj_as_ nn__SemCor appos__ nn__Open nn__Mind nn__Word nn__Expert appos__ amod__eXtended nn__WordNet appos__ nn__Wikipedia appos__ amod__parallel nn__corpora appos__ ' +o,3869,'Giving the increasing sophistication of probabilistic linguistic models -LRB- for example , has a statistical approach to learning gap-threading rules -RRB- a probabilistic extension of our work is attractive -- it will be interesting to see how far an integration of ` logical \' and statistical can go ',Collins,'csubj_attractive_Giving det_sophistication_the amod_sophistication_increasing dobj_Giving_sophistication prep_sophistication_of amod_models_probabilistic amod_models_linguistic pobj_of_models prep_has_for pobj_for_example nsubj_has_ dep_sophistication_has det_approach_a amod_approach_statistical dobj_has_approach prep_approach_to pcomp_to_learning amod_rules_gap-threading dobj_learning_rules det_extension_a amod_extension_probabilistic dep_sophistication_extension prep_extension_of poss_work_our pobj_of_work cop_attractive_is nsubj_interesting_it aux_interesting_will cop_interesting_be parataxis_attractive_interesting aux_see_to xcomp_interesting_see advmod_go_how advmod_how_far det_integration_an nsubj_go_integration prep_integration_of pobj_of_logical cc_logical_and conj_logical_statistical aux_go_can ccomp_see_go ' +o,3870,'The subset was the neighboring alignments of the Viterbi alignments discovered by Model 1 and Model 2 ',Brown,'det_subset_The nsubj__subset cop__was det__the amod__neighboring nn__alignments prep__of det_alignments_the nn_alignments_Viterbi pobj_of_alignments partmod_alignments_discovered prep_discovered_by pobj_by_Model num_Model_1 cc_Model_and conj_Model_Model num_Model_2 ' +o,3871,'To support this claim , first , we used the coefficient to assess the agreement between the classification made by FLSA and the classification from the corpora see Table 8 ',Carletta,'aux_support_To dep_used_support det_claim_this dobj_support_claim advmod_used_first nsubj_used_we det_coefficient_the nsubj__coefficient dep_used_ aux_assess_to xcomp__assess det_agreement_the dobj_assess_agreement prep_agreement_between det_classification_the pobj_between_classification partmod_classification_made prep_made_by pobj_by_FLSA cc__and det_classification_the nsubj_see_classification prep_classification_from det_corpora_the pobj_from_corpora conj__see number_8_Table dobj_see_8 ' +o,3872,' share the goal underlying our own research : improving , rather than replacing , Ochs MERT procedure ',Moore,'num_share_ nn_goal_share det_goal_the dep_goal_underlying poss_research_our amod_research_own dobj_underlying_research parataxis_underlying_improving cc_improving_rather dep_rather_than conj_improving_replacing dep_underlying_Ochs amod_procedure_MERT dobj_Ochs_procedure ' +o,3873,'For example , the topics Sport and Education are important cues for differentiating mentions of Michael Jordan , which may refer to a basketball player , a computer science professor , etc Second , as noted in the top WePS run , feature development is important in achieving good coreference performance ',Chen,'prep_mentions_For pobj_For_example det_topics_the nsubj_mentions_topics nsubj_cues_Sport cc_Sport_and conj_Sport_Education cop_cues_are amod_cues_important rcmod_topics_cues prep_cues_for pcomp_for_differentiating prep_mentions_of nn_Jordan_Michael pobj_of_Jordan nsubj_refer_which aux_refer_may rcmod_Jordan_refer prep_refer_to det_player_a nn_player_basketball pobj_to_player det_professor_a nn_professor_computer nn_professor_science appos_player_professor dep_player_etc advmod_important_Second mark_noted_as dep_important_noted prep_noted_in det__the amod__top nn__WePS nn__run pobj_in_ nn_development_feature nsubj_important_development cop_important_is dep_mentions_important prep_important_in pcomp_in_achieving amod_performance_good nn_performance_coreference dobj_achieving_performance ' +o,3874,'2 Background Default unification has been investigated by many researchers in the context of developing lexical semantics ',Marcus,'num_unification_2 nn_unification_Background nn_unification_Default nsubjpass_investigated_unification aux_investigated_has auxpass_investigated_been prep_investigated_by amod_researchers_many pobj_by_researchers partmod_researchers_ prep__in det_context_the pobj_in_context prep_context_of amod_semantics_developing amod_semantics_lexical pobj_of_semantics ' +o,3875,'The decoder is capable of producing nbest derivations and nbest lists , which are used for Maximum Bleu training ',Och,'det_decoder_The nsubj_capable_decoder cop_capable_is prep_capable_of pcomp_of_producing nn_derivations_nbest dobj_producing_derivations cc_derivations_and nn_lists_nbest conj_derivations_lists tmod_producing_ nsubjpass_used_which auxpass_used_are dep__used prep_used_for nn_training_Maximum nn_training_Bleu pobj_for_training ' +o,3876,'61 Distributional cluster (Brown et al. , 1992): tie, jacket, suit Word \'tie\' (7 alternatives) 0.0000 0.0000 0.0000 1.0000 0.0000 0.0000 0.0000 draw, standoff, tie, stalemate affiliation, association, tie, tie-up: a social or business relationship tie, crosstie, sleeper: subconcept of brace, bracing necktie, tie link, linkup, tie, tie-in: something that serves to join or link drawstring, string, tie: cord used as a fastener tie, tie beam: used to prevent two rafters, e.g., from spreading apart Word \'jacket\' (4 alternatives) 0.0000 book jacket, dust cover: subeoncept of promotional material 0.0000 jacket crown, jacket: artificial crown fitted over a broken or decayed tooth 0.0000 jacket: subconceptofwrapping, wrap, wrapper 1.0000 jacket: a short coat Word \'suit\' (4 alternatives) 0.0000 suit, suing: subconcept of entreaty, prayer, appeal 1.0000 suit, suit of clothes: subconcept of garment 0.0000 suit: any of four sets of13\' cards in a paek 0.0000 legal action, action, case, lawsuit, suit: a judicial proceeding This cluster was derived by Brown et al. using a modification of their algorithm, designed to uncover \'semantically sticky\' clusters.',Brown,'' +o,3877,'Others use sentence cohesion , agreement\\/disagreement between speakers , or structural adjacency ',Pang,'nsubj_use_Others nn__sentence nn__cohesion dobj_use_ conj__agreement\\/disagreement prep_agreement\\/disagreement_between nn__speakers pobj_between_ cc__or amod_adjacency_structural conj__adjacency ' +o,3878,'2 Word Alignment algorithm We use IBM Model 4 as a basis for our word alignment system ',Brown,'num_algorithm_2 nn_algorithm_Word nn_algorithm_Alignment nsubj_use_We dep_algorithm_use nn__IBM nn__Model num__4 dobj_use_ prep_use_as det_basis_a pobj_as_basis prep_basis_for poss_system_our nn_system_word nn_system_alignment pobj_for_system ' +o,3879,'52 Translation experiments with a bigram language model In this section we consider two real translation tasks , namely , translation from English to French , trained on Europarl and translation from German to Spanish training on the NewsCommentary corpus ',Koehn,'num_experiments_52 nn_experiments_Translation prep_experiments_with det_model_a nn_model_bigram nn_model_language pobj_with_model dep_experiments_In det_section_this pobj_In_section nsubj_consider_we rcmod_section_consider num_tasks_two amod_tasks_real nn_tasks_translation dobj_consider_tasks advmod_consider_namely conj_consider_translation prep_translation_from pobj_from_English prep_translation_to pobj_to_French conj_consider_trained prep_trained_on nn__Europarl pobj_on_ cc_consider_and conj_consider_translation prep_translation_from amod_training_German dep_German_to dep_German_Spanish pobj_from_training prep_training_on det_corpus_the nn_corpus_NewsCommentary pobj_on_corpus ' +o,3880,'Example of such algorithms are and that use syntactic features in the vector definition ',Lin,'nsubj__Example prep_Example_of amod_algorithms_such pobj_of_algorithms cop__are cc__and conj__ complm_use_that ccomp__use amod_features_syntactic dobj_use_features prep_features_in det_definition_the nn_definition_vector pobj_in_definition ' +o,3881,'More specifically , two recent works have suggested to use statistical data on lexical relations for resolving ambiguity cases of PP-attachment and pronoun references ',Hindle,'advmod_specifically_More advmod_suggested_specifically num_works_two amod_works_recent nsubj_suggested_works aux_suggested_have aux_use_to xcomp_suggested_use amod_data_statistical dobj_use_data prep_use_on amod_relations_lexical pobj_on_relations prep_use_for pcomp_for_resolving amod_cases_ambiguity dobj_resolving_cases prep_cases_of amod__PP-attachment pobj_of_ cc__and nn__pronoun nn__references conj__ ' +o,3882,'The table also shows the - score , which is another commonly used measure for inter-annotator agreement ',Carletta,'det_table_The nsubj_shows_table advmod_shows_also det_score_the dobj_shows_score nsubj_measure_which cop_measure_is det_measure_another advmod_used_commonly amod_measure_used rcmod_score_measure prep_measure_for amod_agreement_inter-annotator pobj_for_agreement ' +o,3883,'For getting the syntax trees , the latest version of Collins parser was used ',Collins,'pcomp_For_getting det_trees_the nn_trees_syntax nsubjpass_used_trees det_version_the amod_version_latest appos_trees_version prep_version_of nn__Collins nn__parser pobj_of_ auxpass_used_was ccomp_getting_used ' +o,3884,'c2009 Association for Computational Linguistics Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures Lamia Tounsi Mohammed Attia NCLT, School of Computing, Dublin City University, Ireland {lamia.tounsi, mattia, josef}@computing.dcu.ie Josef van Genabith Abstract A number of papers have reported on methods for the automatic acquisition of large-scale, probabilistic LFG-based grammatical resources from treebanks for English (Cahill and al., 2002), (Cahill and al., 2004), German (Cahill and al., 2003), Chinese (Burke, 2004), (Guo and al., 2007), Spanish (ODonovan, 2004), (Chrupala and van Genabith, 2006) and French (Schluter and van Genabith, 2008).',Cahill,'' +o,3885,'Recently , specic probabilistic tree-based models have been proposed not only for machine translation , but also for summarization , paraphrasing , natural language generation , parsing , and language modeling -LRB- Baker 1979 ; Lari and Young 1990 ; 1997 ; Chelba and Jelinek 2000 ; Charniak 2001 ; Klein Information Sciences Institute , 4676 Admiralty Way , Marina del Rey , CA 90292 ',Collins,'advmod_proposed_Recently amod_models_specic amod_models_probabilistic amod_models_tree-based nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been cc_for_not dep_not_only prep_proposed_for nn__machine nn__translation pobj_for_ dep_also_but cc_for_also conj_for_for nn__summarization pobj_for_ amod__paraphrasing conj__ amod__natural nn__language nn__generation conj__ conj__parsing cc__and nn_modeling_language conj__modeling dep_modeling_Baker num_Baker_1979 dep_Baker_Lari cc_Lari_and conj_Lari_Young num_Young_1990 dep__ num__1997 dep__Chelba cc_Chelba_and conj_Chelba_Jelinek num_Jelinek_2000 dep__Charniak num_Charniak_2001 nn_Institute_Klein nn_Institute_Information nn_Institute_Sciences dep__Institute num_Way_4676 nn_Way_Admiralty conj__Way nn_Rey_Marina nn_Rey_del conj__Rey conj__CA num_CA_90292 ' +o,3886,'Monotone Nonmonotone Target B A Positions C D Source Positions Figure 1 : Two Types of Alignment The IBM model 1 -LRB- IBM-1 -RRB- assumes that all alignments have the same probability by using a uniform distribution : p -LRB- fJ1 eI1 -RRB- = 1IJ Jproductdisplay j = 1 Isummationdisplay i = 1 p -LRB- fj ei -RRB- -LRB- 2 -RRB- We use the IBM-1 to train the lexicon parameters p -LRB- f e -RRB- , the training software is GIZA + + ',Brown,'nn_Types_Monotone nn_Types_Nonmonotone nn_Types_Target nn_Types_B nn_Types_A nn_Types_Positions nn_Types_C nn_Types_D nn_Types_Source nn_Types_Positions dep_Types_Figure dep_Types_1 num_Types_Two prep_Types_of pobj_of_Alignment det_model_The nn_model_IBM nn__model num_model_1 abbrev__IBM-1 nsubj_assumes_ dep_Types_assumes complm_have_that det_alignments_all nsubj_have_alignments ccomp_assumes_have det_probability_the amod_probability_same dobj_have_probability prep_probability_by pcomp_by_using det_distribution_a amod_distribution_uniform dobj_using_distribution nsubj_=_p amod_eI1_fJ1 appos_p_eI1 rcmod_probability_= num_Isummationdisplay_1IJ nn_Isummationdisplay_Jproductdisplay nn_Isummationdisplay_j nn_Isummationdisplay_= num_Isummationdisplay_1 nsubj_=_Isummationdisplay dep_Isummationdisplay_i ccomp_=_= num_p_1 dobj_=_p nn_ei_fj appos_p_ei dep_use_2 nsubj_use_We rcmod_p_use det_IBM-1_the nsubj_train_IBM-1 aux_train_to xcomp_use_train det_p_the nn_p_lexicon nn_p_parameters dobj_train_p dep_-LRB-_f dep_f_e det_software_the nn_software_training nsubj_GIZA_software cop_GIZA_is dep_use_GIZA cc_GIZA_+ advmod_GIZA_+ dep_GIZA_ ' +p,3887,'Successful discriminative parsers have used generative models to reduce training time and raise accuracy above generative baselines ',Collins,'amod_parsers_Successful nn_parsers_discriminative nsubj_used_parsers aux_used_have amod_models_generative dobj_used_models aux_reduce_to xcomp_used_reduce nn_time_training dobj_reduce_time cc_reduce_and conj_reduce_raise dobj_raise_accuracy prep_raise_above amod_baselines_generative pobj_above_baselines ' +p,3888,'1 Introduction A hypergraph , as demonstrated by , is a compact data-structure that can encode an exponential number of hypotheses generated by a regular phrase-based machine translation -LRB- MT -RRB- system -LRB- eg , Koehn et al ',Huang,'num_hypergraph_1 nn_hypergraph_Introduction nn_hypergraph_A nsubj_data-structure_hypergraph mark_demonstrated_as dep_hypergraph_demonstrated prep_demonstrated_by pobj_by_ cop_data-structure_is det_data-structure_a amod_data-structure_compact nsubj_encode_that aux_encode_can rcmod_data-structure_encode det_number_an amod_number_exponential dobj_encode_number prep_number_of pobj_of_hypotheses partmod_hypotheses_generated prep_generated_by det_system_a amod_system_regular amod_system_phrase-based nn_system_machine nn_system_translation abbrev_system_MT pobj_by_system dep_system_eg dep_eg_Koehn cc_Koehn_et conj_Koehn_al ' +n,3889,'We preferred the log-likelihood ratio to other statistical scores , such as the association ratio or ; -LRB- 2 , since it adequately takes into account the frequency of the co-occurring words and is less sensitive to rare events and corpussize ',Church,'nsubj_preferred_We det_ratio_the amod_ratio_log-likelihood dobj_preferred_ratio prep_preferred_to amod_such_other amod_such_statistical dep_such_scores pobj_to_such prep_such_as det__the nn__association nn__ratio pobj_as_ cc_such_or nsubj__2 mark_takes_since nsubj_takes_it advmod_takes_adequately parataxis__takes prep_takes_into pobj_into_account det_frequency_the dobj_takes_frequency prep_frequency_of det_words_the amod_words_co-occurring pobj_of_words cc_takes_and cop_sensitive_is advmod_sensitive_less conj_takes_sensitive prep_sensitive_to amod_events_rare pobj_to_events cc_sensitive_and conj_sensitive_corpussize parataxis_preferred_ ' +o,3890,'The system is tested on base noun-phrase -LRB- NP -RRB- chunking using the Wall Street Journal corpus ',Marcus,'det_system_The nsubjpass_tested_system auxpass_tested_is prep_tested_on nn_noun-phrase_base pobj_on_noun-phrase abbrev_noun-phrase_NP partmod_noun-phrase_chunking xcomp_chunking_using det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal dobj_using_corpus ' +o,3891,'1 Introduction Most empirical work in translation analyzes models and algorithms using BLEU and related metrics ',Papineni,'num_work_1 nn_work_Introduction nn_work_Most amod_work_empirical nsubj_analyzes_work prep_work_in pobj_in_translation dobj_analyzes_models cc_models_and conj_models_algorithms partmod_models_using nn__BLEU dobj_using_ cc__and amod_metrics_related conj__metrics ' +o,3892,'This fact is being seriously challenged by current research -LRB- -RRB- , and might not be true in the near future ',Smadja,'det_fact_This nsubjpass_challenged_fact aux_challenged_is auxpass_challenged_being advmod_challenged_seriously prep_challenged_by amod_-LRB-_current nn_-LRB-_research cc_challenged_and aux_true_might neg_true_not cop_true_be conj_challenged_true prep_true_in det_future_the amod_future_near pobj_in_future ' +o,3893,'To perform minimum error rate training to tune the feature weights to maximize the systems BLEU score on development set , we used the script optimizeV5IBMBLEUm ',Och,'aux_perform_To dep_used_perform amod__minimum nn__error nn__rate nn__training dobj_perform_ aux_tune_to xcomp_perform_tune det_weights_the nn_weights_feature nsubj_maximize_weights aux_maximize_to xcomp_tune_maximize det_score_the nn_score_systems nn_score_BLEU dobj_maximize_score prep_maximize_on nn_set_development pobj_on_set nsubj_used_we det__the nn__script nn__optimizeV5IBMBLEUm dobj_used_ ' +p,3894,' compares his method to and shows that for four words the former performs significantly better in distinguishing between two senses ',Yarowsky,'advmod_compares_ poss_method_his dobj_compares_method prep_compares_to pobj_to_ cc__and conj__shows dep_performs_that prep_performs_for num_words_four pobj_for_words det_former_the nsubj_performs_former dep_compares_performs advmod_better_significantly acomp_performs_better prep_better_in pcomp_in_distinguishing prep_distinguishing_between num_senses_two pobj_between_senses ' +o,3895,'These tags are drawn from a tagset which is constructed by extending each argument label by three additional symbols a11 a24 a35 a24a4a12 , following ',Ramshaw,'det_tags_These nsubjpass_drawn_tags auxpass_drawn_are prep_drawn_from det_tagset_a pobj_from_tagset nsubjpass_constructed_which auxpass_constructed_is rcmod_tagset_constructed prep_constructed_by pcomp_by_extending det_label_each nn_label_argument dobj_extending_label prep_extending_by num_symbols_three amod_symbols_additional pobj_by_symbols amod_a24a4a12_a11 amod_a24a4a12_a24 amod_a24a4a12_a35 tmod_extending_a24a4a12 appos_a24a4a12_following ' +o,3896,'The named-entity features are generated by the freely available Stanford NER tagger ',Finkel,'det_features_The amod_features_named-entity nsubjpass_generated_features auxpass_generated_are prep_generated_by det_tagger_the advmod_available_freely amod_tagger_available nn_tagger_Stanford nn_tagger_NER pobj_by_tagger ' +o,3897,'Some of them use human reference translations , eg , the BLEU method , which is based on comparison of N-gram models in MT output and in a set of human reference translations ',Papineni,'nsubj_use_Some prep_Some_of pobj_of_them amod_translations_human nn_translations_reference dobj_use_translations appos_translations_eg det__the amod__BLEU nn__method appos_translations_ nsubjpass_based_which auxpass_based_is rcmod_translations_based prep_based_on pobj_on_comparison prep_comparison_of amod_models_N-gram pobj_of_models prep_models_in nn_output_MT pobj_in_output cc_on_and conj_on_in det_set_a pobj_in_set prep_set_of amod_translations_human nn_translations_reference pobj_of_translations ' +o,3898,' -RRB- , or by using linguistic evidence , mostly lexical similarity -LRB- METEOR , ; MaxSim , Chan and Ng -LRB- 2008 -RRB- -RRB- , or syntactic overlap -LRB- Owczarzak et al ',Banerjee,'dep_by_ cc_by_or nsubj_overlap_by rcmod_by_using amod_evidence_linguistic dobj_using_evidence advmod_similarity_mostly amod_similarity_lexical appos_evidence_similarity dep_similarity_METEOR conj_METEOR_ conj_METEOR_MaxSim conj_METEOR_Chan cc_METEOR_and conj_METEOR_Ng appos_similarity_2008 cc_by_or conj_by_syntactic nn_al_Owczarzak nn_al_et dep_overlap_al ' +o,3899,'Instead of using the NP bracketing information present in the tagged Treebank data , and Marcus modified the data so as to include bracketing information related only to the non-recursive , base NPs present in each sentence while the subject verb phrases were taken as is The data sets include POS tag information generated by and Marcus using Brill \'s transformational part-of-speech tagger ',Ramshaw,'dep_of_Instead prep_present_of pcomp_of_using det_information_the nn_information_NP nn_information_bracketing dobj_using_information prep_present_in det_Treebank_the nn_Treebank_tagged pobj_in_Treebank nsubj_present_data appos_data_ cc__and conj__Marcus partmod__modified det_data_the dobj_modified_data advmod_as_so advmod_modified_as aux_include_to dep_as_include amod_information_bracketing dobj_include_information partmod_information_related advmod_related_only prep_related_to det_present_the amod_present_non-recursive amod_present_base nn_present_NPs pobj_to_present prep_present_in det_sentence_each pobj_in_sentence mark_taken_while det_phrases_the amod_phrases_subject dep_subject_verb nsubjpass_taken_phrases auxpass_taken_were advcl_related_taken mark_data_as cop_data_is det_data_The advcl_taken_data nsubj_include_sets rcmod_data_include dep_information_POS nn_information_tag nsubj_generated_information ccomp_include_generated prep_generated_by pobj_by_ cc__and conj__Marcus xcomp_generated_using poss__Brill possessive_Brill_\'s amod__transformational amod__part-of-speech nn__tagger dobj_using_ ' +o,3900,'Recently , many phrase reordering methods have been proposed , ranging from simple distancebased distortion model , flat reordering model , lexicalized reordering model , to hierarchical phrase-based model and classifier-based reordering model with linear features ',Och,'advmod_proposed_Recently amod_methods_many nn_methods_phrase nn_methods_reordering nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been xcomp_proposed_ranging prep_ranging_from amod__simple amod__distancebased nn__distortion nn__model nn__ amod__flat nn__reordering nn__model pobj_from_ amod__lexicalized nn__reordering nn__model conj__ aux_hierarchical_to conj__hierarchical amod__phrase-based nn__model dobj_hierarchical_ cc__and amod_model_classifier-based nn_model_reordering conj__model prep_model_with amod__linear nn__features pobj_with_ ' +o,3901,'This was done for supervised parsing in different ways by , Klein and Manning -LRB- 2003 -RRB- , and McDonald et al ',Collins,'nsubjpass_done_This auxpass_done_was prep_done_for pobj_for_supervised amod_supervised_parsing prep_supervised_in amod_ways_different pobj_in_ways prep_ways_by pobj_by_ conj__Klein cc__and conj__Manning appos_supervised_2003 cc_supervised_and conj_supervised_McDonald cc_McDonald_et conj_McDonald_al ' +o,3902,'While Schiitze and Pedersen , and Futrelle and Gauch all demonstrate the ability of their systems to identify word similarity using clustering on the most frequently occurring words in their corpus , only Grefenstette demonstrates his system by generating word similarities with respect to a set of target words ',Brown,'mark_demonstrate_While nn__Schiitze cc_Schiitze_and conj_Schiitze_Pedersen nsubj_demonstrate_ conj__ cc__and conj__Futrelle cc__and nn__Gauch conj__ advmod_demonstrate_all advcl_demonstrates_demonstrate det_ability_the dobj_demonstrate_ability prep_ability_of poss_systems_their pobj_of_systems aux_identify_to infmod_systems_identify nn_similarity_word dobj_identify_similarity partmod_similarity_using xcomp_using_clustering prep_clustering_on det_words_the advmod_frequently_most advmod_occurring_frequently amod_words_occurring pobj_on_words prep_words_in poss_corpus_their pobj_in_corpus advmod__only nn__Grefenstette nsubj_demonstrates_ poss_system_his dobj_demonstrates_system prep_demonstrates_by pcomp_by_generating nn_similarities_word dobj_generating_similarities prep_generating_with pobj_with_respect prep_generating_to det_set_a pobj_to_set prep_set_of nn_words_target pobj_of_words ' +p,3903,'So far , SCL has been applied successfully in NLP for Part-of-Speech tagging and Sentiment Analysis ',Blitzer,'dep_far_So advmod_applied_far nsubjpass_applied_SCL aux_applied_has auxpass_applied_been advmod_applied_successfully prep_applied_in pobj_in_NLP prep_NLP_for amod_tagging_Part-of-Speech pobj_for_tagging cc_tagging_and nn_Analysis_Sentiment conj_tagging_Analysis ' +o,3904,'Intercoder reliability was assessed using Cohen \'s Kappa statistic -LRB- ~ -RRB- ',Carletta,'nn_reliability_Intercoder nsubjpass_assessed_reliability auxpass_assessed_was xcomp_assessed_using poss_statistic_Cohen possessive_Cohen_\'s nn_statistic_Kappa dobj_using_statistic appos_statistic_~ ' +o,3905,'This technique is called system combination ',Rosti,'det_technique_This nsubjpass_called_technique auxpass_called_is nn_combination_system dobj_called_combination ' +o,3906,'The agreement on identifying the boundaries of units , using the AK statistic discussed in , was AK BP BMBL -LRB- for two annotators and 500 units -RRB- ; the agreement on features -LRB- 2 annotators and at least 200 units -RRB- was follows : Attribute AK Value utype 76 verbed 9 finite 81 subject 86 NPs Our instructions for identifying NP markables derive from those proposed in the MATE project scheme for annotating anaphoric relations ',Carletta,'det_agreement_The nsubj_BMBL_agreement prep_agreement_on pcomp_on_identifying det_boundaries_the dobj_identifying_boundaries prep_boundaries_of pobj_of_units partmod_BMBL_using det_statistic_the nn_statistic_AK dobj_using_statistic partmod_statistic_discussed prep_discussed_in pobj_in_ cop_BMBL_was nn_BMBL_AK nn_BMBL_BP prep_BMBL_for num_annotators_two pobj_for_annotators cc_annotators_and num_units_500 conj_annotators_units det_agreement_the nsubj_follows_agreement prep_agreement_on pobj_on_features num_annotators_2 dep_features_annotators cc_annotators_and quantmod_200_at dep_at_least num_units_200 conj_annotators_units dep_features_was parataxis_BMBL_follows nn_Value_Attribute nn_Value_AK nsubj_utype_Value ccomp_follows_utype num_instructions_76 amod_instructions_verbed num_instructions_9 amod_instructions_finite num_instructions_81 amod_instructions_subject num_instructions_86 nn_instructions_NPs nn_instructions_Our dobj_utype_instructions prep_utype_for pcomp_for_identifying nn_markables_NP nsubj_derive_markables xcomp_identifying_derive prep_derive_from pobj_from_those partmod_those_proposed prep_proposed_in det_scheme_the nn_scheme_MATE nn_scheme_project pobj_in_scheme prep_proposed_for pcomp_for_annotating amod_relations_anaphoric dobj_annotating_relations advmod_annotating_ ' +o,3907,'As machine learners we used SVM-light1 and the MaxEnt decider from the Stanford Classifier2 ',Manning,'prep_used_As nn_learners_machine pobj_As_learners nsubj_used_we amod__SVM-light1 dobj_used_ cc__and det_decider_the nn_decider_MaxEnt conj__decider prep_used_from det_Classifier2_the nn_Classifier2_Stanford pobj_from_Classifier2 ' +o,3908,'In order to estimate the entropy of English , approximated P -LRB- kI -RRB- by a Poisson distribution whose parameter is the average word length A in the training corpus , and P -LRB- cz cklk , -RRB- by the product of character zerogram probabilities ',Brown,'mark_estimate_In dep_estimate_order aux_estimate_to det_entropy_the dobj_estimate_entropy prep_entropy_of dep__English amod_P_ amod_P_approximated pobj_of_P number__kI dep_P_ prep_estimate_by det_distribution_a amod_distribution_Poisson pobj_by_distribution poss_parameter_whose nsubj_A_parameter cop_A_is det_A_the amod_A_average nn_A_word nn_A_length rcmod_distribution_A prep_A_in det_corpus_the nn_corpus_training pobj_in_corpus cc_A_and conj_A_P nn_cklk_cz appos_P_cklk dep_cklk_ prep_P_by det_product_the pobj_by_product prep_product_of nn_probabilities_character nn_probabilities_zerogram pobj_of_probabilities ' +o,3909,'32 Conversion to Dependencies 321 Syntactic Dependencies There exists no large-scale dependency treebank for English , and we thus had to construct a dependency-annotated corpus automatically from the Penn Treebank ',Marcus,'number_321_32 number_321_Conversion dep_321_to number_321_Dependencies num_Dependencies_321 amod_Dependencies_Syntactic expl_exists_There dep_Dependencies_exists det_treebank_no amod_treebank_large-scale amod_treebank_dependency dobj_exists_treebank prep_treebank_for pobj_for_English cc_exists_and nsubj_had_we advmod_had_thus conj_exists_had aux_construct_to xcomp_had_construct det_corpus_a amod_corpus_dependency-annotated dobj_construct_corpus advmod_construct_automatically prep_construct_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,3910,'Some researchers have concentrated on producing WSD systems that base results on a limited number of words , for example and Schtitze -LRB- 1992 -RRB- who quoted results for 12 words , and a second group , including Leacock , Towell , and Voorhees and Bruce and Wiebe , who gave results for just one , namely interest ',Yarowsky,'det_researchers_Some nsubj_concentrated_researchers aux_concentrated_have dep_gave_concentrated prep_concentrated_on pcomp_on_producing nn_systems_WSD dobj_producing_systems complm_results_that nn_results_base nsubj_results_results prep_results_on det_number_a amod_number_limited pobj_on_number prep_number_of pobj_of_words prep_number_for nn__example pobj_for_ cc__and conj__Schtitze appos_Schtitze_1992 nsubj_quoted_who rcmod_Schtitze_quoted ccomp_producing_results prep_results_for num_words_12 pobj_for_words cc_words_and det_group_a amod_group_second conj_words_group prep_words_including nn_Towell_Leacock pobj_including_Towell cc_Towell_and nn__Voorhees conj_Towell_ cc_Towell_and nn__Bruce cc_Bruce_and conj_Bruce_Wiebe conj_Towell_ nsubj_gave_who iobj_gave_results prep_results_for quantmod_one_just pobj_for_one advmod_one_namely dobj_gave_interest ' +o,3911,'They recover additional latent variables so-called nuisance variablesthat are not of interest to the user1 For example , though machine translation -LRB- MT -RRB- seeks to output a string , typical MT systems 1These nuisance variables may be annotated in training data , but it is more common for them to be latent even there , ie , there is no supervision as to their correct values ',Koehn,'nsubj_recover_They amod_variables_additional amod_variables_latent dobj_recover_variables amod_variablesthat_so-called amod_variablesthat_nuisance nsubj_are_variablesthat rcmod_variables_are neg_are_not prep_are_of pobj_of_interest prep_interest_to det_user1_the pobj_to_user1 dep__For pobj_For_example mark_seeks_though nn_translation_machine nsubj_seeks_translation abbrev_translation_MT advcl__seeks aux_output_to xcomp_seeks_output det_string_a dobj_output_string amod_systems_typical nn_systems_MT nsubj__systems dep_recover_ amod_variables_1These amod_variables_nuisance nsubjpass_annotated_variables aux_annotated_may auxpass_annotated_be ccomp__annotated prep_annotated_in nn_data_training pobj_in_data cc__but nsubj_common_it cop_common_is advmod_common_more conj__common prep_common_for pobj_for_them aux_latent_to cop_latent_be xcomp_common_latent dep_there_even dep_latent_there appos_there_ie expl_is_there conj__is det_supervision_no dep_as_supervision advmod_is_as prep_is_to poss_values_their amod_values_correct pobj_to_values ' +p,3912,'1 Introduction With the introduction of the BLEU metric for machine translation evaluation , the advantages of doing automatic evaluation for various NLP applications have become increasingly appreciated : they allow for faster implement-evaluate cycles -LRB- by by-passing the human evaluation bottleneck -RRB- , less variation in evaluation performance due to errors in human assessor judgment , and , not least , the possibility of hill-climbing on such metrics in order to improve system performance ',Papineni,'num_Introduction_1 prep_become_With det_introduction_the pobj_With_introduction prep_introduction_of det_metric_the amod_metric_BLEU pobj_of_metric prep_metric_for nn__machine nn__translation nn__evaluation pobj_for_ det_advantages_the nsubj_become_advantages prep_advantages_of pcomp_of_doing amod_evaluation_automatic dobj_doing_evaluation prep_doing_for amod_applications_various nn_applications_NLP pobj_for_applications aux_become_have dep_Introduction_become advmod_appreciated_increasingly dep_become_appreciated nsubj_allow_they parataxis_become_allow prep_allow_for amod_cycles_faster amod_cycles_implement-evaluate pobj_for_cycles dep_cycles_by pcomp_by_by-passing det_bottleneck_the amod_bottleneck_human nn_bottleneck_evaluation dobj_by-passing_bottleneck amod_variation_less appos_cycles_variation prep_variation_in nn_performance_evaluation pobj_in_performance amod_performance_due prep_due_to pobj_to_errors prep_errors_in amod_judgment_human nn_judgment_assessor pobj_in_judgment cc_errors_and neg_least_not conj_errors_least det_possibility_the appos_cycles_possibility prep_possibility_of pcomp_of_hill-climbing prep_hill-climbing_on amod_metrics_such pobj_on_metrics mark_improve_in dep_improve_order aux_improve_to purpcl_hill-climbing_improve nn__system nn__performance dobj_improve_ ' +o,3913,'It has been further observed that simply compressing sentences individually and concatenating the results leads to suboptimal summaries ','Daume III','nsubjpass_observed_It aux_observed_has auxpass_observed_been advmod_observed_further complm_leads_that advmod_compressing_simply csubj_leads_compressing dobj_compressing_sentences advmod_compressing_individually cc_compressing_and conj_compressing_concatenating det_results_the dobj_concatenating_results ccomp_observed_leads prep_leads_to amod_summaries_suboptimal pobj_to_summaries ' +o,3914,'Another way to look the algorithm is from the self-training perspective ',McClosky,'det_way_Another nsubj_is_way aux_look_to infmod_way_look det_algorithm_the dobj_look_algorithm prep_is_from det_perspective_the amod_perspective_self-training pobj_from_perspective ' +o,3915,'Stochastic taggers use both contextual and morphological information , and the model parameters are usually defined or updated automatically from tagged texts -LRB- Cerf-Danon and E1-Beze 1991 ; Church 1988 ; Cutting et al 1992 ; Dermatas and Kokkinakis 1988 , 1990 , 1993 , 1994 ; Garside , Leech , and Sampson 1987 ; Kupiec 1992 ; Maltese \\* Department of Electrical Engineering , Wire Communications Laboratory -LRB- WCL -RRB- , University of Patras , 265 00 Patras , Greece ',Marcus,'amod_taggers_Stochastic nsubj_use_taggers det_information_both amod_information_contextual cc_contextual_and conj_contextual_morphological dobj_use_information cc_use_and det_parameters_the nn_parameters_model nsubjpass_defined_parameters auxpass_defined_are advmod_defined_usually conj_use_defined cc_defined_or conj_defined_updated dep_from_automatically prep_defined_from amod_texts_tagged pobj_from_texts dep_texts_Cerf-Danon cc_Cerf-Danon_and conj_Cerf-Danon_E1-Beze num_E1-Beze_1991 dep_Cerf-Danon_Church dep_Church_1988 dep_Cerf-Danon_Cutting cc_Cutting_et conj_Cutting_al dep_Cutting_1992 dep_Cerf-Danon_Dermatas nn_Kokkinakis_and dep_Dermatas_Kokkinakis num_Kokkinakis_1988 num_Kokkinakis_1990 num_Kokkinakis_1993 num_Kokkinakis_1994 dep_Cerf-Danon_Garside conj_Garside_Leech cc_Garside_and conj_Garside_Sampson num_Sampson_1987 dep_Cerf-Danon_Kupiec dep_Kupiec_1992 nn_Department_Maltese nn_Department_\\* dep_Cerf-Danon_Department prep_Department_of amod_Engineering_Electrical pobj_of_Engineering nn_Laboratory_Wire nn_Laboratory_Communications appos_texts_Laboratory abbrev_Laboratory_WCL appos_Laboratory_University prep_University_of pobj_of_Patras number_00_265 num_Patras_00 appos_texts_Patras appos_Patras_Greece ' +o,3916,'In addition , we also made a word alignment available , which was derived using a variant of the current default method for word alignment s refined method ',Och,'prep_made_In pobj_In_addition nsubj_made_we advmod_made_also det_word_a dobj_made_word amod_word_alignment dep_alignment_available nsubjpass_derived_which auxpass_derived_was rcmod_word_derived xcomp_derived_using det_variant_a dobj_using_variant prep_variant_of det_method_the amod_method_current nn_method_default pobj_of_method prep_method_for nn_alignment_word pobj_for_alignment nsubj_method_ dep_method_s amod_method_refined dep_made_method ' +n,3917,'The ubiquitous minimum error rate training -LRB- MERT -RRB- approach optimizes Viterbi predictions , but does not explicitly boost the aggregated posterior probability of desirable n-grams ',Och,'det_training_The amod_training_ubiquitous amod_training_minimum nn_training_error nn_training_rate nsubj_approach_training abbrev_training_MERT amod_predictions_optimizes nn_predictions_Viterbi dobj_approach_predictions cc_approach_but aux_boost_does neg_boost_not advmod_boost_explicitly conj_approach_boost det_probability_the amod_probability_aggregated nn_probability_posterior dobj_boost_probability prep_probability_of amod_n-grams_desirable pobj_of_n-grams ' +o,3918,'In contrast , more recent research has focused on stochastic approaches that model discourse coherence at the local lexical and global levels , while preserving regularities recognized by classic discourse theories ',Och,'prep_focused_In pobj_In_contrast advmod_research_more amod_research_recent nsubj_focused_research aux_focused_has prep_focused_on amod_approaches_stochastic pobj_on_approaches complm__that nn_coherence_model nn_coherence_discourse nsubj__coherence prep_coherence_at det__the amod__local amod__lexical pobj_at_ cc__and amod_levels_global conj__levels ccomp_focused_ dep_preserving_while parataxis__preserving dobj_preserving_regularities partmod_regularities_recognized prep_recognized_by amod_theories_classic nn_theories_discourse pobj_by_theories dobj__ ' +o,3919,'In Smadja \'s collocation algorithm Xtract , the lowest-frequency words are effectively discarded as well ',Smadja,'prep_discarded_In poss_Xtract_Smadja possessive_Smadja_\'s nn_Xtract_collocation nn_Xtract_algorithm pobj_In_Xtract det_words_the amod_words_lowest-frequency nsubjpass_discarded_words auxpass_discarded_are advmod_discarded_effectively advmod_well_as advmod_discarded_well ' +o,3920,'We are encoding the knowledge as axioms in what is for the most part a first-order logic , described by , although quantification over predicates is sometimes convenient ',Hobbs,'nsubj_encoding_We aux_encoding_are det_knowledge_the dobj_encoding_knowledge prep_encoding_as pobj_as_axioms prep_axioms_in nsubj_is_what pcomp_in_is prep_is_for det_part_the amod_part_most pobj_for_part det_logic_a amod_logic_first-order dep_part_logic partmod_part_described prep_described_by pobj_by_ mark_convenient_although nsubj_convenient_quantification prep_quantification_over pobj_over_predicates cop_convenient_is advmod_convenient_sometimes advcl_is_convenient ' +o,3921,'-LRB- 2 -RRB- X1\\/X2 Y1 : r1\\/Y2 : r2 , -LRB- i1 , j1 , i2 , j2 -RRB- , Y1\\/Y2 , -LRB- j1 , k1 , j2 , k2 -RRB- X1\\/X2 Y1 : r1\\/Y2 : r2 , -LRB- i1 , k1 , i2 , k2 -RRB- -LRB- 3 -RRB- X1\\/X2 Y1 : r1\\/Y2 : r2 , -LRB- i1 , j1 , j2 , k2 -RRB- , Y1\\/Y2 , -LRB- j1 , k1 , i2 , j2 -RRB- X1\\/X2 Y1 : r1\\/Y2 : r2 , -LRB- i1 , k1 , i2 , k2 -RRB- Since each inference rule contains six free variables over string positions -LRB- i1 , j1 , k1 , i2 , j2 , k2 -RRB- , we get a parsing complexity of order O -LRB- n6 -RRB- for unlexicalized grammars -LRB- where n is the number of words in the longer of the two strings from language L1 and L2 -RRB- ',Wu,'dep_X1\\/X2_2 prep_X1\\/X2_Y1 dep_Y1_r1\\/Y2 dep_Y1_r2 appos_r2_i1 nn_j2_j1 dep_j2_i2 dep_i1_j2 dep_r2_Y1\\/Y2 appos_Y1\\/Y2_j1 nn_k2_k1 dep_k2_j2 dep_j1_k2 nn_Y1_X1\\/X2 dep_Y1\\/Y2_Y1 dep_Y1_r1\\/Y2 dep_r1\\/Y2_r2 appos_r2_i1 nn_k2_k1 dep_k2_i2 dep_i1_k2 dep_Y1_3 amod_Y1_X1\\/X2 dep_r2_Y1 dep_Y1_r1\\/Y2 dep_r1\\/Y2_r2 appos_r2_i1 nn_k2_j1 dep_k2_j2 dep_i1_k2 appos_r2_Y1\\/Y2 appos_Y1\\/Y2_j1 nn_j2_k1 dep_j2_i2 dep_j1_j2 nn_Y1_X1\\/X2 dep_Y1\\/Y2_Y1 dep_Y1_r1\\/Y2 dep_r1\\/Y2_r2 appos_Y1_i1 nn_k2_k1 dep_k2_i2 dep_i1_k2 mark_contains_Since det_rule_each amod_rule_inference nsubj_contains_rule advcl_get_contains num_variables_six amod_variables_free dobj_contains_variables prep_variables_over nn_positions_string pobj_over_positions nn_k1_i1 dep_k1_j1 appos_positions_k1 nn_k2_i2 dep_k2_j2 dep_k1_k2 nsubj_get_we rcmod_Y1_get det_complexity_a amod_complexity_parsing dobj_get_complexity prep_complexity_of nn_O_order pobj_of_O appos_O_n6 prep_O_for amod_grammars_unlexicalized pobj_for_grammars advmod_number_where nsubj_number_n cop_number_is det_number_the dep_X1\\/X2_number prep_number_of pobj_of_words prep_words_in dep_longer_the dep_in_longer prep_number_of det_strings_the num_strings_two pobj_of_strings prep_strings_from nn_L1_language pobj_from_L1 cc_L1_and conj_L1_L2 ' +o,3922,'It ewduatos the pairwise agreement mnong a set ; of coders making categoryiudgment , correcting tbr expected chance agreement ',Carletta,'nsubj_ewduatos_It det_agreement_the amod_agreement_pairwise dobj_ewduatos_agreement partmod_agreement_mnong det_set_a dobj_mnong_set prep_agreement_of pobj_of_coders partmod_coders_making dobj_making_categoryiudgment xcomp_making_correcting amod_agreement_tbr amod_agreement_expected nn_agreement_chance dobj_correcting_agreement ' +o,3923,'Consequently , the mainstream research in the literature has been focused on the modeling and utilization of local and sentential contexts , either linguistically in a rule-based framework or statistically in a searching and optimization set-up ',Wu,'advmod_focused_Consequently det_research_the nn_research_mainstream nsubjpass_focused_research prep_research_in det_literature_the pobj_in_literature aux_focused_has auxpass_focused_been prep_focused_on det_modeling_the pobj_on_modeling cc_modeling_and conj_modeling_utilization prep_modeling_of amod_contexts_local cc_local_and conj_local_sentential pobj_of_contexts dep_on_either conj_on_linguistically conj_on_in det_framework_a amod_framework_rule-based pobj_in_framework cc_framework_or conj_framework_statistically prep_framework_in det_set-up_a nn_set-up_searching cc_searching_and conj_searching_optimization pobj_in_set-up ' +o,3924,'2 Background The natural language generator used in our experiments is the WSJ-trained system described in and Hogan et al ',Cahill,'num_Background_2 det_generator_The amod_generator_natural nn_generator_language nsubj_system_generator partmod_generator_used prep_used_in poss_experiments_our pobj_in_experiments cop_system_is det_system_the amod_system_WSJ-trained dep_Background_system partmod_system_described prep_described_in pobj_in_ cc__and conj__Hogan cc_Hogan_et conj_Hogan_al ' +p,3925,'SVM has been shown to be useful for text classification tasks , and has previously given good performance in sentiment classification experiments ',Pang,'nsubjpass_shown_SVM aux_shown_has auxpass_shown_been aux_useful_to cop_useful_be xcomp_shown_useful mark__for nn_tasks_text nn_tasks_classification nsubj__tasks advcl_useful_ cc__and aux_given_has advmod_given_previously conj__given amod_performance_good dobj_given_performance prep_given_in nn_experiments_sentiment nn_experiments_classification pobj_in_experiments ' +o,3926,'Nevertheless , the generated rules are strictly required to be derived from the contiguous translational equivalences ',Zhang,'advmod_required_Nevertheless det_rules_the amod_rules_generated nsubjpass_required_rules auxpass_required_are advmod_required_strictly aux_derived_to auxpass_derived_be xcomp_required_derived prep_derived_from det_equivalences_the amod_equivalences_contiguous dep_contiguous_translational pobj_from_equivalences ' +o,3927,'Previous approaches to the problem have all been learning-based ; the primary difference between the present algorithm and earlier ones is that it is not learned , but explicitly incorporates principles of GovernmentBinding theory , since that theory underlies the annotation ',Collins,'amod_approaches_Previous nsubj_learning-based_approaches prep_approaches_to det__the nn__problem pobj_to_ aux_learning-based_have advmod_learning-based_all dep_learning-based_been det_difference_the amod_difference_primary nsubj_is_difference prep_difference_between det_algorithm_the amod_algorithm_present pobj_between_algorithm cc_algorithm_and amod_ones_earlier conj_algorithm_ones parataxis_learning-based_is complm_learned_that nsubjpass_learned_it auxpass_learned_is neg_learned_not ccomp_is_learned dep_incorporates_but advmod_incorporates_explicitly dep_learned_incorporates dobj_incorporates_principles prep_principles_of nn__GovernmentBinding nn__theory pobj_of_ mark_underlies_since det_theory_that nsubj_underlies_theory advcl_learned_underlies det_annotation_the dobj_underlies_annotation ' +o,3928,'This does not seem to be the case , however , for common feature weighting functions , such as Point-wise Mutual Information ',Church,'nsubj_seem_This aux_seem_does neg_seem_not aux_case_to cop_case_be det_case_the xcomp_seem_case advmod_case_however prep_case_for amod_functions_common nn_functions_feature nn_functions_weighting pobj_for_functions dep_as_such prep_case_as nn_Information_Point-wise nn_Information_Mutual pobj_as_Information ' +o,3929,'Many statistical translation models try to model word-toword correspondences between source and target words ',Brown,'amod__Many amod__statistical nn__translation nn__models nsubj_try_ aux_model_to xcomp_try_model amod_correspondences_word-toword dobj_model_correspondences prep_correspondences_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,3930,' approached chucking by using Transformation Based Learning -LRB- TBL -RRB- ',Ramshaw,'nsubj_approached_ xcomp_approached_chucking prep_chucking_by pcomp_by_using nn_Learning_Transformation nn_Learning_Based dobj_using_Learning abbrev_Learning_TBL ' +o,3931,'We use only the words that are content words -LRB- nouns , verbs , or adjectives -RRB- and not in the stopword list used in ROUGE ',Lin,'nsubj_use_We advmod_use_only det_words_the dobj_use_words nsubj_words_that cop_words_are nn_words_content rcmod_words_words dep_words_nouns conj_nouns_verbs cc_nouns_or conj_nouns_adjectives cc_words_and conj_words_not prep_use_in det_list_the amod_list_stopword pobj_in_list partmod_list_used prep_used_in pobj_in_ROUGE ' +o,3932,'However , certain properties of the BLEU metric can be exploited to speed up search , as described in detail by ',Och,'advmod_exploited_However amod_properties_certain nsubjpass_exploited_properties prep_properties_of det_metric_the amod_metric_BLEU pobj_of_metric aux_exploited_can auxpass_exploited_be aux_speed_to purpcl_exploited_speed prt_speed_up dobj_speed_search mark_described_as advcl_exploited_described prep_described_in pobj_in_detail prep_described_by ' +o,3933,'To solve the problem , apply an automatic generation grammar transformation to their training data : they automatically label CFG nodes with additional case information and the model now learns the new improved generation rules of Tables 4 and 5 ',Cahill,'aux_solve_To dep_apply_solve det_problem_the dobj_solve_problem nsubj_apply_ det_transformation_an amod_transformation_automatic nn_transformation_generation nn_transformation_grammar dobj_apply_transformation prep_apply_to poss_data_their nn_data_training pobj_to_data nsubj_label_they advmod_label_automatically parataxis_apply_label nn_nodes_CFG dobj_label_nodes prep_label_with amod_information_additional nn_information_case pobj_with_information cc_apply_and det_model_the nsubj_learns_model advmod_learns_now conj_apply_learns det_rules_the amod_rules_new amod_rules_improved nn_rules_generation dobj_learns_rules prep_rules_of pobj_of_Tables num_Tables_4 cc_Tables_and conj_Tables_5 ' +p,3934,'Also , in a , state-ofthe-art English parser only the words tha , t occur more tha , n d times in training data ',Collins,'advmod_occur_Also prep_occur_in pobj_in_a amod_parser_state-ofthe-art amod_parser_English appos_a_parser partmod_parser_ prep__only det_tha_the nn_tha_words dep_only_tha nsubj_occur_t amod_tha_more dobj_occur_tha nn_times_n nn_times_d appos_tha_times prep_times_in nn_data_training pobj_in_data ' +o,3935,'This obviously does not preclude using the audio-based system together with other features such as utterance position , length , speakers roles , and most others used in the literature ',Penn,'nsubj_preclude_This advmod_preclude_obviously aux_preclude_does neg_preclude_not xcomp_preclude_using det_system_the amod_system_audio-based dobj_using_system advmod_using_together prep_using_with amod_features_other pobj_with_features dep_as_such prep_features_as amod_position_utterance pobj_as_position appos_position_length nn_roles_speakers appos_length_roles cc_preclude_and amod_others_most nsubj_used_others conj_preclude_used prep_used_in det_literature_the pobj_in_literature ' +p,3936,'23 The Averaged Perceptron Reranking Model Averaged perceptron has been successfully applied to several tagging and parsing reranking tasks , and in this paper , we employed it in reranking semantic parses generated by the base semantic parser SCISSOR ',Collins,'det__The nn__Averaged nn__Perceptron nn__Reranking nn__Model nn__Averaged nn__perceptron nsubjpass_applied_ aux_applied_has auxpass_applied_been advmod_applied_successfully dep_23_applied prep_applied_to amod_tagging_several pobj_to_tagging cc_tagging_and nn__parsing nn__reranking nn__tasks conj_tagging_ cc_applied_and prep_generated_in det_paper_this pobj_in_paper nsubj_generated_we partmod_we_employed dobj_employed_it prep_employed_in pcomp_in_reranking amod_parses_semantic dobj_reranking_parses conj_applied_generated prep_generated_by det_SCISSOR_the amod_SCISSOR_base amod_SCISSOR_semantic nn_SCISSOR_parser pobj_by_SCISSOR ' +o,3937,'We used the implementation of MaxEnt classifier described in ',Manning,'nsubj_used_We det_implementation_the nsubj_described_implementation prep_implementation_of nn_classifier_MaxEnt pobj_of_classifier ccomp_used_described prt_described_in ' +o,3938,'Averaged Perceptron Algorithm 5 Experiments We evaluate our method on both Chinese and English syntactic parsing task with the standard division on Chinese Penn Treebank Version 50 and WSJ English Treebank 30 as shown in Table 1 ',Marcus,'nn_Experiments_Averaged nn_Experiments_Perceptron nn_Experiments_Algorithm num_Experiments_5 nsubj_evaluate_We dep_Experiments_evaluate poss_method_our dobj_evaluate_method prep_evaluate_on det_task_both amod_task_Chinese cc_Chinese_and conj_Chinese_English amod_task_syntactic nn_task_parsing pobj_on_task prep_task_with det_division_the amod_division_standard pobj_with_division prep_division_on amod_Version_Chinese nn_Version_Penn nn_Version_Treebank pobj_on_Version num_Version_50 cc_task_and nn_Treebank_WSJ nn_Treebank_English conj_task_Treebank num__30 dep_as_ advmod_shown_as partmod_Treebank_shown prep_shown_in pobj_in_Table num_Table_1 ' +p,3939,'One popular and statistically appealing such measure is Log-Likelihood -LRB- LL -RRB- ',Dunning,'measure_popular_One amod_measure_popular cc_popular_and advmod_appealing_statistically conj_popular_appealing amod_measure_such nsubj_Log-Likelihood_measure cop_Log-Likelihood_is dep_Log-Likelihood_LL ' +o,3940,'1 Introduction In this paper , we study the use of so-called word trigger pairs -LRB- for short : word triggers -RRB- to improve an existing language model , which is typically a trigram model in combination with a cache component ',Brown,'num_Introduction_1 prep_study_In det_paper_this pobj_In_paper nsubj_study_we dep_Introduction_study det_use_the dobj_study_use prep_use_of amod_pairs_so-called nn_pairs_word nn_pairs_trigger pobj_of_pairs dep_pairs_for pobj_for_short dep_short_word rcmod_word_triggers partmod_pairs_ aux_improve_to xcomp__improve det_model_an amod_model_existing nn_model_language dobj_improve_model nsubj_model_which cop_model_is advmod_model_typically det_model_a nn_model_trigram rcmod_model_model prep_model_in pobj_in_combination prep_combination_with det__a nn__cache nn__component pobj_with_ ' +o,3941,'The latter approach has become increasingly popular ',Marcus,'det_approach_The amod_approach_latter nsubj__approach aux__has cop__become advmod_popular_increasingly amod__popular ' +o,3942,'The trigger-based lexicon model used in this work follows the training procedure introduced in and is integrated directly in the decoder instead of being applied in n-best list reranking ',Hasan,'det_model_The amod_model_trigger-based nn_model_lexicon nsubjpass_follows_model partmod_model_used prep_used_in det_work_this pobj_in_work det_procedure_the nn_procedure_training dobj_follows_procedure partmod_procedure_introduced prep_introduced_in pobj_in_ cc_follows_and auxpass_integrated_is conj_follows_integrated advmod_integrated_directly prep_integrated_in det_decoder_the pobj_in_decoder dep_of_instead prep_decoder_of auxpass_applied_being pcomp_of_applied prep_applied_in amod_reranking_n-best nn_reranking_list pobj_in_reranking ' +o,3943,'The problem itself has started to get attention only recently ',Blitzer,'det_problem_The nsubj_started_itself aux_started_has dep_problem_started aux_get_to xcomp_started_get dobj_get_attention advmod_recently_only advmod_get_recently ' +o,3944,'alpha 0 0.1 0.2 0.3 0.4 0.5 Freq=2 13555 13093 12235 11061 10803 10458 Freq=3 4203 3953 3616 3118 2753 2384 Freq=4 1952 1839 1649 1350 1166 960 Freq=5 1091 1019 917 743 608 511 Freq>2 2869 2699 2488 2070 1666 1307 TOTAL 23670 22603 20905 18342 16996 15620 alpha 0.6 0.7 0.8 0.9 1.0 Freq=2 10011 9631 9596 9554 9031 Freq=3 2088 1858 1730 1685 1678 Freq=4 766 617 524 485 468 Freq=5 392 276 232 202 189 Freq>2 1000 796 627 517 439 TOTAL 14257 13178 12709 12443 11805 Table 7: Number of extracted MWUs by frequency 6.2 Qualitative Analysis As many authors assess (Frank Smadja, 1993; John Justeson and Slava Katz, 1995), deciding whether a sequence of words is a multiword unit or not is a tricky problem.',Dunning,'' +o,3945,'A sinfilar approach has been chosen by ',Brown,'det_approach_A nn_approach_sinfilar nsubjpass_chosen_approach aux_chosen_has auxpass_chosen_been prep_chosen_by pobj_by_ ' +o,3946,'5 Discussion As stated above , we aim to build an unsupervised generative model for named entity clustering , since such a model could be integrated with unsupervised coreference models like for joint inference ',Haghighi,'num_Discussion_5 mark_stated_As advcl_aim_stated advmod_stated_above nsubj_aim_we dep_Discussion_aim aux_build_to xcomp_aim_build det_model_an amod_model_unsupervised nn_model_generative dobj_build_model prep_build_for amod_entity_named pobj_for_entity amod_entity_clustering mark_integrated_since predet_model_such det_model_a nsubjpass_integrated_model aux_integrated_could auxpass_integrated_be dep_entity_integrated prep_integrated_with amod_models_unsupervised nn_models_coreference pobj_with_models prep_integrated_like pobj_like_ prep__for amod_inference_joint pobj_for_inference ' +o,3947,'However , there is little agreement on what types of knowledge are helpful : Some suggestions concentrate on lexical information , eg , by the integration of word similarity information as in Meteor or MaxSim ',Banerjee,'advmod_is_However expl_is_there amod_agreement_little nsubj_is_agreement rel_helpful_on pobj_on_what nsubj_helpful_types prep_types_of pobj_of_knowledge cop_helpful_are rcmod_agreement_helpful det_suggestions_Some nsubj_concentrate_suggestions parataxis_is_concentrate prep_concentrate_on amod_information_lexical pobj_on_information appos_information_eg prep_concentrate_by det_integration_the pobj_by_integration prep_integration_of nn_information_word nn_information_similarity pobj_of_information dep_in_as prep_concentrate_in nn__Meteor pobj_in_ cc__or nn__MaxSim conj__ ' +o,3948,'We have explained elsewhere how suitable features can be defined in terms of the a18 word , pos-tag a20 pairs in the context , and how maximum entropy techniques can be used to estimate the probabilities , following ',Ratnaparkhi,'nsubj_explained_We aux_explained_have advmod__elsewhere ccomp_explained_ advmod_defined_how amod_features_suitable nsubjpass_defined_features aux_defined_can auxpass_defined_be dep__defined prep_defined_in pobj_in_terms prep_terms_of det_word_the amod_word_a18 pobj_of_word amod_pairs_pos-tag amod_pairs_a20 appos_word_pairs prep_pairs_in det_context_the pobj_in_context cc_defined_and advmod_used_how amod_techniques_maximum nn_techniques_entropy nsubjpass_used_techniques aux_used_can auxpass_used_be conj_defined_used aux_estimate_to purpcl_used_estimate det_probabilities_the dobj_estimate_probabilities dep__following ' +o,3949,'This statistic is given by -2 log A = 2 -LRB- log L -LRB- p1 , kl , hi -RRB- log L -LRB- p2 , k2 , n2 -RRB- - log L -LRB- p , kl , R1 -RRB- -- log L -LRB- p , k2 , n2 -RRB- -RRB- , where log LCo , k , n -RRB- = k logp + -LRB- n k -RRB- log -LRB- 1 - p -RRB- , and Pl = ~ , P2 = ~ , P = , ~ \' , ~ ; -LRB- For a detailed description of the statistic used , see -RRB- ',Dunning,'det_statistic_This nsubjpass_given_statistic auxpass_given_is prep_given_by pobj_by_-2 rcmod_-2_log nsubj_2_A dep_2_= ccomp_log_2 parataxis_log_log nsubj_log_L nn_hi_p1 dep_hi_kl appos_L_hi ccomp_log_log dobj_log_L nn_n2_p2 dep_n2_k2 appos_L_n2 nn_L_log dep_L_L nn_R1_p dep_R1_kl appos_L_R1 nn_L_log dep_L_L nn_n2_p dep_n2_k2 appos_L_n2 advmod_=_where nn_LCo_log nsubj_=_LCo dep_n_k dep_LCo_n dep_given_= nn_logp_k dobj_=_logp prep_logp_+ nn_k_n dep_+_k pobj_+_log appos_log_1 dep_1_p cc_log_and nn_~_Pl nn_~_= conj_log_~ nn_~_P2 nn_~_= conj_log_~ nn_=_P conj_log_= conj_log_~ conj_log_~ dep_=_For det_description_a amod_description_detailed pobj_For_description prep_description_of det_statistic_the pobj_of_statistic partmod_statistic_used ccomp_used_see acomp_see_ ' +o,3950,'High-performance taggers typically also include joint three-tag counts in some way , either as tag trigrams or tag-triple features ',Ratnaparkhi,'dep_taggers_High-performance advmod_taggers_typically advmod_taggers_also dep_taggers_include amod_counts_joint amod_counts_three-tag dobj_include_counts prep_include_in det_way_some pobj_in_way dep_in_either conj_in_as nn_trigrams_tag pobj_as_trigrams nsubj_taggers_ cc__or amod_features_tag-triple conj__features ' +o,3951,'Initial estimates of lexical translation probabilities came from the IBM Model 4 translation tables produced by GIZA + + ',Brown,'amod_estimates_Initial nsubj_came_estimates prep_estimates_of amod_probabilities_lexical nn_probabilities_translation pobj_of_probabilities prep_came_from det_tables_the nn_tables_IBM nn_tables_Model num_tables_4 nn_tables_translation pobj_from_tables partmod_tables_produced prep_produced_by nn_+_GIZA nn_+_+ pobj_by_+ ' +o,3952,'1 Introduction In the field of statistical parsing , various probabilistic evaluation models have been proposed where different models use different feature types ',Collins,'num_Introduction_1 dep_In_Introduction prep_proposed_In det_field_the pobj_In_field prep_field_of amod_parsing_statistical pobj_of_parsing amod_models_various amod_models_probabilistic nn_models_evaluation nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been dep__proposed advmod_use_where amod_models_different nsubj_use_models advcl_proposed_use amod_types_different nn_types_feature dobj_use_types nn__ nn__ nn__ nn__ nn__ nn__ nn__ nn__ nsubj__ ' +o,3953,'For the evaluation of translation quality , we applied standard automatic evaluation metrics , ie , BLEU and METEOR ',Papineni,'prep_applied_For det_evaluation_the pobj_For_evaluation prep_evaluation_of nn_quality_translation pobj_of_quality nsubj_applied_we amod_metrics_standard amod_metrics_automatic nn_metrics_evaluation dobj_applied_metrics conj_metrics_ie nn__BLEU conj_metrics_ cc_metrics_and nn__METEOR conj_metrics_ ' +o,3954,'Since we need knowledge-poor Daille , 1996 -RRB- induction , we can not use human-suggested filtering Chi-squared -LRB- G24 -RRB- 2 Z-Score Students t-Score n-gram list in accordance to each probabilistic algorithm ',Smadja,'mark_need_Since nsubj_need_we advcl_use_need amod_induction_knowledge-poor dep_knowledge-poor_Daille appos_Daille_1996 dobj_need_induction nsubj_use_we aux_use_can neg_use_not amod_Chi-squared_human-suggested amod_Chi-squared_filtering dobj_use_Chi-squared appos_Chi-squared_G24 number__2 num_Students_ amod_Students_Z-Score nn_Students_ nsubj__Students advmod__t-Score rcmod_Chi-squared_ amod_list_n-gram dobj__list prep__in pobj_in_accordance prep__to det_algorithm_each amod_algorithm_probabilistic pobj_to_algorithm ' +o,3955,'As we remarked earlier , however , the input data required by our method -LRB- triples -RRB- could be generated automatically from unparsed corpora making use of existing heuristic rules , although for the experiments we report here we used a parsed corpus ',Smadja,'mark_remarked_As nsubj_remarked_we advcl_generated_remarked advmod_remarked_earlier advmod_generated_however det_data_the nn_data_input nsubjpass_generated_data partmod_data_required prep_required_by poss_method_our pobj_by_method appos_data_triples aux_generated_could auxpass_generated_be advmod_generated_automatically prep_generated_from amod_corpora_unparsed pobj_from_corpora purpcl_generated_making dobj_making_use prep_use_of amod__existing amod__heuristic nn__rules pobj_of_ mark_report_although prep_report_for det_experiments_the pobj_for_experiments nsubj_report_we advcl_generated_report advmod_report_here nsubj_used_we dep_generated_used det_corpus_a amod_corpus_parsed dobj_used_corpus ' +o,3956,'Second , in keeping with ontological promiscuity , we represent the importance of attributes by the salience of events and states in the discourse model -- these states and events now have the same status in the discourse model as any other entities ',Hobbs,'advmod_represent_Second prep_represent_in pcomp_in_keeping prep_keeping_with amod__ontological nn__promiscuity pobj_with_ nsubj_represent_we det_importance_the dobj_represent_importance prep_importance_of pobj_of_attributes prep_represent_by det_salience_the pobj_by_salience prep_salience_of pobj_of_events cc_events_and conj_events_states prep_events_in det_model_the nn_model_discourse pobj_in_model det_states_these nsubj_have_states cc_states_and conj_states_events advmod_have_now parataxis_represent_have det_status_the amod_status_same dobj_have_status prep_status_in det_model_the nn_model_discourse pobj_in_model prep_model_as det_entities_any amod_entities_other pobj_as_entities ' +o,3957,'It has been claimed that content analysis researchers usually regard a -RRB- 8 to demonstrate good reliability and 67 -LRB- ~ -LRB- 8 alf16 lows tentative conclusions to be drawn -LRB- see -RRB- ',Carletta,'nsubjpass_claimed_It aux_claimed_has auxpass_claimed_been complm_regard_that nn_researchers_content nn_researchers_analysis nsubj_regard_researchers advmod_regard_usually ccomp_claimed_regard dep_8_a dobj_regard_8 prep_8_to amod_reliability_demonstrate amod_reliability_good pobj_to_reliability cc_8_and conj_8_67 dep_67_~ num_lows_8 amod_lows_alf16 dep_~_lows amod_conclusions_tentative dep_lows_conclusions aux_drawn_to auxpass_drawn_be xcomp_regard_drawn parataxis_drawn_see acomp_see_ ' +o,3958,'In recent years , many researchers have employed statistical models or association measures to build alignment links ',Wu,'prep_employed_In amod_years_recent pobj_In_years amod_researchers_many nsubj_employed_researchers aux_employed_have amod_models_statistical dobj_employed_models nn_measures_ cc__or conj__association nsubj__measures rcmod_models_ aux_build_to xcomp__build amod_links_alignment dobj_build_links ' +o,3959,'Breidt alsopointedouta coupleof problemsthatmakes extractionfor Germanmoredifficultthanfor English : the stronginflectionfor verbs , the variable word-order , andthepositionalambiguityoftheargumentsSheshowsthatevendistinguishingsubjectsfromobjectsisverydifficultwithoutparsing ',Dunning,'nn_English_Breidt nn_English_ nn_English_alsopointedouta nn_English_coupleof nn_English_problemsthatmakes nn_English_extractionfor nn_English_Germanmoredifficultthanfor det_verbs_the nn_verbs_stronginflectionfor nsubj_andthepositionalambiguityoftheargumentsSheshowsthatevendistinguishingsubjectsfromobjectsisverydifficultwithoutparsing_verbs det_word-order_the amod_word-order_variable appos_verbs_word-order dep_English_andthepositionalambiguityoftheargumentsSheshowsthatevendistinguishingsubjectsfromobjectsisverydifficultwithoutparsing ' +o,3960,'TB TBR JJ , JJR , JJS JJ RB , RBR , RBS RB CD , LS CD CC CC DT , WDT , PDT DT FW FW MD , VB , VBD , VBG , VBN , VBP , VBZ , VH , VHD , VHG , VHN , VHP , VHZ MD NN , NNS , NP , NPS NN PP , WP , PP$ , WP$ , EX , WRB PP IN , TO IN POS PO RP RP SYM SY UH UH VV , VVD , VVG , VVN , VVP , VVZ VB ',Marcus,'nn_JJ_TB nn_JJ_TBR appos_JJ_JJR nn_RB_JJS nn_RB_JJ appos_JJ_RB appos_JJ_RBR nn_CD_RBS nn_CD_RB appos_JJ_CD nn_DT_LS nn_DT_CD nn_DT_CC nn_DT_CC appos_JJ_DT appos_JJ_WDT nn_MD_PDT nn_MD_DT nn_MD_FW nn_MD_FW appos_JJ_MD nn_NN_VB appos_NN_VBD appos_NN_VBG appos_NN_VBN appos_NN_VBP appos_NN_VBZ appos_NN_VH appos_NN_VHD appos_NN_VHG appos_NN_VHN appos_NN_VHP appos_NN_VHZ nn_NN_MD appos_JJ_NN appos_JJ_NNS appos_JJ_NP nn_PP_NPS nn_PP_NN appos_JJ_PP nn_IN_WP appos_IN_PP$ appos_IN_WP$ appos_IN_EX appos_IN_WRB nn_IN_PP appos_JJ_IN appos_JJ_TO prep_TO_IN dep_VV_POS nn_VV_PO nn_VV_RP nn_VV_RP nn_VV_SYM nn_VV_SY nn_VV_UH amod_VV_UH pobj_IN_VV nn_VB_VVD appos_VB_VVG appos_VB_VVN appos_VB_VVP appos_VB_VVZ appos_JJ_VB ' +o,3961,'The applications range from simple classification tasks such as text classification and history-based tagging to more complex structured prediction tasks such as partof-speech -LRB- POS -RRB- tagging , syntactic parsing and semantic role labeling ',Ratnaparkhi,'det_applications_The nsubj_range_applications prep_range_from amod_tasks_simple nn_tasks_classification pobj_from_tasks dep_as_such prep_tasks_as nn_classification_text pobj_as_classification cc_classification_and amod__history-based nn__tagging conj_classification_ prep_range_to advmod_complex_more amod_tasks_complex amod_tasks_structured nn_tasks_prediction pobj_to_tasks dep_as_such prep_tasks_as amod__partof-speech appos__POS amod__tagging nn__ amod__syntactic dep_syntactic_parsing pobj_as_ cc_tasks_and amod__semantic nn__role nn__labeling conj_tasks_ ' +p,3962,'Our MT experiments use a re-implementation of Moses called Phrasal , which provides an easier API for adding features ',Koehn,'poss_experiments_Our nn_experiments_MT nsubj_use_experiments det_re-implementation_a nsubj_called_re-implementation prep_re-implementation_of nn__Moses pobj_of_ ccomp_use_called dobj_called_Phrasal nsubj_provides_which rcmod_Phrasal_provides det_API_an amod_API_easier dobj_provides_API prep_provides_for pcomp_for_adding dobj_adding_features ' +o,3963,'22 Automatic metrics Similarly to the Pyramid method , ROUGE and Basic Elements require multiple topics and model summaries to produce optimal results ',Lin,'num_metrics_22 nn_metrics_Automatic nsubj_require_metrics dep_to_Similarly prep_metrics_to det_method_the nn_method_Pyramid pobj_to_method nn__ROUGE conj_method_ cc_method_and nn__Basic nn__Elements conj_method_ amod_topics_multiple nsubj_produce_topics cc_topics_and nn_summaries_model conj_topics_summaries aux_produce_to xcomp_require_produce amod_results_optimal dobj_produce_results ' +o,3964,'Previous work on building hybrid systems includes , among others , approaches using reranking , regeneration with an SMT decoder , and confusion networks ',Rosti,'amod_work_Previous nsubj_includes_work prep_work_on pcomp_on_building amod_systems_hybrid dobj_building_systems dep_includes_among pobj_among_others dobj_includes_approaches partmod_approaches_using dobj_using_reranking conj_reranking_regeneration prep_regeneration_with det__an nn__SMT nn__decoder pobj_with_ cc_reranking_and nn_networks_confusion conj_reranking_networks ' +o,3965,'The probabilities from these back-off levels are interpolated using the techniques in ',Collins,'det_probabilities_The nsubjpass_interpolated_probabilities prep_probabilities_from det_levels_these amod_levels_back-off pobj_from_levels auxpass_interpolated_are xcomp_interpolated_using det_techniques_the dobj_using_techniques prep_using_in ' +p,3966,'Recent several years have witnessed the rapid development of system combination methods based on confusion networks -LRB- eg , -RRB- , which show state-of-theart performance in MT benchmarks ',Rosti,'amod_years_Recent amod_years_several nsubj_witnessed_years aux_witnessed_have det_development_the amod_development_rapid dobj_witnessed_development prep_development_of nn_methods_system nn_methods_combination pobj_of_methods prep_witnessed_based dep_based_on nn_networks_confusion pobj_on_networks appos_networks_eg dep_eg_ nsubj_show_which rcmod_networks_show amod_performance_state-of-theart dobj_show_performance prep_performance_in nn_benchmarks_MT pobj_in_benchmarks ' +o,3967,'The idea of word class gives a general solution to this problem ',Brown,'det_idea_The nsubj_gives_idea prep_idea_of nn__word nn__class pobj_of_ det_solution_a amod_solution_general dobj_gives_solution prep_gives_to det_problem_this pobj_to_problem ' +o,3968,' and Nivre and can be seen as methods to combine separately defined models ',McDonald,'nsubjpass_seen_ cc__and conj__Nivre cc_Nivre_and conj_Nivre_ aux_seen_can auxpass_seen_be prep_seen_as pobj_as_methods aux_combine_to purpcl_seen_combine advmod_defined_separately amod_models_defined dobj_combine_models ' +o,3969,'To be able identify that adjacent blocks -LRB- eg , the development and and progress -RRB- can be merged into larger blocks , our model infers binary -LRB- non-linguistic -RRB- trees reminiscent of ',Wu,'aux_able_To cop_able_be dep_infers_able ccomp_able_identify complm_merged_that amod_blocks_adjacent nsubjpass_merged_blocks dep_blocks_eg det_development_the conj_eg_development cc_eg_and nn_progress_and conj_eg_progress aux_merged_can auxpass_merged_be ccomp_identify_merged prep_merged_into amod_blocks_larger pobj_into_blocks poss_model_our nsubj_infers_model amod_trees_binary dep_trees_non-linguistic nsubj_reminiscent_trees xcomp_infers_reminiscent prep_reminiscent_of ' +o,3970,'To train models , we used projectivized versions of the training dependency trees2 1We are grateful to the providers of the treebanks that constituted the data for the shared task ',Marcus,'aux_train_To dep_used_train dobj_train_models nsubj_used_we amod_versions_projectivized nsubj_grateful_versions prep_versions_of det_1We_the nn_1We_training nn_1We_dependency nn_1We_trees2 pobj_of_1We cop_grateful_are ccomp_used_grateful prep_grateful_to det_providers_the pobj_to_providers prep_providers_of det_treebanks_the pobj_of_treebanks nsubj_constituted_that rcmod_treebanks_constituted det_data_the dobj_constituted_data prep_data_for det_task_the amod_task_shared pobj_for_task ' +o,3971,'Some approaches have used syntax at the core while others have integrated syntax into existing phrase-based frameworks ',Wu,'det_approaches_Some nsubj_used_approaches aux_used_have dobj_used_syntax prep_syntax_at det__the amod__core pobj_at_ mark_integrated_while nsubj_integrated_others aux_integrated_have advcl_used_integrated dobj_integrated_syntax prep_integrated_into amod__existing amod__phrase-based nn__frameworks pobj_into_ ' +o,3972,'Based on IBM Model 1 lexical parameters , providing a complementary probability for each tuple in the translation table ',Brown,'prep_Based_on nn__IBM nn__Model num__1 nn__lexical nn__parameters pobj_on_ xcomp_Based_providing det_probability_a amod_probability_complementary dobj_providing_probability prep_providing_for det_tuple_each pobj_for_tuple prep_tuple_in det_table_the nn_table_translation pobj_in_table ' +n,3973,'However , work in that direction has so far addressed only parse reranking ',Collins,'advmod_addressed_However nsubj_addressed_work prep_work_in det_direction_that pobj_in_direction aux_addressed_has advmod_far_so advmod_addressed_far advmod_reranking_only amod_reranking_parse dobj_addressed_reranking ' +o,3974,'In addition , a number of approaches have focused on developing discriminative approaches for unsupervised and semi-supervised tagging ',Haghighi,'prep_focused_In pobj_In_addition det_number_a nsubj_focused_number prep_number_of pobj_of_approaches aux_focused_have prep_focused_on pcomp_on_developing amod_approaches_discriminative dobj_developing_approaches prep_developing_for amod_tagging_unsupervised cc_unsupervised_and conj_unsupervised_semi-supervised pobj_for_tagging ' +o,3975,'However for remedy , many of the current word alignment methods combine the results of both alignment directions , via intersection or 249 grow-diag-final heuristic , to improve the alignment reliability ',Koehn,'mark_combine_for nn_methods_remedy amod_methods_many prep_many_of det_word_the amod_word_current pobj_of_word nn_methods_alignment nsubj_combine_methods dep_However_combine det_results_the nsubj_improve_results prep_results_of det_directions_both amod_directions_alignment pobj_of_directions prep_results_via pobj_via_intersection cc_intersection_or num_heuristic_249 amod_heuristic_grow-diag-final conj_intersection_heuristic aux_improve_to xcomp_combine_improve det_reliability_the amod_reliability_alignment dobj_improve_reliability ' +o,3976,'Correspondences between MALTUS and other tagsets were also provided ',Manning,'nsubjpass_provided_Correspondences prep_Correspondences_between pobj_between_MALTUS cc_MALTUS_and amod__other nn__tagsets conj_MALTUS_ auxpass_provided_were advmod_provided_also dep_provided_ ' +o,3977,'The extraction procedure consists of three steps : First , the bracketing of the trees in the Penn Treebank is corrected and extended based on the approaches of Magerman and ',Collins,'det_procedure_The nn_procedure_extraction nsubj_consists_procedure prep_consists_of num_steps_three pobj_of_steps advmod_corrected_First det_bracketing_the nsubjpass_corrected_bracketing prep_bracketing_of det_trees_the pobj_of_trees prep_trees_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank auxpass_corrected_is parataxis_consists_corrected cc_corrected_and conj_corrected_extended prep_corrected_based dep_based_on det_approaches_the pobj_on_approaches prep_approaches_of nn_and_Magerman nn_and_ pobj_of_and ' +n,3978,'The combination is significantly better than at a very high level , but more importantly , Shens results -LRB- currently representing the replicable state-of-the-art in POS tagging -RRB- have been significantly surpassed also by the semisupervised Morce -LRB- at the 99 \% confidence level -RRB- ',Shen,'det_combination_The nsubj_results_combination cop_results_is dep_importantly_significantly dep_than_better quantmod__than dep_significantly_ prep__at det_level_a advmod_high_very amod_level_high pobj_at_level cc_importantly_but advmod_importantly_more advmod_results_importantly nn_results_Shens advmod_representing_currently dep_results_representing det_state-of-the-art_the amod_state-of-the-art_replicable dobj_representing_state-of-the-art prep_state-of-the-art_in dep_tagging_POS pobj_in_tagging aux_surpassed_have auxpass_surpassed_been advmod_surpassed_significantly rcmod_results_surpassed advmod_surpassed_also prep_surpassed_by det_Morce_the amod_Morce_semisupervised pobj_by_Morce dep_Morce_at det_level_the dep_\%_99 amod_level_\% nn_level_confidence pobj_at_level ' +o,3979,'1 Introduction Word alignment , which can be defined as an object for indicating the corresponding words in a parallel text , was first introduced as an intermediate result of statistical translation models ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubjpass_introduced_alignment nsubjpass_defined_which aux_defined_can auxpass_defined_be rcmod_alignment_defined prep_defined_as det_object_an pobj_as_object prep_defined_for pcomp_for_indicating det_words_the amod_words_corresponding dobj_indicating_words prep_indicating_in det_text_a amod_text_parallel pobj_in_text auxpass_introduced_was advmod_introduced_first prep_introduced_as det_result_an amod_result_intermediate pobj_as_result prep_result_of amod_models_statistical nn_models_translation pobj_of_models ' +o,3980,'1 Introduction Most recent approaches in SMT , eg , use a log-linear model to combine probabilistic features ',Koehn,'num_approaches_1 nn_approaches_Introduction nn_approaches_Most amod_approaches_recent nsubj_use_approaches prep_approaches_in pobj_in_SMT nn__eg appos_approaches_ det_model_a amod_model_log-linear dobj_use_model aux_combine_to infmod_model_combine amod_features_probabilistic dobj_combine_features ' +o,3981,'Thispaperfocusesontheframeworkintroduced in Figure 2 for two reasons : -LRB- a -RRB- cautious al50 gorithms were shown to perform best for several NLP problems -LRB- including acquisition of IE patterns -RRB- , and -LRB- b -RRB- it has nice theoretical properties : showed that , regardless of the selection procedure , sequential bootstrapping algorithms converge to a local minimum of K , where K is an upper bound of the negative log likelihood of the data ',Abney,'prep_Thispaperfocusesontheframeworkintroduced_in pobj_in_Figure num_Figure_2 prep_Thispaperfocusesontheframeworkintroduced_for num_reasons_two pobj_for_reasons dep_-RRB-_a amod_gorithms_cautious nn_gorithms_al50 nsubjpass_shown_gorithms auxpass_shown_were dep_reasons_shown aux_perform_to xcomp_shown_perform dobj_perform_best prep_best_for amod_problems_several nn_problems_NLP pobj_for_problems dep_problems_including pobj_including_acquisition prep_acquisition_of nn_patterns_IE pobj_of_patterns cc_shown_and dep_-RRB-_b nsubj_has_it conj_shown_has amod_properties_nice amod_properties_theoretical dobj_has_properties nsubj_showed_ parataxis_shown_showed complm_converge_that advmod_converge_regardless dep_regardless_of det_procedure_the nn_procedure_selection pobj_of_procedure amod_algorithms_sequential nn_algorithms_bootstrapping nsubj_converge_algorithms ccomp_showed_converge prep_converge_to det_minimum_a amod_minimum_local pobj_to_minimum prep_minimum_of pobj_of_K advmod_bound_where nsubj_bound_K cop_bound_is det_bound_an amod_bound_upper advcl_converge_bound prep_bound_of det_likelihood_the amod_likelihood_negative nn_likelihood_log pobj_of_likelihood prep_likelihood_of det_data_the pobj_of_data ' +o,3982,'The information content of this set is defined as mutual information I -LRB- F -LRB- w -RRB- -RRB- ',Church,'det_content_The nn_content_information prep_content_of det_set_this pobj_of_set auxpass_defined_is rcmod_content_defined prep_defined_as amod_information_mutual pobj_as_information dep_content_I dep_I_F appos_I_w ' +o,3983,'The MT systems of learn to generate text in the target language straight from the source language , without the aid of an explicit semantic representation ',Berger,'det_systems_The nn_systems_MT nsubj_learn_systems prep_systems_of pobj_of_ aux_generate_to xcomp_learn_generate dobj_generate_text prep_text_in det_language_the nn_language_target pobj_in_language dep_from_straight prep_generate_from det_language_the nn_language_source pobj_from_language prep_generate_without det_aid_the pobj_without_aid prep_aid_of det_representation_an amod_representation_explicit amod_representation_semantic pobj_of_representation ' +o,3984,'Many of the current approaches of domain modeling collapse together different instances and make the decision on what information is important for a domain based on this generalized corpus ',Barzilay,'nsubj_modeling_Many prep_Many_of det_approaches_the amod_approaches_current pobj_of_approaches prep_approaches_of pobj_of_domain dep_modeling_collapse prt_collapse_together amod_instances_different dobj_collapse_instances cc_collapse_and conj_collapse_make det_decision_the dobj_make_decision rel_important_on pobj_on_what nsubj_important_information cop_important_is rcmod_decision_important prep_important_for det_domain_a pobj_for_domain partmod_domain_based prep_based_on det_corpus_this amod_corpus_generalized pobj_on_corpus ' +o,3985,'The benefits of using grammatical information for automatic WSD were first explored by and Resnik -LRB- 1996 -RRB- in unsupervised approaches to disambiguating single words in context ',Yarowsky,'det_benefits_The nsubjpass_explored_benefits prep_benefits_of pcomp_of_using amod_information_grammatical dobj_using_information prep_using_for amod_WSD_automatic pobj_for_WSD auxpass_explored_were advmod_explored_first prep_explored_by pobj_by_ cc__and conj__Resnik appos__1996 prep__in amod_approaches_unsupervised pobj_in_approaches aux_disambiguating_to xcomp_explored_disambiguating amod_words_single dobj_disambiguating_words prep_disambiguating_in pobj_in_context ' +o,3986,'determining document orientation -LRB- or polarity -RRB- , as in deciding if a given Subjective text expresses a Positive or a Negative opinion on its subject matter ; 3 ',Pang,'nn_orientation_document dobj_determining_orientation dep_polarity_or dep_orientation_polarity dep_in_as prep_determining_in pcomp_in_deciding mark_expresses_if det_text_a amod_text_given nn_text_Subjective nsubj_expresses_text advcl_deciding_expresses dep_Positive_a dobj_expresses_Positive cc_Positive_or det_opinion_a amod_opinion_Negative conj_Positive_opinion prep_opinion_on poss__its nn__subject nn__matter pobj_on_ dep_Positive_3 ' +o,3987,'A variety of algorithms -LRB- eg , bootstrapping , co-training , alternating structure optimization , etc -RRB- ',Yarowsky,'det_variety_A prep_variety_of pobj_of_algorithms dep_algorithms_eg amod__bootstrapping appos_eg_ amod__co-training appos_eg_ amod__alternating nn__structure nn__optimization appos_eg_ dep_eg_etc ' +o,3988,'Since Chinese text is not orthographically separated into words , the standard methodology is to first preproce ~ input texts through a segmentation module ',Brown,'mark_separated_Since amod_text_Chinese nsubj_separated_text cop_separated_is neg_separated_not advmod_separated_orthographically advcl_is_separated prep_separated_into pobj_into_words det_methodology_the amod_methodology_standard nsubj_is_methodology aux_preproce_to advmod_preproce_first xcomp_is_preproce amod_texts_~ nn_texts_input dobj_preproce_texts prep_preproce_through det__a nn__segmentation nn__module pobj_through_ ' +o,3989,'Several authors have used mutual information and similar statistics as an objective function for word clustering , for automatic determination of phonemic baseforms , and for language modeling for speech recognition ',Dagan,'amod_authors_Several nsubj_used_authors aux_used_have amod_information_mutual dobj_used_information cc_information_and amod_statistics_similar conj_information_statistics prep_information_as det_function_an amod_function_objective pobj_as_function prep_used_for nn__word nn__clustering pobj_for_ prep_used_for amod_determination_automatic pobj_for_determination prep_determination_of amod__phonemic nn__baseforms pobj_of_ cc_for_and conj_for_for nn_modeling_language pobj_for_modeling prep_modeling_for nn__speech nn__recognition pobj_for_ ' +o,3990,'Uses Maximum Entropy classification , trained on JNLPBA -LRB- NER -RRB- ',Berger,'nn_classification_Maximum nn_classification_Entropy nn_classification_ dep_Uses_classification partmod_classification_trained prep_trained_on nn__JNLPBA pobj_on_ abbrev__NER ' +o,3991,'Word-aligned corpora have been found to be an excellent source for translation-related knowledge , not only for phrase-based models , but also for syntax-based models -LRB- eg , -RRB- ',Koehn,'amod_corpora_Word-aligned nsubjpass_found_corpora aux_found_have auxpass_found_been aux_source_to cop_source_be det_source_an amod_source_excellent xcomp_found_source prep_source_for amod_knowledge_translation-related pobj_for_knowledge cc_for_not dep_not_only prep_source_for amod__phrase-based nn__models pobj_for_ dep_also_but cc_for_also conj_for_for amod_models_syntax-based pobj_for_models appos_models_eg dep_eg_ ' +o,3992,'One judge annotated allarticles in four datasets of the Wall Street Journal Treebank corpus -LRB- W9-4 , W9-10 , W9-22 , and W933 , each approximately 160K words -RRB- as well as thecorpusofWall Street Journal articles used in -LRB- called WSJ-SE below -RRB- ',Marcus,'num_allarticles_One nn_allarticles_judge amod_allarticles_annotated prep_allarticles_in num_datasets_four pobj_in_datasets prep_datasets_of det__the nn__Wall nn__Street nn__Journal nn__Treebank nn__corpus pobj_of_ nn_W933_W9-4 conj_W933_W9-10 conj_W933_W9-22 cc_W933_and dep__W933 det_words_each advmod_words_approximately num_words_160K appos_W933_words dep_well_as cc_datasets_well dep_well_as amod_articles_thecorpusofWall nn_articles_Street nn_articles_Journal conj_datasets_articles partmod_articles_used prep_used_in pobj_in_ dep_allarticles_called dobj_called_WSJ-SE advmod_called_below ' +o,3993,'We set all feature weights by optimizing Bleu directly using minimum error rate training -LRB- MERT -RRB- on the tuning part of the development set ',Papineni,'nsubj_set_We advmod_set_all dep_set_feature dobj_feature_weights prep_feature_by pcomp_by_optimizing nn__Bleu dobj_optimizing_ advmod_using_directly partmod__using amod__minimum nn__error nn__rate nn__training abbrev__MERT dobj_using_ prep__on det_part_the amod_part_tuning pobj_on_part prep_part_of det__the nn__development nn__set pobj_of_ ' +o,3994,'We use the Penn Treebank Wall Street Journal corpus as the large corpus and individual sections of the Brown corpus as the target corpora ',Marcus,'nsubj_use_We det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_Wall nn_corpus_Street nn_corpus_Journal dobj_use_corpus prep_corpus_as det_corpus_the amod_corpus_large pobj_as_corpus cc_corpus_and amod_sections_individual conj_corpus_sections prep_sections_of det_corpus_the nn_corpus_Brown pobj_of_corpus prep_use_as det_corpora_the nn_corpora_target pobj_as_corpora ' +o,3995,'The most common answer is component testing , where the component is compared against a standard of goodness , usually the Penn Treebank for English , allowing a numerical score of precision and recall ',Marcus,'det_answer_The advmod_answer_most amod_answer_common nsubj_testing_answer cop_testing_is nn_testing_component advmod_is_where det_component_the nsubj_is_component rcmod_testing_is prep_is_compared dep_compared_against det_standard_a nsubj_allowing_standard prep_standard_of pobj_of_goodness advmod_standard_usually det_Treebank_the nn_Treebank_Penn dep_standard_Treebank prep_Treebank_for nn__English pobj_for_ pcomp_against_allowing det_score_a amod_score_numerical dobj_allowing_score prep_score_of nn__precision cc_precision_and conj_precision_recall pobj_of_ ' +o,3996,'In the present work , the approach taken by is used to derive such values for selected phrases in the text ',Turney,'prep_used_In det_work_the amod_work_present pobj_In_work det_approach_the nsubjpass_used_approach partmod_approach_taken prep_taken_by pobj_by_ auxpass_used_is aux_derive_to xcomp_used_derive amod_values_such dobj_derive_values prep_derive_for amod_phrases_selected pobj_for_phrases prep_phrases_in det_text_the pobj_in_text ' +o,3997,'Mutual infornaation involves a problem in that it is overestimated for low-frequency terms -LRB- I -RRB- unning 1993 -RRB- ',Dunning,'amod_infornaation_Mutual nsubj_involves_infornaation det_problem_a dobj_involves_problem rel_overestimated_in pobj_in_that nsubjpass_overestimated_it auxpass_overestimated_is rcmod_problem_overestimated prep_overestimated_for amod_terms_low-frequency pobj_for_terms dep_terms_I xcomp_overestimated_unning dobj_unning_1993 ' +o,3998,' uses the mutual information clustering algorithm described in ',Brown,'nsubj_uses_ det_algorithm_the amod_algorithm_mutual nn_algorithm_information amod_algorithm_clustering nsubj_described_algorithm ccomp_uses_described prt_described_in ' +o,3999,'The second attempts to instill knowledge of collocations in the data ; we use the technique described by to compute multi-word expressions and then mark words that are commonly used as such with a feature that expresses this fact ',Dunning,'det_second_The nsubj_attempts_second aux_instill_to xcomp_attempts_instill dobj_instill_knowledge prep_knowledge_of pobj_of_collocations prep_collocations_in det_data_the pobj_in_data nsubj_use_we parataxis_attempts_use det_technique_the dobj_use_technique partmod_technique_described prep_described_by pobj_by_ aux_compute_to xcomp_described_compute amod_expressions_multi-word dobj_compute_expressions cc_compute_and advmod_mark_then conj_compute_mark dobj_mark_words nsubjpass_used_that auxpass_used_are advmod_used_commonly rcmod_words_used dep_with_as dep_with_such prep_used_with det_feature_a pobj_with_feature nsubj_expresses_that rcmod_technique_expresses det_fact_this dobj_expresses_fact ' +o,4000,'Several sentiment information retrieval models were proposed in the framework of probabilistic language models by ',Eguchi,'amod_models_Several nn_models_sentiment nn_models_information nn_models_retrieval nsubjpass_proposed_models auxpass_proposed_were prep_proposed_in det_framework_the pobj_in_framework prep_framework_of amod_models_probabilistic nn_models_language pobj_of_models prep_proposed_by ' +o,4001,'Word alignments were produced by GIZA + + with a standard training regimen of five iterations of Model 1 , five iterations of the HMM Model , and five iterations of Model 4 , in both directions ',Och,'amod_alignments_Word nsubjpass_produced_alignments auxpass_produced_were prep_produced_by pobj_by_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_produced_with det_regimen_a amod_regimen_standard nn_regimen_training pobj_with_regimen prep_regimen_of num_iterations_five pobj_of_iterations prep_iterations_of pobj_of_Model num_Model_1 num_iterations_five conj_iterations_iterations prep_iterations_of det_Model_the nn_Model_HMM pobj_of_Model cc_iterations_and num_iterations_five conj_iterations_iterations prep_iterations_of pobj_of_Model num_Model_4 prep_iterations_in det_directions_both pobj_in_directions ' +o,4002,'32 Details To learn alignments , translation probabilities , etc in the first method we used work that has been done in statistical machine translation , where the translation process is considered to be equivalent to a corruption of the source language text to the target language text due to a noisy channel ',Brown,'num_Details_32 aux_learn_To infmod_Details_learn dep_etc_alignments nn_probabilities_translation appos_alignments_probabilities dobj_learn_etc prep_learn_in det_method_the amod_method_first pobj_in_method nsubj_used_we dep_Details_used dobj_used_work nsubjpass_done_that aux_done_has auxpass_done_been rcmod_work_done prep_done_in amod__statistical nn__machine nn__translation pobj_in_ advmod_considered_where det_process_the nn_process_translation nsubjpass_considered_process auxpass_considered_is rcmod__considered aux_equivalent_to cop_equivalent_be xcomp_considered_equivalent prep_equivalent_to det_corruption_a pobj_to_corruption prep_corruption_of det_text_the nn_text_source nn_text_language pobj_of_text prep_equivalent_to det_text_the nn_text_target nn_text_language pobj_to_text amod_text_due prep_due_to det_channel_a amod_channel_noisy pobj_to_channel ' +o,4003,'We apply the log likelihood principle to compute this score ',Dunning,'nsubj_apply_We det_principle_the nn_principle_log nn_principle_likelihood nsubj__principle ccomp_apply_ aux_compute_to xcomp__compute det_score_this dobj_compute_score ' +o,4004,'For automatic evaluation , we employed BLEU by following ',Papineni,'prep_employed_For amod_evaluation_automatic pobj_For_evaluation nsubj_employed_we number__BLEU dobj_employed_ prep_employed_by amod__following pobj_by_ ' +o,4005,'The MBT 180 Tagger Type Standard Trigram MBT Rule-based Maximum-Entropy Full Second-Order HMM SNOW Voting Constraints Full Second-Order HMM Known Unknown Overall Open\\/Closed Lexicon ? ',Marcus,'det_MBT_The number_180_ num_Overall_180 nn_Overall_Tagger nn_Overall_Type nn_Overall_Standard nn_Overall_Trigram nn_Overall_ nn_Overall_MBT nn_Overall_ nn_Overall_Rule-based nn_Overall_ nn_Overall_Maximum-Entropy nn_Overall_ nn_Overall_Full nn_Overall_Second-Order nn_Overall_HMM nn_Overall_SNOW nn_Overall_ nn_Overall_Voting nn_Overall_Constraints nn_Overall_ nn_Overall_Full nn_Overall_Second-Order nn_Overall_HMM nn_Overall_Known nn_Overall_Unknown dep_MBT_Overall nn_Lexicon_Open\\/Closed dep_Overall_Lexicon ' +o,4006,'1 Introduction The Inversion Transduction Grammar or ITG formalism , which historically was developed in the context of translation and alignment , hypothesizes strong expressiveness restrictions that constrain paraphrases to vary word order only in certain allowable nested permutations of arguments ',Wu,'num_Grammar_1 nn_Grammar_Introduction nn_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction nsubj_hypothesizes_Grammar cc_Grammar_or nn_formalism_ITG conj_Grammar_formalism nsubjpass_developed_which advmod_developed_historically auxpass_developed_was rcmod_Grammar_developed prep_developed_in det_context_the pobj_in_context prep_context_of pobj_of_translation cc_translation_and conj_translation_alignment amod_restrictions_strong nn_restrictions_expressiveness dobj_hypothesizes_restrictions complm_paraphrases_that nsubj_paraphrases_constrain ccomp_hypothesizes_paraphrases aux_vary_to xcomp_paraphrases_vary nn_order_word dobj_vary_order advmod_vary_only prep_vary_in amod_permutations_certain amod_permutations_allowable amod_permutations_nested pobj_in_permutations prep_permutations_of pobj_of_arguments ' +o,4007,'Table 6 contrasts our results with those from ',Collins,'nsubj_contrasts_Table num_Table_6 poss_results_our dobj_contrasts_results prep_contrasts_with pobj_with_those prep_contrasts_from ' +o,4008,'Because of this , it is generally accepted that some kind of postprocessing should be performed to improve the final result , by shortening , fusing , or otherwise revising the material ',Jing,'prep_accepted_Because dep_Because_of pobj_of_this nsubjpass_accepted_it auxpass_accepted_is advmod_accepted_generally complm_performed_that det_kind_some nsubjpass_performed_kind prep_kind_of pobj_of_postprocessing aux_performed_should auxpass_performed_be ccomp_accepted_performed aux_improve_to purpcl_performed_improve det_result_the amod_result_final dobj_improve_result prep_performed_by pcomp_by_shortening conj_shortening_fusing cc_shortening_or advmod_revising_otherwise conj_shortening_revising det_material_the dobj_revising_material ' +o,4009,'The other approach selected was ',Yarowsky,'det_approach_The amod_approach_other partmod_approach_selected dep_approach_was ' +p,4010,'The averaged 1555 perceptron has a solid theoretical fundamental and was proved to be effective across a variety of NLP tasks ',Collins,'det_perceptron_The number_1555_averaged num_perceptron_1555 nsubjpass_has_perceptron det_fundamental_a amod_fundamental_solid amod_fundamental_theoretical dobj_has_fundamental cc_has_and auxpass_proved_was conj_has_proved aux_effective_to cop_effective_be xcomp_proved_effective prep_effective_across det_variety_a pobj_across_variety prep_variety_of nn_tasks_NLP pobj_of_tasks ' +p,4011,'23 Classifier Training We chose maximum entropy as our primary classifier , since it had been successfully applied by the highest performing systems in both the SemEval-2007 preposition sense disambiguation task and the general word sense disambiguation task ',Berger,'num_Training_23 nn_Training_Classifier nsubj_chose_We dep_Training_chose amod__maximum nn__entropy dobj_chose_ prep_chose_as poss_classifier_our amod_classifier_primary pobj_as_classifier mark_applied_since nsubjpass_applied_it aux_applied_had auxpass_applied_been advmod_applied_successfully advcl_chose_applied prep_applied_by det_systems_the amod_systems_highest amod_systems_performing pobj_by_systems prep_systems_in preconj__both det__the amod__SemEval-2007 nn__preposition nn__sense nn__disambiguation nn__task pobj_in_ cc__and det__the amod__general nn__word nn__sense nn__disambiguation nn__task conj__ ' +p,4012,'It could be shown that such methods , of which BLEU is the most common , can deliver evaluation results that show a high agreement with human judgments ',Papineni,'nsubjpass_shown_It aux_shown_could auxpass_shown_be complm_deliver_that amod_methods_such nsubj_deliver_methods rel_common_of pobj_of_which nn__BLEU nsubj_common_ cop_common_is det_common_the advmod_common_most rcmod_methods_common aux_deliver_can ccomp_shown_deliver nn_results_evaluation dobj_deliver_results nsubj_show_that rcmod_results_show det_agreement_a amod_agreement_high dobj_show_agreement prep_show_with amod_judgments_human pobj_with_judgments ' +o,4013,'The tagger from first annotates sentences of raw text with a sequence of partof-speech tags ',Ratnaparkhi,'det_tagger_The prep_tagger_from amod_annotates_ amod_annotates_first pobj_from_annotates dep_tagger_sentences prep_sentences_of amod_text_raw pobj_of_text prep_text_with det_sequence_a pobj_with_sequence prep_sequence_of amod_tags_partof-speech pobj_of_tags ' +o,4014,' used both supervised and unsupervised WSD for correct phonetizitation of words in speech synthesis ',Yarowsky,'nsubj_used_ det_WSD_both amod_WSD_supervised cc_supervised_and conj_supervised_unsupervised dobj_used_WSD prep_used_for amod_phonetizitation_correct pobj_for_phonetizitation prep_phonetizitation_of pobj_of_words prep_words_in nn_synthesis_speech pobj_in_synthesis ' +o,4015,'The former is a task of identifying positive and negative sentiments from a text which can be a passage , a sentence , a phrase and even a word ',Pang,'det_former_The nsubj_task_former cop_task_is det_task_a prep_task_of pcomp_of_identifying amod_sentiments_positive cc_positive_and conj_positive_negative dobj_identifying_sentiments prep_identifying_from det_text_a pobj_from_text nsubj_passage_which aux_passage_can cop_passage_be det_passage_a rcmod_text_passage det_sentence_a conj_passage_sentence det_phrase_a conj_passage_phrase cc_passage_and advmod_word_even det_word_a conj_passage_word ' +o,4016,'2 Statistical Word Alignment According to the IBM models , the statistical word alignment model can be generally represented as in Equation -LRB- 1 -RRB- ',Dunning,'num_Alignment_2 nn_Alignment_Statistical nn_Alignment_Word prep_represented_According dep_According_to det__the nn__IBM nn__models pobj_to_ det_model_the amod_model_statistical nn_model_word nn_model_alignment nsubjpass_represented_model aux_represented_can auxpass_represented_be advmod_represented_generally dep_Alignment_represented prep_represented_as dep_as_in pobj_in_Equation appos_Equation_1 ' +o,4017,'Further work will look at how to integrate probabilities such as p -LRB- clv , r -RRB- into a model of dependency structure , similar to that of and Collins -LRB- 1997 -RRB- , which can be used -LRB- ` or parse selection ',Collins,'amod_work_Further nsubj_look_work aux_look_will prep_look_at advmod_integrate_how aux_integrate_to pcomp_at_integrate dobj_integrate_probabilities dep_as_such prep_probabilities_as pobj_as_p dep_p_clv appos_clv_r prep_integrate_into det_model_a pobj_into_model prep_model_of amod_structure_dependency pobj_of_structure amod_model_similar prep_similar_to pobj_to_that prep_that_of pobj_of_ cc__and conj__Collins appos__1997 nsubjpass_used_which aux_used_can auxpass_used_be rcmod_model_used cc_used_or conj_used_parse dobj_parse_selection ' +o,4018,'If POS denotes the POS of the English word , we can define the word-to-word distance measure -LRB- Equation 4 -RRB- as POS POS -LRB- 15 -RRB- Ratnaparkhis POS tagger was used to obtain POS tags for each word in the English sentence ',Ratnaparkhi,'mark_denotes_If nsubj_denotes_POS advcl_define_denotes det_POS_the dobj_denotes_POS prep_POS_of det_word_the amod_word_English pobj_of_word nsubj_define_we aux_define_can det_measure_the amod_measure_word-to-word nn_measure_distance dobj_define_measure appos_measure_Equation num_Equation_4 prep_measure_as dep_POS_POS pobj_as_POS appos_POS_15 nn__Ratnaparkhis dep__POS nn__tagger nsubjpass_used_ auxpass_used_was dep_define_used aux_obtain_to xcomp_used_obtain dep_tags_POS dobj_obtain_tags prep_obtain_for det_word_each pobj_for_word prep_word_in det_sentence_the amod_sentence_English pobj_in_sentence ' +o,4019,'43 Baseline We use a standard log-linear phrase-based statistical machine translation system as a baseline : GIZA + + implementation of IBM word alignment model 4 ,8 the refinement and phrase-extraction heuristics described in , minimum-error-rate training 7More specifically , we choose the first English reference from the 7 references and the Chinese sentence to construct new sentence pairs ',Koehn,'num_Baseline_43 nsubj_use_We dep_Baseline_use det_system_a amod_system_standard amod_system_log-linear amod_system_phrase-based amod_system_statistical nn_system_machine nn_system_translation dobj_use_system prep_system_as det_baseline_a pobj_as_baseline amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dep_system_implementation prep_implementation_of nn_model_IBM nn_model_word nn_model_alignment pobj_of_model number__4 num_,8_ prep_model_,8 det_refinement_the dep_,8_refinement cc_refinement_and amod_heuristics_phrase-extraction conj_refinement_heuristics partmod_heuristics_described prep_described_in amod_7More_ amod_7More_minimum-error-rate nn_7More_training pobj_in_7More advmod_choose_specifically nsubj_choose_we dep_Baseline_choose det_reference_the amod_reference_first amod_reference_English dobj_choose_reference prep_choose_from det_references_the num_references_7 pobj_from_references cc_references_and det_sentence_the amod_sentence_Chinese conj_references_sentence aux_construct_to xcomp_choose_construct amod_pairs_new nn_pairs_sentence dobj_construct_pairs ' +o,4020,'Surprisingly , though , rather little work has been devoted to learning local syntactic patterns , mostly noun phrases ',Ramshaw,'advmod_devoted_Surprisingly advmod_devoted_though advmod_work_rather amod_work_little nsubjpass_devoted_work aux_devoted_has auxpass_devoted_been prep_devoted_to pcomp_to_learning amod_patterns_local amod_patterns_syntactic dobj_learning_patterns advmod_patterns_mostly nn_phrases_noun dep_patterns_phrases ' +o,4021,'1 Introduction Finite-state parsing -LRB- also called chunking or shallow parsing -RRB- has typically been motivated as a fast firstpass for or approximation to more expensive context-free parsing ',Ramshaw,'num_parsing_1 nn_parsing_Introduction nn_parsing_Finite-state nsubjpass_motivated_parsing advmod_called_also dep_parsing_called amod_parsing_chunking cc_chunking_or conj_chunking_shallow dobj_called_parsing aux_motivated_has advmod_motivated_typically auxpass_motivated_been prep_motivated_as det_firstpass_a amod_firstpass_fast pobj_as_firstpass prep_motivated_for cc_motivated_or conj_motivated_approximation prep_approximation_to advmod_expensive_more amod_parsing_expensive amod_parsing_context-free pobj_to_parsing ' +o,4022,'This approach gave an improvement of 27 in BLEU score on the IWSLT05 Japanese to English evaluation corpus -LRB- improving the score from 524 to 551 -RRB- ',Papineni,'det_approach_This nsubj_gave_approach det_improvement_an dobj_gave_improvement prep_improvement_of pobj_of_27 prep_gave_in amod_score_BLEU amod_score_ pobj_in_score prep_score_on det_Japanese_the nn_Japanese_IWSLT05 pobj_on_Japanese prep_gave_to amod_corpus_English nn_corpus_evaluation pobj_to_corpus parataxis_gave_improving det_score_the dobj_improving_score prep_improving_from number_551_524 dep_551_to pobj_from_551 ' +o,4023,'For instance , the Penn Treebank policy is to annotate the lowest node that is unfinished with an - UNF tag as in Figure 4 -LRB- a -RRB- ',Marcus,'prep_is_For pobj_For_instance det__the nn__Penn nn__Treebank nn__policy nsubj_is_ aux_annotate_to xcomp_is_annotate det_node_the amod_node_lowest dobj_annotate_node nsubjpass_unfinished_that auxpass_unfinished_is rcmod_node_unfinished prep_unfinished_with det_tag_an nn_tag_UNF pobj_with_tag prep_unfinished_as dep_as_in pobj_in_Figure num_Figure_4 dep_-LRB-_a ' +o,4024,'This definition is similar to that of minimal translation units as described in , although they allow null words on either side ',Quirk,'det_definition_This nsubj_similar_definition cop_similar_is prep_similar_to pobj_to_that prep_that_of amod_units_minimal nn_units_translation pobj_of_units mark_described_as advcl_similar_described prep_described_in pobj_in_ mark_allow_although nsubj_allow_they advcl_described_allow amod_words_null dobj_allow_words prep_allow_on det_side_either pobj_on_side ' +o,4025,'122 SPECIFIC SYNTACTIC AND SEMANTIC ASSUMPTIONS The basic scheme , or some not too distant relative , is the one used in many large-scale implemented systems ; as examples , we can quote TEAM , PUNDIT , TACITUS , MODL , CLE , and SNACK-85 ',Hobbs,'nn_SYNTACTIC_122 nn_SYNTACTIC_SPECIFIC nsubj_one_SYNTACTIC cc_SYNTACTIC_AND nn_ASSUMPTIONS_SEMANTIC conj_SYNTACTIC_ASSUMPTIONS det_scheme_The amod_scheme_basic dep_SYNTACTIC_scheme cc_SYNTACTIC_or det_relative_some amod_relative_not advmod_not_too amod_relative_distant conj_SYNTACTIC_relative cop_one_is det_one_the partmod_one_used prep_used_in amod_systems_many amod_systems_large-scale nn_systems_implemented pobj_in_systems prep_quote_as pobj_as_examples nsubj_quote_we aux_quote_can parataxis_one_quote nn__TEAM dobj_quote_ nn__PUNDIT conj__ nn__TACITUS conj__ nn__MODL conj__ nn__CLE conj__ cc__and nn__SNACK-85 conj__ ' +n,4026,'a time-consuming process ',Marcus,'det_process_a amod_process_time-consuming ' +o,4027,'Of particular relevance is other work on parsing the Penn WSJ Treebank ',Collins,'prep_is_Of amod_relevance_particular pobj_Of_relevance amod_work_other nsubj_is_work prep_work_on pcomp_on_parsing det_Treebank_the nn_Treebank_Penn nn_Treebank_WSJ dobj_parsing_Treebank ' +o,4028,'In practice , we used MMR in our experiments , since the original MEAD considers also sentence positions 3 , which can always been added later as in ',Penn,'prep_used_In pobj_In_practice nsubj_used_we dobj_used_MMR prep_used_in poss_experiments_our pobj_in_experiments mark_considers_since det_MEAD_the amod_MEAD_original nsubj_considers_MEAD advcl_used_considers advmod_considers_also nn_positions_sentence dobj_considers_positions dep_positions_3 nsubj_can_which rcmod_positions_can advmod_added_always auxpass_added_been ccomp_can_added advmod_as_later advmod_added_as prep_added_in ' +o,4029,'The loglinear model weights are learned using Chiangs implementation of the maximum BLEU training algorithm , both for the baseline , and the WSD-augmented system ',Och,'det_weights_The nn_weights_loglinear nn_weights_model nsubjpass_learned_weights auxpass_learned_are xcomp_learned_using nn_implementation_Chiangs dobj_using_implementation prep_implementation_of det__the amod__maximum amod__BLEU nn__training nn__algorithm pobj_of_ conj_implementation_both prep_both_for det_baseline_the pobj_for_baseline cc_implementation_and det_system_the amod_system_WSD-augmented conj_implementation_system ' +p,4030,'The most widely used are Word Error Rate -LRB- WER -RRB- , Position Independent Word Error Rate -LRB- PER -RRB- , the BLEU score and the NIST score ',Papineni,'dep_used_The advmod_used_most advmod_used_widely nsubj_Rate_used cop_Rate_are nn_Rate_Word nn_Rate_Error abbrev_Rate_WER nn_Rate_Position nn_Rate_Independent nn_Rate_Word nn_Rate_Error appos_Rate_Rate abbrev_Rate_PER det__the amod__BLEU nn__score conj_Rate_ cc_Rate_and det__the amod__NIST nn__score conj_Rate_ ' +o,4031,'On the other hand , structural annotation such as that used in syntactic treebanks assigns a syntactic category to a contiguous sequence of corpus positions ',Marcus,'prep_assigns_On det_hand_the amod_hand_other pobj_On_hand amod_annotation_structural nsubj_assigns_annotation dep_as_such prep_annotation_as pobj_as_that partmod_that_used prep_used_in amod__syntactic nn__treebanks pobj_in_ det_category_a amod_category_syntactic dobj_assigns_category prep_assigns_to det_sequence_a amod_sequence_contiguous pobj_to_sequence prep_sequence_of nn_positions_corpus pobj_of_positions ' +o,4032,'Evaluations are typically carried out on newspaper texts , ie on section 23 of the Penn Treebank -LRB- PTB -RRB- ',Marcus,'nsubjpass_carried_Evaluations auxpass_carried_are advmod_carried_typically prt_carried_out prep_carried_on nn_texts_newspaper pobj_on_texts appos_texts_ie prep_carried_on pobj_on_section num_section_23 prep_section_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank abbrev_Treebank_PTB ' +o,4033,'We use the finite-state parses of FaSTU $ for recognizing these entities , but the method extends to any basic phrasal parser 4 ',Marcus,'nsubj_use_We det_parses_the amod_parses_finite-state dobj_use_parses prep_parses_of pobj_of_FaSTU dep_for_$ num_$_ prep_use_for pcomp_for_recognizing det_entities_these dobj_recognizing_entities cc_use_but det_method_the nsubj_extends_method conj_use_extends prep_extends_to det_parser_any amod_parser_basic nn_parser_phrasal pobj_to_parser num_parser_4 ' +o,4034,'The last line shows the results of -LRB- recognizing NP \'s -RRB- with the same train\\/test data ',Ramshaw,'det_line_The amod_line_last nsubj_shows_line det_results_the dobj_shows_results prep_results_of pobj_of_ parataxis_shows_recognizing dobj_recognizing_NP possessive_NP_\'s prep_shows_with det_data_the amod_data_same amod_data_train\\/test pobj_with_data ' +o,4035,'Prior to running the parsers , we trained the POS tagger described in ',Collins,'dep_to_Prior advcl_trained_to pcomp_to_running det_parsers_the dobj_running_parsers nsubj_trained_we det_tagger_the dep_tagger_POS nsubj_described_tagger ccomp_trained_described prt_described_in ' +n,4036,'Other statistical machine translation systems such as and also produce a tree a15 given a sentence a16 Their models are based on mechanisms that generate two languages at the same time , so an English tree a15 is obtained as a subproduct of parsing a16 However , their use of the LM is not mathematically motivated , since their models do not decompose into Pa4a5a2a9a8a3a10a6 and a12a14a4a5a3a7a6 unlike the noisy channel model ',Wu,'amod_systems_Other amod_systems_statistical nn_systems_machine nn_systems_translation nsubj_produce_systems dep_as_such prep_systems_as pobj_as_ cc__and conj__ advmod_produce_also det_a15_a nn_a15_tree nsubjpass_based_a15 partmod_a15_given det_a16_a nn_a16_sentence iobj_given_a16 poss_models_Their dobj_given_models auxpass_based_are ccomp_produce_based prep_based_on pobj_on_mechanisms nsubj_generate_that rcmod_mechanisms_generate num_languages_two dobj_generate_languages prep_languages_at det_time_the amod_time_same pobj_at_time dep_produce_so det_a15_an amod_a15_English nn_a15_tree nsubjpass_obtained_a15 auxpass_obtained_is dep_produce_obtained prep_obtained_as det_subproduct_a pobj_as_subproduct prep_subproduct_of amod_a16_parsing pobj_of_a16 advmod_motivated_However poss_use_their nsubjpass_motivated_use prep_use_of det_LM_the pobj_of_LM auxpass_motivated_is neg_motivated_not advmod_motivated_mathematically dep_produce_motivated mark_decompose_since poss_models_their nsubj_decompose_models aux_decompose_do neg_decompose_not advcl_motivated_decompose prep_decompose_into pobj_into_Pa4a5a2a9a8a3a10a6 cc_Pa4a5a2a9a8a3a10a6_and conj_Pa4a5a2a9a8a3a10a6_a12a14a4a5a3a7a6 prep_decompose_unlike det_model_the amod_model_noisy nn_model_channel pobj_unlike_model ' +o,4037,'9 The definition of BLEU used in this training was the original IBM definition , which defines the effective reference length as the reference length that is closest to the test sentence length ',Papineni,'det_definition_The nsubj__definition prep_definition_of pobj_of_BLEU partmod_BLEU_used prep_used_in det_training_this pobj_in_training cop__was det__the amod__original nn__IBM nn__definition dep_9_ nsubj_defines_which rcmod__defines det_length_the amod_length_effective nn_length_reference dobj_defines_length prep_defines_as det_length_the nn_length_reference pobj_as_length nsubj_closest_that cop_closest_is rcmod_length_closest prep_closest_to det_length_the nn_length_test nn_length_sentence pobj_to_length ' +o,4038,'287 System Train + base Test + base 1 Baseline 8789 8789 2 Contrastive 8870 082 8845 056 -LRB- 5 trials\\/fold -RRB- 3 Contrastive 8882 093 8855 066 -LRB- greedy selection -RRB- Table 1 : Average F1 of 7-way cross-validation To generate the alignments , we used Model 4 , as implemented in GIZA + + ',Brown,'nn_Train_System nsubj_+_Train ccomp_287_+ amod_base_base nn_base_Test amod_base_+ iobj_+_base num_Baseline_1 dobj_+_Baseline number_8789_8789 num_Baseline_8789 num_Contrastive_2 dobj_+_Contrastive number_082_8870 num_Contrastive_082 number_056_8845 dobj_+_056 num_trials\\/fold_5 appos_056_trials\\/fold num_Contrastive_3 dobj_+_Contrastive number_093_8882 num_Contrastive_093 number_066_8855 dobj_+_066 amod_selection_greedy appos_066_selection prep_+_Table pobj_Table_1 amod_F1_Average dobj_+_F1 prep_F1_of amod_cross-validation_7-way pobj_of_cross-validation aux_generate_To dep_287_generate det_alignments_the dobj_generate_alignments nsubj_used_we parataxis_287_used nn__Model num__4 dobj_used_ mark_implemented_as dep_287_implemented prep_implemented_in pobj_in_GIZA amod_GIZA_+ cc_+_+ conj_+_ ' +p,4039,' has proved to be a simple yet powerful observation and has been successfully used in word sense disambiguation -LRB- WSD -RRB- and related tasks -LRB- eg , ; Agirre and Rigau The author was partially funded by GALE DARPA Contract No ',Yarowsky,'nsubjpass_proved_ aux_proved_has aux_observation_to cop_observation_be det_observation_a amod_observation_simple cc_simple_yet conj_simple_powerful xcomp_proved_observation cc_proved_and aux_used_has auxpass_used_been advmod_used_successfully conj_proved_used prep_used_in nn_disambiguation_word nn_disambiguation_sense pobj_in_disambiguation abbrev_disambiguation_WSD cc_disambiguation_and amod_tasks_related conj_disambiguation_tasks dep_disambiguation_eg dep_eg_ nsubjpass_funded_Agirre cc_Agirre_and conj_Agirre_Rigau det_author_The dep_Agirre_author auxpass_funded_was advmod_funded_partially parataxis_proved_funded prep_funded_by nn_No_GALE nn_No_DARPA nn_No_Contract pobj_by_No ' +n,4040,'WSD systems have been far more successful in distinguishing coarsegrained senses than fine-grained ones , but does that approach neglect necessary meaning differences ? ',Navigli,'nn_systems_WSD nsubj_successful_systems aux_successful_have dep_successful_been advmod_successful_far advmod_successful_more prep_successful_in pcomp_in_distinguishing amod_senses_coarsegrained dobj_distinguishing_senses mark__than amod_ones_fine-grained nsubj__ones ccomp_successful_ cc__but conj__does det_neglect_that nn_neglect_approach nsubj_meaning_neglect dep_meaning_necessary xcomp_does_meaning dobj_meaning_differences ' +o,4041,'More rare words rather than common words are found even in standard dictionaries ',Church,'amod_words_More amod_words_rare nsubjpass_found_words dep_than_rather prep_words_than amod_words_common pobj_than_words auxpass_found_are advmod_found_even prep_found_in amod_dictionaries_standard pobj_in_dictionaries ' +o,4042,'Several researchers also studied feature\\/topicbased sentiment analysis ',Titov,'amod_researchers_Several nsubj_studied_researchers advmod_studied_also amod_analysis_feature\\/topicbased nn_analysis_sentiment dobj_studied_analysis ' +o,4043,'The noun phrase chunking -LRB- NP chunking -RRB- module uses the basic NP chunker software from 483 to recognize the noun phrases in the question ',Ramshaw,'det_module_The amod_module_noun nn_module_phrase nn_module_chunking nn_chunking_NP abbrev_module_chunking nsubj_uses_module det_software_the amod_software_basic nn_software_NP nn_software_chunker dobj_uses_software prep_uses_from number__483 pobj_from_ aux_recognize_to xcomp_uses_recognize det_phrases_the nn_phrases_noun dobj_recognize_phrases prep_recognize_in det_question_the pobj_in_question ' +o,4044,'221 BLEU Evaluation The BLEU score was defined to measure overlap between a hypothesized translation and a set of human references ',Papineni,'number_BLEU_221 num_Evaluation_BLEU det__The amod__BLEU nn__score nsubjpass_defined_ auxpass_defined_was dep_Evaluation_defined aux_measure_to xcomp_defined_measure dobj_measure_overlap prep_overlap_between det_translation_a amod_translation_hypothesized pobj_between_translation cc_overlap_and det_set_a conj_overlap_set prep_set_of amod_references_human pobj_of_references ' +p,4045,'Furthermore , end-to-end systems like speech recognizers and automatic translators use increasingly sophisticated discriminative models , which generalize well to new data that is drawn from the same distribution as the training data ',Och,'advmod_translators_Furthermore amod_systems_end-to-end nsubj_translators_systems prep_systems_like nn_recognizers_speech pobj_like_recognizers amod_recognizers_ cc__and conj__automatic nsubj_use_ ccomp_translators_use advmod_sophisticated_increasingly amod_models_sophisticated nn_models_discriminative dobj_use_models nsubj_generalize_which rcmod_models_generalize advmod_generalize_well prep_generalize_to amod_data_new pobj_to_data nsubjpass_drawn_that auxpass_drawn_is rcmod_data_drawn prep_drawn_from det_distribution_the amod_distribution_same pobj_from_distribution prep_distribution_as det_data_the nn_data_training pobj_as_data ' +o,4046,'We tune all feature weights automatically to maximize the BLEU score on the dev set ',Och,'nsubj_tune_We det_weights_all nn_weights_feature nsubj__weights advmod__automatically ccomp_tune_ aux_maximize_to xcomp__maximize det_score_the amod_score_BLEU amod_score_ dobj_maximize_score prep_maximize_on det_set_the nn_set_dev pobj_on_set ' +o,4047,'Other linear time algorithms for rank reduction are found in the literature , but they are restricted to the case of synchronous context-free grammars , a strict subclass of the LCFRS with f = 2 ',Zhang,'amod_algorithms_Other amod_algorithms_linear nn_algorithms_time nsubjpass_found_algorithms prep_algorithms_for amod_reduction_rank pobj_for_reduction auxpass_found_are prep_found_in det__the nn__literature pobj_in_ cc_found_but nsubjpass_restricted_they auxpass_restricted_are conj_found_restricted prep_restricted_to det_case_the pobj_to_case prep_case_of amod_grammars_synchronous amod_grammars_context-free pobj_of_grammars det_subclass_a amod_subclass_strict appos_grammars_subclass prep_subclass_of det_LCFRS_the pobj_of_LCFRS prep_restricted_with pobj_with_f dep_2_= rcmod_f_2 ' +o,4048,'In the above equation , P -LRB- ti -RRB- and P -LRB- wi ; t -RRB- are estimated by the maximum-likelihood method , and the probability of a POC tag ti , given a character wi -LRB- P -LRB- tijwi ; ti 2 TPOC -RRB- -RRB- is estimated using ME models ',Berger,'prep_estimated_In det_equation_the amod_equation_above pobj_In_equation nsubjpass_estimated_P appos_P_ti cc_P_and conj_P_P nn_t_wi appos_P_t auxpass_estimated_are prep_estimated_by det_method_the amod_method_maximum-likelihood pobj_by_method cc_method_and det_probability_the conj_method_probability prep_probability_of det_ti_a nn_ti_POC nn_ti_tag pobj_of_ti prep_estimated_given det_wi_a nn_wi_character dep_given_wi dep_wi_P dep_wi_tijwi nn_TPOC_ti num_TPOC_2 dep_tijwi_TPOC auxpass_estimated_is rcmod_wi_estimated xcomp_estimated_using nn_models_ME dobj_using_models ' +o,4049,'Hyperparameter is automatically selected from 2Although that for their dataset similar to ours was 083 , this value can not be directly compared with our value because their dataset includes both individual words and pairs of words ',Kanayama,'nsubjpass_selected_Hyperparameter auxpass_selected_is advmod_selected_automatically prep_selected_from amod__2Although pobj_from_ complm_compared_that mark_083_for poss_dataset_their nsubj_083_dataset amod_dataset_similar prep_similar_to pobj_to_ours cop_083_was advcl_compared_083 det_value_this nsubjpass_compared_value aux_compared_can neg_compared_not auxpass_compared_be advmod_compared_directly ccomp_selected_compared prep_compared_with poss_value_our pobj_with_value mark_includes_because poss_dataset_their nsubj_includes_dataset advcl_compared_includes preconj_words_both amod_words_individual dobj_includes_words cc_words_and conj_words_pairs prep_words_of pobj_of_words ' +o,4050,'-LRB- This example is adapted from Resnik -RRB- ',Dunning,'det_example_This nsubjpass_adapted_example auxpass_adapted_is prep_adapted_from nn__Resnik pobj_from_ ' +o,4051,'as follows : p -LRB- synI1 trgI1 -RRB- = -LRB- Iproductdisplay i = 1 p -LRB- syni trgi -RRB- -LRB- 4 -RRB- pprime -LRB- trgi syni -RRB- prime pw -LRB- syni trgi -RRB- w pwprime -LRB- trgi syni -RRB- wprime pd -LRB- syni , trgi -RRB- d -RRB- lw -LRB- synI1 -RRB- l c -LRB- synI1 -RRB- c pLM -LRB- synI1 -RRB- LM For estimation of the feature weights vector defined in equation -LRB- 4 -RRB- we employed minimum error rate -LRB- MER -RRB- training under the BLEU measure ',Och,'advmod_follows_as measure_=_p amod_trgI1_synI1 appos_p_trgI1 amod_lw_= nn_i_Iproductdisplay nsubj_=_i dep_=_= num_p_1 dobj_=_p nn_trgi_syni appos_p_trgi appos_p_4 dep_p_pprime nn_syni_trgi appos_pprime_syni amod_d_prime amod_d_pw nn_trgi_syni appos_d_trgi nn_d_w amod_d_pwprime nn_syni_trgi appos_d_syni nn_d_wprime nn_d_pd dep_d_syni appos_syni_trgi dep_pprime_d dobj_follows_lw appos_lw_synI1 dep_lw_l dep_lw_c dep_follows_synI1 dep_LM_c nsubj_LM_pLM appos_pLM_synI1 dep_follows_LM prep_employed_For pobj_For_estimation prep_estimation_of det_vector_the nn_vector_feature nn_vector_weights pobj_of_vector partmod_vector_defined prep_defined_in pobj_in_equation appos_equation_4 nsubj_employed_we ccomp_LM_employed amod_training_minimum nn_training_error nn_training_rate abbrev_training_MER dobj_employed_training prep_training_under det_measure_the amod_measure_BLEU pobj_under_measure ' +p,4052,'Moreover , log likelihood ratios are regarded as a more effective method to identify collocations especially when the occurrence count is very low ',Dunning,'advmod_regarded_Moreover nn_ratios_log nn_ratios_likelihood nsubjpass_regarded_ratios auxpass_regarded_are prep_regarded_as det_method_a advmod_effective_more amod_method_effective pobj_as_method aux_collocations_to dep_collocations_identify xcomp_regarded_collocations advmod_collocations_especially advmod_low_when det_count_the nn_count_occurrence nsubj_low_count cop_low_is advmod_low_very advcl_collocations_low ' +o,4053,' also used windows of size 2 , which corresponds to word bigrams ',Dunning,'advmod__also dep__used dobj_used_windows prep_windows_of pobj_of_size nsubj__2 nsubj_corresponds_which rcmod_2_corresponds prep_corresponds_to nn_bigrams_word pobj_to_bigrams ' +o,4054,'Ralph Weischedel et al 1993 ',Marcus,'nn_Weischedel_Ralph cc_Weischedel_et conj_Weischedel_al dep_Weischedel_1993 ' +o,4055,'Unlexicalized methods refine the grammar in a more conservative fashion , splitting each non-terminal or pre-terminal symbol into a much smaller number of subsymbols ',Matsuzaki,'amod_methods_Unlexicalized nsubj_refine_methods det_grammar_the dobj_refine_grammar prep_grammar_in det_fashion_a advmod_conservative_more amod_fashion_conservative pobj_in_fashion xcomp_refine_splitting det_symbol_each amod_symbol_non-terminal cc_non-terminal_or conj_non-terminal_pre-terminal dobj_splitting_symbol prep_splitting_into det_number_a advmod_smaller_much amod_number_smaller pobj_into_number prep_number_of pobj_of_subsymbols ' +o,4056,' describe the voted perceptron applied to the named-entity data in this paper , but using kernel-based features rather than the explicit features described in this paper ',Collins,'nsubj_describe_ det_perceptron_the amod_perceptron_voted dobj_describe_perceptron partmod_perceptron_applied prep_applied_to det_data_the amod_data_named-entity pobj_to_data prep_data_in det_paper_this pobj_in_paper prep_data_but dep_but_using amod_features_kernel-based dobj_using_features cc_perceptron_rather dep_rather_than det_features_the amod_features_explicit conj_perceptron_features partmod_features_described prep_described_in det_paper_this pobj_in_paper ' +o,4057,'The algorithm is slightly different from other online training algorithms in that we keep and update oracle translations , which is a set of good translations reachable by a decoder according to a metric , ie BLEU ',Papineni,'det_algorithm_The nsubj_different_algorithm cop_different_is advmod_different_slightly prep_different_from amod_algorithms_other amod_algorithms_online nn_algorithms_training pobj_from_algorithms partmod_algorithms_ prep__in complm_keep_that nsubj_keep_we pcomp_in_keep cc_keep_and conj_keep_update amod_translations_oracle dobj_keep_translations nsubj_set_which cop_set_is det_set_a rcmod_algorithms_set prep_set_of amod_translations_good pobj_of_translations partmod_translations_reachable prep_reachable_by det_decoder_a pobj_by_decoder prep_reachable_according dep_according_to det_BLEU_a amod_BLEU_metric amod_BLEU_ie pobj_to_BLEU ' +o,4058,'For instance , the resulting word graph can be used in the prediction engine of a CAT system ',Och,'prep_used_For pobj_For_instance det_graph_the amod_graph_resulting nn_graph_word nsubjpass_used_graph aux_used_can auxpass_used_be prep_used_in det_engine_the nn_engine_prediction pobj_in_engine prep_engine_of det_system_a nn_system_CAT pobj_of_system ' +o,4059,'2 \' -RRB- ` he WSJ corpus ',Marcus,'poss_-RRB-_2 nsubj_WSJ_he dep_-RRB-_WSJ dobj_WSJ_corpus ' +o,4060,'291 31 Level of Analysis Research on sentiment annotation is usually conducted at the text or at the sentence levels ',Turney,'num_Level_291 num_Level_31 nsubjpass_conducted_Level prep_Level_of nn_Research_Analysis pobj_of_Research prep_Research_on nn_annotation_sentiment pobj_on_annotation auxpass_conducted_is advmod_conducted_usually prep_conducted_at det__the nn__text pobj_at_ cc_at_or conj_at_at det__the nn__sentence nn__levels pobj_at_ ' +o,4061,'Automatic measures like BLEU or NIST do so by counting sequences of words in such paraphrases ',Papineni,'nsubj_measures_Automatic mark_do_like nn__BLEU nsubj_do_ cc__or nn__NIST conj__ advcl_measures_do advmod_do_so prep_do_by amod_sequences_counting pobj_by_sequences prep_sequences_of pobj_of_words prep_words_in amod_paraphrases_such pobj_in_paraphrases ' +o,4062,'CIT -RRB- ',Brown,'nn_-RRB-_CIT ' +o,4063,'The basic model uses the following features , analogous to Pharaohs default feature set : P -LRB- -RRB- and P -LRB- -RRB- the lexical weights Pw -LRB- -RRB- and Pw -LRB- -RRB- ; 1 a phrase penalty exp -LRB- 1 -RRB- ; a word penalty exp -LRB- l -RRB- , where l is the number of terminals in ',Koehn,'det_model_The amod_model_basic nsubj_uses_model det_features_the amod_features_following dobj_uses_features amod_default_analogous prep_analogous_to pobj_to_Pharaohs appos_features_default nn_set_feature dep_default_set dep_default_P cc_-RRB-_and conj_-RRB-_P det__the amod__lexical nn__weights nn__Pw cc_-RRB-_and conj_-RRB-_Pw dep_P_ dep_default_1 det_exp_a nn_exp_phrase nn_exp_penalty dep_1_exp appos_exp_1 det_exp_a nn_exp_word nn_exp_penalty dep_default_exp dep_exp_l advmod_number_where dep_number_l cop_number_is det_number_the rcmod_features_number prep_number_of pobj_of_terminals prep_terminals_in ' +o,4064,'Other researchers , use clustering techniques coupled with syntactic dependency features to identify IS-A relations in large text collections ',Snow,'amod_researchers_Other amod_techniques_ amod_techniques_ nn_techniques_use nn_techniques_clustering nsubj_features_techniques partmod_techniques_coupled prep_coupled_with amod_dependency_syntactic pobj_with_dependency dep_researchers_features aux_identify_to xcomp_features_identify amod_relations_IS-A dobj_identify_relations prep_identify_in amod_collections_large nn_collections_text pobj_in_collections ' +o,4065,'INTRODUCTION Class-based language models have been proposed for dealing with two problems confronted by the well-known word n-gram language models -LRB- 1 -RRB- data sparseness : the amount of training data is insufficient for estimating the huge number of parameters ; and -LRB- 2 -RRB- domain robustness : the model is not adaptable to new application domains ',Brown,'amod_models_Class-based nn_models_language nsubjpass_proposed_models advmod_proposed_ aux_proposed_have auxpass_proposed_been dep_INTRODUCTION_proposed prep_proposed_for pcomp_for_dealing prep_dealing_with num_problems_two pobj_with_problems partmod_problems_confronted prep_confronted_by det_word_the amod_word_well-known pobj_by_word amod_models_n-gram nn_models_language dep_word_models dep_sparseness_1 nsubj_sparseness_data dep_INTRODUCTION_sparseness det_amount_the nsubj_insufficient_amount prep_amount_of nn_data_training pobj_of_data cop_insufficient_is parataxis_sparseness_insufficient prep_insufficient_for pcomp_for_estimating det_number_the amod_number_huge dobj_estimating_number prep_number_of pobj_of_parameters cc_sparseness_and dep_adaptable_2 nn_robustness_domain nsubj_adaptable_robustness det_model_the dep_robustness_model cop_adaptable_is neg_adaptable_not conj_sparseness_adaptable prep_adaptable_to amod_domains_new nn_domains_application pobj_to_domains ' +o,4066,'The construction is defined in Fillmore \'s Construction Grammar as \' a pairing of a syntactic pattern with a meaning structure \' ; they are similar to signs in HPSG and pattern-concept pairs ',Hobbs,'det_construction_The nsubjpass_defined_construction auxpass_defined_is prep_defined_in poss_Grammar_Fillmore possessive_Fillmore_\'s nn_Grammar_ nn_Grammar_Construction pobj_in_Grammar prep_defined_as det_pairing_a pobj_as_pairing prep_pairing_of det_pattern_a amod_pattern_syntactic pobj_of_pattern prep_pattern_with det_structure_a nn_structure_meaning pobj_with_structure nsubj_similar_they cop_similar_are parataxis_defined_similar prep_similar_to pobj_to_signs prep_signs_in pcomp_in_HPSG amod__ cc__and conj__pattern-concept nn__pairs dobj_HPSG_ ' +o,4067,'7Following , we measure agreement in Kappa , which follows the formula K = P -LRB- A -RRB- P -LRB- E -RRB- 1P -LRB- E -RRB- where P -LRB- A -RRB- is observed , and P -LRB- E -RRB- expected agreement ',Carletta,'nsubj__7Following nsubj_measure_we ccomp__measure dobj_measure_agreement prep_agreement_in pobj_in_Kappa nsubj_follows_which rcmod_Kappa_follows det_K_the nn_K_formula nsubj_P_K dep_P_= ccomp_follows_P appos_agreement_A dep_agreement_P appos_P_E dep_agreement_1P appos_1P_E advmod_observed_where nsubjpass_observed_P appos_P_A auxpass_observed_is rcmod_1P_observed cc_agreement_and conj_agreement_P appos_P_E partmod_P_expected dobj_expected_agreement ' +o,4068,'In , automatically extracted collocations are judged by a lexicographer ',Smadja,'prep_extracted_In pobj_In_ nsubj_extracted_automatically nsubjpass_judged_collocations auxpass_judged_are ccomp_extracted_judged prep_judged_by det_lexicographer_a pobj_by_lexicographer ' +o,4069,'The mutual information of a pair of words is defined in terms of their co-occurrence frequency and respective occurrence frequencies ',Church,'det_information_The amod_information_mutual nsubjpass_defined_information prep_information_of det_pair_a pobj_of_pair prep_pair_of pobj_of_words auxpass_defined_is prep_defined_in pobj_in_terms prep_terms_of poss_frequency_their amod_frequency_co-occurrence pobj_of_frequency cc_frequency_and amod_frequencies_respective nn_frequencies_occurrence conj_frequency_frequencies ' +o,4070,'We carefully implemented the original Grammar Association system described in (Vidal et al. , 1993), tuned empirically a couple of smoothing parameters, trained the models and, finally, obtained an a119a21a120 a100 a104a122a121 of correct translations.9 Then, we studied the impact of: (1) sorting, as proposed in Section 3, the set of sentences presented to ECGI; (2) making language models deterministic and minimum; (3) constraining the best translation search to those sentences whose lengths have been seen, in the training set, related to the length of the input sentence.',Brown,'' +o,4071,'In order to improve translation quality , this tuning can be effectively performed by minimizing translation error over a development corpus for which manually translated references are available ',Och,'mark_improve_In dep_improve_order aux_improve_to dep_performed_improve nn_quality_translation dobj_improve_quality det_tuning_this nsubjpass_performed_tuning aux_performed_can auxpass_performed_be advmod_performed_effectively prep_performed_by pcomp_by_minimizing nn_error_translation dobj_minimizing_error prep_minimizing_over det_corpus_a nn_corpus_development pobj_over_corpus rel_available_for pobj_for_which advmod_translated_manually amod_references_translated nsubj_available_references cop_available_are rcmod_corpus_available ' +o,4072,'Recently , there have been several discriminative approaches at training large parameter sets including and ',Tillmann,'advmod_been_Recently expl_been_there aux_been_have amod_approaches_several nn_approaches_discriminative dobj_been_approaches prep_been_at pcomp_at_training amod_sets_large nn_sets_parameter dobj_training_sets prep_sets_including pobj_including_ cc__and conj__ ' +p,4073,'After the success in syntactic -LRB- Penn TreeBank -RRB- and propositional encodings -LRB- Penn PropBank -RRB- , more sophisticated semantic data -LRB- such as temporal or opinion annotations -RRB- and discourse data -LRB- eg , for anaphora resolution and rhetorical parsing -RRB- are being generated ',Marcus,'prep_generated_After det_success_the pobj_After_success prep_success_in pobj_in_syntactic nn__Penn nn__TreeBank appos_syntactic_ cc_syntactic_and amod_encodings_propositional conj_syntactic_encodings nn__Penn nn__PropBank appos_syntactic_ advmod_sophisticated_more amod_data_sophisticated amod_data_semantic nsubjpass_generated_data dep_as_such dep_data_as amod__temporal pobj_as_ cc__or nn__opinion nn__annotations conj__ cc_data_and nn_data_discourse conj_data_data dep_data_eg prep_eg_for nn__anaphora nn__resolution pobj_for_ cc__and amod__rhetorical nn__parsing conj__ aux_generated_are auxpass_generated_being ' +o,4074,'5 Datasets and Evaluation We train our models with verb instances extracted from three parsed corpora : -LRB- 1 -RRB- the Wall Street Journal section of the Penn Treebank -LRB- PTB -RRB- , which was parsed by human annotators , -LRB- 2 -RRB- the Brown Laboratory for Linguistic Information Processing corpus of Wall Street Journal text -LRB- BLLIP -RRB- , which was parsed automatically by the Charniak parser , and -LRB- 3 -RRB- the Gigaword corpus of raw newswire text -LRB- GW -RRB- , which we parsed ourselves with the Stanford parser ',Marcus,'num_Datasets_5 cc_Datasets_and conj_Datasets_Evaluation nsubj_train_We ccomp_Evaluation_train poss_models_our dobj_train_models prep_train_with nn_instances_verb pobj_with_instances partmod_instances_extracted prep_extracted_from num_corpora_three amod_corpora_parsed pobj_from_corpora dep_section_1 det_section_the nn_section_Wall nn_section_Street nn_section_Journal dep_corpora_section prep_section_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank abbrev_Treebank_PTB nsubjpass_parsed_which auxpass_parsed_was rcmod_Treebank_parsed prep_parsed_by amod__human nn__annotators pobj_by_ dep_Laboratory_2 det_Laboratory_the nn_Laboratory_Brown conj_section_Laboratory prep_Laboratory_for nn_corpus_Linguistic nn_corpus_Information nn_corpus_Processing pobj_for_corpus prep_corpus_of nn_text_Wall nn_text_Street nn_text_Journal pobj_of_text abbrev_text_BLLIP nsubjpass_parsed_which auxpass_parsed_was rcmod_text_parsed advmod_parsed_automatically prep_parsed_by det__the nn__Charniak nn__parser pobj_by_ cc_section_and dep_corpus_3 det_corpus_the nn_corpus_Gigaword conj_section_corpus prep_corpus_of amod_text_raw nn_text_newswire pobj_of_text abbrev_text_GW rel_parsed_which nsubj_parsed_we rcmod_text_parsed dobj_parsed_ourselves prep_parsed_with det_parser_the nn_parser_Stanford pobj_with_parser ' +o,4075,'The original formulation of statistical machine translation was defined as a word-based operation ',Brown,'det_formulation_The amod_formulation_original nsubjpass_defined_formulation prep_formulation_of amod__statistical nn__machine nn__translation pobj_of_ auxpass_defined_was prep_defined_as det_operation_a amod_operation_word-based pobj_as_operation ' +p,4076,'We choose those sections because several state-of-thwart parsers are trained on Section 2-21 and tested on Section 23 ',Collins,'nsubj_sections_We dep_sections_choose det_sections_those mark_trained_because amod_parsers_several amod_parsers_state-of-thwart nsubjpass_trained_parsers advmod_trained_ auxpass_trained_are advcl_sections_trained prep_trained_on pobj_on_Section num_Section_2-21 cc_trained_and conj_trained_tested prep_tested_on pobj_on_Section num_Section_23 ' +o,4077,'This operation can be used in applications like Minimum Error Rate Training , or optimizing system combination as described by Hillard et al ',Och,'det_operation_This nsubjpass_used_operation aux_used_can auxpass_used_be prep_used_in pobj_in_applications prep_applications_like nn__Minimum nn__Error nn__Rate nn__Training pobj_like_ cc_used_or conj_used_optimizing nn_combination_system dobj_optimizing_combination advmod_described_as amod_combination_described prep_optimizing_by pobj_by_Hillard cc_Hillard_et conj_Hillard_al ' +o,4078,'For evaluation we use a state-of-the-art baseline system -LRB- Moses -RRB- which works with a log-linear interpolation of feature functions optimized by MERT ',Och,'prep_use_For pobj_For_evaluation nsubj_use_we det_system_a amod_system_state-of-the-art nn_system_baseline nsubj__system appos_system_Moses ccomp_use_ nsubj_works_which ccomp__works prep_works_with det_interpolation_a amod_interpolation_log-linear pobj_with_interpolation prep_interpolation_of nn_functions_feature pobj_of_functions partmod_functions_optimized prep_optimized_by pobj_by_MERT ' +o,4079,'Others try to accommodate both syntactic and lexical differences between the candidate translation and the reference , like CDER , which employs a version of edit distance for word substitution and reordering ; or METEOR , which uses stemming and WordNet synonymy ',Banerjee,'nsubj_try_Others aux_accommodate_to xcomp_try_accommodate preconj_syntactic_both amod_differences_syntactic cc_syntactic_and conj_syntactic_lexical dobj_accommodate_differences prep_differences_between det_translation_the nn_translation_candidate pobj_between_translation cc_differences_and det_reference_the conj_differences_reference prep_reference_like nn__CDER pobj_like_ nsubj_employs_which rcmod__employs det_version_a dobj_employs_version prep_version_of amod_distance_edit pobj_of_distance prep_distance_for nn_substitution_word pobj_for_substitution cc_substitution_and conj_substitution_reordering cc_differences_or nn__METEOR conj_differences_ nsubj_uses_which rcmod__uses xcomp_uses_stemming cc_accommodate_and conj_accommodate_WordNet advmod_WordNet_synonymy ' +o,4080,'Generative methods treat word alignment as a hidden process and maximize the likelihood of bilingual training corpus using the expectation maximization -LRB- EM -RRB- algorithm ',Brown,'amod_methods_Generative nsubj__methods dep__treat nn_alignment_word dobj_treat_alignment prep_treat_as det_process_a amod_process_hidden pobj_as_process cc_treat_and conj_treat_maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_of amod_corpus_bilingual nn_corpus_training pobj_of_corpus partmod_corpus_using det_algorithm_the nn_algorithm_expectation nn_algorithm_maximization abbrev_algorithm_EM dobj_using_algorithm ' +o,4081,'42 Experiments on SRL dataset We used two different corpora : PropBank -LRB- wwwcisupennedu\\/ace -RRB- along with Penn Treebank 2 and FrameNet ',Marcus,'num_Experiments_42 nsubj_dataset_Experiments prep_Experiments_on pobj_on_SRL nsubj_used_We ccomp_dataset_used num_corpora_two amod_corpora_different dobj_used_corpora dep_along_PropBank appos_PropBank_wwwcisupennedu\\/ace advmod__along dep_along_with nn_Treebank_Penn pobj_with_Treebank num__2 dep_corpora_ cc__and conj__FrameNet ' +o,4082,'These constraints tie words in such a way that the space of alignments can not be enumerated as in IBM models 1 and 2 ',Dunning,'det_constraints_These nsubj_tie_constraints dobj_tie_words prep_tie_in predet_way_such det_way_a pobj_in_way complm_enumerated_that det_space_the nsubjpass_enumerated_space prep_space_of pobj_of_alignments aux_enumerated_can neg_enumerated_not auxpass_enumerated_be ccomp_tie_enumerated advmod_enumerated_as dep_as_in nn_models_IBM pobj_in_models num__1 cc_1_and conj_1_2 dobj_enumerated_ ' +o,4083,'In the literature on the kappa statistic , most authors address only category data ; some can handle more general data , such as data in interval scales or ratio scales ',Carletta,'prep_address_In det_literature_the pobj_In_literature prep_literature_on det_statistic_the nn_statistic_kappa pobj_on_statistic amod_authors_most nsubj_address_authors amod_data_only nn_data_category dobj_address_data nsubj_handle_some aux_handle_can parataxis_address_handle advmod_general_more amod_data_general dobj_handle_data dep_as_such prep_data_as pobj_as_data prep_data_in amod_scales_interval pobj_in_scales cc_scales_or nn_scales_ratio conj_scales_scales ' +o,4084,'Movie and product reviews have been the main focus of many of the recent studies in this area ',Pang,'nn_reviews_Movie cc_Movie_and conj_Movie_product nsubj_been_reviews aux_been_have det_focus_the amod_focus_main dobj_been_focus prep_focus_of pobj_of_many prep_many_of det_studies_the amod_studies_recent pobj_of_studies prep_been_in det_area_this pobj_in_area ' +o,4085,'The number of weights wi is 3 plus the number of source languages , and they are trained using minimum error-rate training -LRB- MERT -RRB- to maximize the BLEU score on a development set ',Och,'det_number_The nsubj_3_number prep_number_of nn_wi_weights pobj_of_wi cop_3_is prep_3_plus det_number_the dep_plus_number prep_number_of nn_languages_source pobj_of_languages cc_3_and nsubjpass_trained_they auxpass_trained_are conj_3_trained xcomp_trained_using amod_training_minimum amod_training_error-rate dobj_using_training abbrev_training_MERT aux_maximize_to xcomp_using_maximize det__the amod__BLEU nn__score dobj_maximize_ prep_maximize_on det_set_a nn_set_development pobj_on_set ' +o,4086,'Part-of-speech tags are assigned by the MXPOST maximum-entropy based part-of-speech tagger ',Ratnaparkhi,'nn_tags_Part-of-speech nsubjpass_assigned_tags auxpass_assigned_are prep_assigned_by det_maximum-entropy_the amod_maximum-entropy_MXPOST pobj_by_maximum-entropy partmod_maximum-entropy_based amod_tagger_part-of-speech dobj_based_tagger ' +o,4087,'3.1 The traditional IBM alignment model IBM Model 4 (Brown et al. , 1993) learns a set of 4 probability tables to compute p(f|e) given a foreign sentence f and its target translation e via the following (greatly simplified) generative story: 361 NP-C NPB NPB NNP taiwan POS s NN surplus PP IN in NP-C NPB NN trade PP IN between NP-C NPB DT the CD two NNS shores FTD0 GR G4E7 DYBG EL DIDV TAIWAN IN TWO-SHORES TRADE MIDDLE SURPLUS R1: NP-C NPB x0:NPB x1:NN x2:PP x0 x2EL x1 R10: NP-C NPB x0:NPB x1:NN x2:PP x0 x2 x1 R10: NP-C NPB x0:NPB x1:NN x2:PP x0 x2 x1 R2: NPB NNP taiwan POS s FTD0 R11: NPB x0:NNP POS s x0 R17: NPB NNP taiwan x0:POS x0 R12: NNP taiwan FTD0 R18: POS s FTD0 R3: PP x0:IN x1:NP-C x0 x1 R13: PP IN in x0:NP-C GR x0EL R19: PP IN in x0:NP-C x0 R4: IN in GR R5: NP-C x0:NPB x1:PP x1 x0 R5: NP-C x0:NPB x1:PP x1 x0 R20: NP-C x0:NPB PP x1:IN x2:NP-C x2 x0 x1 R6: PP IN between NP-C NPB DT the CD two NNS shores G4E7 R14: PP IN between x0:NP-C x0 R21: IN between EL R15: NP-C x0:NPB x0 R15: NP-C x0:NPB x0 R16: NPB DT the CD two NNS shores G4E7 R22: NPB x0:DT CD two x1:NNS x0 x1 R23: NNS shores G4E7 R24: DT the GR R7: NPB x0:NN x0 R7: NPB x0:NN x0 R7: NPB x0:NN x0 R8: NN trade DYBG R9: NN surplus DIDV R8: NN trade DYBG R9: NN surplus DIDV R8: NN trade DYBG R9: NN surplus DIDV Figure 2: A (English tree, Chinese string) pair and three different sets of multilevel tree-to-string rules that can explain it; the first set is obtained from bootstrap alignments, the second from this papers re-alignment procedure, and the third is a viable, if poor quality, alternative that is not learned.',Brown,'' +o,4088,' -RRB- simplify these probability distributions , as given in Equations 9 and 10 ',Marcus,'amod_probability_simplify det_probability_these dobj__probability advmod__distributions mark_given_as advcl__given prep_given_in pobj_in_Equations num_Equations_9 cc_Equations_and conj_Equations_10 ' +o,4089,'By using 8-bit floating point quantization 1 , N-gram language models are compressed into 10 GB , which is comparable to a lossy representation ',Talbot,'prep_compressed_By pcomp_By_using amod_point_8-bit amod_point_floating dobj_using_point amod_point_quantization tmod_quantization_1 amod_models_N-gram nn_models_language nsubjpass_compressed_models auxpass_compressed_are prep_compressed_into num_GB_10 pobj_into_GB nsubj_comparable_which cop_comparable_is rcmod_GB_comparable prep_comparable_to det_representation_a amod_representation_lossy pobj_to_representation ' +o,4090,'3 Previous Work on Subjectivity Tagging In previous work , a corpus of sentences from the Wall Street Journal Treebank Corpus was manually anno - tated with subjectivity classifications by multiple judges ',Marcus,'num_Work_3 amod_Work_Previous nsubj_anno_Work prep_Work_on nn_Tagging_Subjectivity pobj_on_Tagging prep_anno_In amod__previous nn__work pobj_In_ det_corpus_a nsubj_anno_corpus prep_corpus_of pobj_of_sentences prep_sentences_from det__the nn__Wall nn__Street nn__Journal nn__Treebank nn__Corpus pobj_from_ cop_anno_was advmod_anno_manually partmod_anno_tated prep_tated_with amod_classifications_subjectivity pobj_with_classifications prep_tated_by amod_judges_multiple pobj_by_judges ' +n,4091,'Secondly , while most pronoun resolution evaluations simply exclude non-referential pronouns , recent unsupervised approaches must deal with all pronouns in unrestricted text , and therefore need robust modules to automatically handle non-referential instances ',Haghighi,'advmod_deal_Secondly mark_exclude_while advmod_evaluations_most amod_evaluations_pronoun nn_evaluations_resolution nsubj_exclude_evaluations advmod_exclude_simply dep_deal_exclude amod_pronouns_non-referential dobj_exclude_pronouns amod__recent amod__unsupervised nn__approaches nsubj_deal_ aux_deal_must prep_deal_with det_pronouns_all pobj_with_pronouns prep_pronouns_in amod_text_unrestricted pobj_in_text cc_deal_and advmod_deal_therefore conj_deal_need amod_modules_robust dobj_need_modules aux_handle_to advmod_handle_automatically xcomp_need_handle amod_instances_non-referential dobj_handle_instances ' +o,4092,'In this paper , we bring forward the first idea by studying the issue of how to utilize structured syntactic features for phrase reordering in a phrase-based SMT system with BTG -LRB- Bracketing Transduction Grammar -RRB- constraints ',Wu,'prep_bring_In det_paper_this pobj_In_paper nsubj_bring_we prt_bring_forward det_idea_the amod_idea_first dobj_bring_idea prep_bring_by pcomp_by_studying det_issue_the dobj_studying_issue prep_issue_of advmod_utilize_how aux_utilize_to pcomp_of_utilize amod_features_structured amod_features_syntactic dobj_utilize_features prep_features_for nn_reordering_phrase pobj_for_reordering prep_utilize_in det_system_a amod_system_phrase-based nn_system_SMT pobj_in_system prep_system_with amod_constraints_BTG nn_Grammar_Bracketing nn_Grammar_Transduction appos_constraints_Grammar pobj_with_constraints ' +o,4093,'While showed that this technique was effective when testing on WSJ , the true distribution was closer to WSJ so it made sense to emphasize it ',McClosky,'mark_showed_While nsubj_showed_ advcl_closer_showed complm_effective_that det_technique_this nsubj_effective_technique cop_effective_was ccomp_showed_effective advmod_testing_when advcl_effective_testing prep_testing_on pobj_on_WSJ det_distribution_the amod_distribution_true nsubj_closer_distribution cop_closer_was prep_closer_to pobj_to_WSJ mark_made_so nsubj_made_it advcl_closer_made dobj_made_sense aux_emphasize_to xcomp_made_emphasize dobj_emphasize_it ' +o,4094,'4This was a straightforward task ; two annotators annotated independently , with very high agreementkappa score of over 095 ',Carletta,'nsubj_task_4This cop_task_was det_task_a amod_task_straightforward num_annotators_two dep_task_annotators partmod_annotators_annotated advmod_annotated_independently prep_annotated_with advmod_high_very amod_score_high nn_score_agreementkappa pobj_with_score prep_score_of quantmod_095_over pobj_of_095 ' +o,4095,'There have been a lot of prol -RRB- OS ~ fls for statistical analysis , in ninny languages , in particular in English and Japanese -LRB- I\\/atnal -RRB- arkhi , 1997 -RRB- ',Collins,'expl_been_There aux_been_have det_lot_a dobj_been_lot prep_lot_of pobj_of_prol nn_~_OS nsubj_fls_~ rcmod_prol_fls prep_fls_for amod_analysis_statistical pobj_for_analysis prep_fls_in amod_languages_ninny pobj_in_languages prep_been_in amod__particular prep_particular_in amod__English cc_English_and conj_English_Japanese amod__ nn__ pobj_in_ appos__I\\/atnal dep_particular_arkhi appos_arkhi_1997 amod__ amod__ nn__ pobj_in_ ' +o,4096,'1 Introduction Word associations -LRB- co-occurrences -RRB- have a wide range of applications including : Speech Recognition , Optical Character Recognition and Information Retrieval -LRB- IR -RRB- ',Dunning,'num_associations_1 nn_associations_Introduction nn_associations_Word nsubj_have_associations appos_associations_co-occurrences det_range_a amod_range_wide dobj_have_range prep_range_of pobj_of_applications prep_have_including nn_Recognition_Speech pobj_including_Recognition nn_Recognition_Optical nn_Recognition_Character appos_Recognition_Recognition cc_Recognition_and nn_Retrieval_Information conj_Recognition_Retrieval abbrev_Recognition_IR ' +o,4097,'1 Full Morphological Tagging English Part of Speech -LRB- POS -RRB- tagging has been widely described in the recent past , starting with the paper , followed by numerous others using various methods : neural networks , HMM tagging , decision trees , transformation-based error-driven learning , and maximum entropy , to select just a few ',Ratnaparkhi,'dep_described_1 nn_Part_Full nn_Part_Morphological nn_Part_Tagging nn_Part_English nsubjpass_described_Part prep_Part_of nn_tagging_Speech appos_tagging_POS pobj_of_tagging aux_described_has auxpass_described_been advmod_described_widely prep_described_in det_past_the amod_past_recent pobj_in_past prep_described_starting dep_starting_with det_paper_the amod_paper_ pobj_with_paper prep_described_followed dep_followed_by amod_others_numerous pobj_by_others partmod_others_using amod_methods_various dobj_using_methods amod_networks_neural nsubj__networks parataxis_described_ dep__HMM amod__tagging dobj__ nn__decision nn__trees conj__ amod__transformation-based amod__error-driven nn__learning conj__ cc__and amod__maximum nn__entropy conj__ aux_select_to xcomp__select quantmod_few_just quantmod_few_a dobj_select_few ' +o,4098,'1LDC2002E18 -LRB- 4,000 sentences -RRB- , LDC2002T01 , LDC2003E07 , LDC2003E14 , LDC2004T07 , LDC2005T10 , LDC2004T08 HK Hansards -LRB- 500,000 sentences -RRB- 2http : \\/ \\/ wwwstatmtorg\\/wmt07\\/shared-taskhtml For both the tasks , the word alignment were trained by GIZA + + in two translation directions and refined by grow-diag-final method ',Koehn,'nsubj_\\/_1LDC2002E18 num_sentences_4,000 appos_1LDC2002E18_sentences appos_1LDC2002E18_LDC2002T01 appos_LDC2002T01_LDC2003E07 appos_1LDC2002E18_LDC2003E14 appos_1LDC2002E18_LDC2004T07 appos_1LDC2002E18_LDC2005T10 nn_2http_LDC2004T08 nn_2http_HK nn_2http_Hansards num_sentences_500,000 appos_2http_sentences appos_1LDC2002E18_2http advmod_wwwstatmtorg\\/wmt07\\/shared-taskhtml_\\/ acomp_\\/_wwwstatmtorg\\/wmt07\\/shared-taskhtml prep_trained_For predet_tasks_both det_tasks_the pobj_For_tasks det_alignment_the nn_alignment_word nsubjpass_trained_alignment auxpass_trained_were dep_\\/_trained prep_trained_by nn_+_GIZA nn_+_+ pobj_by_+ prep_+_in num_directions_two nn_directions_translation pobj_in_directions cc_trained_and conj_trained_refined prep_refined_by amod_method_grow-diag-final pobj_by_method ' +o,4099,'input pegging a ?transfer correct partially correct b incorrect 1 raw no M4 decoding c 7 4 4 2 stemmed yes M4 decoding 8 3 4 3 stemmed no M4 decoding 13 2 0 4 raw no gloss 13 1 1 5a stemmed yes gloss 8 3 4 5b stemmed yes gloss 12 2 1 6 stemmed no gloss 11 2 2 a pegging causes the training algorithm to consider a larger search space b correct top level category but incorrect sub-category c translation by maximizing the IBM Model 4 probability of the source/translation pair (Brown et al. , 1993; Brown et al. , 1995) classification might be performed by automatic procedures rather than humans.',Brown,'' +o,4100,'In addition , explicitly using the left context symbols allows easy use of smoothing techniques , such as deleted interpolation , clustering techniques , and model refinement techniques to estimate the probabilities more reliably by changing the window sizes of the context and weighting the various estimates dynamically ',Brown,'prep_allows_In pobj_In_addition advmod_using_explicitly dep_allows_using det_context_the amod_context_left dobj_using_context nsubj_allows_symbols amod_use_easy dobj_allows_use prep_use_of amod_techniques_smoothing pobj_of_techniques dep_as_such prep_allows_as amod__deleted nn__interpolation pobj_as_ amod_techniques_clustering nsubj__techniques ccomp_allows_ cc_allows_and nn_techniques_model nn_techniques_refinement nsubj__techniques conj_allows_ aux_estimate_to xcomp__estimate det_probabilities_the dobj_estimate_probabilities advmod_estimate_more advmod_estimate_reliably prep_estimate_by pcomp_by_changing det_sizes_the nn_sizes_window dobj_changing_sizes prep_sizes_of det_context_the pobj_of_context cc_changing_and conj_changing_weighting det_estimates_the amod_estimates_various dobj_weighting_estimates advmod_weighting_dynamically ' +o,4101,'If we assign a probability a15a17a16a19a18 a12 a13a7a21a20a4a6a5a7a23a22 to each pair of strings a18 a12a14a13a7a25a24 a4 a5a7 a22, then according to Bayes decision rule, we have to choose the target string that maximizes the product of the target language model a15a17a16a19a18 a12a14a13a7 a22 and the string translation model a15a17a16a19a18a26a4a6a5 a7 a20 a12 a13 a7 a22 . Many existing systems for statistical machine translation (Berger et al. , 1994; Wang and Waibel, 1997; Tillmann et al. , 1997; Nieen et al. , 1998) make use of a special way of structuring the string translation model like proposed by (Brown et al. , 1993): The correspondence between the words in the source and the target string is described by alignments that assign one target word position to each source word position.',Brown,'' +o,4102,'We employ the phrase-based SMT framework , and use the Moses toolkit , and the SRILM language modelling toolkit , and evaluate our decoded translations using the BLEU measure , using a single reference translation ',Papineni,'nsubj_employ_We det__the amod__phrase-based nn__SMT nn__framework dobj_employ_ cc_employ_and conj_employ_use det__the nn__Moses nn__toolkit dobj_use_ cc_employ_and det_language_the nn_language_SRILM nsubj_modelling_language conj_employ_modelling amod__toolkit dobj_modelling_ cc_employ_and conj_employ_evaluate poss_translations_our amod_translations_decoded dobj_evaluate_translations xcomp_evaluate_using det__the amod__BLEU nn__measure dobj_using_ dep_evaluate_using det_translation_a amod_translation_single nn_translation_reference dobj_using_translation ' +o,4103,'Some researchers classify terms by similarities based on their distributional syntactic patterns ',Lin,'det_researchers_Some nsubj__researchers amod_terms_classify dobj__terms prep__by pobj_by_similarities prep__based dep_based_on poss_patterns_their amod_patterns_distributional amod_patterns_syntactic pobj_on_patterns ' +o,4104,'Other representative collocation research can be found in and Smadja -LRB- 1993 -RRB- ',Church,'amod_research_Other nn_research_representative nn_research_collocation nsubjpass_found_research aux_found_can auxpass_found_be prep_found_in pobj_in_ cc__and conj__Smadja dep_found_1993 ' +o,4105,'Previous studies defined a probabilistic model of unification-based grammars as a log-linear model or maximum entropy model ',Berger,'amod__Previous nn__studies nsubj_defined_ det_model_a amod_model_probabilistic dobj_defined_model prep_model_of amod_grammars_unification-based pobj_of_grammars prep_defined_as det_model_a amod_model_log-linear pobj_as_model cc_model_or amod_model_maximum nn_model_entropy conj_model_model ' +o,4106,'These results confirm the observed figures in the previous subsection and reinforce the sight that clustering is a worthless effort for automatic paraphrase corpora construction , contrarily to what suggest ',Barzilay,'det_results_These nsubj_confirm_results det_figures_the amod_figures_observed dobj_confirm_figures prep_confirm_in det_subsection_the amod_subsection_previous pobj_in_subsection cc_confirm_and dep_sight_reinforce det_sight_the conj_confirm_sight complm_effort_that nsubj_effort_clustering cop_effort_is det_effort_a amod_effort_worthless ccomp_sight_effort prep_effort_for amod_construction_automatic amod_construction_paraphrase nn_construction_corpora pobj_for_construction appos_construction_contrarily rel_suggest_to pobj_to_what nsubj_suggest_ rcmod_contrarily_suggest ' +p,4107,'441 N-gram Co-Occurrence Statistics for Answer Extraction N-gram co-occurrence statistics have been successfully used in automatic evaluation , and more recently as training criteria in statistical machine translation ',Papineni,'num_Statistics_441 amod_Statistics_N-gram nn_Statistics_Co-Occurrence nsubjpass_used_Statistics prep_Statistics_for nn_statistics_Answer nn_statistics_Extraction nn_statistics_N-gram amod_statistics_co-occurrence pobj_for_statistics aux_used_have auxpass_used_been advmod_used_successfully prep_used_in amod__automatic nn__evaluation pobj_in_ cc_in_and advmod_recently_more conj_in_recently conj_in_as nn_criteria_training pobj_as_criteria prep_criteria_in amod__statistical nn__machine nn__translation pobj_in_ ' +p,4108,'However , since work in this direction has started , a significant progress has also been made in the research on statistical learning of full parsers , both in terms of accuracy and processing time ',Collins,'advmod_made_However mark_started_since nsubj_started_work prep_work_in det_direction_this pobj_in_direction aux_started_has dep_made_started det_progress_a amod_progress_significant nsubjpass_made_progress aux_made_has advmod_made_also auxpass_made_been prep_made_in det_research_the pobj_in_research prep_research_on amod_learning_statistical pobj_on_learning prep_learning_of amod_parsers_full pobj_of_parsers dep_in_both conj_in_in pobj_in_terms prep_terms_of nn_time_accuracy cc_accuracy_and conj_accuracy_processing pobj_of_time ' +o,4109,'Evaluating the algorithm on the output of Charniaks parser and the Penn treebank shows that the patternmatching algorithm does surprisingly well on the most frequently occuring types of empty nodes given its simplicity ',Marcus,'det_algorithm_the dobj_Evaluating_algorithm prep_Evaluating_on det_output_the pobj_on_output prep_output_of nn__Charniaks nn__parser pobj_of_ cc__and det__the nn__Penn nn__treebank conj__ dep_Evaluating_shows complm_does_that det_algorithm_the amod_algorithm_patternmatching nsubj_does_algorithm ccomp_shows_does advmod_well_surprisingly advmod_does_well prep_does_on det_most_the pobj_on_most advmod_does_frequently amod_types_occuring nsubj_Evaluating_types prep_types_of amod_nodes_empty pobj_of_nodes partmod_nodes_given poss_simplicity_its dobj_given_simplicity ' +o,4110,'52 Evaluation Metrics The commonly used criteria to evaluate the translation results in the machine translation community are : WER -LRB- word error rate -RRB- , PER -LRB- positionindependent word error rate -RRB- , BLEU , and NIST ',Papineni,'num_Evaluation_52 dep_Evaluation_Metrics det_criteria_The advmod_used_commonly amod_criteria_used nsubj_evaluate_criteria aux_evaluate_to xcomp_Metrics_evaluate det_results_the nn_results_translation dobj_evaluate_results prep_evaluate_in det_community_the nn_community_machine nn_community_translation pobj_in_community dep_community_are dep_Evaluation_WER nn_rate_word nn_rate_error dep_WER_rate conj_WER_PER amod_rate_positionindependent nn_rate_word nn_rate_error appos_PER_rate nn__BLEU conj_WER_ cc_WER_and nn__NIST conj_WER_ ' +o,4111,'2 Related Work A large amount of previous research on clustering has been focused on how to find the best clusters ',Brown,'num_Work_2 amod_Work_Related det_amount_A amod_amount_large nsubjpass_focused_amount prep_amount_of amod_research_previous pobj_of_research prep_research_on pobj_on_clustering aux_focused_has auxpass_focused_been dep_Work_focused prep_focused_on advmod_find_how aux_find_to pcomp_on_find det_clusters_the amod_clusters_best dobj_find_clusters ' +o,4112,'Thus , we propose a bootstrapping approach to train the stochastic transducer iteratively as it extracts transliterations from a bitext ',Yarowsky,'advmod_propose_Thus nsubj_propose_we det__a amod__bootstrapping nn__approach nsubj_train_ aux_train_to xcomp_propose_train det_transducer_the amod_transducer_stochastic dobj_train_transducer advmod_train_iteratively mark_transliterations_as dep_extracts_it nsubj_transliterations_extracts advcl_train_transliterations prep_transliterations_from det_bitext_a pobj_from_bitext ' +n,4113,'12Poon and Domingos outperformed ',Haghighi,'nsubj_outperformed_12Poon cc_12Poon_and nn__Domingos conj_12Poon_ ' +o,4114,'Nevertheless , in the problem described in this article , the source and the target sentences are given , and we are focusing on the optimization of the aligment a The translation probability Pr -LRB- f , a e -RRB- can be rewritten as follows : Pr -LRB- f , a e -RRB- = Jproductdisplay j = 1 Pr -LRB- fj , aj fj11 , aj11 , eI1 -RRB- = Jproductdisplay j = 1 Pr -LRB- aj fj11 , aj11 , eI1 -RRB- Pr -LRB- fj fj11 , aj1 , eI1 -RRB- -LRB- 2 -RRB- The probability Pr -LRB- f , a e -RRB- can be estimated by using the word-based IBM statistical alignment models ',Brown,'advmod_given_Nevertheless prep_given_in det_problem_the pobj_in_problem partmod_problem_described prep_described_in det_article_this pobj_in_article det_source_the nsubjpass_given_source cc_source_and det_sentences_the nn_sentences_target conj_source_sentences auxpass_given_are cc_given_and nsubj_focusing_we aux_focusing_are conj_given_focusing prep_focusing_on det_optimization_the pobj_on_optimization prep_optimization_of det_a_the amod_a_aligment pobj_of_a det_Pr_The nn_Pr_translation nn_Pr_probability nsubjpass_rewritten_Pr dep_Pr_f det_e_a appos_f_e aux_rewritten_can auxpass_rewritten_be rcmod_a_rewritten mark_follows_as advcl_rewritten_follows nsubj_=_Pr dep_=_f det_e_a appos_f_e ccomp_follows_= nn_j_Jproductdisplay nsubj_=_j ccomp_=_= num_Pr_1 dobj_=_Pr appos_Pr_fj nn_aj11_aj nn_aj11_fj11 dep_fj_aj11 dep_fj_eI1 dep_Pr_= nn_j_Jproductdisplay iobj_=_j number_1_= num_Pr_1 nn_Pr_Pr nn_fj11_aj appos_Pr_fj11 dep_fj11_aj11 dep_fj11_eI1 dobj_=_Pr nn_fj11_fj appos_Pr_fj11 dep_fj11_aj1 dep_fj11_eI1 dep_estimated_2 det_Pr_The nn_Pr_probability nsubjpass_estimated_Pr dep_Pr_f det_e_a appos_f_e aux_estimated_can auxpass_estimated_be dep_given_estimated prep_estimated_by pcomp_by_using det_models_the amod_models_word-based nn_models_IBM nn_models_statistical nn_models_alignment dobj_using_models ' +o,4115,'To closely reproduce the experiment with the best performance carried out in using SVM , we use unigram with the presence feature ',Pang,'aux_reproduce_To advmod_reproduce_closely dep_use_reproduce det_experiment_the dobj_reproduce_experiment prep_experiment_with det_performance_the amod_performance_best pobj_with_performance partmod_performance_carried prt_carried_out prep_carried_in nsubj_using_ pcomp_in_using dobj_using_SVM nsubj_use_we dobj_use_unigram prep_use_with det_feature_the nn_feature_presence pobj_with_feature ' +o,4116,'The literature on relational similarity , on the other hand , has focused on pairs of words , devising various methods to compare how similar the contexts in which target pairs appear are to the contexts of other pairs that instantiate a relation of interest ',Turney,'det_literature_The nsubj_focused_literature prep_literature_on amod_similarity_relational pobj_on_similarity prep_literature_on det_hand_the amod_hand_other pobj_on_hand aux_focused_has prep_focused_on pobj_on_pairs prep_pairs_of pobj_of_words xcomp_focused_devising amod_methods_various dobj_devising_methods aux_compare_to xcomp_devising_compare advmod_are_how amod_contexts_similar det_contexts_the nsubj_are_contexts rel_appear_in pobj_in_which nn_pairs_target nsubj_appear_pairs rcmod_contexts_appear ccomp_compare_are prep_are_to det_contexts_the pobj_to_contexts prep_contexts_of amod_pairs_other pobj_of_pairs nsubj_instantiate_that rcmod_contexts_instantiate det_relation_a dobj_instantiate_relation prep_relation_of pobj_of_interest ' +o,4117,'For process -LRB- 2 -RRB- , existing methods aim to distinguish between subjective and objective descriptions in texts ',Pang,'prep_aim_For pobj_For_process appos_process_2 amod_methods_existing nsubj_aim_methods aux_distinguish_to xcomp_aim_distinguish prep_distinguish_between amod_descriptions_subjective cc_subjective_and conj_subjective_objective pobj_between_descriptions prep_distinguish_in pobj_in_texts ' +o,4118,'25 Model Training We adapt the Minimum Error Rate Training -LRB- MERT -RRB- algorithm to estimate parameters for each member model in co-decoding ',Och,'num_Training_25 nn_Training_Model nsubj_adapt_We dep_Training_adapt det_algorithm_the nn_Training_Minimum nn_Training_Error nn_Training_Rate nn_algorithm_Training abbrev_Training_MERT nn_algorithm_ dobj_adapt_algorithm aux_estimate_to xcomp_adapt_estimate dobj_estimate_parameters prep_estimate_for det_model_each nn_model_member pobj_for_model prep_model_in pobj_in_co-decoding ' +o,4119,'In Step 3 , a simple perceptron update is performed ',Collins,'prep_performed_In pobj_In_Step num_Step_3 det__a amod__simple nn__perceptron nn__update nsubjpass_performed_ auxpass_performed_is ' +p,4120,'Phrases extracted using these heuristics are also shown to perform better than syntactically motivated phrases , the joint model , and IBM model 4 ',Koehn,'nsubjpass_shown_Phrases partmod_Phrases_extracted xcomp_extracted_using det_heuristics_these dobj_using_heuristics auxpass_shown_are advmod_shown_also aux_perform_to xcomp_shown_perform advmod_perform_better prep_perform_than advmod_motivated_syntactically amod_phrases_motivated pobj_than_phrases det_model_the amod_model_joint conj_phrases_model cc_phrases_and nn_model_IBM conj_phrases_model num_model_4 ' +o,4121,'Identifying subjectivity helps separate opinions from fact , which may be useful in question answering , summarization , etc Sentiment detection is the task of determining positive or negative sentiment of words , phrases and sentences , or documents ',Turney,'advmod_Identifying_subjectivity dep_Identifying_helps amod_opinions_separate dobj_helps_opinions prep_helps_from pobj_from_fact nsubj_useful_which aux_useful_may cop_useful_be rcmod_fact_useful prep_useful_in nn_answering_question pobj_in_answering nsubj_Identifying_summarization nn_detection_etc nn_detection_Sentiment appos_summarization_detection cop_task_is det_task_the rcmod_detection_task prep_task_of pcomp_of_determining amod_sentiment_positive cc_positive_or conj_positive_negative dobj_determining_sentiment prep_sentiment_of nn__words pobj_of_ conj__phrases cc_phrases_and nn__sentences conj_phrases_ cc__or conj__documents ' +o,4122,'ROUGE-S ROUGE-S is an extension of ROUGE-2 defined as follows (Lin, 2004b): ROUGE-Sa59a61a146a31a62a98a147a49a65a68a67 a59a68a101a161a128a104a162 a2 a65a161a163 a157 a134a61a135a93a245a246 a2 a59a61a146a31a62a98a147a49a65a161a163 a145 a134a61a135a89a245a246 a2 a59a61a146a31a62a164a147a49a65 a157 a134a136a135a93a245a246 a2 a59a61a146a31a62a90a147a49a65a51a128a104a162 a2 a145 a134a61a135a89a245a246 a2 a59a61a146a31a62a98a147a49a65 (11) Where a166a168a169a78a170a248a247a250a249 a26 and a171a138a169a90a170a158a247a250a249 a26 are defined as follows: a251 a134a61a135a89a245a246 a2 a59a61a146a31a62a90a147a49a65a68a67 a252a248a253a85a254a255 a1 a59a61a146a31a62a90a147a49a65 # of skip bigram a2a23a147 (12) a3 a134a136a135a93a245a246 a2 a59a61a146a31a62a90a147a49a65a68a67 a252a83a253a118a254a255 a1 a59a61a146a31a62a90a147a49a65 # of skip bigram a2 a146 (13) Here, function Skip2 returns the number of skipbi-grams that are common to a141 and a139 . ROUGE-SU ROUGE-SU is an extension of ROUGE-S, which includes unigrams as a feature defined as follows (Lin, 2004b): ROUGE-SUa59a61a146a31a62a90a147a49a65a68a67 a59a68a101a161a128a49a162 a2 a65a117a163 a157 a134a5a4 a59a61a146a31a62a98a147a49a65a71a163 a145 a134a6a4 a59a61a146a31a62a98a147a49a65 a157 a134a5a4 a59a61a146a31a62a90a147a49a65a47a128a49a162 a2 a145 a134a5a4 a59a61a146a31a62a164a147a49a65 (14) Where a166 a169a8a7 and a171 a169a8a7 are defined as follows: a251 a134a5a4 a59a61a146a31a62a98a147a49a65a68a67 a252 a9 a59a61a146a31a62a90a147a49a65 (# of skip bigrams + # of unigrams) a2 a147 (15) a3 a134a5a4 a59a61a146a31a62a90a147a49a65a68a67 a252 a9 a59a61a146a31a62a90a147a49a65 (# of skip bigrams + # of unigrams) a2 a146 (16) Here, function SU returns the number of skip-bigrams and unigrams that are common to a141 and a139 . ROUGE-L ROUGE-L is an LCS-based evaluation measure defined as follows (Lin, 2004b): ROUGE-La59a61a146a31a62a90a147a49a65a68a67 a59a68a101a161a128a49a162 a2 a65a161a163 a157a11a10 a225a90a134 a59a61a146a31a62a90a147a49a65a161a163 a145a12a10 a225a90a134 a59a61a146a31a62a98a147a49a65 a157a11a10 a225a90a134 a59a61a146a31a62a90a147a49a65a47a128a49a162 a2 a145a12a10 a225a98a134 a59a61a146a31a62a90a147a49a65 (17) where a166a14a13a250a241a132a169 and a171a15a13a250a241a130a169 are defined as follows: a157a11a10 a225a98a134 a59a61a146a31a62a98a147a49a65a68a67 a101 a91 a16 a75 a77a29a216 LCSa17a244a59a61a156 a88 a62a90a146a21a65 (18) a145a18a10 a225a98a134 a59a61a146a31a62a98a147a49a65a68a67 a101 a95 a16 a75a78a77a83a216 LCSa17 a59a61a156a34a88a78a62a98a146a21a65 (19) Here, LCSa19a244a28a78a144a183a114a93a32a93a139a102a36 is the LCS score of the union longest common subsequence between reference sentences a144a25a114 and a139 . a115 and a122 are the number of words contained in a141, and a139, respectively.',Lin,'' +o,4123,'145 2 The Latent Variable Architecture In this section we will begin by briefly introducing the class of graphical models we will be using , Incremental Sigmoid Belief Networks ',Titov,'number_2_145 det_Architecture_The nn_Architecture_Latent nn_Architecture_Variable dep_2_Architecture dep_2_In det_section_this pobj_In_section nsubj_begin_we aux_begin_will rcmod_section_begin prep_begin_by advmod_introducing_briefly pcomp_by_introducing det_class_the dobj_introducing_class prep_class_of amod_models_graphical pobj_of_models nsubj_using_we aux_using_will aux_using_be rcmod_class_using nn_Networks_Incremental nn_Networks_Sigmoid nn_Networks_Belief dep_2_Networks ' +p,4124,'Recently , various works have improved the quality of statistical machine translation systems by using phrase translation ',Koehn,'advmod_improved_Recently amod_works_various nsubj_improved_works aux_improved_have det_quality_the dobj_improved_quality prep_quality_of amod_systems_statistical nn_systems_machine nn_systems_translation pobj_of_systems prep_improved_by pcomp_by_using nn_translation_phrase dobj_using_translation ' +o,4125,'The segmentation is based on the guidelines , given in the Chinese national standard GB13715 , and the POS tagging specification was developed according to the Grammatical Knowledge-base of contemporary Chinese ',Marcus,'det_segmentation_The nsubjpass_based_segmentation auxpass_based_is prep_based_on det_guidelines_the pobj_on_guidelines prep_based_given dep_given_in det_GB13715_the amod_GB13715_Chinese amod_GB13715_national amod_GB13715_standard pobj_in_GB13715 conj_GB13715_ cc_GB13715_and det_POS_the conj_GB13715_POS amod_specification_tagging nsubjpass_developed_specification auxpass_developed_was dep_based_developed prep_developed_according dep_according_to det_Knowledge-base_the nn_Knowledge-base_Grammatical pobj_to_Knowledge-base prep_Knowledge-base_of amod_Chinese_contemporary pobj_of_Chinese ' +o,4126,'For example , in this work we use loglikelihood ratio to determine the SoA between a word sense and co-occurring words , and cosine to determine the distance between two DPWSs log likelihood vectors ',Dunning,'prep_use_For pobj_For_example prep_example_in det_work_this pobj_in_work nsubj_use_we amod__loglikelihood nn__ratio nsubj_determine_ aux_determine_to xcomp_use_determine det_SoA_the dobj_determine_SoA prep_SoA_between det_sense_a nn_sense_word pobj_between_sense cc_sense_and amod_words_co-occurring conj_sense_words cc_determine_and conj_determine_cosine aux_determine_to xcomp_cosine_determine det_distance_the dobj_determine_distance prep_distance_between num__two nn__DPWSs amod__log nn__likelihood nn__vectors pobj_between_ ' +o,4127,'For example , the words test and exam are similar because both of them can follow verbs such as administer , cancel , cheat on , conduct , etc Many methods have been proposed to compute distributional similarity between words , eg , ',Hindle,'prep_similar_For pobj_For_example det_test_the nn_test_words nsubj_similar_test cc_test_and conj_test_exam cop_similar_are mark_follow_because nsubj_follow_both prep_both_of pobj_of_them aux_follow_can ccomp_similar_follow dobj_follow_verbs dep_as_such prep_verbs_as pobj_as_administer dep_follow_cancel dep_follow_cheat prt_cheat_on dep_follow_conduct dobj_follow_etc amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been rcmod_etc_proposed aux_compute_to xcomp_proposed_compute amod_similarity_distributional dobj_compute_similarity prep_similarity_between pobj_between_words appos_words_eg ' +o,4128,'25 Evaluation Minnen and Carroll -LRB- Under review -RRB- report an evaluation of the accuracy of the morphological generator with respect to the CELEX lexical database ',Marcus,'num_Minnen_25 nn_Minnen_Evaluation nsubj_report_Minnen cc_Minnen_and conj_Minnen_Carroll dep_Carroll_Under pobj_Under_review det_evaluation_an dobj_report_evaluation prep_evaluation_of det_accuracy_the pobj_of_accuracy prep_accuracy_of det_generator_the amod_generator_morphological pobj_of_generator prep_generator_with pobj_with_respect prep_report_to det__the nn__CELEX amod__lexical nn__database pobj_to_ ' +o,4129,'We use two state-of-the-art POS taggersa maximum entropy based English POS tagger , and an HMM based Chinese POS tagger ',Ratnaparkhi,'nsubj_use_We num_maximum_two amod_maximum_state-of-the-art dep_maximum_POS nn_maximum_taggersa nsubj_based_maximum advmod_based_entropy ccomp_use_based nn__English dep__POS nn__tagger dobj_based_ cc_use_and det_HMM_an nsubj_based_HMM conj_use_based amod_tagger_Chinese dep_tagger_POS dobj_based_tagger ' +o,4130,'We used a publicly available tagger to provide the part-of-speech tags for each word in the sentence ',Ratnaparkhi,'nsubj_used_We det__a advmod_available_publicly amod__available nn__tagger dobj_used_ aux_provide_to infmod__provide det_tags_the amod_tags_part-of-speech dobj_provide_tags prep_provide_for det_word_each pobj_for_word prep_word_in det_sentence_the pobj_in_sentence ' +o,4131,'1 Introduction The use of various synchronous grammar based formalisms has been a trend for statistical machine translation -LRB- SMT -RRB- ',Wu,'num_Introduction_1 det_use_The nsubj_been_use prep_use_of amod_grammar_various amod_grammar_synchronous pobj_of_grammar partmod_grammar_based dobj_based_formalisms aux_been_has dep_Introduction_been det_trend_a dobj_been_trend prep_trend_for amod_translation_statistical nn_translation_machine pobj_for_translation abbrev_translation_SMT ' +p,4132,'He has achieved state-of-the art results by applying ME to parsing , part-of-speech tagging , and sentence-boundary detection ',Ratnaparkhi,'nsubj_achieved_He aux_achieved_has amod_results_state-of-the nn_results_art dobj_achieved_results prep_achieved_by pcomp_by_applying dobj_applying_ME prep_applying_to amod__parsing pobj_to_ amod__part-of-speech amod__tagging conj__ cc__and amod_detection_sentence-boundary conj__detection ' +o,4133,'The closest work is that of Jing and McKeown and , in which multiple sentences are processed , with fragments within them being recycled to generate the novel generated text ','Daume III','det_work_The amod_work_closest nsubj_that_work cop_that_is prep_that_of pobj_of_Jing cc_Jing_and conj_Jing_McKeown amod_Jing_ cc__and conj__ rel_processed_in pobj_in_which amod_sentences_multiple nsubj_processed_sentences cop_processed_are rcmod_Jing_processed prep_processed_with pobj_with_fragments prep_fragments_within nsubjpass_recycled_them auxpass_recycled_being pcomp_within_recycled aux_generate_to xcomp_recycled_generate det_novel_the dobj_generate_novel partmod_novel_generated dobj_generated_text ' +n,4134,'As with similar work , the size of the corpus makes preprocessing such as lemmatization , POS tagging or partial parsing , too costly ',Brown,'prep_makes_As dep_As_with amod__similar nn__work pobj_with_ det_size_the nsubj_makes_size prep_size_of det_corpus_the pobj_of_corpus dobj_makes_preprocessing dep_as_such prep_preprocessing_as pobj_as_lemmatization dep_tagging_POS conj_lemmatization_tagging cc_lemmatization_or amod_parsing_partial conj_lemmatization_parsing advmod_costly_too amod_preprocessing_costly ' +p,4135,'There also have been prior work on maintaining approximate counts for higher-order language models -LRB- LMs -RRB- -LRB- -RRB- operates under the model that the goal is to store a compressed representation of a disk-resident table of counts and use this compressed representation to answer count queries approximately ',Talbot,'expl_been_There advmod_been_also aux_been_have dep_operates_been amod_work_prior dobj_been_work prep_work_on pcomp_on_maintaining amod_counts_approximate dobj_maintaining_counts prep_maintaining_for amod_models_higher-order nn_models_language pobj_for_models abbrev_models_LMs appos_models_ prep_operates_under det_model_the pobj_under_model complm_is_that det_goal_the nsubj_is_goal ccomp_operates_is aux_store_to xcomp_is_store det_representation_a amod_representation_compressed dobj_store_representation prep_representation_of det_table_a amod_table_disk-resident pobj_of_table prep_table_of pobj_of_counts cc_store_and conj_store_use det_representation_this amod_representation_compressed dobj_use_representation aux_answer_to xcomp_use_answer nn_queries_count dobj_answer_queries nsubj_operates_approximately ' +p,4136,'By default , the log-likelihood ratio measure -LRB- LLR -RRB- is proposed , since it was shown to be particularly suited to language data ',Dunning,'prep_proposed_By pobj_By_default det_measure_the amod_measure_log-likelihood nn_measure_ratio nsubjpass_proposed_measure abbrev_measure_LLR auxpass_proposed_is mark_shown_since nsubjpass_shown_it auxpass_shown_was advcl_proposed_shown aux_suited_to auxpass_suited_be advmod_suited_particularly xcomp_shown_suited prep_suited_to nn_data_language pobj_to_data ' +o,4137,'2 Literature Survey The task of sentiment analysis has evolved from document level analysis -LRB- eg , ; -RRB- to sentence level analysis -LRB- eg , ; ; -RRB- ',Turney,'num_Survey_2 nn_Survey_Literature det_task_The nsubj_evolved_task prep_task_of nn_analysis_sentiment pobj_of_analysis aux_evolved_has dep_Survey_evolved prep_evolved_from nn_analysis_document nn_analysis_level pobj_from_analysis appos_analysis_eg nn__ dep_eg_ prep_evolved_to nn_analysis_sentence nn_analysis_level pobj_to_analysis nn__eg dep_analysis_ dep__ dep__ ' +o,4138,'Thus , a lot of alignment techniques have been suggested at ; the sentence , phrase , nomt t -RRB- hrase , word , collocation and terminology level ',Brown,'advmod_suggested_Thus det_lot_a nsubjpass_suggested_lot prep_lot_of amod_techniques_alignment pobj_of_techniques aux_suggested_have auxpass_suggested_been prep_suggested_at det__the nn__sentence pobj_at_ nn__phrase conj__ amod__nomt nn__t amod__hrase conj__ nn__word conj__ nn__collocation conj__ cc__and amod_level_terminology conj__level ' +o,4139,'The class based disambiguation operator is the Mutual Conditioned Plausibility -LRB- MCPI -RRB- ',Marcus,'det_class_The partmod_class_based nn_operator_disambiguation nsubj_Mutual_operator cop_Mutual_is det_Mutual_the ccomp_based_Mutual nn__Conditioned nn__Plausibility abbrev__MCPI dep_class_ ' +o,4140,'Parsing has been also used after extraction for filtering out invalid results ',Smadja,'aux_used_has auxpass_used_been advmod_used_also dep_Parsing_used prep_used_after pobj_after_extraction nsubj_Parsing_ prep__for amod_results_filtering dep_invalid_out amod_results_invalid pobj_for_results ' +o,4141,'The significance values are obtained using the loglikelihood measure assuming a binomial distribution for the unrelatedness hypothesis ',Dunning,'det_values_The nn_values_significance nsubjpass_obtained_values auxpass_obtained_are xcomp_obtained_using det_measure_the amod_measure_loglikelihood dobj_using_measure partmod_measure_assuming det_distribution_a amod_distribution_binomial dobj_assuming_distribution prep_distribution_for det_hypothesis_the amod_hypothesis_unrelatedness pobj_for_hypothesis ' +o,4142,'Ramshaw and Marcus views chunking as a tagging problem ',Ramshaw,'nn_views_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nn_views_ dep_chunking_views prep_chunking_as det_problem_a amod_problem_tagging pobj_as_problem ' +o,4143,'This approach will generally take advantage of language-specific -LRB- eg in -RRB- and domain-specific knowledge , of any external resources -LRB- eg database , names dictionaries , etc -RRB- , and of any information about the entities to process , eg their type -LRB- person name , organization , etc -RRB- , or internal structure -LRB- eg in -RRB- ',Freeman,'det_approach_This nsubj_take_approach aux_take_will advmod_take_generally dobj_take_advantage prep_advantage_of pobj_of_language-specific dep_language-specific_eg prep_eg_in pobj_in_ cc_language-specific_and amod_knowledge_domain-specific conj_language-specific_knowledge prep_advantage_of det_resources_any amod_resources_external pobj_of_resources dep_resources_eg dep_eg_database nn_dictionaries_names appos_eg_dictionaries dep_eg_etc cc_advantage_and conj_advantage_of det_information_any pobj_of_information prep_information_about det_entities_the pobj_about_entities prep_entities_to pobj_to_process appos_process_eg poss_type_their dep_advantage_type dep_type_person dep_person_name appos_person_organization dep_person_etc cc_type_or amod_structure_internal conj_type_structure dep_structure_eg prep_eg_in pobj_in_ ' +o,4144,'6 Bracketing of Compound Nouns The first analysis task we consider is the syntactic disambiguation of compound nouns , which has received a fair amount of attention in the NLP literature ',Marcus,'num_Bracketing_6 nsubj_Nouns_Bracketing prep_Bracketing_of pobj_of_Compound det_task_The amod_task_first nn_task_analysis dobj_Nouns_task nsubj_consider_we dep_Nouns_consider cop_disambiguation_is det_disambiguation_the amod_disambiguation_syntactic ccomp_consider_disambiguation prep_disambiguation_of amod_nouns_compound pobj_of_nouns nsubj_received_which aux_received_has rcmod_nouns_received det_amount_a amod_amount_fair dobj_received_amount prep_amount_of pobj_of_attention prep_received_in det__the nn__NLP nn__literature pobj_in_ ' +o,4145,'7 Related Work There has been a recent interest in training methods that enable the use of first-order features ','Daume III','num_Work_7 amod_Work_Related expl_been_There aux_been_has dep_Work_been det_interest_a amod_interest_recent dobj_been_interest prep_interest_in nn_methods_training pobj_in_methods nsubj_enable_that rcmod_interest_enable det_use_the dobj_enable_use prep_use_of amod_features_first-order pobj_of_features ' +o,4146,'Ontologies are formal specifications of a conceptualization so that it seems straightforward to formalize annotation schemes as ontologies and make use of semantic annotation tools such as OntoMat for the purpose of linguistic annotation ',Marcus,'nsubj_specifications_Ontologies cop_specifications_are amod_specifications_formal prep_specifications_of det__a nn__conceptualization pobj_of_ advmod_straightforward_so dep_straightforward_that nsubj_straightforward_it cop_straightforward_seems dep_specifications_straightforward aux_formalize_to xcomp_straightforward_formalize nn_schemes_annotation dobj_formalize_schemes prep_formalize_as pobj_as_ontologies cc_formalize_and conj_formalize_make dobj_make_use prep_use_of amod_tools_semantic nn_tools_annotation pobj_of_tools dep_as_such prep_tools_as nn__OntoMat pobj_as_ prep__for det_purpose_the pobj_for_purpose prep_purpose_of amod_annotation_linguistic pobj_of_annotation ' +o,4147,'Attempts to alleviate this tagbottleneck i ~ lude tmotstr ~ ias and unsupervised algorith ~ -LRB- , 199s -RRB- Dictionary-based approaches rely on linguistic knowledge sources such as ma ~ l ~ i , ~ e-readable dictionaries and WordNet and e0 -LRB- ploit these for word sense disaznbiguation ',Yarowsky,'aux_alleviate_to xcomp_Attempts_alleviate det_lude_this nn_lude_tagbottleneck nn_lude_i nn_lude_~ dobj_alleviate_lude nn__tmotstr nn__~ nn__ias nsubj_Attempts_ cc__and amod_~_unsupervised nn_~_algorith conj__~ appos__ dep__199s amod_approaches_Dictionary-based dep__approaches rcmod_approaches_rely prep_rely_on amod_sources_linguistic nn_sources_knowledge pobj_on_sources dep_as_such prep_sources_as pobj_as_ma amod_l_~ tmod_rely_l xcomp_rely_~ dobj_~_i nn__~ amod__e-readable nn__dictionaries conj_i_ cc_i_and nn__WordNet conj_i_ cc_i_and conj_i_e0 dep_e0_ploit dep_ploit_these prep_ploit_for nn_disaznbiguation_word nn_disaznbiguation_sense pobj_for_disaznbiguation ' +o,4148,' and Chan et al ',Carpuat,'cc__and conj__Chan cc_Chan_et conj_Chan_al ' +o,4149,'Feature selection methods have been proposed in the maximum-entropy literature by several authors ',Berger,'amod_methods_Feature nn_methods_selection nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been prep_proposed_in det_literature_the amod_literature_maximum-entropy pobj_in_literature prep_proposed_by amod_authors_several pobj_by_authors ' +o,4150,'Prototype-drive learning specifies prior knowledge by providing a few prototypes -LRB- ie , canonical example words -RRB- for each label ',Haghighi,'amod__Prototype-drive nn__learning nsubj_specifies_ amod_knowledge_prior dobj_specifies_knowledge prep_specifies_by pcomp_by_providing det_prototypes_a amod_prototypes_few dobj_providing_prototypes dep_prototypes_ie amod_words_canonical nn_words_example appos_ie_words prep_prototypes_for det_label_each pobj_for_label ' +o,4151,'Probabilities based on relative frequencies , or derived fl ` om the measure defined in , for example , allow to take this fact into account ',Dunning,'prep_Probabilities_based dep_based_on amod_frequencies_relative pobj_on_frequencies cc_Probabilities_or amod_fl_derived conj_Probabilities_fl dep_Probabilities_om det_measure_the dobj_om_measure partmod_measure_defined prep_defined_in pobj_in_ prep_defined_for pobj_for_example dep_om_allow aux_take_to xcomp_allow_take det_fact_this dobj_take_fact prep_take_into pobj_into_account ' +o,4152,'There have been considerable amount of efforts to improve the reordering model in SMT systems , ranging from the fundamental distance-based distortion model , flat reordering model , to lexicalized reordering model , hierarchical phrase-based model , and maximum entropy-based phrase reordering model ',Koehn,'expl_been_There aux_been_have amod_amount_considerable dobj_been_amount prep_amount_of pobj_of_efforts aux_improve_to infmod_efforts_improve det_model_the nn_model_reordering dobj_improve_model prep_model_in nn_systems_SMT pobj_in_systems xcomp_improve_ranging prep_ranging_from det__the amod__fundamental amod__distance-based nn__distortion nn__model pobj_from_ amod__flat nn__reordering nn__model appos__ prep_ranging_to amod__lexicalized nn__reordering nn__model pobj_to_ amod__hierarchical amod__phrase-based nn__model conj__ cc__and amod__maximum amod__entropy-based nn__phrase nn__reordering nn__model conj__ ' +o,4153,'ISBNs , originally proposed for constituent parsing in , use vectors of binary latent variables to encode information about the parse history ',Titov,'nsubj_use_ISBNs advmod_proposed_originally partmod_ISBNs_proposed prep_proposed_for amod_parsing_constituent pobj_for_parsing prep_parsing_in pobj_in_ nsubj_encode_vectors prep_vectors_of advmod_latent_binary amod_variables_latent pobj_of_variables aux_encode_to xcomp_use_encode dobj_encode_information prep_encode_about det_history_the amod_history_parse pobj_about_history ' +o,4154,'The same Powells method has been used to estimate feature weights of a standard feature-based phrasal MT decoder in ',Och,'det_method_The amod_method_same nn_method_Powells nsubjpass_used_method aux_used_has auxpass_used_been aux_estimate_to xcomp_used_estimate nn_weights_feature dobj_estimate_weights prep_weights_of det_decoder_a amod_decoder_standard amod_decoder_feature-based nn_decoder_phrasal nn_decoder_MT pobj_of_decoder prep_estimate_in ' +p,4155,'1 Introduction Very large corpora obtained from the Web have been successfully utilized for many natural languageprocessing -LRB- NLP -RRB- applications , suchasprepositional phrase -LRB- PP -RRB- attachment , other-anaphora resolution , spellingcorrection , confusablewordsetdisambiguation and machine translation ',Brants,'num_corpora_1 nn_corpora_Introduction nn_corpora_Very amod_corpora_large nsubjpass_utilized_corpora partmod_corpora_obtained prep_obtained_from det_Web_the pobj_from_Web aux_utilized_have auxpass_utilized_been advmod_utilized_successfully prep_utilized_for amod_applications_many amod_applications_natural nn_applications_languageprocessing abbrev_applications_NLP pobj_for_applications amod_attachment_suchasprepositional nn_attachment_phrase abbrev_attachment_PP appos_applications_attachment amod_translation_other-anaphora nn_translation_resolution conj_resolution_spellingcorrection conj_resolution_confusablewordsetdisambiguation cc_resolution_and conj_resolution_machine appos_applications_translation ' +o,4156,'Parameter tuning is done with Minimum Error Rate Training -LRB- MERT -RRB- ',Och,'nn_tuning_Parameter nsubjpass_done_tuning auxpass_done_is prep_done_with nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_with_Training abbrev_Training_MERT ' +o,4157,'The lexical scores are computed as the -LRB- unnormalized -RRB- log probability of the Viterbi alignment for a phrase pair under IBM word-translation Model 1 ',Brown,'det_scores_The amod_scores_lexical nsubjpass_computed_scores auxpass_computed_are prep_computed_as det_probability_the appos_probability_unnormalized nn_probability_log pobj_as_probability prep_probability_of det_alignment_the nn_alignment_Viterbi pobj_of_alignment prep_computed_for det_pair_a nn_pair_phrase pobj_for_pair prep_computed_under nn_Model_IBM nn_Model_word-translation pobj_under_Model num_Model_1 ' +o,4158,'Previous work has shown it to be appropriate to large-scale language modeling ',Brants,'amod__Previous nn__work nsubj_shown_ aux_shown_has nsubj_appropriate_it aux_appropriate_to cop_appropriate_be xcomp_shown_appropriate prep_appropriate_to amod_modeling_large-scale nn_modeling_language pobj_to_modeling ' +o,4159,'To perform minimum error rate training to tune the feature weights to maximize the systems BLEU score on development set , we used optimizeV5IBMBLEUm ',Och,'aux_perform_To dep_used_perform amod__minimum nn__error nn__rate nn__training dobj_perform_ aux_tune_to xcomp_perform_tune det_weights_the nn_weights_feature nsubj_maximize_weights aux_maximize_to xcomp_tune_maximize det_score_the nn_score_systems nn_score_BLEU dobj_maximize_score prep_maximize_on nn_set_development pobj_on_set nsubj_used_we amod__optimizeV5IBMBLEUm dobj_used_ ' +o,4160,'These lists are rescored with the different models described above , a character penalty , and three different features based on IBM Models 1 and 2 calculated in both translation directions ',Brown,'det_lists_These nsubjpass_rescored_lists auxpass_rescored_are prep_rescored_with det_models_the amod_models_different pobj_with_models partmod_models_described prep_described_above det_penalty_a nn_penalty_character conj_models_penalty cc_models_and num_features_three amod_features_different conj_models_features partmod_features_based prep_based_on nn_Models_IBM pobj_on_Models num_Models_1 cc_Models_and num__2 conj_Models_ partmod_features_calculated prep_calculated_in det_directions_both nn_directions_translation pobj_in_directions ' +o,4161,'what does student want to write your Figure 3 : A derivation tree of lexicalized parse trees , such as the distinction of arguments\\/modifiers and unbounded dependencies , are elegantly represented in derivation trees ',Collins,'dobj_want_what aux_want_does nsubj_want_student aux_write_to xcomp_want_write poss_Figure_your nsubjpass_represented_Figure dep_Figure_3 det_tree_A nn_tree_derivation dep_Figure_tree prep_tree_of amod_trees_lexicalized nn_trees_parse pobj_of_trees dep_as_such prep_Figure_as det_distinction_the pobj_as_distinction prep_distinction_of pobj_of_arguments\\/modifiers cc_arguments\\/modifiers_and amod__unbounded nn__dependencies conj_arguments\\/modifiers_ auxpass_represented_are advmod_represented_elegantly ccomp_write_represented prep_represented_in nn_trees_derivation pobj_in_trees ' +o,4162,'Within the machine learning paradigm , IL has been incorporated as a technique for bootstrapping an extensional learning algorithm , as in ',Yarowsky,'prep_incorporated_Within det_machine_the nsubj_learning_machine pcomp_Within_learning dobj_learning_paradigm nsubjpass_incorporated_IL aux_incorporated_has auxpass_incorporated_been prep_incorporated_as det_technique_a pobj_as_technique prep_incorporated_for pcomp_for_bootstrapping det_algorithm_an amod_algorithm_extensional nn_algorithm_learning dobj_bootstrapping_algorithm prep_bootstrapping_as dep_as_in ' +o,4163,'Algorithm 1 The RRM Decoding Algorithm foreacha26a29a27a67a42 foreacha68 a1a20a23a69a10a11a10a12a10a45 a60 a48a22a70a26a22a71 a1a73a72a2a25 a57a38a50 a7 a56 a48a54a57 a64a74a30 a57 a31a33a26a17a34 a5a11a75 a60a77a76a74a76 a31a78a26a35a34a66a79a81a80a83a82a38a84a69a85a86a80a24a87a88a48 a60 a48 a70a26a61a71 Somewhat similarly, the MaxEnt algorithm has an associated set of weights a31a33a89 a48a54a57 a34a48a90a50 a7a53a52a54a52a54a52a15 a57a38a50 a7a58a52a54a52a54a52 a25, which are estimated during the training phase so as to maximize the likelihood of the data (Berger et al. , 1996).',Berger,'' +o,4164,'By treating a letter\\/character as a word and a group of letters\\/characters as a phrase or token unit in SMT , one can easily apply the traditional SMT models , such as the IBM generative model or the phrase-based translation model to transliteration ',Brown,'prep_apply_By pcomp_By_treating det_letter\\/character_a dobj_treating_letter\\/character prep_treating_as det_word_a pobj_as_word cc_word_and det_group_a conj_word_group prep_group_of pobj_of_letters\\/characters prep_letters\\/characters_as det_phrase_a pobj_as_phrase cc_letters\\/characters_or amod_unit_token conj_letters\\/characters_unit prep_unit_in pobj_in_SMT nsubj_apply_one aux_apply_can advmod_apply_easily det_models_the amod_models_traditional nn_models_SMT dobj_apply_models dep_as_such prep_models_as det__the nn__IBM amod__generative nn__model pobj_as_ cc__or det__the amod__phrase-based nn__translation nn__model conj__ prep_apply_to pobj_to_transliteration ' +o,4165,'lscript1-regularized log-linear models -LRB- lscript1-LLMs -RRB- , on the other hand , provide sparse solutions , in which weights of irrelevant features are exactly zero , by assumingaLaplacianpriorontheweight ',Goodman,'amod_models_lscript1-regularized amod_models_log-linear nsubj_provide_models appos_models_lscript1-LLMs prep_models_on det_hand_the amod_hand_other pobj_on_hand amod_solutions_sparse dobj_provide_solutions rel_zero_in pobj_in_which nsubj_zero_weights prep_weights_of amod_features_irrelevant pobj_of_features cop_zero_are advmod_zero_exactly rcmod_solutions_zero prep_zero_by pobj_by_assumingaLaplacianpriorontheweight ' +o,4166,'Two major research topics in this field are Named Entity Recognition -LRB- NER -RRB- and Word Sense Disambiguation -LRB- WSD -RRB- ',Yarowsky,'num_topics_Two amod_topics_major nn_topics_research nsubjpass_Named_topics prep_topics_in det_field_this pobj_in_field auxpass_Named_are nn__Entity nn__Recognition abbrev__NER dobj_Named_ cc__and nn_Disambiguation_Word nn_Disambiguation_Sense conj__Disambiguation abbrev_Disambiguation_WSD ' +p,4167,'To estimate the parameters of the MEMM + pred model we turn to the successful Maximum Entropy parameter estimation method ',Berger,'aux_estimate_To det_parameters_the dobj_estimate_parameters prep_parameters_of det_model_the amod_model_MEMM cc_MEMM_+ conj_MEMM_pred pobj_of_model nsubj_turn_we rcmod_model_turn prep_turn_to det_method_the amod_method_successful nn_method_Maximum nn_method_Entropy nn_method_ nn_method_parameter nn_method_estimation pobj_to_method ' +p,4168,'For the full parser , we use the one developed by Michael Collins one of the most accurate full parsers around ',Collins,'prep_use_For det_parser_the amod_parser_full pobj_For_parser nsubj_use_we det_one_the dobj_use_one partmod_one_developed prep_developed_by nn__Michael nn__Collins pobj_by_ num__one prep__of det_parsers_the advmod_accurate_most amod_parsers_accurate amod_parsers_full pobj_of_parsers advmod_parsers_around ' +o,4169,'Finally , other approaches rely on reviews with numeric ratings from websites and train -LRB- semi - -RRB- supervised learning algorithms to classify reviews as positive or negative , or in more fine-grained scales ',Pang,'advmod_rely_Finally amod_approaches_other nsubj_rely_approaches prep_rely_on pobj_on_reviews prep_reviews_with amod_ratings_numeric pobj_with_ratings prep_ratings_from pobj_from_websites dobj_rely_ cc__and conj__train dep_-LRB-_semi partmod__supervised xcomp_supervised_learning dobj_learning_algorithms aux_classify_to xcomp_learning_classify nsubj_positive_reviews advmod_positive_as xcomp_classify_positive cc_positive_or conj_positive_negative cc__or conj__in amod_scales_more amod_scales_fine-grained pobj_in_scales ' +p,4170,'This algorithm appears fairly widely known : it was described by Goodman and and used by Ding et al -LRB- 2005 -RRB- , and is very similar to other dynamic programming algorithms for CFGs , so we only summarize it here ',Finkel,'det_algorithm_This nsubj_known_algorithm dep_known_appears advmod_known_fairly advmod_known_widely nsubjpass_described_it auxpass_described_was parataxis_known_described prep_described_by nn__Goodman pobj_by_ cc__and conj__ cc_described_and conj_described_used prep_used_by pobj_by_Ding cc_Ding_et conj_Ding_al appos_Ding_2005 cc_described_and cop_similar_is advmod_similar_very conj_described_similar prep_similar_to amod_algorithms_other amod_algorithms_dynamic nn_algorithms_programming pobj_to_algorithms prep_algorithms_for pobj_for_CFGs dep_described_so nsubj_summarize_we advmod_summarize_only dep_described_summarize dobj_summarize_it advmod_summarize_here ' +o,4171,'We have : -RRB- -LRB- -RRB- , -LRB- -RRB- , -LRB- -RRB- , , -LRB- -RRB- -LRB- 21 21 trictrictric trictritri erpercpercp ecrcpecp = = -LRB- 6 -RRB- Assumption 2 : For an English triple tri e , assume that i c only depends on -LCB- 1,2 -RCB- -RRB- -LRB- i i e , and c r only depends on e r Equation -LRB- 6 -RRB- is rewritten as : -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- , -LRB- -RRB- , -LRB- -RRB- -LRB- 2211 21 ec trietrictrictritri rrpecpecp erpercpercpecp = = -LRB- 7 -RRB- Notice that -RRB- -LRB- 11 ecp and -RRB- -LRB- 22 ecp are translation probabilities within triples , they are different from the unrestricted probabilities such as the ones in IBM models ',Brown,'nsubj_have_We number_21_21 num_trictritri_21 amod_trictritri_trictrictric dep_-RRB-_trictritri nn_=_erpercpercp nn_=_ecrcpecp nn_=_= nsubj_assume_= appos_=_6 dep_=_Assumption num_Assumption_2 prep_=_For det_tri_an amod_tri_English amod_tri_triple pobj_For_tri dep_For_e rcmod_trictritri_assume complm_depends_that nn_c_i nsubj_depends_c advmod_depends_only ccomp_assume_depends prep_trictritri_on pobj_on_1,2 nn_e_i nn_e_i nsubj_r_e cc_e_and conj_e_c dep_-RRB-_r advmod_depends_only dep_r_depends prt_depends_on dep_r_e dep_rewritten_r nsubjpass_rewritten_Equation appos_Equation_6 auxpass_rewritten_is dep_depends_rewritten prep_rewritten_as number_21_2211 num_erpercpercpecp_21 nn_erpercpercpecp_ec nn_erpercpercpecp_trietrictrictritri nn_erpercpercpecp_rrpecpecp dep_-LRB-_erpercpercpecp nn_Notice_= nn_Notice_= appos_Notice_7 dep_erpercpercpecp_Notice dep_erpercpercpecp_that appos_-LRB-_11 cc_11_ecp conj_11_and num_ecp_22 nsubj_probabilities_ecp cop_probabilities_are nn_probabilities_translation rcmod_-LRB-_probabilities prep_probabilities_within pobj_within_triples nsubj_different_they cop_different_are dep_have_different prep_different_from det_probabilities_the amod_probabilities_unrestricted pobj_from_probabilities dep_as_such prep_probabilities_as det_ones_the pobj_as_ones prep_ones_in nn_models_IBM pobj_in_models ' +o,4172,'Towards a Meaning-Full Comparison of Lexieal Resources Kenneth C Lltkowska CL Research 9208 Gue Road Damascus, MD 20872 ken@clres corn http//www tires tom Abstract The mapping from WordNet to Hector senses m Senseval provides a \'gold standard\' against wluch to judge our ability to compare lexlcal resources The \'gold standard\' is provided through a word overlap analysis (with and without a stop list) for flus mapping, achieving at most a 36 percent correct mapping (inflated by 9 percent from \'empty\' assignments) An alternaUve componenttal analysis of the defimtaons, using syntacUc, collocatmnal, and semantac component and relation identification (through the use ofdefimng patterns integrated seamlessly mto the parsing thclaonary), provides an almost 41 percent correct mapping, with an additaonal 4 percent by recogmzmg semantic components not used in the Senseval mapping Defimtion sets of the Senseval words from three pubhshed thclaonanes and Dorr\'s lextcal knowledge base were added to WordNet and the Hector database to exanune the nature of the mapping process between defimtton sets of more and less sco[~e The tecbauques described here consUtute only an maaal implementation of the componenUal analysis approach and suggests that considerable further improvements can be aclueved Introduction The difficulty of companng lemcal resources, long a s~gnfficant challenge in computauonal hnguistlcs (Atlans, 1991), came to the fore in the recent Senseval competatton (IOlgarnff, 1998), when some systems that relied heavily on the WordNet (Miller, et al, 1990) sense inventory were faced with the necessity of using another sense inventory (Hecto0 A hasty solutaon to the problem was the \' development of a map between the two inventories, but some part~cipants expressed concerns that use of flus map may have degraded their performance to an unknown degree Although there were disclaimers about the WordNet-Hector map, it nonetheless stands as a usable gold standard for efforts to compare lexical resources Moreover, we have a usable baseline (a word overlap method suggested m (Lesk, 1986)) against which to compare whether we are able to make improvements m the mapping (since flus method has been shown to perform not as well as expected (Krovetz, 1992)) We first describe the lextcal resources used m the study (Hector, WordNet, other dicUonanes, and a lex~cal knowledge base), first characterizing them in terms ofpolysemy and the types of leracal mformaUon each contmns (syntacUc properties and features, semantac components and relaUons, and collocaUonal properties) We then present results of perfornung the word overlap analysis of the 18 verbs used m Senseval, analyzing the definitions m WordNet and Hector We then expand our analysis to include other dictionaries We describe our methods of analysis, particularly the methods of parsing defimtaons and identff)qng semantic relations (semrels) based on defimng patterns, essentially takang first steps m Implementing the program described by Atkms and focusmg on the use of\'meamng\' full mformataon rather than statistical mformaUon We identify the results that have been achieved thus far and outline further steps that may add more \'meanmg\' to the analysis IAll analyses described m this paper were performed automatically using functlonahty incorporated m DIMAP (Dictionary Maintenance Programs) (available for immediate download at (CL Research, 1999a)) This includes automatac extracuon of WordNet reformation for the selected words (mtegrated m DIMAP) Hector defimtlons were uploaded into DIMAP dicUonanes after use of a conversmn program Defimtlons for other 30 The Lexical Resources Tlus analysis focuses on the mmn verb senses used In Senseval (not ichoms and phrases), specifically the followmg AMAZE, BAND, BET, BOTHER, BURY, CALCULATE, CONSUME, DERIVE, FLOAT, HURDLE, INVADE, PROMISE, SACK, SANCTION, SCRAP, SEIZE, SHAKE, SLIGHT The Hector database used In Senseval consists of a tree of senses, each of which contains defimttons, syntactic properties, example usages, and \'clues\' (collocational information about the syntactic and semantic enwronment in wluch a word appears in the spectfic sense) The WordNet database contmns synonyms (synsets), perhaps a defimtton or example usages (gloss), some syntactic mformaUon (verb frames), hypernyms, hyponyms, and some other semrels (ENTAILS, CAUSES) To extend our analysis In order to look at other issues of lexacal resource comparison, we have included the defirauons or leracal information from the following additional sources Webster\'s 3 ra New International Dictionary (W3) Oxford Advanced l.earners D~ctlonary (OALD) American Hentage DlcUonary (AI-ID) Dorr\'s Lexacal Knowledge Base (Dorr) We used only the defimuons from W3, OALD, and AHD (which also contmn sample usages and some collocattonal information m the form of usage notes, not used at the present tame) Dorr\'s database contains thematic grids wluch characterize the thematic roles of obligatory and optional semanuc components, frequently identifying accompanying preposmons (Olsen, et al, 1998) The following table identities the number of senses and average overall polysemy for each of these resources dictionaries were entered by hand Word amaze band bet bother bury calculate consume denve float hurdle invade pronuse sack sanction scrap seize shake shght Average Polysemy o o o 1 2 4 2 3 1 II 4 4 2 5 5 7 6 9 7 12 6 14 5 5 5 10 9 6 6 8 8 6 5 15 5 16 4 41 14 2 1 4 3 6 2 10 5 5 4 7 4 4 4 6 3 2 2 5 2 3 1 3 3 11 6 21 13 8 8 37 17 1 1 6 3 O 1 2 2 4 1 3 4 4 8 1 3 1 3 1 3 2 10 5 1 0 3 1 3 2 2 0 1 1 1 0 7 1 7 12 I 0 57 37 120 62 34 22 Word Overlap Analysis We first estabhsh a baseline for automatic replication of the lexicographer\'s mappmg from WordNet 1 6 to Hector, using a s~mple word overlap analysis smular to (Lesk, 1986) The lextcographer mapped the 66 WordNet senses (each synset m which a test occurred) Into 102 Hector senses A total of 86 assignments were made, 9 WordNet senses were gwen no assignments, 40 recewed exactly one, and 17 senses received 2 or 3 asssgnments The WordNet senses contained 348 words (about half of wluch were common words appeanng on our stop list, which contained 165 words, mostly preposmons, pronouns, and conjunctions) The Hector senses selected m the word overlap analysis contained about 960 words (all Hector senses contained 1878 words) We performed a strict word overlap analysts (with and wsthout a stop hst) between tile definlUons in WordNet and the Hector senses, that is, we did not attempt to ldenttfy root forms of Inflected words We took each word m a WordNet sense and determined whether ~t appeared in a Hector sense, we selected a Hector sense based on the highest percentage of words over all Hector senses An 31 empty selection was made ff all the words in the WordNet sense did not appear in any Hector sense, only content words were considered when the stop hst was used For example, for bet, WordNet sense 2 (stake (money) on the outcome of an issue) mapped into Hector sense 4 ((of a person) to risk (a sum of money or property) m thts way) In this case, there was an overlap on two words (money, 039 in the Hector defimtlon (0 13 of its 15 words) without the stop list When the stop list was invoked, there was an overlap of only one word (money, 0 07 of the Hector defimtion) In this case, the lexicographer had made three assignments (Hector senses 2, 3, and 4), our scoring method treated flus as only 1 out of 3 correct (not using the relaxed method employed in Senseval of treating flus as completely correct) Without the stop hst, our selections matched the lexicographer\'s in 28 of 86 cases (32 6\%), using the stop list, we were successful in 31 of 86 cases (36 1\%) The improvement arising when the stop list was used is deceptive, where 8 cases were due to empty assignments (so that only 23 cases, 26 7\%, were due to matching content words) Overall, only 41 content words were involved in these 23 successes when the stop list was used, an average of I 8 content words To summanze the word overlap analysis (1) despite a ncher set of defimtions in Hector, 9 of 66 WordNet senses (13 6\%) could not be assigned, (2) despite the greater detail in Hector senses compared to WordNet senses (2 8 times as many words), only 1 8 content words participated in the assignments, and (3) therefore, the defimng vocabulary between these two definition sets seems to be somewhat divergent Although it might appear as if the word overlap analysis does not perform well, this is not the case The analysis provides a broad overview of the defimuon companson process between two definmon sets and frames a deeper analysis of the differences Moreover, it appears that the accuracy of a \'gold standard\' mapping is not crucially important The quality of the mapping may help frame the subsequent analysis more precisely, but it seems sufficient that any reasonable mapping will suffice This will be discussed further after presenting the results of the componentlal analysis of the defimtlons 32 Meaning-Full Analysis of Definitions The deeper analysis of the mapping between two defimtion sets relies primarily on two major steps (1) parsing definitions and using defimng patterns to identify semrels present m the definitions and (2) relaxing values to these relations by allowing \'synonymic\' substitution (using WordNet) Thus, for example, ffwe identify hypernyms or instruments from parsing a defimtion, we would say that the defimtions are \'equal\' not just ffthe hypernym or instrument is the same word, but also Lf the hypernyms or instruments are members of the same synset This approach is based on the finding (Litkowski, 1978) that a dictionary induces a semantic network where nodes represent \'concepts\' that may be lexicahzed and verbalized in more than one way This finding implies, in general, the absence of true synonyms, and instead the kind of \'concept\' embodied in WordNet synsets (with several lexical items and phraseologles) A slmdar approach, parsing defimtlons and relaxing semrel values, was followed in (Dolan, 1994) for clnstenng related senses w~thin a single dictionary The ideal toward which this approach strives is a complete identification of the meamng components included in a defimtion The meaning components can include syntactic features and charactenstlcs (including subcategonzation patterns), semantm components (realized through identification of semrels), selectional restrictions, and coUocational specifications The first stage of the analysis parses the definitions (CL Research, 1999b, Litkowski, to appear) and uses the parse results to extract (via defining patterns) semrels Since definitions have many idiosyncrasies (that do not follow ordinary text), an important first step in this stage is preprocessmg the definition text to put it into a sentence frame that facilitates the extraction of semrels 2 2Note that the stop hst is not applicable to the definition parsing The parser is a full-scale sentence parser, where prepositmns and other words on the stop list are necessary for successful parsing Moreover, inclusion of the prepositions is cmcml to the method, since they are the bearers of much semrel information The extractmn of semrels examines the parse results, a e, a tree whose mtermedaate nodes represent non-ternunals and whose leaves represent the lextcal atems that compnse the defimuons, where any node may also include annotations such as characterizations of number and tense For all noun or verb defimttons, flus includes Identification of the head noun (with recogmtton of\'empty\' heads) or verb, for verbs, we signal whether the defimtaon contmned any selecttonal restnctmus (that as, pamcular parenthesazed expressaons) for the subject and object We then exanune preposattonal phrases In the defimUon and deterrmne whether we have a \'defining pattern\' for the preposaUon whach we can use as mdacaUve of a partacular semrel We also identify adverbs m the parse tree and look these up in WordNet to adentffy an adjecuve synset from wluch they are derived (if one is gwen) The defimng pattems are actually part of the dictionary used by the parser That is, we do not have to develop specafic routines to look for specLfic patterns A defimng pattern ~s a regular expressaon that arlaculates a syntactac pattern to be matched Thus, to recograze a \'manner\' semrel, we have the foUowmg entry for \'m\' m(dpat((~ rep0 l(det(0)) adj manner(0) st(manner)))) This allows us to recognize \'m\' as possibly gwmg rise to a \'manner\' component, where we recogmze \'m\' (the tdde, which allows us to specify partacular elements before the \'m\' as well), vath a noun phrase that consasts of 0 or 1 determiner, an adjectwe, and the lateral \'manner\' The \'0 after the detenmner and the hteral mdacate that these words are not copied into the value for a \'manner\' role, so that the value to the \'manner\' semrel becomes only the adjectwe that as recogmzed The second stage of the analysis uses the populated lexacal database to compare senses and make the selectaons This process follows the general methodology used m Senseval (Lltkowska, to appear) Specifically, m the defimtaon comparison, we first exanune exclusaon cntena to rule out specific mappings These criteria include syntacUc properUes (e g, a verb sense that Is only transluve cannot map into one that Is only mtransRave) and collocataonal propertaes (e g, a sense that is used with a parUcle cannot map into one that uses a different particle) At the present tune, these are used only rmmmally 33 We next score each viable sense based on rots semrels We increment the score ff the senses have a common hypernym or If a sense\'s hypernyms belong to the same synset as the other sense\'s hypernyms If a parUcular sense con~ns a large number of synonyms (that as, no differentiae on the hypernym) and they overlap consaderably m the synsets they evoke, the score can be increased substanUally Currently, we add 5 points for each match 3 We increment the score based on common semrels In tins amtml tmplementaUon, we have defimng patterns (usually qmte nummal) for recogmzmg Instrument, means, location, purpose, source, manner, has-constituents, has-members, is-part-of, locale, and goal 4 We Increment the score by 2 points when we have a common semrel and then by another 5 points when the value Is ~dentacal or m the same synset After all possable increments to the scores have been made, we then select the sense(s) w~th the lughest score Finally, we compare our selecuon with that of the gold standard to assess our mapping over all senses Another way an wluch our methodology follows the Senseval process as that at proceeds incrementally Thus, ~t ms not necessary to have a \'final\' perfect parse and mapping rouUne We can make conUnual refinements at any stage of the process and exarmne the overall effect As m Senseval, we may make changes to deal wath a particular phenomenon with the result that overall performance dechnes, but w~th a sounder basis for making subsequent amprovements Results of Componential Analysis The \'gold standard\' analysis Involves mapping 66 WordNet senses with 348 words into 102 Hector senses with 1878 words Using the method described above, we obtained 35 out of 86 correct 3At the present tame, we use WordNet to adentffy semreis We envaslon usmg the full semanlac network created by parsing all a dlcUonary\'s defimtaons Thas would include a richer set of semrels than currently included m WordNet 4The defimng patterns are developed by hand We have onlyJust begun this effort, so the current set ms somewhat Impoverished mappmgs (407\%), a shght improvement over the 31 correct assignments usmg the stop-last word overlap techmque However, as mentioned above, the stophst techmque had aclueved 8 of its successes by matclung null assignments Consadered on tlus basins, ~t seems that the componentaal analysis techmque provides substantial ~mprovement In addition, our technique \'erred\' on 4 cases by malang assagnments where none were made by the leracographer We suggest that these cases do con~n some common elements of meaning and may conceivably not be construed as errors The mapping from WordNet to Hector had relatavely few empty mappings, senses for wtuch It was not possable to make an assignment These are the cases where at appears that the chetmnanes do not overlap and thus prowde a tentative mdacataon of where two dictionaries may have different coverage The cases of multiple assignments mchcate the degree ofamblgmty m the mapping The average m both darecUons between Hector and WordNet were donunated by the mabdaty to obtain good dascnnunatton for the word \'semze\' Thus, tlus method identifies individual words where the &scnnunatwe ablhty needs to be further refined Perhaps more importantly, the componentml analysis method exploits consaderably more WordNet Hector mformauon than the word overlap methods Whereas the stop-hst word overlap mapping was based on only 41 content words, the componenual ~ approach (In the selected mappings) had 228 hits in ~.~ developing ats scores, with only a small number of ~ .~ ~ defining patterns Comparison of Dictionaries tel O ~3 0\'3 We next exanuned the nature of the mterrelalaons between parrs of chctaonanes w~thout use of a \'gold standard\' to assess the process of mapping For t/us purpose, we mapped m both &recttons between the paars {WordNet, Hector}, {W3, OALD}, and {W3, AHD We exanune Dorr\'s lexacal knowledge base for the amphcatlons It may have m the mapping process Neither WordNet nor Hector are properly v~ewed as chcuonanes, since there was no mtenuon to pubhsh them as such WordNet \'glosses\' are generally smaller (53 words per sense) compared to Hector (184 words per sense), whach contains many words specff3nng selectmnal restnct~ons on the subject and object of the verbs Hector was used primarily for a large-scale sense tagging project The three formal d~ctmnanes were subject to rigorous pubhslung and style standards The average number of words per sense were 87 (OALD), 7 1 (AHD), and 9 9 (W3), w~th an average of 3 4, 62, and 120 senses per word Each table shows the average number of senses being mapped, the average number of assignments m the target dlCtmnary, the average number of senses for which no assagnment could be made, the average number of mulUple assignments per word, and the average score of the assignments that were made WN-Hector 37 47 06 17 119 Hector-WN 57 64 14 22 113 These points are further emphasized m the mapping between W3 and OALD, where the disparity between the empty and mulUple assagnments indicate that we are mapping between dictionaries qmte disparate This tends to be the case not only for the enUre set of words, but also is evident for individual words where there is a considerable d~spanty m the number of senses, wtuch then dominate the overall dlspanty Thus, for example, W3 has 41 defimUons for \'float\', while OALD has 10 We tend to be unable to find the specific sense m going from W3 to OALD, because at is likely that we have many more specific defimtlons that are not present In the other direction, we are hkely to have considerable ambiguity and multiple assignments W3-OALD OALD-W3 W3 OALD 120 78 60 18 99 34 60 07 32 86 34 A Between W3 and AHD, there ss less overall daspanty between the defimtaon sets, although since W3 Is tmabndged, we stall have a relatavely lugh number of senses m W3 that do not appear to be present m AHD Finally, It should be noted that the scores for the published dictaonanes tend to be a little lower than for WordNet and Hector Tlus reflects the hkehhood that we have not extracted as much mformataon as we dad m parsing and analyzmg the defimtaon sets used m Senseval W3 AHD oJ \'q O W3-AHD 120 115 40 36 90 AHD-W3 6 2 9 1 1 2 4 1 9 1 We next considered Dorr\'s lexacal database We first transformed her theta grids to syntactic spectflcataons (transttave or lntransmttve) and identtficataon of semreis (e g, where she Identified an instr component, we added such a semrel to the DIMAP sense) We were able to identify a mappmg from WordNet to her senses for two words (\'float\' and \'shake\') for wluch Dorr has several entries However, smce she has considerably more semanuc components than we are currently able to recogmze, we dad not pursue this avenue any further at flus time More important than just mappmg between two words, Dorr\'s data mdacates the posstbday of further exploitation of a richer set of semanUc components Spectfically, as reported m (Olsen, et al, 1998), m descnbmg procedures for automatically acqumng thematic grids for Mandann Chinese, ~t was noted that \'verbs that incorporate themaUc elements m their meamng would not allow that element to appear m the complement structure\' Thus, by usmg Dorr\'s thematic grids when verb are parsed m defimtaons, it ~s possible to ~dentffy where partacular semantac components are lexicahzed and which others are transnutted through to the themaUc grid (complement or subcategonzataon pattern) for the defimendum The transmiss~on of semantic components to the thematic gnd ~s also reflected overtly m many defimtlons For example, shake has one definition, \'to bnng to a specified condatton by or as ffby repeated qmck jerky movements\' We would thus expect that the thematac grid for this defimtaon should include a \'goal\' And, deed, Dorr\'s database has two senses whch reqmre a \'goal\' as part of their thematic grid Smularly, for many defimtaons m the sample set, we ~dentLfied a source defimng pattern based on the word \'from,\' frequently, the object of the preposmon was the word \'source\' ttseff, mdacatmg that the subcategonzaUon, properties of the defimendum should elude a source component Discussion Wlule the improvement m mapping by using the componentaal analysis techmque (over the word overlap methods) is modest, we consider these results qmte slgmficant m wew of the very small number of defimng patterns we have Implemented Most of the improvement stems from the word substatuUon pnnclple described earlier (as ewdenced by the preponderance of 5 point scores) This techmque also provides a mechamsm for bnngmg back the stop words, wz, the preposmons, wluch are the careers of mformatmn about semrels (the 2 point scores) The more general conclusion (from the word subsutuuon) is that the success arises from no longer considenng a defimtmn m ~solation The proper context for a word and its defimtions consists not .lUSt of the words that make up the definition, but also the total semantac network represented by the dictaonary We have aclueved our results by explomng only a small part of that network We have moved only a few steps to that network beyond the mdawdual words and their definitions We would expect that further expansmn, first by the addon of further and ~mproved semrel defining patterns, and second, through the identaficataon of more pnmmve semanuc components, will add considerably to our abflay to map between lexacal resources We also expect ~mprovements from consideration of other techniques, such as attempts at ontology ahgnment (Hovy, 1998) Although tile definition analysis provlded here was performed on definmons with a stogie language, the vanous meamng components m m m m m m m m 35 correspond to those used in an Interhngua The use of the exUncuon method (developed m order to charactenze verbs m another language, Clunese) can frmtfully be applied here as well Two further observaUons about tlus process can be made The first is that rchance on a wellestablished semantic network such as WordNet,s not necessary The componenUal analysis method rehes on the local neighborhood of words m the defimUons, not on the completeness of the network Indeed, the network ~tsel can be bootstrapped based on the parsing results The method can work vath any semanUc network or ontology and may be used to refine or flesh out the network or ontology The second observation is that it is not necessary to have a well-estabhshed \'gold standard\' Any mapping vail do All that Is necessary is for any mvesugator (lemcographer or not) to create a judgmental mappmg The methods employed here can then quanufy ttus mapping based on a word overlap analysis and then further examine tt based on the componenaal analysis The componenUal analysis method can then be used to exanune underlying subtleUes and nuances tn the defimUous, wluch a lemcographer or analyst can then examine m further detail to assess the mapping Future Work Tlus work has marked the first ume that all the necessary mfrastructure has been combmed tn a rudimentary form Because of its rudimentary status, the opportumUes for improvement are quite extensive In addlUon, there are many opportumUes for using the techmques descnbed here m further NLP apphcatlons First, the techmques described here have immediate apphcabtllty as part of a lexicographer\'s workstaUon When defimUons are parsed and semrels are zdenttfied, the resulUng data structures can be apphed against a corpus of instances for parUcular words (as m Senseval) for improving word-sense disamblguaUon The techmques will also permit comparing an entry vath Itself to deternune the mterrelattonshtps among ~ts defimUons and of companng the defimUons of two \'synonyms\' to deternune the amount of overlap between them on a defimtlon by defimUon bas~s Although the analys,s here has focused on the parsing of defimUous, the development of defimng patterns clearly extends to generalized text parsing since the defimng patterns have been incorporated mto the same chcttonary used for parsing free text, the patterns can be used threctly to identify the presence of parUcular semrels among sentenual consUtuents We are working to integrate th~s funcUonahty into our word-sense &sambiguaUon techruques (both the defimng patterns and the semrels) Even further, mt seems that matclung defimng patterns in free text can be used for lextcal acquisition Textual matenal that contains these patterns could concewably be flagged as providing defimUonal matenal which can then be compared to emstmg defimUons to assess whether their use ts cous,stent vath these defimUons, and ff not, at least to flag the inconsistency The tecluuques descnbed here can be apphed directly to the fields of ontology development and analysis of ternunologlcal databases For ontoiogles, vath or w~thout defimuons, the methods employed can be used to compare entries m dai\'erent ontologles based pnmanly on the relattous m the ontology, both luerarclucal and other For ternunologlcal databases, the methods descnbed here can be used to exanune the set of conceptual relaUons lmphed by the defimtmus The defimuon parsing wall facd~tate the development of the termmolog~ca I network tn the pamcular field covered by the database The componenUal analysts methods result m a richer semantic network that can be used m other apphcattous Thus, for example, ~t ts possible to extend the leracal chatmng methods described m (Green, 1997), which are based on the semrels used m WordNet The semrels developed with the componenttal analysis method would provide additional detad available for apphcauon of lexlcal cohesion methods In particular, addtUonal relattous would penmt some structunng wmthm the individual leracal chams, rather than just consldenng each cham as an amorphous set (Green, 1999) Finally, we are currently investigating the use of the componenUal analysts techmque for mformauon extracUon The techmque identifies (from defimtlous) slots that can be used as slots or fields m template generataon Once these slots are identified, we wall be attemptmg to extract slot values from Items m large catalog databases (mdhons of items) 36 In conclusion, it would seem that, instead of a paucity of tnformation allovang us to compare lexmal resources, by bnngmg m the full semantic network of the lexicon, we are overwhelmed with a plethora of data Acknowledgments I would like to thank Bonnie Dorr, Chnstiane Fellbaum, Steve Green, Ed Hovy, Ramesh Knshnamurthy, Bob Krovetz, Thomas Potter, Lucy Vanderwende, and an anonymous reviewer for their comments on an earlier draft of this paper References Atlans, B T S (1991) Bmldmga lexicon The contribution of lexicography lnternattonal Journal of Lextcography, 4(3), 167-204 CL Research (1999a) CL Research Demos http//www clres com/Demo html CL Research (1999b) Dmtlonary Parsing Project http//www clres com/dpp html Dolan, W B (1994, 5-9 Aug) Word Sense Amblguation Chistenng Related Senses COLING-94, The 15th International Conference on Computational Linguistics Kyoto, Japan Green, S J (1997) Automatically generating hypertext by computing semantic smulanty [Dlss], Toronto, Canada Umverstty of Toronto Green, S J (Sjgreen@mn mq edu au) (1999, 1 June) (Rich semantic networks) Hovy, E (1998, May) Combining and Standardizing Large-Scale, Practical Ontologms for Machine Translation and Other Uses Language Resources and Evaluation Conference Granada, Spam Kalgarnff, A (1998) SENSEVAL Home Page http//www itn bton ac uk/events/senseval/ Krovetz, R (1992, June) Sense-Linking m a Machine Readable Dictionary 30th Annual Meeting of the Association for Computational Lmgu~stics Newark, Delaware Association for Computational Lmgtustics Lesk, M (1986) Automatic Sense Dlsamblguation Using Machine Readable Dmttonanes How to Tell a Pine Cone from an Ice Cream Cone Proceechngs of SIGDOC Lttkowski, K C (1978) Models of the semantic structure of dictionaries American Journal of Computattonal Lmgutsttcs, Atf 81, 25-74 Lttkowskl, K C (to appear) SENSEVAL The CL Research Expenence Computers and the Humamttes Mtller, G A, Beckwlth, R, Fellbaum, C, Gross, D, & Miller, K J (1990) Introduction to WordNet An on-hne lexical database lnternatwnal Journal of Lexicography, 3(4), 235-244 Olsen, M B, Dorr, B J, & Thomas, S C (1998, 28-31 October) Enhancmg Automatic Acqulsmon of Thematic Structure in a Large-Scale Lexacon for Mandann Chinese Tlurd Conference of the Association for Machine Translation m the Americas, AMTA-98 Langhorne, PA',Dolan,'' +o,4173,'1 Introduction In phrase-based statistical machine translation phrases extracted from word-aligned parallel data are the fundamental unit of translation ',Koehn,'num_Introduction_1 dep_Introduction_In amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_In_translation amod_phrases_ nsubj_unit_phrases partmod_phrases_extracted prep_extracted_from amod_data_word-aligned amod_data_parallel pobj_from_data cop_unit_are det_unit_the amod_unit_fundamental rcmod_translation_unit prep_unit_of pobj_of_translation ' +o,4174,'Some of these methods make use of prior knowledge in the form of an existing thesaurus , while others do not rely on any prior knowledge ',Marcus,'nsubj_make_Some prep_Some_of det_methods_these pobj_of_methods dobj_make_use prep_use_of amod_knowledge_prior pobj_of_knowledge prep_make_in det_form_the pobj_in_form prep_form_of det__an amod__existing nn__thesaurus pobj_of_ mark_rely_while nsubj_rely_others aux_rely_do neg_rely_not advcl_make_rely prep_rely_on det__any amod__prior nn__knowledge pobj_on_ ' +o,4175,'(Snow et al., 2006; Nakov & Hearst, 2008).',Snow,'' +p,4176,' -RRB- are best known and studied ',Brown,'nn_-RRB-_ auxpass_known_are advmod_known_best cc_known_and conj_known_studied ' +o,4177,'There is a vast literature on language modeling ; see , eg , ',Brants,'expl_is_There det_literature_a amod_literature_vast nsubj_is_literature prep_literature_on nn_modeling_language pobj_on_modeling rcmod_modeling_see ccomp_see_eg ' +p,4178,'Introduction The Penn Treebank initiated a new paradigm in corpus-based research ',Marcus,'nn__Introduction nn__The nn__Penn nn__Treebank nsubj_initiated_ det_paradigm_a amod_paradigm_new dobj_initiated_paradigm prep_initiated_in amod_research_corpus-based pobj_in_research ' +o,4179,'2 Related Work The Yarowsky algorithm , originally proposed for word sense disambiguation , makes the assumption that it is very unlikely for two occurrences of a word in the same discourse to have different senses ',Yarowsky,'num_Work_2 amod_Work_Related det__The nn__Yarowsky nn__algorithm nsubj_assumption_ advmod_proposed_originally partmod__proposed prep_proposed_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation dep_assumption_makes det_assumption_the dep_Work_assumption complm_unlikely_that nsubj_unlikely_it cop_unlikely_is advmod_unlikely_very ccomp_assumption_unlikely prep_unlikely_for num_occurrences_two pobj_for_occurrences prep_occurrences_of det_word_a pobj_of_word prep_word_in det_discourse_the amod_discourse_same pobj_in_discourse aux_have_to xcomp_unlikely_have amod_senses_different dobj_have_senses ' +o,4180,'The annotation consists of four parts : 1 -RRB- a context-free structure augmented with traces to mark movement and discontinuous constituents , 2 -RRB- phrasal categories that are annotated as node labels , 3 -RRB- a small set of grammatical functions that are annotated as extensions to the node labels , and 4 -RRB- part-of-speech tags ',Marcus,'det_annotation_The nsubj_consists_annotation prep_consists_of num_parts_four pobj_of_parts dep_augmented_1 det_structure_a amod_structure_context-free nsubj_augmented_structure parataxis_consists_augmented prep_augmented_with pobj_with_traces aux_mark_to xcomp_augmented_mark nn_constituents_movement cc_movement_and conj_movement_discontinuous dobj_mark_constituents dep_labels_2 amod_categories_phrasal nsubj_labels_categories nsubjpass_annotated_that auxpass_annotated_are rcmod_categories_annotated prep_annotated_as amod_labels_node pobj_as_labels dep_set_3 det_set_a amod_set_small appos_labels_set prep_set_of amod_functions_grammatical pobj_of_functions nsubjpass_annotated_that auxpass_annotated_are rcmod_functions_annotated prep_annotated_as pobj_as_extensions prep_annotated_to det_node_the pobj_to_node ccomp_consists_labels cc_consists_and dep_tags_4 nsubj_tags_part-of-speech conj_consists_tags ' +o,4181,'We report on ROUGE-1 -LRB- unigrams -RRB- , ROUGE-2 -LRB- bigrams -RRB- , ROUGE W-12 -LRB- weighted LCS -RRB- , and ROUGE-S \\* -LRB- skip bigrams -RRB- as they appear to correlate well with human judgments for longer multi-document summaries , particularly ROUGE-1 ',Lin,'nsubj_report_We prep_report_on pobj_on_ROUGE-1 appos_ROUGE-1_unigrams conj_ROUGE-1_ROUGE-2 appos_ROUGE-2_bigrams nn_W-12_ROUGE conj_ROUGE-1_W-12 amod_LCS_weighted abbrev_W-12_LCS cc_ROUGE-1_and nn_\\*_ROUGE-S conj_ROUGE-1_\\* num_bigrams_skip appos_\\*_bigrams mark_appear_as nsubj_appear_they advcl_report_appear aux_correlate_to xcomp_appear_correlate advmod_correlate_well prep_correlate_with amod_judgments_human pobj_with_judgments prep_judgments_for amod_summaries_longer amod_summaries_multi-document pobj_for_summaries advmod_ROUGE-1_particularly amod_judgments_ROUGE-1 ' +o,4182,'The word alignments were created with Giza + + applied to a parallel corpus containing the complete Europarl training data , plus sets of 4,051 sentence pairs created by pairing the test sentences with the reference translations , and the test sentences paired with each of the system translations ',Och,'det_alignments_The nn_alignments_word nsubjpass_created_alignments auxpass_created_were prep_created_with nn__Giza nn__+ nn__+ pobj_with_ partmod__applied prep_applied_to det_corpus_a amod_corpus_parallel pobj_to_corpus partmod_corpus_containing det_data_the amod_data_complete nn_data_Europarl nn_data_training dobj_containing_data cc_data_plus conj_data_sets prep_sets_of num_pairs_4,051 nn_pairs_sentence pobj_of_pairs partmod_pairs_created prep_created_by pcomp_by_pairing det_sentences_the nn_sentences_test dobj_pairing_sentences prep_pairing_with det_translations_the nn_translations_reference pobj_with_translations cc_data_and det_sentences_the nn_sentences_test conj_data_sentences partmod_sentences_paired prep_paired_with pobj_with_each prep_each_of det_translations_the nn_translations_system pobj_of_translations ' +o,4183,'6 Evaluation 61 Data The data used for our comparison experiments were developed as part of the OntoNotes project , which uses the WSJ part of the Penn Treebank ',Marcus,'num_Evaluation_6 num_data_61 nn_data_Data nn_data_The nsubjpass_developed_data partmod_data_used prep_used_for poss_experiments_our nn_experiments_comparison pobj_for_experiments auxpass_developed_were dep_Evaluation_developed prep_developed_as pobj_as_part prep_part_of det__the nn__OntoNotes nn__project pobj_of_ nsubj_uses_which rcmod__uses det_part_the nn_part_WSJ dobj_uses_part prep_part_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,4184,' and et al ',Blitzer,'cc__and conj__ cc__et conj__al ' +o,4185,'The system was trained in a standard manner , using a minimum error-rate training -LRB- MERT -RRB- procedure with respect to the BLEU score on held-out development data to optimize the loglinear model weights ',Och,'det_system_The nsubjpass_trained_system auxpass_trained_was prep_trained_in det_manner_a amod_manner_standard pobj_in_manner partmod_manner_using det_training_a amod_training_minimum amod_training_error-rate nn__training abbrev__MERT nn__procedure dobj_using_ prep__with pobj_with_respect prep_using_to det__the amod__BLEU nn__score pobj_to_ prep__on amod_data_held-out nn_data_development pobj_on_data aux_optimize_to xcomp_using_optimize det_weights_the amod_weights_loglinear nn_weights_model dobj_optimize_weights ' +o,4186,'Probabilistic generative models like IBM 1-5 (Brown et al., 1993), HMM (Vogel et al., 1996), ITG (Wu, 1997), and LEAF (Fraser and Marcu, 2007) define formulas for P(f | e) or P(e, f), with ok-voon ororok sprok at-voon bichat dat erok sprok izok hihok ghirok totat dat arrat vat hilat ok-drubel ok-voon anok plok sprok at-drubel at-voon pippat rrat dat ok-voon anok drok brok jok at-voon krat pippat sat lat wiwok farok izok stok totat jjat quat cat lalok sprok izok jok stok wat dat krat quat cat lalok farok ororok lalok sprok izok enemok wat jjat bichat wat dat vat eneat lalok brok anok plok nok iat lat pippat rrat nnat wiwok nok izok kantok ok-yurp totat nnat quat oloat at-yurp lalok mok nok yorok ghirok clok wat nnat gat mat bat hilat lalok nok crrrok hihok yorok zanzanok wat nnat arrat mat zanzanat lalok rarok nok izok hihok mok wat nnat forat arrat vat gat Figure 1: Word alignment exercise (Knight, 1997).',Brown,'' +o,4187,'For comparison to previous results , table 2 lists the results on the testing set for our best model -LRB- TOP-Efficient-Freq20 -RRB- and several other statistical parsers ',Collins,'prep_lists_For pobj_For_comparison prep_comparison_to amod_results_previous pobj_to_results nsubj_lists_table num_table_2 det_results_the dobj_lists_results prep_lists_on det_testing_the pobj_on_testing partmod_testing_set prep_set_for poss_model_our amod_model_best pobj_for_model abbrev_model_TOP-Efficient-Freq20 cc_model_and amod_parsers_several amod_parsers_other amod_parsers_statistical conj_model_parsers ' +o,4188,'2 Background 21 Phrase Table Extraction Phrasal decoders require a phrase table , which contains bilingual phrase pairs and 17 scores indicating their utility ',Koehn,'num_decoders_2 nn_decoders_Background num_decoders_21 nn_decoders_Phrase nn_decoders_Table nn_decoders_Extraction nn_decoders_Phrasal nsubj_require_decoders det__a nn__phrase nn__table dobj_require_ nsubj_contains_which rcmod__contains amod_pairs_bilingual nn_pairs_phrase nsubj_indicating_pairs cc_pairs_and num_scores_17 conj_pairs_scores dep_contains_indicating poss_utility_their dobj_indicating_utility ' +o,4189,'This set of words -LRB- rooted primarily in the verbs of the set -RRB- corresponds to the Characterize -LRB- class 292 -RRB- , Declare -LRB- 294 -RRB- , Admire -LRB- 312 -RRB- , and Judgment verbs -LRB- 33 -RRB- and hence may have particular syntactic and semantic patterning ',Dunning,'det_set_This nsubj_corresponds_set prep_set_of pobj_of_words dep_words_rooted advmod_rooted_primarily prep_rooted_in det_verbs_the pobj_in_verbs prep_verbs_of det_set_the pobj_of_set prep_corresponds_to det_Characterize_the nn_Characterize_ pobj_to_Characterize number_292_class dep_Characterize_292 conj_Characterize_Declare appos_Declare_294 conj_Characterize_Admire appos_Admire_312 cc_Characterize_and amod_verbs_Judgment conj_Characterize_verbs appos_verbs_33 cc_corresponds_and advmod_have_hence aux_have_may conj_corresponds_have amod_patterning_particular amod_patterning_syntactic cc_syntactic_and conj_syntactic_semantic dobj_have_patterning ' +o,4190,'212 Research on Syntax-Based SMT A number of researchers have proposed models where the translation process involves syntactic representations of the source and\\/or target languages ',Galley,'nn_Research_212 dep_number_Research prep_Research_on pobj_on_Syntax-Based dep_number_SMT det_number_A dep_have_number prep_number_of pobj_of_researchers dep_have_ amod_models_proposed dobj_have_models advmod_involves_where det_process_the nn_process_translation nsubj_involves_process advcl_have_involves amod_representations_syntactic dobj_involves_representations prep_representations_of det_source_the pobj_of_source cc_have_and\\/or conj_have_target dobj_target_languages ' +p,4191,' then extended their method and established a sound probabilistic model series , relying on different parameters describing how words within parallel sentences are aligned to each other ',Brown,'nsubj_extended_ advmod_extended_then poss_method_their dobj_extended_method cc_extended_and conj_extended_established det_series_a amod_series_sound amod_series_probabilistic nn_series_model dobj_established_series conj_extended_relying prep_relying_on amod_parameters_different pobj_on_parameters partmod_parameters_describing advmod_aligned_how nsubjpass_aligned_words prep_words_within amod_sentences_parallel pobj_within_sentences auxpass_aligned_are ccomp_describing_aligned prep_aligned_to det_other_each pobj_to_other ' +o,4192,'The reported results for the full parse tree -LRB- on section 23 -RRB- are recall\\/precision of 881\\/875 ',Collins,'det_results_The amod_results_reported nsubj_recall\\/precision_results prep_results_for det_tree_the amod_tree_full nn_tree_parse pobj_for_tree dep_tree_on pobj_on_section num_section_23 cop_recall\\/precision_are prep_recall\\/precision_of pobj_of_881\\/875 ' +o,4193,'1 Introduction Estimating the degree of semantic relatedness between words in a text is deemed important in numerous applications : word-sense disambiguation , story segmentation , error correction , summarization ',Pedersen,'num_Introduction_1 nsubjpass_deemed_Introduction partmod_Introduction_Estimating det_degree_the dobj_Estimating_degree prep_degree_of amod_relatedness_semantic pobj_of_relatedness prep_relatedness_between pobj_between_words prep_words_in det_text_a pobj_in_text auxpass_deemed_is acomp_deemed_important prep_important_in amod_applications_numerous pobj_in_applications amod__word-sense nn__disambiguation dep_important_ nn__story nn__segmentation appos__ nn__error nn__correction appos__ nn__summarization appos__ ' +o,4194,'Many studies and improvements have been conducted for Presently with Service Media Laboratory , Corporate ResearchandDevelopmentCenter , OkiElectricIndustry Co , Ltd POS tagging , and major methods of POS tagging achieve an accuracy of 9697 \% on the Penn Treebank WSJ corpus , but obtaining higher accuracies is difficult ',Ratnaparkhi,'amod_studies_Many nsubjpass_conducted_studies cc_studies_and conj_studies_improvements aux_conducted_have auxpass_conducted_been prep_conducted_for pobj_for_Presently prep_conducted_with nn_Laboratory_Service nn_Laboratory_Media pobj_with_Laboratory nn_ResearchandDevelopmentCenter_Corporate conj_Laboratory_ResearchandDevelopmentCenter nn_Co_OkiElectricIndustry conj_Laboratory_Co nn_tagging_Ltd dep_tagging_POS conj_Laboratory_tagging cc_Laboratory_and amod_methods_major conj_Laboratory_methods prep_methods_of pobj_of_POS csubj_difficult_tagging ccomp_tagging_achieve det_accuracy_an dobj_achieve_accuracy prep_accuracy_of num_\%_9697 pobj_of_\% prep_\%_on det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_WSJ pobj_on_corpus cc_tagging_but conj_tagging_obtaining amod_accuracies_higher dobj_obtaining_accuracies cop_difficult_is dep_conducted_difficult ' +o,4195,'4 Pattern switching The compositional translation presents problems which have been reported by : Fertility SWTs and MWTs are not translated by a term of a same length ',Dunning,'dep_presents_4 nsubj_presents_Pattern partmod_Pattern_switching det_translation_The amod_translation_compositional dobj_switching_translation dobj_presents_problems nsubjpass_reported_which aux_reported_have auxpass_reported_been rcmod_problems_reported prep_reported_by pobj_by_ nn_SWTs_Fertility nsubjpass_translated_SWTs cc_SWTs_and conj_SWTs_MWTs auxpass_translated_are neg_translated_not parataxis_presents_translated prep_translated_by det_term_a pobj_by_term prep_term_of det_length_a amod_length_same pobj_of_length ' +o,4196,'We attribute the difference in M3\\/4 scores to the fact we use a Viterbi-like training procedure -LRB- ie , we consider a single configuration of the hidden variables in EM training -RRB- while GIZA uses pegging to sum over a set of likely hidden variable configurations in EM ',Brown,'nsubj_attribute_We det_difference_the dobj_attribute_difference prep_difference_in nn_scores_M3\\/4 pobj_in_scores prep_attribute_to det_fact_the pobj_to_fact nsubj_use_we dep_attribute_use det_procedure_a amod_procedure_Viterbi-like nn_procedure_training dobj_use_procedure parataxis_use_ie nsubj_consider_we ccomp_ie_consider det_configuration_a amod_configuration_single dobj_consider_configuration prep_configuration_of det_variables_the amod_variables_hidden pobj_of_variables prep_variables_in nn_training_EM pobj_in_training mark_uses_while nsubj_uses_GIZA advcl_use_uses xcomp_uses_pegging nsubj_sum_ aux_sum_to xcomp_pegging_sum prt_sum_over det_set_a dobj_sum_set prep_set_of amod_configurations_likely amod_configurations_hidden amod_configurations_variable pobj_of_configurations prep_configurations_in pobj_in_EM ' +o,4197,'In particular , ROUGE-2 is the recall in bigrams with a set of human-written abstractive summaries ',Lin,'prep_recall_In pobj_In_particular nsubj_recall_ROUGE-2 cop_recall_is det_recall_the prep_recall_in pobj_in_bigrams prep_bigrams_with det_set_a pobj_with_set prep_set_of amod_summaries_human-written nn_summaries_abstractive pobj_of_summaries ' +o,4198,'There have been a number of methods proposed in the literature to address the word clustering problem -LRB- eg , -RRB- ',Brown,'expl_been_There aux_been_have det_number_a dobj_been_number prep_number_of pobj_of_methods partmod_methods_proposed prep_proposed_in det_literature_the pobj_in_literature aux_address_to infmod_literature_address det_problem_the nn_problem_word nn_problem_clustering dobj_address_problem appos_problem_eg dep_eg_ ' +o,4199,'Supervised methods include hidden Markov model -LRB- HMM -RRB- , maximum entropy , conditional random fields -LRB- CRF -RRB- , and support vector machines -LRB- SVM -RRB- ',Galley,'amod_methods_Supervised nsubj_include_methods amod_model_hidden amod_model_Markov dobj_include_model abbrev_model_HMM amod_fields_maximum amod_fields_entropy dep_entropy_conditional amod_fields_random conj_model_fields abbrev_fields_CRF cc_model_and nn_machines_support nn_machines_vector conj_model_machines abbrev_machines_SVM ' +o,4200,'As a final note , following , we used the averaged parameters from the training algorithm in decoding test examples in our experiments ',Collins,'prep_used_As det_note_a amod_note_final pobj_As_note prep_used_following pobj_following_ nsubj_used_we det_parameters_the amod_parameters_averaged dobj_used_parameters prep_used_from det_algorithm_the nn_algorithm_training pobj_from_algorithm prep_used_in pcomp_in_decoding nn_examples_test dobj_decoding_examples prep_decoding_in poss_experiments_our pobj_in_experiments ' +o,4201,'SO can be used to classify reviews -LRB- eg , movie reviews -RRB- as positive or negative , and applied to subjectivity analysis such as recognizing hostile messages , classifying emails , mining reviews ',Turney,'nsubjpass_used_SO aux_used_can auxpass_used_be aux_classify_to xcomp_used_classify dobj_classify_reviews appos_reviews_eg nn_reviews_movie dep_eg_reviews prep_reviews_as amod__positive cc_positive_or conj_positive_negative pobj_as_ cc_used_and conj_used_applied aux_subjectivity_to xcomp_applied_subjectivity dobj_subjectivity_analysis dep_as_such prep_analysis_as pcomp_as_recognizing amod_messages_hostile dobj_recognizing_messages dep_recognizing_classifying dobj_classifying_emails dep_recognizing_mining dobj_mining_reviews dobj_recognizing_ ' +p,4202,'On the other hand , the best available parsers trained on the Penn Treebank , those of and Charniak -LRB- 2000 -RRB- , use statistical models for disambiguation that make crucial use of dependency relations ',Collins,'prep_use_On det_hand_the amod_hand_other pobj_On_hand det_parsers_the dep_available_best amod_parsers_available nsubj_use_parsers partmod_parsers_trained prep_trained_on det_Treebank_the nn_Treebank_Penn pobj_on_Treebank appos_parsers_those prep_those_of pobj_of_ cc__and conj__Charniak appos__2000 amod_models_statistical dobj_use_models prep_models_for pobj_for_disambiguation nsubj_make_that rcmod_models_make amod_use_crucial dobj_make_use prep_use_of amod_relations_dependency pobj_of_relations ' +o,4203,'A boundary-based model of co-occurrence assumes that both halves of the bitext have been segmented into s segments, so that segment Ui in one half of the bitext and segment Vi in the other half are mutual translations, 1 < i < s. Under the boundary-based model of co-occurrence, there are several ways to compute co-occurrence counts cooc(u, v) between word types u and v. In the models of Brown, Della Pietra, Della Pietra, and Mercer (1993), reviewed in Section 4.3, s COOC(R, V) = ~ ei(u) .j~(V), (12) i=1 where ei and j5 are the unigram frequencies of u and v, respectively, in each aligned text segment i. For most translation models, this method produces suboptimal results, however, when ei(u) > 1 and )~(v) > 1.',Dunning,'' +o,4204,'We guess it is an acronym for the authors of : Michel Galley , Mark Hopkins , Kevin Knight and Daniel Marcu ',Galley,'nsubj_guess_We nsubj_acronym_it cop_acronym_is det_acronym_an ccomp_guess_acronym prep_acronym_for det_authors_the pobj_for_authors prep_authors_of pobj_of_ nn_Galley_Michel dep_acronym_Galley nn_Hopkins_Mark conj_Galley_Hopkins nn_Knight_Kevin conj_Galley_Knight cc_Galley_and nn_Marcu_Daniel conj_Galley_Marcu ' +n,4205,'The size of the development set used to generate 1 and 2 compensates the tendency of the unsmoothed MERT algorithm to overfit by providing a high ratio between number of variables and number of parameters to be estimated ',Och,'det_size_The nsubj_compensates_size prep_size_of det_set_the nn_set_development pobj_of_set partmod_set_used aux_generate_to xcomp_used_generate num__1 cc_1_and conj_1_2 dobj_generate_ det_tendency_the nsubj_overfit_tendency prep_tendency_of det_algorithm_the amod_algorithm_unsmoothed nn_algorithm_MERT pobj_of_algorithm aux_overfit_to xcomp_compensates_overfit dobj_overfit_ prep_overfit_by pcomp_by_providing det_ratio_a amod_ratio_high dobj_providing_ratio prep_ratio_between pobj_between_number prep_number_of pobj_of_variables cc_variables_and conj_variables_number prep_ratio_of pobj_of_parameters aux_estimated_to auxpass_estimated_be xcomp_providing_estimated ' +o,4206,'1 Introduction The rapid and steady progress in corpus-based machine translation has been supported by large parallel corpora such as the Arabic-English and Chinese-English parallel corpora distributed by the Linguistic Data Consortium and the Europarl corpus , which consists of 11 European languages ',Brown,'num_Introduction_1 det_progress_The amod_progress_rapid cc_rapid_and conj_rapid_steady nsubjpass_supported_progress prep_progress_in amod__corpus-based nn__machine nn__translation pobj_in_ aux_supported_has auxpass_supported_been dep_Introduction_supported prep_supported_by amod_corpora_large amod_corpora_parallel pobj_by_corpora dep_as_such prep_corpora_as dep_Arabic-English_the pobj_as_Arabic-English cc_Arabic-English_and amod_corpora_Chinese-English nn_corpora_parallel conj_Arabic-English_corpora partmod_Arabic-English_distributed prep_distributed_by det_Consortium_the nn_Consortium_Linguistic nn_Consortium_Data pobj_by_Consortium cc_Consortium_and det__the nn__Europarl nn__corpus conj_Consortium_ nsubj_consists_which rcmod_Arabic-English_consists prep_consists_of num_languages_11 amod_languages_European pobj_of_languages ' +o,4207,'A few exceptions are the hierarchical -LRB- possibly syntax-based -RRB- transduction models and the string transduction models ',Wu,'det_exceptions_A amod_exceptions_few nsubj__exceptions cop__are det__the amod__hierarchical dep_syntax-based_possibly dep_hierarchical_syntax-based nn__transduction nn__models cc__and det_string_the conj__string nn_models_transduction nsubj__models dep__ ' +o,4208,' also detailed techniques for collocation extraction and developed a program called XTRACT , which is capable of computing flexible collocations based on elaborated statistical calculation ',Smadja,'advmod__also amod_techniques_detailed dobj__techniques prep_techniques_for nn_extraction_collocation pobj_for_extraction cc__and conj__developed det_program_a dobj_developed_program dep__called nsubj__XTRACT nsubj_capable_which cop_capable_is rcmod_XTRACT_capable prep_capable_of pcomp_of_computing amod_collocations_flexible dobj_computing_collocations prep_computing_based dep_based_on amod_calculation_elaborated amod_calculation_statistical pobj_on_calculation ' +o,4209,'Our observation is that this situation is ideal for so-called bootstrapping , co-training , or minimally supervised learning methods ',Yarowsky,'poss_observation_Our nsubj_is_observation complm_ideal_that det_situation_this nsubj_ideal_situation cop_ideal_is ccomp_is_ideal prep_ideal_for amod_bootstrapping_so-called pobj_for_bootstrapping conj_ideal_co-training cc_ideal_or advmod_supervised_minimally conj_ideal_supervised xcomp_ideal_learning dobj_learning_methods ' +o,4210,'This is contrastive to the one dimensional models used by Collinss perceptronbased sequence method which our algorithms are based upon , and by the linear-chain CRFs ',Collins,'nsubj_contrastive_This cop_contrastive_is prep_contrastive_to det_models_the num_models_one amod_models_dimensional pobj_to_models partmod_models_used prep_used_by nn__Collinss nn__perceptronbased nn__sequence nn__method pobj_by_ dobj_based_which poss_algorithms_our nsubjpass_based_algorithms auxpass_based_are rcmod__based prep_based_upon cc_to_and conj_to_by det_CRFs_the amod_CRFs_linear-chain pobj_by_CRFs ' +o,4211,'The phrase translation table is learned in the following manner : The parallel corpus is word-aligned bidirectionally , and using various heuristics -LRB- see for details -RRB- phrase correspondences are established ',Koehn,'det_table_The nn_table_phrase nn_table_translation nsubjpass_learned_table auxpass_learned_is prep_learned_in det_manner_the amod_manner_following pobj_in_manner det_corpus_The amod_corpus_parallel nsubjpass_word-aligned_corpus auxpass_word-aligned_is parataxis_learned_word-aligned advmod_word-aligned_bidirectionally cc_learned_and dep_established_using amod_heuristics_various dobj_using_heuristics dep_established_see dobj_see_ prep__for pobj_for_details nn_correspondences_phrase nsubjpass_established_correspondences auxpass_established_are conj_learned_established ' +p,4212,'Experiments show that the resulting rule set significantly improves the speed and accuracy over monolingual binarization -LRB- see Table 1 -RRB- in a stateof-the-art syntax-based machine translation system ',Galley,'nsubj_show_Experiments complm_improves_that det_set_the amod_set_resulting nn_set_rule nsubj_improves_set advmod_improves_significantly ccomp_show_improves det_speed_the dobj_improves_speed cc_speed_and conj_speed_accuracy prep_speed_over amod_binarization_monolingual pobj_over_binarization parataxis_improves_see dobj_see_Table num_Table_1 prep_improves_in det_system_a amod_system_stateof-the-art amod_system_syntax-based nn_system_machine nn_system_translation pobj_in_system ' +p,4213,'An important aspect of web search is to be able to narrow down search results by distinguishing among people with the same name leading to multiple efforts focusing on web person name disambiguation in the literature ',Cucerzan,'det_aspect_An amod_aspect_important nsubj_is_aspect prep_aspect_of nn_search_web pobj_of_search aux_able_to cop_able_be xcomp_is_able aux_narrow_to xcomp_able_narrow prt_narrow_down nn_results_search dobj_narrow_results prep_narrow_by pcomp_by_distinguishing prep_distinguishing_among pobj_among_people prep_distinguishing_with det_name_the amod_name_same pobj_with_name partmod_name_leading prep_leading_to amod_efforts_multiple pobj_to_efforts xcomp_leading_focusing prep_focusing_on nn_disambiguation_web nn_disambiguation_person nn_disambiguation_name pobj_on_disambiguation prep_disambiguation_in det_literature_the pobj_in_literature ' +o,4214,'For example , the words test and exam are similar because both of them follow verbs such as administer , cancel , cheat on , conduct , and both of them can be preceded by adjectives such as academic , comprehensive , diagnostic , difficult , Many methods have been proposed to compute distributional similarity between words ',Hindle,'dep_cheat_For pobj_For_example det_test_the nn_test_words nsubj_similar_test cc_test_and conj_test_exam cop_similar_are dep_cheat_similar mark_follow_because nsubj_follow_both prep_both_of pobj_of_them ccomp_similar_follow dobj_follow_verbs dep_as_such prep_verbs_as pobj_as_administer dep_follow_cancel prt_cheat_on dep_cheat_conduct cc_cheat_and nsubjpass_preceded_both prep_both_of pobj_of_them aux_preceded_can auxpass_preceded_be conj_cheat_preceded prep_preceded_by pobj_by_adjectives dep_as_such prep_adjectives_as amod_methods_academic amod_methods_comprehensive amod_methods_diagnostic amod_methods_difficult amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been pobj_as_proposed aux_compute_to xcomp_proposed_compute amod_similarity_distributional dobj_compute_similarity prep_similarity_between pobj_between_words ' +n,4215,' have proposed a rule-based algorithm for sentence combination , but no results have been reported ',Jing,'nsubj_proposed_ aux_proposed_have det_algorithm_a amod_algorithm_rule-based dobj_proposed_algorithm prep_algorithm_for nn_combination_sentence pobj_for_combination cc_proposed_but det_results_no nsubjpass_reported_results aux_reported_have auxpass_reported_been conj_proposed_reported ' +o,4216,'In the future , we plan to explore our discriminative framework on a full distortion model or even a hierarchical model ',Koehn,'prep_plan_In det_future_the pobj_In_future nsubj_plan_we aux_explore_to xcomp_plan_explore poss_framework_our amod_framework_discriminative dobj_explore_framework prep_explore_on det__a amod__full nn__distortion nn__model pobj_on_ cc__or advmod__even det__a amod__hierarchical nn__model conj__ ' +o,4217,'The MBT 180 Tagger Type Standard Trigram MBT Rule-based Maximum-Entropy Full Second-Order HMM SNOW Voting Constraints Full Second-Order HMM Known Unknown Overall Open\\/Closed Lexicon ? ',Ratnaparkhi,'det_MBT_The number_180_ num_Overall_180 nn_Overall_Tagger nn_Overall_Type nn_Overall_Standard nn_Overall_Trigram nn_Overall_ nn_Overall_MBT nn_Overall_ nn_Overall_Rule-based nn_Overall_ nn_Overall_Maximum-Entropy nn_Overall_ nn_Overall_Full nn_Overall_Second-Order nn_Overall_HMM nn_Overall_SNOW nn_Overall_ nn_Overall_Voting nn_Overall_Constraints nn_Overall_ nn_Overall_Full nn_Overall_Second-Order nn_Overall_HMM nn_Overall_Known nn_Overall_Unknown dep_MBT_Overall nn_Lexicon_Open\\/Closed dep_Overall_Lexicon ' +o,4218,'Step 2 involves extracting minimal xRS rules from the set of string\\/tree\\/alignments triplets ',Galley,'nsubj_involves_Step num_Step_2 xcomp_involves_extracting amod_rules_minimal nn_rules_xRS dobj_extracting_rules advmod_extracting_ prep_extracting_from det_set_the pobj_from_set prep_set_of amod_triplets_string\\/tree\\/alignments pobj_of_triplets ' +p,4219,'1 Introduction Statistical approaches to machine translation , pioneered by , achieved impressive performance by leveraging large amounts of parallel corpora ',Brown,'num_approaches_1 nn_approaches_Introduction nn_approaches_Statistical nsubj_achieved_approaches prep_approaches_to nn_translation_machine pobj_to_translation dep_approaches_pioneered prep_pioneered_by pobj_by_ amod_performance_impressive dobj_achieved_performance prep_achieved_by pcomp_by_leveraging amod_amounts_large dobj_leveraging_amounts prep_amounts_of amod_corpora_parallel pobj_of_corpora ' +o,4220,'? ? queries : The queries of are made up of a pair of adjectives , and in our approach the query contains the content words of the headline and an emotion ',Turney,'det_queries_The nsubjpass_made_queries prep_queries_of pobj_of_ auxpass_made_are dep_queries_made prt_made_up prep_made_of det_pair_a pobj_of_pair prep_pair_of pobj_of_adjectives cc_made_and prep_contains_in poss_approach_our pobj_in_approach det_query_the nsubj_contains_query conj_made_contains det_words_the nn_words_content dobj_contains_words prep_words_of det_headline_the pobj_of_headline cc_headline_and det_emotion_an conj_headline_emotion ' +o,4221,'First , it recognizes non-recursive Base Noun Phrase -LRB- BNP -RRB- ',Ramshaw,'advmod_recognizes_First nsubj_recognizes_it amod_Phrase_non-recursive nn_Phrase_Base nn_Phrase_Noun dobj_recognizes_Phrase abbrev_Phrase_BNP ' +p,4222,'2 Maximum Entropy Models Maximum entropy -LRB- ME -RRB- models , also known as log-linear and exponential learning models , provideageneralpurposemachinelearningtechnique for classification and prediction which has been successfully applied to natural language processing including part of speech tagging , named entity recognition etc Maximum entropy models can integrate features from many heterogeneous information sources for classification ',Berger,'number_Maximum_2 num_models_Maximum nn_models_Entropy nn_models_Models nn_models_Maximum nn_models_entropy abbrev_models_ME nsubj_named_ advmod_known_also partmod__known prep_known_as amod_models_log-linear cc_log-linear_and conj_log-linear_exponential nn_models_learning pobj_as_models appos_models_provideageneralpurposemachinelearningtechnique prep_provideageneralpurposemachinelearningtechnique_for pobj_for_classification cc_classification_and conj_classification_prediction nsubjpass_applied_which aux_applied_has auxpass_applied_been advmod_applied_successfully rcmod_provideageneralpurposemachinelearningtechnique_applied prep_applied_to amod_processing_natural nn_processing_language pobj_to_processing prep_processing_including pobj_including_part prep_part_of nn_tagging_speech pobj_of_tagging dep_models_named nn_recognition_entity iobj_named_recognition nn_entropy_etc nn_entropy_Maximum dobj_named_entropy nsubj_integrate_models aux_integrate_can rcmod_entropy_integrate dobj_integrate_features prep_integrate_from amod_sources_many amod_sources_heterogeneous nn_sources_information pobj_from_sources prep_sources_for pobj_for_classification ' +o,4223,'Statistical or probabilistic methods are often used to extract semantic clusters from corpora in order to build lexical resources for ANLP tools , , , or for automatic thesaurus generation ',Hindle,'nsubjpass_used_Statistical cc_Statistical_or amod_methods_probabilistic conj_Statistical_methods auxpass_used_are advmod_used_often aux_extract_to xcomp_used_extract amod_clusters_semantic dobj_extract_clusters prep_extract_from pobj_from_corpora mark_build_in dep_build_order aux_build_to purpcl_extract_build amod_resources_lexical dobj_build_resources prep_build_for amod_tools_ANLP pobj_for_tools amod_tools_ dep__ dep__ cc_for_or conj_for_for amod__automatic nn__thesaurus nn__generation pobj_for_ ' +o,4224,'Examples of such early work include ',Turney,'nsubj_include_Examples prep_Examples_of amod_work_such amod_work_early pobj_of_work ' +o,4225,'CRP-based samplers have served the communitywellinrelatedlanguagetasks , suchaswordsegmentation and coreference resolution ',Haghighi,'amod_samplers_CRP-based nsubj_served_samplers aux_served_have det_resolution_the amod_resolution_communitywellinrelatedlanguagetasks conj_communitywellinrelatedlanguagetasks_suchaswordsegmentation cc_communitywellinrelatedlanguagetasks_and conj_communitywellinrelatedlanguagetasks_coreference dobj_served_resolution ' +o,4226,'Apart from the fact that we present an alternative model , our work differs from in two important ways ',Marcu,'dep_differs_Apart prep_Apart_from det_fact_the pobj_from_fact complm_present_that nsubj_present_we ccomp_Apart_present det_model_an amod_model_alternative dobj_present_model poss_work_our nsubj_differs_work prep_differs_from pcomp_from_ prep__in num_ways_two amod_ways_important pobj_in_ways ' +o,4227,'2 Baseline DP Decoder The translation model used in this paper is a phrasebased model , where the translation units are so-called blocks : a block b is a pair consisting of a source phrase s and a target phrase t which are translations of each other ',Koehn,'num_Decoder_2 nn_Decoder_Baseline nn_Decoder_DP det_model_The nn_model_translation nsubj__model partmod_model_used prep_used_in det_paper_this pobj_in_paper cop__is det__a amod__phrasebased nn__model dep_Decoder_ advmod_blocks_where det_units_the nn_units_translation nsubj_blocks_units cop_blocks_are amod_blocks_so-called rcmod__blocks det_block_a dep__block nsubj_pair_b cop_pair_is det_pair_a rcmod_block_pair partmod_pair_consisting prep_consisting_of det_source_a pobj_of_source nn_s_phrase dep__s cc__and det_t_a nn_t_target nn_t_phrase conj__t nsubj_translations_which cop_translations_are rcmod_t_translations prep_translations_of det_other_each pobj_of_other ' +o,4228,'2 Related Work Starting with the IBM models , researchers have developed various statistical word alignment systems based on different models , such as hidden Markov models -LRB- HMM -RRB- , log-linear models , and similarity-based heuristic methods ',Brown,'num_Work_2 amod_Work_Related prep_developed_Starting dep_Starting_with det__the nn__IBM nn__models pobj_with_ nsubj_developed_researchers aux_developed_have dep_Work_developed amod_systems_various amod_systems_statistical nn_systems_word nn_systems_alignment dobj_developed_systems prep_developed_based dep_based_on amod_models_different pobj_on_models dep_as_such prep_developed_as amod_models_hidden nn_models_Markov pobj_as_models abbrev_models_HMM advmod__ amod_models_log-linear nsubj__models dep_Work_ cc__and amod_methods_similarity-based amod_methods_heuristic nsubj__methods conj__ ' +o,4229,'Text similarity has been also used for relevance feedback and text classification , word sense disambiguation , and more recently for extractive summarization , and methods for automatic evaluation of machine translation or text summarization ',Papineni,'amod_similarity_Text nsubjpass_used_similarity aux_used_has auxpass_used_been advmod_used_also prep_used_for nn_feedback_relevance nn__feedback cc_feedback_and nn__text nn__classification conj_feedback_ nn__word nn__sense amod__disambiguation appos_feedback_ cc_feedback_and advmod_recently_more conj_feedback_recently prep_recently_for amod_summarization_extractive pobj_for_summarization pobj_for_ cc__and conj__methods prep_methods_for amod_evaluation_automatic pobj_for_evaluation prep_evaluation_of nn__machine nn__translation pobj_of_ cc__or nn__text nn__summarization conj__ ' +o,4230,'Models describing these types of dependencies are referred to as alignrnent models , ',Brown,'nsubjpass_referred_Models partmod_Models_describing det_types_these dobj_describing_types prep_types_of pobj_of_dependencies auxpass_referred_are prep_referred_to advmod_alignrnent_as amod_models_alignrnent pobj_to_models amod_models_ dep__ ' +p,4231,'In this work we use the averaged perceptron algorithm since it is an online algorithm much simpler and orders of magnitude faster than Boosting and MaxEnt methods ',Collins,'prep_use_In det_work_this pobj_In_work nsubj_use_we det_algorithm_the amod_algorithm_averaged nn_algorithm_perceptron nsubj__algorithm ccomp_use_ mark_algorithm_since nsubj_algorithm_it cop_algorithm_is det_algorithm_an amod_algorithm_online advcl__algorithm advmod_simpler_much amod_algorithm_simpler cc_simpler_and measure_faster_orders prep_orders_of pobj_of_magnitude conj_simpler_faster prep_algorithm_than nn_methods_Boosting cc_Boosting_and conj_Boosting_MaxEnt pobj_than_methods ' +o,4232,'Another way of doing the parameter estimation for this matching task would have been to use an averaged perceptron method , as in ',Collins,'det_way_Another nsubj_been_way prep_way_of pcomp_of_doing det_estimation_the nn_estimation_parameter dobj_doing_estimation prep_doing_for det_task_this amod_task_matching pobj_for_task aux_been_would aux_been_have aux_use_to xcomp_been_use det_method_an amod_method_averaged nn_method_perceptron dobj_use_method mark_in_as advcl_use_in ' +o,4233,'Phrases of up to 10 in length on the French side were extracted from the parallel text , and minimum-error-rate training was 8 We can train on the full training data shown if tighter constraints are placed on rule extraction for the United Nations data ',Och,'nsubjpass_extracted_Phrases prep_Phrases_of quantmod_10_up dep_10_to pobj_of_10 prep_10_in pobj_in_length prep_length_on det_side_the amod_side_French pobj_on_side auxpass_extracted_were prep_extracted_from det_text_the amod_text_parallel pobj_from_text cc_extracted_and amod__minimum-error-rate nn__training nsubj_8_ cop_8_was conj_extracted_8 nsubj_train_We aux_train_can dep_extracted_train prep_train_on det_data_the amod_data_full nn_data_training pobj_on_data partmod_data_shown mark_placed_if amod_constraints_tighter nsubjpass_placed_constraints auxpass_placed_are advcl_shown_placed prep_placed_on nn_extraction_rule pobj_on_extraction prep_extraction_for det_data_the nn_data_United nn_data_Nations pobj_for_data ' +o,4234,'We have investigated this and our results are in line with showing that the translation quality does not improve if we utilize phrases beyond a certain length ',Koehn,'nsubj_investigated_We aux_investigated_have nsubj_are_this cc_this_and poss_results_our conj_this_results ccomp_investigated_are prep_are_in nsubj_showing_line prep_line_with pobj_with_ pcomp_in_showing complm_improve_that det_quality_the nn_quality_translation nsubj_improve_quality aux_improve_does neg_improve_not ccomp_showing_improve mark_utilize_if nsubj_utilize_we advcl_improve_utilize dobj_utilize_phrases prep_utilize_beyond det_length_a amod_length_certain pobj_beyond_length ' +o,4235,'3 Related work Word collocation Various collocation metrics have been proposed , including mean and variance , the t-test , the chi-square test , pointwise mutual information -LRB- MI -RRB- , and binomial loglikelihood ratio test -LRB- BLRT -RRB- ',Church,'num_work_3 amod_work_Related nsubj_Word_work amod_metrics_collocation amod_metrics_Various nn_metrics_collocation nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been ccomp_Word_proposed prep_proposed_including nn__mean cc_mean_and conj_mean_variance pobj_including_ det__the amod__t-test conj__ det_test_the amod_test_chi-square conj__test amod__pointwise amod__mutual nn__information abbrev__MI appos_test_ cc__and amod_test_binomial amod_test_loglikelihood nn_test_ratio conj__test abbrev_test_BLRT advmod_proposed_ ' +o,4236,'(Papineni et al. , 2002).',Papineni,'' +o,4237,'By introducing the hidden word alignment variable a , the optimal translation can be searched for based on the following criterion : \\* 1 , arg max -LRB- -LRB- , , -RRB- -RRB- M mm m ea eh = = efa -LRB- 1 -RRB- where is a string of phrases in the target language , e f fa is the source language string of phrases , he are feature functions , weights -LRB- , , -RRB- m m are typically optimized to maximize the scoring function ',Och,'prep_mm_By pcomp_By_introducing det_variable_the amod_variable_hidden nn_variable_word nn_variable_alignment dobj_introducing_variable det__a nsubjpass_searched_ det_translation_the amod_translation_optimal nsubjpass_searched_translation aux_searched_can auxpass_searched_be dep_By_searched prep_searched_for prep_searched_based dep_based_on det_criterion_the amod_criterion_following pobj_on_criterion dep_criterion_\\* num_\\*_1 appos_\\*_arg partmod_arg_max nsubj_mm_M nn_eh_m nn_eh_ea nsubj_efa_eh dep_efa_= amod_efa_= ccomp_mm_efa dep_string_1 advmod_string_where cop_string_is det_string_a dep_string_string prep_string_of pobj_of_phrases prep_phrases_in det_language_the nn_language_target pobj_in_language dep_f_e dep_string_f nsubj_string_fa cop_string_is det_string_the nn_string_source nn_string_language parataxis_mm_string prep_string_of pobj_of_phrases nsubj_functions_he cop_functions_are nn_functions_feature ccomp_string_functions nn_-LRB-_weights nn_m_m nsubjpass_optimized_m auxpass_optimized_are advmod_optimized_typically dep_mm_optimized aux_maximize_to xcomp_optimized_maximize det_function_the amod_function_scoring dobj_maximize_function ' +o,4238,'To do this , we first identify initial phrase pairs using the same criterion as previous systems : Definition 1 ',Koehn,'aux_do_To dep_identify_do dobj_do_this nsubj_identify_we advmod_identify_first amod_pairs_initial nn_pairs_phrase dobj_identify_pairs partmod_pairs_using det_criterion_the amod_criterion_same dobj_using_criterion prep_criterion_as amod__previous nn__systems pobj_as_ dep_criterion_Definition num_Definition_1 ' +o,4239,'We use the minimum-error rate training procedure by as implemented in the Moses toolkit to set the weights of the various translation and language models , optimizing for BLEU ',Och,'nsubj_use_We det_procedure_the amod_procedure_minimum-error nn_procedure_rate nn_procedure_training dobj_use_procedure prep_use_by pobj_by_ mark_implemented_as advcl_use_implemented prep_implemented_in det_toolkit_the nn_toolkit_Moses pobj_in_toolkit aux_set_to xcomp_implemented_set det_weights_the dobj_set_weights prep_weights_of det_translation_the amod_translation_various pobj_of_translation cc_translation_and nn_models_language conj_translation_models dep_implemented_optimizing prep_optimizing_for pobj_for_BLEU ' +p,4240,'Still , it is in our next plans and part of our future work to embed in our model some of the interesting WSD approaches , like knowledgebased , corpus-based , or combinations with very high accuracy ',Mihalcea,'advmod_is_Still nsubj_is_it prep_is_in poss_plans_our amod_plans_next pobj_in_plans cc_plans_and conj_plans_part prep_part_of poss_work_our amod_work_future pobj_of_work aux_embed_to xcomp_is_embed prep_embed_in poss_model_our pobj_in_model nsubj_approaches_some prep_some_of det_WSD_the amod_WSD_interesting pobj_of_WSD dep_is_approaches prep_approaches_like amod__knowledgebased pobj_like_ amod__corpus-based conj__ cc__or conj__combinations prep_approaches_with advmod_high_very amod__high nn__accuracy pobj_with_ ' +o,4241,'Word association norms , mutual information , and lexicography , Computational Linguistics , 16 -LRB- 1 -RRB- : 22-29 , M et al 1993 ',Marcus,'nn_norms_association dobj_Word_norms amod_information_mutual conj_norms_information cc_norms_and conj_norms_lexicography nn_Linguistics_Computational nsubj_Word_Linguistics appos_Linguistics_16 appos_16_1 nn__22-29 dep_16_ conj__M cc__et conj__al dep__1993 ' +o,4242,'These forest rescoring algorithms have potential applications to other computationally intensive tasks involving combinations of different models , for example , head-lexicalized parsing ; joint parsing and semantic role labeling ; or tagging and parsing with nonlocal features ',Collins,'det_algorithms_These nn_algorithms_forest amod_algorithms_rescoring nsubj_have_algorithms amod_applications_potential dobj_have_applications prep_have_to amod_tasks_other advmod_intensive_computationally amod_tasks_intensive pobj_to_tasks dep_have_involving dobj_involving_combinations prep_combinations_of amod_models_different pobj_of_models prep_involving_for pobj_for_example amod__head-lexicalized nn__parsing appos_example_ amod__joint amod__parsing cc_parsing_and conj_parsing_semantic nn__role nn__labeling dep_example_ cc_involving_or conj_involving_tagging cc_tagging_and conj_tagging_parsing prep_tagging_with amod_features_nonlocal pobj_with_features ' +o,4243,'For each word in the LDV , we consulted three existing thesauri : Rogets Thesaurus , Collins COBUILD Thesaurus , and WordNet ',Lin,'prep_consulted_For det_word_each pobj_For_word prep_word_in det_LDV_the pobj_in_LDV nsubj_consulted_we num_thesauri_three amod_thesauri_existing dobj_consulted_thesauri nn__Rogets nn__Thesaurus dep_thesauri_ nn__Collins nn__COBUILD nn__Thesaurus conj__ cc__and nn__WordNet conj__ ' +o,4244,'Such methods were presented in and ~ flensky , 1978 -RRB- ',Marcus,'amod_methods_Such nsubjpass_presented_methods auxpass_presented_were prep_presented_in pobj_in_ cc__and conj__~ num_~_flensky num_~_1978 ' +o,4245,'This direction has been forming the mainstream of research on opinion-sensitive text processing ',Pang,'det_direction_This nsubj_forming_direction aux_forming_has aux_forming_been det_mainstream_the dobj_forming_mainstream prep_mainstream_of pobj_of_research prep_forming_on amod_processing_opinion-sensitive nn_processing_text pobj_on_processing ' +o,4246,'The up-arrows and down-arrows are shorthand for -LRB- M -LRB- ni -RRB- -RRB- = -LRB- ni -RRB- where ni is the c-structure node annotated with the equation2 Treebest : = argmaxTreeP -LRB- Tree F-Str -RRB- -LRB- 1 -RRB- P -LRB- Tree F-Str -RRB- : = productdisplay X Y in Tree Feats = -LCB- ai vj -LRB- -LRB- X -RRB- -RRB- ai = vj -RCB- P -LRB- X Y X , Feats -RRB- -LRB- 2 -RRB- The generation model of maximises the probability of a tree given an f-structure -LRB- Eqn ',Cahill,'det_up-arrows_The nsubj_shorthand_up-arrows cc_up-arrows_and conj_up-arrows_down-arrows cop_shorthand_are prep_shorthand_for nn_ni_M pobj_for_ni rcmod_ni_= dobj_=_ni advmod_node_where nsubj_node_ni cop_node_is det_node_the amod_node_c-structure rcmod_ni_node amod_P_annotated prep_annotated_with det_equation2_the pobj_with_equation2 dep_equation2_Treebest dep_argmaxTreeP_= dep_equation2_argmaxTreeP nn_F-Str_Tree appos_equation2_F-Str appos_equation2_1 dep_node_P nn_F-Str_Tree appos_P_F-Str dep_node_= nn_Y_productdisplay nn_Y_X dobj_=_Y prep_Y_in nn_=_Tree nn_=_Feats pobj_in_= amod_vj_ai dep_node_vj nsubj_ai_X rcmod_vj_ai dep_ai_= dobj_=_vj dep_node_P nn_X_X nn_X_Y appos_P_X dep_X_Feats dep_maximises_2 det_model_The nn_model_generation nsubj_maximises_model prep_model_of pobj_of_ dep_shorthand_maximises det_probability_the dobj_maximises_probability prep_probability_of det_tree_a pobj_of_tree partmod_tree_given dep_Eqn_an nn_Eqn_f-structure dobj_given_Eqn ' +p,4247,'Thus, we obtain the following second-order model: a36a39a38a41a40 a17 a5a7 a42a4 a5a7 a44 a8 a5a57 a15a27a58 a7 a36a39a38a41a40 a17a20a15a59a42a17 a15a41a49 a7 a7 a60 a4 a5a7 a44 a8 ma61a63a62a65a64a33a66 a5a57 a15a27a58 a7a68a67 a40 a17 a15 a42a17 a15a50a49 a7 a15a50a49a51a48 a60 a4 a15a27a47a55a48 a15a50a49a54a48 a44 a11 A well-founded framework for directly modeling the posterior probability a67 a40 a17 a15 a42a17 a15a50a49 a7 a15a50a49a54a48 a60 a4 a15a12a47a55a48 a15a50a49a54a48 a44 is maximum entropy (Berger et al. , 1996).',Berger,'' +o,4248,'where they are expected to be maximally discriminative ',Tillmann,'advmod_expected_where nsubjpass_expected_they auxpass_expected_are aux_discriminative_to cop_discriminative_be advmod_discriminative_maximally xcomp_expected_discriminative ' +n,4249,' provides anecdotal evidence that only incorrect alignments are eliminated by ITG constraints ',Wu,'advmod_provides_ amod_evidence_anecdotal dobj_provides_evidence dep_eliminated_that advmod_alignments_only amod_alignments_incorrect nsubjpass_eliminated_alignments auxpass_eliminated_are dep_provides_eliminated prep_eliminated_by nn_constraints_ITG pobj_by_constraints ' +o,4250,'In our own work on document compression models , both of which extend the sentence compression model of Knight and Marcu , we assume that sentences and documents can be summarized exclusively through deletion of contiguous text segments ','Daume III','prep_assume_In poss_work_our amod_work_own pobj_In_work prep_work_on nn__document nn__compression nn__models pobj_on_ dep_of_both nsubj_extend_of pobj_of_which rcmod__extend det_model_the nn_model_sentence nn_model_compression dobj_extend_model prep_model_of nn__Knight cc_Knight_and conj_Knight_Marcu pobj_of_ nsubj_assume_we complm_summarized_that nsubjpass_summarized_sentences cc_sentences_and conj_sentences_documents aux_summarized_can auxpass_summarized_be ccomp_assume_summarized dep_through_exclusively prep_summarized_through pobj_through_deletion prep_deletion_of amod_segments_contiguous nn_segments_text pobj_of_segments ' +o,4251,'4 Optimizing Metric Parameters The original version of Meteor has instantiated values for three parameters in the metric : one for controlling the relative weight of precision and recall in computing the Fmean score -LRB- -RRB- ; one governing the shape of the penalty as a function of fragmentation -LRB- -RRB- and one for the relative weight assigned to the fragmentation penalty -LRB- -RRB- ',Banerjee,'num_Parameters_4 nn_Parameters_Optimizing nn_Parameters_Metric det_version_The amod_version_original nsubj_instantiated_version prep_version_of nn__Meteor pobj_of_ aux_instantiated_has rcmod_Parameters_instantiated dobj_instantiated_values prep_instantiated_for num_parameters_three pobj_for_parameters prep_parameters_in det_metric_the pobj_in_metric dep_metric_one prep_one_for pcomp_for_controlling det_weight_the amod_weight_relative dobj_controlling_weight prep_weight_of pobj_of_precision cc_precision_and conj_precision_recall prep_controlling_in pcomp_in_computing det_-LRB-_the nn_-LRB-_Fmean nn_-LRB-_score dep_metric_one partmod_one_governing det_shape_the dobj_governing_shape prep_shape_of det_penalty_the pobj_of_penalty prep_penalty_as det_function_a pobj_as_function prep_function_of pobj_of_fragmentation cc_-RRB-_and conj_-RRB-_one prep_-RRB-_for det_weight_the amod_weight_relative pobj_for_weight partmod_-RRB-_assigned prep_assigned_to det_-LRB-_the nn_-LRB-_fragmentation nn_-LRB-_penalty ' +o,4252,'It has been lately incorporated into computational lexicography in , , , , -RRB- ',Ostler,'nsubjpass_incorporated_It aux_incorporated_has auxpass_incorporated_been advmod_incorporated_lately prep_incorporated_into amod__computational amod__lexicography prep_lexicography_in nn__ appos__ appos__ pobj_in_ pobj_into_ ' +o,4253,'5 Related Work We already discussed the relation of our work to in Section 24 ','Daume III','num_Work_5 amod_Work_Related nsubj_discussed_We advmod_discussed_already dep_Work_discussed det_relation_the dobj_discussed_relation prep_relation_of poss_work_our pobj_of_work aux__to infmod_work_ prep__in pobj_in_Section num_Section_24 ' +p,4254,'2 Related Work The popular IBM models for statistical machine translation are described in and the HMM-based alignment model was introduced in ',Brown,'num_Work_2 amod_Work_Related det_models_The amod_models_popular nn_models_IBM nsubjpass_described_models prep_models_for amod_translation_statistical nn_translation_machine pobj_for_translation auxpass_described_are dep_Work_described prep_described_in pobj_in_ cc_described_and det_model_the amod_model_HMM-based nn_model_alignment nsubjpass_introduced_model auxpass_introduced_was conj_described_introduced prep_introduced_in pobj_in_ ' +o,4255,'In the proposed method , the statistical machine translation -LRB- SMT -RRB- is deeply incorporated into the question answering process , instead of using the SMT as the preprocessing before the mono-lingual QA process as in the previous work ',Brown,'prep_incorporated_In det_method_the amod_method_proposed pobj_In_method det_translation_the amod_translation_statistical nn_translation_machine appos_method_translation abbrev_translation_SMT nsubj_incorporated_ cop_incorporated_is advmod_incorporated_deeply prep_incorporated_into det_process_the nn_process_question nn_process_answering pobj_into_process dep_of_instead prep_incorporated_of pcomp_of_using det_SMT_the dobj_using_SMT prep_using_as det_preprocessing_the pobj_as_preprocessing prep_using_before det_process_the amod_process_mono-lingual nn_process_QA pobj_before_process dep_in_as prep_process_in det_work_the amod_work_previous pobj_in_work ' +o,4256,'For the factored language models , a feature-based word representation was obtained by tagging the text with Rathnaparkis maximum-entropy tagger and by stemming words using the Porter stemmer ',Ratnaparkhi,'prep_obtained_For det_models_the amod_models_factored nn_models_language pobj_For_models det_representation_a amod_representation_feature-based nn_representation_word nsubjpass_obtained_representation auxpass_obtained_was prep_obtained_by pcomp_by_tagging det_text_the dobj_tagging_text prep_tagging_with nn__Rathnaparkis amod__maximum-entropy nn__tagger pobj_with_ cc_by_and conj_by_by pcomp_by_stemming dobj_stemming_words xcomp_stemming_using det__the nn__Porter nn__stemmer dobj_using_ ' +n,4257,'By segmenting words into morphemes , we can improve the performance of natural language systems including machine translation and information retrieval ',Brown,'prep_improve_By pcomp_By_segmenting dobj_segmenting_words prep_segmenting_into pobj_into_morphemes nsubj_improve_we aux_improve_can det_performance_the dobj_improve_performance prep_performance_of amod_systems_natural nn_systems_language pobj_of_systems prep_systems_including nn__machine nn__translation pobj_including_ cc__and nn__information nn__retrieval conj__ ' +o,4258,'Due to the positive results in Ando , et al ',Blitzer,'dep_to_Due det_results_the amod_results_positive pobj_to_results prep_results_in nn__Ando pobj_in_ conj__ cc__et conj__al ' +o,4259,'Many-to-many alignments can be created by combining two GIZA + + alignments , one where English generates Foreign and another with those roles reversed ',Och,'amod_alignments_Many-to-many nsubjpass_created_alignments aux_created_can auxpass_created_be prep_created_by pcomp_by_combining num_alignments_two amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ dobj_combining_alignments appos_alignments_one advmod_generates_where nsubj_generates_English rcmod_one_generates nsubj_reversed_Foreign cc_Foreign_and conj_Foreign_another prep_another_with det_roles_those pobj_with_roles ccomp_generates_reversed ' +p,4260,'In the news article domain , ROUGE scores have been shown to be generally highly correlated with human evaluation in content match ',Lin,'prep_shown_In det_domain_the nn_domain_news nn_domain_article pobj_In_domain nn_scores_ROUGE nsubjpass_shown_scores aux_shown_have auxpass_shown_been aux_correlated_to cop_correlated_be advmod_correlated_generally advmod_correlated_highly xcomp_shown_correlated prep_correlated_with amod_evaluation_human pobj_with_evaluation prep_correlated_in nn_match_content pobj_in_match ' +o,4261,'Be-Comp Following the general idea in , we identify the ISA pattern in the definition sentence by extracting nominal complements of the verb be , taking 451 No ',Kazama,'dep_identify_Be-Comp prep_Be-Comp_Following det_idea_the amod_idea_general pobj_Following_idea prep_idea_in pobj_in_ nsubj_identify_we det_pattern_the nn_pattern_ISA dobj_identify_pattern prep_pattern_in det_sentence_the nn_sentence_definition pobj_in_sentence prep_identify_by pcomp_by_extracting amod_complements_nominal nsubj_,_complements prep_complements_of det_verb_the pobj_of_verb aux_,_be xcomp_,_taking dobj_taking_451 dep_451_No ' +o,4262,'2 Data and annotation Yahoo ! s image query API was used to obtain a corpus of pairs of semantically ambiguous images , in thumbnail and true size , and their corresponding web sites for three ambiguous keywords inspired by : BASS , CRANE , and SQUASH ',Yarowsky,'dep_query_2 nsubj_query_Data cc_Data_and nn_Yahoo_annotation conj_Data_Yahoo dep_query_s nn_query_image nsubjpass_used_API auxpass_used_was dep_query_used aux_obtain_to xcomp_used_obtain det_corpus_a dobj_obtain_corpus prep_corpus_of pobj_of_pairs prep_pairs_of amod_images_semantically amod_images_ambiguous pobj_of_images conj_pairs_in amod_size_thumbnail cc_thumbnail_and conj_thumbnail_true pobj_in_size cc_pairs_and poss_sites_their amod_sites_corresponding nn_sites_web conj_pairs_sites prep_sites_for num_keywords_three amod_keywords_ambiguous pobj_for_keywords partmod_keywords_inspired prep_inspired_by pobj_by_ dep_sites_BASS conj_BASS_CRANE cc_BASS_and conj_BASS_SQUASH ' +o,4263,'32 Wall Street Journal Our out-of-domain data is the Wall Street Journal -LRB- WSJ -RRB- portion of the Penn Treebank which consists of about 40,000 sentences -LRB- one million words -RRB- annotated with syntactic information ',Marcus,'num_Journal_32 nn_Journal_Wall nn_Journal_Street poss_data_Our amod_data_out-of-domain nsubj_portion_data cop_portion_is det_portion_the nn_Journal_Wall nn_Journal_Street nn_portion_Journal abbrev_Journal_WSJ rcmod_Journal_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ nsubj_consists_which rcmod_portion_consists prep_Journal_of quantmod_40,000_about num_sentences_40,000 pobj_of_sentences number_million_one num_words_million appos_sentences_words partmod_sentences_annotated prep_annotated_with amod_information_syntactic pobj_with_information ' +o,4264,'production rules are typically learned from alignment structures or from alignment structures and derivation trees for the source string ',Wu,'nn_rules_production nsubjpass_learned_rules auxpass_learned_are advmod_learned_typically prep_learned_from amod__alignment nn__structures pobj_from_ cc_from_or conj_from_from amod_structures_alignment pobj_from_structures cc_structures_and nn_trees_derivation conj_structures_trees prep_structures_for det__the nn__source nn__string pobj_for_ ' +o,4265,'A token can be a word or a punctuation symbol , and each of these neighboring tokens must be in the same sentence as a2 We use a sentence segmentation program and a POS tagger to segment the tokens surrounding a2 into sentences and assign POS tags to these tokens ',Ratnaparkhi,'det_token_A nsubj_word_token aux_word_can cop_word_be det_word_a cc_word_or det_symbol_a nn_symbol_punctuation conj_word_symbol cc_word_and nsubj_be_each prep_each_of det_tokens_these amod_tokens_neighboring pobj_of_tokens aux_be_must conj_word_be prep_be_in det_sentence_the amod_sentence_same pobj_in_sentence prep_sentence_as pobj_as_a2 nsubj_use_We dep_word_use det__a nn__sentence nn__segmentation nn__program nsubj_segment_ cc__and det__a dep__POS nn__tagger conj__ aux_segment_to xcomp_use_segment det_a2_the amod_a2_tokens amod_a2_surrounding dobj_segment_a2 prep_segment_into pobj_into_sentences cc_segment_and conj_segment_assign dep_tags_POS dobj_assign_tags prep_assign_to det_tokens_these pobj_to_tokens ' +p,4266,'1 Introduction Phrase-based systems are probably the most widespread class of Statistical Machine Translation systems , and arguably one of the most successful ',Koehn,'num__1 nn__Introduction amod__Phrase-based nn__systems nsubj_class_ cop_class_are advmod_class_probably det_class_the advmod_widespread_most amod_class_widespread prep_class_of nn_systems_Statistical nn_systems_Machine nn_systems_Translation pobj_of_systems cc_class_and advmod_one_arguably conj_class_one prep_one_of dep_successful_the advmod_successful_most pobj_of_successful ' +o,4267,'POS disambiguation has usually been performed by statistical approaches , mainly using the hidden Markov model -LRB- HMM -RRB- in English research communities ',Cutting,'dep_disambiguation_POS nsubjpass_performed_disambiguation aux_performed_has advmod_performed_usually auxpass_performed_been prep_performed_by amod_approaches_statistical pobj_by_approaches advmod_using_mainly xcomp_performed_using det_model_the amod_model_hidden nn_model_Markov dobj_using_model abbrev_model_HMM prep_model_in amod_communities_English nn_communities_research pobj_in_communities ' +o,4268,'Optimization and measurement were done with the NIST implementation of case-insensitive BLEU 4n4r 4 41 Baseline We compared translation by pattern matching with a conventional exact model representation using external prefix trees ',Papineni,'nsubjpass_done_Optimization cc_Optimization_and conj_Optimization_measurement auxpass_done_were prep_done_with det_implementation_the amod_implementation_NIST pobj_with_implementation prep_implementation_of amod__case-insensitive nn__BLEU nn__4n4r pobj_of_ number_41_4 num_Baseline_41 dep__Baseline nsubj_compared_We dep_done_compared dobj_compared_translation prep_compared_by nn_matching_pattern pobj_by_matching prep_compared_with det_representation_a amod_representation_conventional amod_representation_exact nn_representation_model pobj_with_representation partmod_representation_using amod_trees_external nn_trees_prefix dobj_using_trees advmod_using_ ' +o,4269,'2 Bilingual Bracketing In , the Bilingual Bracketing PCFG was introduced , which can be simplified as the following production rules : A ! -LRB- AA -RRB- -LRB- 1 -RRB- A ! -LRB- AA -RRB- -LRB- 2 -RRB- A ! f = e -LRB- 3 -RRB- A ! f = null -LRB- 4 -RRB- A ! null = e -LRB- 5 -RRB- Where f and e are words in the target vocabulary Vf and source vocabulary Ve respectively ',Wu,'num_Bracketing_2 nn_Bracketing_Bilingual prep_introduced_In pobj_In_ det_PCFG_the nn_PCFG_Bilingual nn_PCFG_Bracketing nsubjpass_introduced_PCFG auxpass_introduced_was rcmod_Bracketing_introduced nsubjpass_simplified_which aux_simplified_can auxpass_simplified_be dep_A_simplified prep_simplified_as det_rules_the amod_rules_following nn_rules_production pobj_as_rules dobj_introduced_A abbrev_A_AA dep_A_1 det_A_A dep_!_AA appos_A_2 dep_A_A dep_A_f nn_A_= nn_A_e appos_A_3 dep_Bracketing_A dobj_!_f dep_f_= iobj_=_null dep_=_4 det_=_A nn_=_null dobj_=_= dep_=_e appos_f_5 advmod_words_Where nsubj_words_f cc_f_and conj_f_e cop_words_are dep_f_words prep_words_in det_Vf_the nn_Vf_target nn_Vf_vocabulary pobj_in_Vf cc_Vf_and nn_Ve_source nn_Ve_vocabulary conj_Vf_Ve advmod_words_respectively ' +o,4270,' also used re-decoding to do system combination by extracting sentence-specific phrase translation tables from the outputs of different MT systems and running a phrase-based decoding with this new translation table ',Rosti,'nsubj_used_ advmod_used_also ccomp_used_re-decoding aux_do_to xcomp_re-decoding_do nn_combination_system dobj_do_combination prep_do_by pcomp_by_extracting amod_tables_sentence-specific nn_tables_phrase nn_tables_translation dobj_extracting_tables prep_extracting_from det_outputs_the pobj_from_outputs prep_outputs_of amod_systems_different nn_systems_MT pobj_of_systems cc_re-decoding_and conj_re-decoding_running det_decoding_a amod_decoding_phrase-based dobj_running_decoding prep_running_with det_table_this amod_table_new nn_table_translation pobj_with_table ' +o,4271,'82 2 Aggregate Markov models In this section we consider how to construct classbased bigram models ',Brown,'number_2_82 num_models_2 nn_models_Aggregate nn_models_Markov prep_consider_In det_section_this pobj_In_section nsubj_consider_we rcmod_models_consider advmod_construct_how aux_construct_to ccomp_consider_construct amod_models_classbased nn_models_bigram dobj_construct_models ' +o,4272,'The average senior high school student achieves 57 \% correct ',Turney,'det_student_The amod_student_average amod_student_senior amod_student_high nn_student_school nsubj_achieves_student num_\%_57 measure_correct_\% acomp_achieves_correct ' +o,4273,'Secondly , we used the Kappa coefficient , which has become the standard evaluation metric and the score obtained was 0905 ',Carletta,'advmod_used_Secondly nsubj_used_we det__the nn__Kappa nn__coefficient dobj_used_ nsubj_metric_which aux_metric_has cop_metric_become det_metric_the amod_metric_standard nn_metric_evaluation rcmod__metric cc_used_and det_score_the nsubj_0905_score partmod_score_obtained cop_0905_was conj_used_0905 ' +o,4274,' and Chiang , in terms of what alignments they induce , has been discussed in and Wellington et al ',Wu,'nn__ cc__and conj__Chiang nsubjpass_discussed_ prep__in pobj_in_terms prep_terms_of nsubj_alignments_what pcomp_of_alignments nsubj_induce_they ccomp_alignments_induce aux_discussed_has auxpass_discussed_been prep_discussed_in pobj_in_ cc__and conj__Wellington cc_Wellington_et conj_Wellington_al ' +o,4275,'The basic engine used to perform the tagging in these experiments is a direct descendent of the maximum entropy -LRB- ME -RRB- tagger of which in turn is related to the taggers of and ',Ratnaparkhi,'det_engine_The amod_engine_basic nsubj_descendent_engine partmod_engine_used aux_perform_to xcomp_used_perform det_tagging_the dobj_perform_tagging prep_tagging_in det_experiments_these pobj_in_experiments cop_descendent_is det_descendent_a amod_descendent_direct prep_descendent_of det_entropy_the amod_entropy_maximum pobj_of_entropy abbrev_descendent_ME dep_descendent_tagger prep_tagger_of pobj_of_ nsubjpass_related_which prep_related_in pobj_in_turn auxpass_related_is rcmod_tagger_related prep_related_to det_taggers_the pobj_to_taggers prep_taggers_of pobj_of_ cc__and conj__ ' +p,4276,'21 Log-Linear Models The log-linear model -LRB- LLM -RRB- , or also known as maximum-entropy model , is a linear classifier widely used in the NLP literature ',Berger,'num_Log-Linear_21 nsubj_Models_Log-Linear det_model_The amod_model_log-linear dobj_Models_model abbrev_model_LLM cc_Models_or advmod_known_also conj_Models_known prep_known_as amod__maximum-entropy nn__model pobj_as_ cop_classifier_is det_classifier_a amod_classifier_linear conj_Models_classifier advmod_used_widely amod_classifier_used prep_classifier_in det_literature_the nn_literature_NLP pobj_in_literature ' +p,4277,'Although we see statistically significant improvements -LRB- at the 05 level on a paired permutation test -RRB- , the quality of the parsers is still quite poor , in contrast to other applications of bootstrapping which rival supervised methods ',Yarowsky,'mark_see_Although nsubj_see_we advcl_poor_see advmod_significant_statistically amod_improvements_significant dobj_see_improvements dep_improvements_at det_level_the num_level_05 pobj_at_level prep_level_on det_test_a amod_test_paired nn_test_permutation pobj_on_test det_quality_the nsubj_poor_quality prep_quality_of det_parsers_the pobj_of_parsers cop_poor_is advmod_poor_still advmod_poor_quite prep_poor_in dep_in_contrast dep_in_to amod_applications_other pobj_in_applications prep_applications_of pcomp_of_bootstrapping dobj_supervised_which nsubj_supervised_rival ccomp_bootstrapping_supervised dobj_supervised_methods ' +o,4278,'In recent years , many researchers have tried to automatically evaluate the quality of MT and improve the performance of automatic MT evaluations because improving the performance of automatic MT evaluation is expected to enable us to use and improve MT systems efficiently ',Papineni,'prep_tried_In amod_years_recent pobj_In_years amod_researchers_many nsubj_tried_researchers aux_tried_have aux_evaluate_to advmod_evaluate_automatically xcomp_tried_evaluate det_quality_the dobj_evaluate_quality prep_quality_of pobj_of_MT cc_evaluate_and conj_evaluate_improve det_performance_the dobj_improve_performance prep_performance_of amod_evaluations_automatic nn_evaluations_MT pobj_of_evaluations xcomp_improve_ mark_expected_because csubjpass_expected_improving det_performance_the dobj_improving_performance prep_performance_of amod_evaluation_automatic nn_evaluation_MT pobj_of_evaluation auxpass_expected_is advcl__expected aux_enable_to xcomp_expected_enable nsubj_use_us aux_use_to xcomp_enable_use cc_use_and conj_use_improve nn_systems_MT dobj_use_systems advmod_use_efficiently ' +o,4279,'52 Results on the Newsblaster data We measured how well the models trained on DUC data perform with current news labeled using human 4http : \\/ \\/ newsblastercscolumbiaedu 5a20 -LRB- kappa -RRB- is a measure of inter-annotator agreement over and above what might be expected by pure chance -LRB- See for discussion of its use in NLP -RRB- a20a22a21a24a23 if there is perfect agreement between annotators anda20a25a21a27a26 if the annotators agree only as much as you would expect by chance ',Carletta,'num_Results_52 nsubj_perform_Results prep_Results_on det_data_the nn_data_Newsblaster pobj_on_data nsubj_measured_We rcmod_data_measured advmod_trained_how advmod_how_well det_models_the nsubj_trained_models ccomp_measured_trained prep_trained_on nn_data_DUC pobj_on_data prep_perform_with amod_news_current pobj_with_news partmod_news_labeled xcomp_labeled_using amod_4http_human dobj_using_4http amod_5a20_\\/ amod_5a20_\\/ amod_5a20_newsblastercscolumbiaedu nsubj_measure_5a20 appos_5a20_kappa cop_measure_is det_measure_a parataxis_perform_measure prep_measure_of amod_agreement_inter-annotator pobj_of_agreement advmod_measure_over cc_over_and conj_over_above nsubjpass_expected_what aux_expected_might auxpass_expected_be dep_measure_expected prep_expected_by amod_a20a22a21a24a23_pure nn_a20a22a21a24a23_chance nn__See dep_a20a22a21a24a23_ prep__for pobj_for_discussion prep_discussion_of poss_use_its pobj_of_use prep_use_in pobj_in_NLP pobj_by_a20a22a21a24a23 mark_is_if expl_is_there dep_a20a22a21a24a23_is amod_agreement_perfect nsubj_is_agreement prep_agreement_between amod_anda20a25a21a27a26_annotators pobj_between_anda20a25a21a27a26 mark_agree_if det_annotators_the nsubj_agree_annotators advcl_is_agree advmod_expect_only advmod_expect_as advmod_expect_much dep_expect_as nsubj_expect_you aux_expect_would ccomp_agree_expect prep_expect_by pobj_by_chance ' +p,4280,'This paper demonstrates several of the characteristics and benefits of SemFrame , a system that produces such a resource ',Green,'det_paper_This nsubj_demonstrates_paper dobj_demonstrates_several prep_several_of det_characteristics_the pobj_of_characteristics cc_characteristics_and conj_characteristics_benefits prep_characteristics_of nn__SemFrame pobj_of_ det_system_a appos__system nsubj_produces_that rcmod_system_produces predet_resource_such det_resource_a dobj_produces_resource ' +p,4281,'3 The Perceptron The perceptron algorithm introduced into NLP by , is a simple but effective discriminative training method ',Collins,'nsubj_method_3 det_algorithm_The nn_algorithm_Perceptron nn_algorithm_The nn_algorithm_perceptron nsubj_introduced_algorithm rcmod_3_introduced prep_introduced_into pobj_into_NLP prep_introduced_by pobj_by_ cop_method_is det_method_a amod_method_simple cc_simple_but conj_simple_effective amod_method_discriminative nn_method_training ' +p,4282,'3 CLaC-NB System : Nave Bayes Supervised statistical methods have been very successful in sentiment tagging of texts and in subjectivity detection at sentence level : on movie review texts they reach an accuracy of 85-90 \% and up to 92 \% accuracy on classifying movie review snippets into subjective and objective using both Nave Bayes and SVM ',Pang,'num_System_3 nn_System_CLaC-NB nn_Supervised_Nave nn_Supervised_Bayes dep_System_Supervised amod_methods_statistical nsubj_successful_methods aux_successful_have dep_successful_been advmod_successful_very rcmod_Supervised_successful prep_successful_in pobj_in_sentiment xcomp_successful_tagging prep_tagging_of pobj_of_texts cc_of_and conj_of_in amod_detection_subjectivity pobj_in_detection prep_tagging_at nn_level_sentence pobj_at_level prep_tagging_on nn_texts_movie nn_texts_review pobj_on_texts nsubj_reach_they rcmod_texts_reach det_accuracy_an dobj_reach_accuracy prep_accuracy_of num_\%_85-90 pobj_of_\% tmod_reach_ dep__and advmod__up dep_up_to num_\%_92 pobj_to_\% dep__accuracy prep_tagging_on pcomp_on_classifying nn_snippets_movie nn_snippets_review dobj_classifying_snippets prep_classifying_into pobj_into_subjective cc_subjective_and conj_subjective_objective xcomp_successful_using preconj_Bayes_both nn_Bayes_Nave dobj_using_Bayes cc_Bayes_and conj_Bayes_SVM ' +o,4283,'After unioning the Viterbi alignments , the stems were replaced with their original words , and phrase-pairs of up to five foreign words in length were extracted in the usual fashion ',Koehn,'prep_replaced_After pcomp_After_unioning det_alignments_the nn_alignments_Viterbi dobj_unioning_alignments det_stems_the nsubjpass_replaced_stems auxpass_replaced_were prep_replaced_with poss_words_their amod_words_original pobj_with_words cc_replaced_and nsubjpass_extracted_phrase-pairs prep_phrase-pairs_of quantmod_five_up dep_five_to num_words_five amod_words_foreign pobj_of_words prep_words_in pobj_in_length auxpass_extracted_were conj_replaced_extracted prep_extracted_in det_fashion_the amod_fashion_usual pobj_in_fashion ' +o,4284,'Past work has examined the use of monolingual parallel corpora for paraphrase extraction ',Barzilay,'amod__Past nn__work nsubj_examined_ aux_examined_has det_use_the dobj_examined_use prep_use_of amod_corpora_monolingual amod_corpora_parallel pobj_of_corpora prep_examined_for amod_extraction_paraphrase pobj_for_extraction ' +o,4285,'Instead of analyzing sentences directly , AUCONTRAIRE relies on the TEXTRUNNER Open Information Extraction system to map each sentence to one or more tuples that represent the entities in the sentences and the relationships between them -LRB- eg , was born in -LRB- Mozart , Salzburg -RRB- -RRB- ',Banko,'dep_of_Instead advcl_relies_of pcomp_of_analyzing dobj_analyzing_sentences advmod_analyzing_directly nsubj_relies_AUCONTRAIRE prep_relies_on det__the nn__TEXTRUNNER nn__Open nn__Information nn__Extraction nn__system pobj_on_ aux_map_to xcomp_relies_map det_sentence_each dobj_map_sentence prep_map_to num_tuples_one cc_one_or conj_one_more pobj_to_tuples nsubj_represent_that rcmod_tuples_represent det_entities_the dobj_represent_entities prep_entities_in det_sentences_the pobj_in_sentences cc_entities_and det_relationships_the conj_entities_relationships prep_relationships_between pobj_between_them nsubjpass_born_eg auxpass_born_was parataxis_relies_born prep_born_in dep_in_Mozart dep_Mozart_Salzburg ' +p,4286,'1 Introduction Raw parallel data need to be preprocessed in the modern phrase-based SMT before they are aligned by alignment algorithms , one of which is the wellknown tool , GIZA + + , for training IBM models -LRB- 1-4 -RRB- ',Och,'num_data_1 nn_data_Introduction nn_data_Raw nn_data_parallel nsubj_need_data aux_preprocessed_to auxpass_preprocessed_be xcomp_need_preprocessed prep_preprocessed_in det_SMT_the amod_SMT_modern amod_SMT_phrase-based pobj_in_SMT mark_aligned_before nsubjpass_aligned_they auxpass_aligned_are dep_SMT_aligned prep_aligned_by amod_algorithms_alignment pobj_by_algorithms nsubj_tool_one prep_one_of pobj_of_which cop_tool_is det_tool_the nn_tool_wellknown rcmod_algorithms_tool appos_tool_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_tool_for amod_models_training nn_models_IBM pobj_for_models appos_SMT_1-4 ' +o,4287,'The training data for the French\\/English data set is taken from the LDC Canadian Hansard data set , from which the word aligned data was also taken ',Och,'det_data_The nn_data_training nsubjpass_taken_data prep_data_for det_set_the nn_set_French\\/English nn_set_data pobj_for_set auxpass_taken_is prep_taken_from det_set_the nn_set_LDC nn_set_Canadian nn_set_Hansard nn_set_data pobj_from_set rel_taken_from pobj_from_which det_word_the nsubjpass_taken_word partmod_word_aligned nn__data dobj_aligned_ auxpass_taken_was advmod_taken_also rcmod_set_taken ' +p,4288,'3 Language modelling with Bloom filters Recentwork presenteda scheme for associating static frequency information with a set of n-grams in a BF efficiently1 31 Log-frequency Bloom filter The efficiency of the scheme for storing n-gram statistics within a BF presented in Talbot and Osborne relies on the Zipf-like distribution of n-gramfrequencies : mosteventsoccuranextremely small number of times , while a small number are very frequent ',Talbot,'nsubj_modelling_Language dep_3_modelling prep_modelling_with nn_scheme_Bloom nn_scheme_filters nn_scheme_Recentwork nn_scheme_ nn_scheme_presenteda pobj_with_scheme prep_modelling_for pcomp_for_associating amod_information_static nn_information_frequency dobj_associating_information prep_associating_with det_set_a pobj_with_set prep_set_of pobj_of_n-grams prep_n-grams_in det_efficiently1_a nn_efficiently1_BF pobj_in_efficiently1 num_efficiently1_31 nn_Bloom_Log-frequency nsubj_filter_Bloom dep_3_filter det_efficiency_The dobj_filter_efficiency prep_efficiency_of det_scheme_the pobj_of_scheme prep_filter_for pcomp_for_storing amod_statistics_n-gram dobj_storing_statistics prep_storing_within det_BF_a pobj_within_BF partmod_BF_presented prep_presented_in pobj_in_Talbot cc_filter_and nn__Osborne nsubj_relies_ conj_filter_relies prep_relies_on det_distribution_the amod_distribution_Zipf-like pobj_on_distribution prep_distribution_of pobj_of_n-gramfrequencies amod_number_mosteventsoccuranextremely amod_number_small dep_n-gramfrequencies_number prep_number_of pobj_of_times mark_frequent_while det_number_a amod_number_small nsubj_frequent_number cop_frequent_are advmod_frequent_very advcl_relies_frequent ' +o,4289,'The feature weights are learned by maximizing the BLEU score on held-out data,usingminimum-error-ratetraining as implemented by ',Koehn,'det_weights_The nn_weights_feature nsubjpass_learned_weights auxpass_learned_are prep_learned_by pcomp_by_maximizing det__the amod__BLEU nn__score dobj_maximizing_ prep_maximizing_on amod__held-out amod__data,usingminimum-error-ratetraining pobj_on_ mark_implemented_as advcl_maximizing_implemented prep_implemented_by ' +o,4290,'We base our work partly on previous work done by Bagga and Baldwin , which has also been used in later work ',Chen,'nsubj_base_We poss_work_our dobj_base_work dep_on_partly prep_base_on amod_work_previous pobj_on_work partmod_work_done prep_done_by nn__Bagga cc_Bagga_and conj_Bagga_Baldwin pobj_by_ nsubjpass_used_which aux_used_has advmod_used_also auxpass_used_been rcmod__used prep_used_in amod_work_later pobj_in_work ' +o,4291,'We have developed a set of extensions to a probabilistic translation model that enable us to successfully merge oversegmented regions into coherent objects ',Brown,'nsubj_developed_We aux_developed_have det_set_a dobj_developed_set prep_set_of pobj_of_extensions prep_developed_to det__a amod__probabilistic nn__translation nn__model pobj_to_ nsubj_enable_that rcmod__enable nsubj_merge_us aux_merge_to advmod_merge_successfully xcomp_enable_merge amod_regions_oversegmented dobj_merge_regions prep_merge_into amod_objects_coherent pobj_into_objects ' +o,4292,'Evaluation metrics such as BLEU have a built-in preference for shorter translations ',Papineni,'nn_metrics_Evaluation nsubj_have_metrics dep_as_such prep_metrics_as nn__BLEU pobj_as_ det_preference_a amod_preference_built-in dobj_have_preference prep_preference_for amod_translations_shorter pobj_for_translations ' +o,4293,'4 POS Tagger and Named Entity Recognizer For the POS tagging task , the tagger is built based on the work of which was applied for English POS tagging ',Ratnaparkhi,'num_Tagger_4 dep_Tagger_POS cc_Tagger_and nn_Recognizer_Named nn_Recognizer_Entity conj_Tagger_Recognizer prep_built_For det_task_the dep_task_POS amod_task_tagging pobj_For_task det_tagger_the nsubjpass_built_tagger auxpass_built_is dep_Tagger_built prep_built_based dep_based_on det_work_the pobj_on_work prep_work_of pobj_of_ nsubjpass_applied_which auxpass_applied_was rcmod_work_applied prep_applied_for nn_POS_English pobj_for_POS partmod_POS_tagging ' +o,4294,'Early experiments with syntactically-informed phrases , and syntactic reranking of K-best lists produced mostly negative results ',Och,'advmod_experiments_Early nsubj__experiments prep_experiments_with amod_phrases_syntactically-informed pobj_with_phrases cc__and amod_reranking_syntactic nsubj_produced_reranking prep_reranking_of amod_lists_K-best pobj_of_lists advmod_produced_ conj__produced advmod_negative_mostly amod_results_negative dobj_produced_results ' +o,4295,'1 Specifically , MIMIC uses an n-dimensional call router front-end , which is a generalization of the vector-based call-routing paradigm of semantic interpretation ; that is , instead of detecting one concept per utterance , MIMIC \'s semantic interpretation engine detects multiple -LRB- n -RRB- concepts or classes conveyed by a single utterance , by using n call touters in parallel ',Chu-Carroll,'advmod_uses_Specifically nsubj_uses_MIMIC dep_1_uses det_router_an amod_router_n-dimensional nn_router_call dobj_uses_router amod__front-end tmod_uses_ nsubj_generalization_which cop_generalization_is det_generalization_a dep__generalization prep_generalization_of det_paradigm_the amod_paradigm_vector-based amod_paradigm_call-routing pobj_of_paradigm prep_paradigm_of amod__semantic nn__interpretation pobj_of_ nsubj_is_that rcmod_paradigm_is dep_of_instead advcl_conveyed_of pcomp_of_detecting num_concept_one dobj_detecting_concept prep_concept_per pobj_per_utterance poss_concepts_MIMIC possessive_MIMIC_\'s amod_concepts_semantic nn_concepts_interpretation nn_concepts_engine nn_concepts_detects nn_concepts_multiple appos_concepts_n nsubj_conveyed_concepts cc_concepts_or conj_concepts_classes ccomp_is_conveyed prep_conveyed_by det_utterance_a amod_utterance_single pobj_by_utterance prep_conveyed_by pcomp_by_using nn_touters_n nn_touters_call dobj_using_touters prep_using_in pobj_in_parallel ' +p,4296,'In the past five years , important research on the automatic acquisition of word classes based on lexical distribution has been published ',Church,'prep_published_In det_years_the amod_years_past num_years_five pobj_In_years amod_research_important nsubjpass_published_research prep_research_on det_acquisition_the amod_acquisition_automatic pobj_on_acquisition prep_acquisition_of nn_classes_word pobj_of_classes partmod_classes_based prep_based_on amod_distribution_lexical pobj_on_distribution aux_published_has auxpass_published_been ' +o,4297,'42 Cast3LB Function Tagging For the task of Cast3LB function tag assignment we experimented with three generic machine learning algorithms : a memory-based learner , a maximum entropy classifier and a Support Vector Machine classifier ',Berger,'num_Cast3LB_42 partmod_Cast3LB_Function xcomp_Function_Tagging prep_experimented_For det_task_the pobj_For_task prep_task_of amod_assignment_Cast3LB nn_assignment_function nn_assignment_tag pobj_of_assignment nsubj_experimented_we ccomp_Tagging_experimented prep_experimented_with num_machine_three amod_machine_generic pobj_with_machine partmod_machine_learning dobj_learning_algorithms det__a amod__memory-based nn__learner dep_Cast3LB_ det__a amod__maximum amod__entropy nn__classifier conj__ cc__and det__a nn__Support nn__Vector nn__Machine nn__classifier conj__ ' +o,4298,'Specifically , we explore the statistical term weighting features of the word generation model with Support Vector machine -LRB- SVM -RRB- , faithfully reproducing previous work as closely as possible ',Pang,'advmod_explore_Specifically nsubj_explore_we det_features_the amod_features_statistical nn_features_term nn_features_weighting dobj_explore_features prep_features_of det_model_the nn_model_word nn_model_generation pobj_of_model prep_model_with nn_machine_Support nn_machine_Vector pobj_with_machine abbrev_machine_SVM advmod_reproducing_faithfully xcomp_explore_reproducing amod_work_previous dobj_reproducing_work advmod_closely_as advmod_reproducing_closely dep_closely_as pobj_as_possible ' +o,4299,'5 Evaluation 51 Datasets We used two datasets , customer reviews 1 and movie reviews 2 to evaluate sentiment classification of sentences ',Pang,'number_Evaluation_5 num_Datasets_Evaluation num_Datasets_51 nsubj_used_We dep_Datasets_used num_datasets_two dobj_used_datasets nn_reviews_customer conj_datasets_reviews num__1 dep_reviews_ cc_datasets_and nn_reviews_movie conj_datasets_reviews num__2 dep_reviews_ aux_evaluate_to xcomp_used_evaluate nn_classification_sentiment dobj_evaluate_classification prep_classification_of pobj_of_sentences ' +o,4300,'Semantic DSN : The construction of this network is inspired by ',Lin,'nn_DSN_Semantic dep_inspired_DSN det_construction_The nsubjpass_inspired_construction prep_construction_of det_network_this pobj_of_network auxpass_inspired_is prep_inspired_by ' +o,4301,'distance -LRB- MSD -RRB- and the maximum swap segment size -LRB- MSSS -RRB- ranging from 0 to 10 and evaluated the translations with the BLEU7 metric ',Papineni,'abbrev_distance_MSD cc_distance_and det_size_the amod_size_maximum nn_size_swap nn_size_segment conj_distance_size abbrev_size_MSSS dep_distance_ranging prep_ranging_from num_10_0 dep_10_to pobj_from_10 cc_ranging_and conj_ranging_evaluated det_translations_the dobj_evaluated_translations prep_evaluated_with det_metric_the nn_metric_BLEU7 pobj_with_metric ' +o,4302,'An application of the idea of alternative targets can be seen in work on election prediction ',Kim,'det_application_An nsubjpass_seen_application prep_application_of det_idea_the pobj_of_idea prep_idea_of amod_targets_alternative pobj_of_targets aux_seen_can auxpass_seen_be prep_seen_in amod_work_ pobj_in_work prep_work_on nn_prediction_election pobj_on_prediction ' +o,4303,' from Sections 2-21 of the Wall Street Journal -LRB- WSJ -RRB- in the Penn Treebank and its subsets3 We then converted them into strongly equivalent HPSG-style grammars using the grammar conversion described in Section 21 ',Marcus,'prep__from pobj_from_Sections num_Sections_2-21 prep_Sections_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal abbrev_Journal_WSJ prep_Journal_in det__the nn__Penn nn__Treebank pobj_in_ cc__and poss_subsets3_its conj__subsets3 nsubj_converted_We advmod_converted_then rcmod_subsets3_converted dobj_converted_them prep_converted_into advmod_equivalent_strongly amod_grammars_equivalent amod_grammars_HPSG-style pobj_into_grammars xcomp_converted_using det_conversion_the nn_conversion_grammar dobj_using_conversion partmod_conversion_described prep_described_in pobj_in_Section num_Section_21 ' +o,4304,'As referring dataset , we used the PropBank corpora available at wwwcisupennedu\\/ace , along with the Penn TreeBank 2 -LRB- wwwcisupennedu\\/treebank -RRB- ',Marcus,'prep_used_As pcomp_As_referring dobj_referring_dataset nsubj_used_we det_corpora_the nn_corpora_PropBank nsubj_available_corpora xcomp_used_available prep_available_at pobj_at_wwwcisupennedu\\/ace advmod_wwwcisupennedu\\/ace_along dep_along_with det_TreeBank_the nn_TreeBank_Penn pobj_with_TreeBank dep_wwwcisupennedu\\/ace_2 appos_wwwcisupennedu\\/ace_wwwcisupennedu\\/treebank ' +o,4305,'See and Collins and Duffy -LRB- 2001 , 2002 -RRB- for applications of the perceptron algorithm ',Collins,'dobj_See_ cc__and conj__Collins cc_Collins_and conj_Collins_Duffy dep_Collins_2001 appos_2001_2002 prep_Collins_for pobj_for_applications prep_applications_of det_algorithm_the nn_algorithm_perceptron pobj_of_algorithm ' +o,4306,'There has been a growing interest in corpus-based approaches which retrieve collocations from large corpora , , , , , , ',Smadja,'expl_been_There aux_been_has det_interest_a amod_interest_growing dobj_been_interest prep_interest_in amod_approaches_corpus-based pobj_in_approaches nsubj_retrieve_which rcmod_interest_retrieve dobj_retrieve_collocations prep_retrieve_from amod_corpora_large pobj_from_corpora amod__ nsubj__ num__ num__ num__ num__ num__ dep_been_ ' +o,4307,'42 Smoothing : Gaussian Priors Since NLP maximum entropy models usually have lots of features and lots of sparseness -LRB- eg features seen in testing not occurring in training -RRB- , smoothing is essential as a way to optimize the feature weights ',Manning,'num_Smoothing_42 nn_Priors_Gaussian dep_Smoothing_Priors mark_have_Since nn_models_NLP nn_models_maximum nn_models_entropy nsubj_have_models advmod_have_usually advcl_essential_have dobj_have_lots prep_lots_of pobj_of_features cc_features_and conj_features_lots prep_lots_of pobj_of_sparseness nn_features_eg dep_sparseness_features partmod_features_seen prep_seen_in pcomp_in_testing neg_occurring_not acomp_testing_occurring prep_occurring_in pobj_in_training nsubj_essential_smoothing cop_essential_is rcmod_Priors_essential prep_essential_as det_way_a pobj_as_way aux_optimize_to xcomp_essential_optimize det_weights_the nn_weights_feature dobj_optimize_weights ' +o,4308,'It is possible that there is a better automated method for finding such phrases , such as the methods in ',Kanayama,'nsubj_possible_It cop_possible_is complm_is_that expl_is_there ccomp_possible_is det_method_a amod_method_better amod_method_automated nsubj_is_method prep_method_for pcomp_for_finding amod_phrases_such dobj_finding_phrases dep_as_such prep_method_as det_methods_the pobj_as_methods prep_methods_in ' +o,4309,'We then train word alignment models using 6 Model-1 iterations and 6 HMM iterations ',Och,'nsubj_train_We advmod_train_then nn_models_word nn_models_alignment nsubj__models ccomp_train_ xcomp__using num_iterations_6 amod_iterations_Model-1 dobj_using_iterations cc_iterations_and num_iterations_6 nn_iterations_HMM conj_iterations_iterations ' +o,4310,'Sang used the IOB tagging method proposed by Ramshow and memory-based learning for each level of chunking and achieved an f-score of 8049 on the Penn Treebank corpus ',Ramshaw,'dep_Sang_used det_method_the nn_method_IOB nn_method_tagging nsubj_Sang_method dep_method_proposed prep_proposed_by nn__Ramshow pobj_by_ cc_proposed_and conj_proposed_memory-based xcomp_memory-based_learning prep_learning_for det_level_each pobj_for_level prep_level_of pobj_of_chunking cc_memory-based_and conj_memory-based_achieved det_f-score_an dobj_achieved_f-score prep_f-score_of pobj_of_8049 prep_achieved_on det_corpus_the nn_corpus_Penn nn_corpus_Treebank pobj_on_corpus ' +o,4311,' present Factored Translation Models as an extension to phrase-based statistical machine translation models ',Koehn,'amod_Models_present nn_Models_Factored nn_Models_Translation nsubj__Models advmod_Models_as det_extension_an dep_as_extension dep_as_to amod_models_phrase-based amod_models_statistical nn_models_machine nn_models_translation pobj_to_models ' +o,4312,'We discriminatively trained our parser in an on-line fashion using a variant of the voted perceptron ',Collins,'nsubj_trained_We advmod_trained_discriminatively poss_parser_our dobj_trained_parser prep_parser_in det_fashion_an amod_fashion_on-line pobj_in_fashion partmod_fashion_using det_variant_a dobj_using_variant prep_variant_of det_perceptron_the amod_perceptron_voted pobj_of_perceptron ' +o,4313,'This criticism leads us to automatic approaches for building thesauri from large corpora ',Hindle,'det_criticism_This nsubj_leads_criticism dobj_leads_us prep_leads_to amod_approaches_automatic pobj_to_approaches prep_approaches_for nn_thesauri_building pobj_for_thesauri prep_thesauri_from amod_corpora_large pobj_from_corpora ' +o,4314,'In this paper , we use IBM model 1 in order to get the probability P -LRB- Q DA -RRB- as follows ',Brown,'prep_use_In det_paper_this pobj_In_paper nsubj_use_we nn_model_IBM nsubj__model num_model_1 ccomp_use_ mark_get_in dep_get_order aux_get_to purpcl__get det_P_the nn_P_probability dobj_get_P nn_DA_Q abbrev_P_DA mark_follows_as dep_P_follows ' +o,4315,'MXPOST , and in order to discover more general patterns , we map the tag set down after tagging , eg NN , NNP , NNPS and NNS all map to NN ',Ratnaparkhi,'nsubj__MXPOST cc__and mark_discover_in dep_discover_order aux_discover_to dep_map_discover amod_patterns_more amod_patterns_general dobj_discover_patterns nsubj_map_we conj__map det_tag_the dobj_map_tag partmod_tag_set prt_set_down prep_set_after amod_NN_tagging amod_NN_eg pobj_after_NN conj_NN_NNP conj_NN_NNPS cc_NN_and conj_NN_NNS nsubj_map_all dep__map prep_map_to pobj_to_NN ' +o,4316,'For instance , the to-PP frame is poorly \' represented in the syntactically annotated version of the Penn Treebank ',Dunning,'prep_is_For pobj_For_instance det_frame_the amod_frame_to-PP nsubj_is_frame advmod_is_poorly ccomp_is_represented prep_represented_in det_version_the advmod_annotated_syntactically amod_version_annotated pobj_in_version prep_version_of det__the nn__Penn nn__Treebank pobj_of_ ' +o,4317,'It also shows that DOP \'s frontier lexicalization is a viable alternative to constituent lexicalization ',Collins,'nsubj_shows_It advmod_shows_also complm_alternative_that poss_lexicalization_DOP possessive_DOP_\'s nn_lexicalization_frontier nsubj_alternative_lexicalization cop_alternative_is det_alternative_a amod_alternative_viable ccomp_shows_alternative prep_alternative_to amod_lexicalization_constituent pobj_to_lexicalization ' +o,4318,'They have used the representation as well -LRB- IOB1 -RRB- ',Ramshaw,'nsubj_used_They aux_used_have det_representation_the amod_representation_ dobj_used_representation dep_well_as cc_representation_well conj_representation_IOB1 ' +o,4319,'This sort of problem can be solved in principle by conditional variants of the Expectation-Maximization algorithm ',Brown,'det_sort_This nsubjpass_solved_sort prep_sort_of pobj_of_problem aux_solved_can auxpass_solved_be prep_solved_in pobj_in_principle prep_solved_by amod_variants_conditional pobj_by_variants prep_variants_of det__the amod__Expectation-Maximization nn__algorithm pobj_of_ ' +o,4320,'They are most commonly used for parsing and linguistic analysis , but are now commonly seen in applications like machine translation and question answering ',Johnson,'nsubjpass_used_They auxpass_used_are advmod_commonly_most advmod_used_commonly prep_used_for amod__parsing cc_parsing_and conj_parsing_linguistic nn__analysis pobj_for_ cc_used_but auxpass_seen_are advmod_seen_now advmod_seen_commonly conj_used_seen prep_seen_in pobj_in_applications prep_applications_like nn__machine nn__translation pobj_like_ cc__and nn__question nn__answering conj__ ' +n,4321,'It has been difficult to identify all and only those cases where a token functions as a discourse connective , and in many cases , the syntactic analysis in the Penn TreeBank provides no help ',Marcus,'nsubj_difficult_It aux_difficult_has dep_difficult_been aux_identify_to xcomp_difficult_identify dobj_identify_all cc_all_and advmod_all_only det_cases_those dep_all_cases advmod_connective_where det_functions_a amod_functions_token nsubj_connective_functions prep_functions_as det_discourse_a pobj_as_discourse advcl_difficult_connective cc_difficult_and prep_provides_in amod_cases_many pobj_in_cases det_analysis_the amod_analysis_syntactic nsubj_provides_analysis prep_analysis_in det__the nn__Penn nn__TreeBank pobj_in_ conj_difficult_provides det_help_no dobj_provides_help ' +o,4322,'We run Maximum BLEU for 25 iterations individually for each system ',Och,'nsubj_run_We nn__Maximum nn__BLEU dobj_run_ prep__for num_iterations_25 pobj_for_iterations advmod_iterations_individually dep_individually_for det_system_each pobj_for_system ' +n,4323,'For example , 10 million words of the American National Corpus will have manually corrected POS tags , a tenfold increase over the Penn Treebank , currently used for training POS taggers ',Marcus,'prep_corrected_For pobj_For_example number_million_10 num_words_million nsubj_corrected_words prep_words_of det__the nn__American nn__National nn__Corpus pobj_of_ aux_corrected_will aux_corrected_have advmod_corrected_manually dep_tags_POS nsubj_used_tags det_increase_a amod_increase_tenfold appos_tags_increase prep_increase_over det__the nn__Penn nn__Treebank pobj_over_ advmod_used_currently dep_corrected_used prep_used_for nn_taggers_training dep_taggers_POS pobj_for_taggers ' +p,4324,'Ranking algorithms , such as Kleinbergs HITS algorithm or Googles PageRank , have been traditionally and successfully used in Web-link analysis , social networks , and more recently in text processing applications , , ',Mihalcea,'amod_algorithms_Ranking nsubj_been_algorithms dep_as_such prep_algorithms_as nn__Kleinbergs nn__HITS nn__algorithm pobj_as_ cc__or nn__Googles nn__PageRank conj__ aux_been_have advmod_been_traditionally cc_been_and advmod_used_successfully conj_been_used prep_used_in amod__Web-link nn__analysis nn_applications_ amod_networks_social appos__networks cc__and advmod_recently_more conj__recently prep_recently_in nn_processing_text pobj_in_processing pobj_in_applications amod_applications_ dep__ partmod_applications_ ' +o,4325,'GIZA + + refined alignments have been used in state-of-the-art phrase-based statistical MT systems such as ; variations on the refined heuristic have been used by -LRB- diag and diag-and -RRB- and by the phrase-based system Moses -LRB- grow-diag-final -RRB- ',Koehn,'amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ amod_alignments_refined nsubjpass_used_alignments aux_used_have auxpass_used_been prep_used_in amod_systems_state-of-the-art amod_systems_phrase-based amod_systems_statistical nn_systems_MT pobj_in_systems dep_as_such prep_systems_as pobj_as_ nsubjpass_used_variations prep_variations_on det_heuristic_the amod_heuristic_refined pobj_on_heuristic aux_used_have auxpass_used_been parataxis_used_used prep_used_by pobj_by_ dep__diag cc_diag_and conj_diag_diag-and cc_by_and conj_by_by det_Moses_the amod_Moses_phrase-based nn_Moses_system pobj_by_Moses appos_Moses_grow-diag-final ' +o,4326,'By introducing the hidden word alignment variable a, the following approximate optimization criterion can be applied for that purpose: e = argmaxe Pr(e | f) = argmaxe summationdisplay a Pr(e,a | f) argmaxe,a Pr(e,a | f) Exploiting the maximum entropy (Berger et al. , 1996) framework, the conditional distribution Pr(e,a | f) can be determined through suitable real valued functions (called features) hr(e,f,a),r = 1R, and takes the parametric form: p(e,a | f) exp Rsummationdisplay r=1 rhr(e,f,a)} The ITC-irst system (Chen et al. , 2005) is based on a log-linear model which extends the original IBM Model 4 (Brown et al. , 1993) to phrases (Koehn et al. , 2003; Federico and Bertoldi, 2005).',Brown,'' +o,4327,'We consider three class models , models S , M , and L , defined as pS -LRB- cj c1cj1 , w1wj1 -RRB- = png -LRB- cj cj2cj1 -RRB- pS -LRB- wj c1cj , w1wj1 -RRB- = png -LRB- wj cj -RRB- pM -LRB- cj c1cj1 , w1wj1 -RRB- = png -LRB- cj cj2cj1 , wj2wj1 -RRB- pM -LRB- wj c1cj , w1wj1 -RRB- = png -LRB- wj wj2wj1cj -RRB- pL -LRB- cj c1cj1 , w1wj1 -RRB- = png -LRB- cj wj2cj2wj1cj1 -RRB- pL -LRB- wj c1cj , w1wj1 -RRB- = png -LRB- wj wj2cj2wj1cj1cj -RRB- Model S is an exponential version of the class-based n-gram model from ; model M is a novel model introduced in ; and model L is an exponential version of the model indexpredict from ',Brown,'nsubj_consider_We num_models_three nn_models_class nsubj_defined_models nn_S_models appos_models_S conj_S_M cc_S_and conj_S_L ccomp_consider_defined mark_=_as nsubj_=_pS nn_c1cj1_cj appos_pS_c1cj1 dep_c1cj1_w1wj1 advcl_defined_= dobj_=_png nn_cj2cj1_cj appos_png_cj2cj1 nsubj_=_pS nn_c1cj_wj appos_pS_c1cj dep_c1cj_w1wj1 rcmod_png_= dobj_=_png nn_cj_wj appos_png_cj nsubj_=_pM nn_c1cj1_cj appos_pM_c1cj1 dep_c1cj1_w1wj1 rcmod_png_= dobj_=_png nn_cj2cj1_cj appos_png_cj2cj1 dep_cj2cj1_wj2wj1 nsubj_=_pM nn_c1cj_wj appos_pM_c1cj dep_c1cj_w1wj1 rcmod_png_= dobj_=_png nn_wj2wj1cj_wj appos_png_wj2wj1cj nsubj_=_pL nn_c1cj1_cj appos_pL_c1cj1 dep_c1cj1_w1wj1 rcmod_png_= dobj_=_png nn_wj2cj2wj1cj1_cj appos_png_wj2cj2wj1cj1 nsubj_=_pL nn_c1cj_wj appos_pL_c1cj dep_c1cj_w1wj1 rcmod_png_= dobj_=_png nn_wj2cj2wj1cj1cj_wj appos_png_wj2cj2wj1cj1cj nn_S_Model nsubj_version_S cop_version_is det_version_an amod_version_exponential rcmod_png_version prep_version_of det_model_the amod_model_class-based amod_model_n-gram pobj_of_model prep_=_from pobj_from_ nn_M_model nsubj_model_M cop_model_is det_model_a amod_model_novel parataxis_consider_model partmod_model_introduced prep_introduced_in pobj_in_ cc_consider_and nn_L_model nsubj_version_L cop_version_is det_version_an amod_version_exponential conj_consider_version prep_version_of det_indexpredict_the nn_indexpredict_model pobj_of_indexpredict prep_indexpredict_from pobj_from_ ' +o,4328,'Errors from the sentence boundary detector in GATE were especially problematic because they caused the Collins parser to fail , resulting in no dependency tree information ',Pang,'nsubj_problematic_Errors prep_Errors_from det_detector_the nn_detector_sentence nn_detector_boundary pobj_from_detector prep_detector_in nn__GATE pobj_in_ cop_problematic_were advmod_problematic_especially mark_caused_because nsubj_caused_they advcl_problematic_caused det_parser_the nn_parser_Collins dobj_caused_parser aux_fail_to infmod_parser_fail xcomp_caused_resulting prep_resulting_in det_information_no amod_information_dependency nn_information_tree pobj_in_information ' +p,4329,'Some notable efforts in this direction for other languages have been the Penn Tree Bank for English and the Prague Dependency Bank for Czech ',Marcus,'det_efforts_Some amod_efforts_notable nsubj_been_efforts prep_efforts_in det_direction_this pobj_in_direction prep_direction_for amod_languages_other pobj_for_languages aux_been_have det__the nn__Penn nn__Tree nn__Bank dobj_been_ prep__for pobj_for_English cc__and det__the nn__Prague nn__Dependency nn__Bank conj__ prep__for pobj_for_Czech ' +o,4330,'The system combination weights one for each system , LM weight , and word and NULL insertion penalties were tuned to maximize the BLEU score on the tuning set -LRB- newssyscomb2009 -RRB- ',Papineni,'det_combination_The nn_combination_system nsubj_weights_combination dobj_weights_one prep_weights_for det_system_each pobj_for_system nn_weight_LM appos_system_weight cc_weights_and nn_penalties_word cc_word_and conj_word_NULL nn_penalties_insertion nsubjpass_tuned_penalties auxpass_tuned_were conj_weights_tuned aux_maximize_to xcomp_tuned_maximize det_score_the amod_score_BLEU amod_score_ dobj_maximize_score prep_maximize_on det_set_the amod_set_tuning pobj_on_set appos_set_newssyscomb2009 ' +o,4331,'For instance , the frequency collected from the data can be used to bias initial transition and emission probabilities in an HMM model ; the tagged words in IGT can be used to label the resulting clusters produced by the word clustering approach ; the frequent and unambiguous words in the target lines can serve as prototype examples in the prototype-driven approach ',Haghighi,'prep_used_For pobj_For_instance det_frequency_the nsubjpass_used_frequency partmod_frequency_collected prep_collected_from det_data_the pobj_from_data aux_used_can auxpass_used_be prep_used_to nn_transition_bias amod_transition_initial pobj_to_transition cc_transition_and nn_probabilities_emission conj_transition_probabilities prep_transition_in det_model_an nn_model_HMM pobj_in_model det_words_the amod_words_tagged nsubjpass_used_words prep_words_in pobj_in_IGT aux_used_can auxpass_used_be parataxis_used_used aux_label_to purpcl_used_label det_clusters_the amod_clusters_resulting dobj_label_clusters partmod_clusters_produced prep_produced_by det_approach_the nn_approach_word amod_approach_clustering pobj_by_approach det_words_the amod_words_frequent cc_frequent_and conj_frequent_unambiguous nsubj_serve_words prep_words_in det_lines_the nn_lines_target pobj_in_lines aux_serve_can parataxis_used_serve prep_serve_as nn_examples_prototype pobj_as_examples prep_examples_in det_approach_the amod_approach_prototype-driven pobj_in_approach ' +o,4332,'The learning algorithm follows the global strategy introduced in and adapted in for partial parsing tasks ',Collins,'det_algorithm_The nn_algorithm_learning nsubj_follows_algorithm det_strategy_the amod_strategy_global dobj_follows_strategy dep_strategy_introduced prep_introduced_in pobj_in_ cc_introduced_and conj_introduced_adapted prep_adapted_in pobj_in_ prep_introduced_for amod_tasks_partial amod_tasks_parsing pobj_for_tasks ' +o,4333,'In a different work , argued that the measured reliability of metrics can be due to averaging effects but might not be robust across translations ',Banerjee,'prep_argued_In det_work_a amod_work_different pobj_In_work nsubj_argued_ complm_due_that det_reliability_the amod_reliability_measured nsubj_due_reliability prep_reliability_of pobj_of_metrics aux_due_can cop_due_be ccomp_argued_due prep_due_to pcomp_to_averaging dobj_averaging_effects cc_argued_but aux_robust_might neg_robust_not cop_robust_be conj_argued_robust prep_robust_across pobj_across_translations ' +o,4334,'We automatically measure performance by comparing the produced headlines against one reference headline produced by a human using ROUGEa129 ',Lin,'nsubj_measure_We advmod_measure_automatically dobj_measure_performance prep_measure_by pcomp_by_comparing det_headlines_the amod_headlines_produced dobj_comparing_headlines prep_comparing_against pobj_against_one nn_headline_reference nsubj_produced_headline dep_measure_produced prep_produced_by det_human_a pobj_by_human xcomp_produced_using dobj_using_ROUGEa129 ' +p,4335,'In informal experiments described elsewhere , I found that the G 2 statistic suggested by slightly outperforms 2 ',Dunning,'prep_found_In amod_experiments_informal pobj_In_experiments partmod_experiments_described advmod_described_elsewhere dep_elsewhere_ nsubj_found_I complm_outperforms_that det_statistic_the nn_statistic_G num_statistic_2 nsubj_outperforms_statistic partmod_statistic_suggested prep_suggested_by pobj_by_ advmod_outperforms_slightly ccomp_found_outperforms dobj_outperforms_2 ' +o,4336,' effectively utilized unlabeled data to improve parsing accuracy on the standard WSJ training set , but they used a two-stage parser comprised of Charniaks lexicalized probabilistic parser with n-best parsing and a discriminative reranking parser , and thus it would be better categorized as co-training ',McClosky,'nsubj_utilized_ advmod_utilized_effectively amod_data_unlabeled nsubj_improve_data aux_improve_to xcomp_utilized_improve amod_accuracy_parsing dobj_improve_accuracy prep_accuracy_on det_set_the amod_set_standard nn_set_WSJ nn_set_training pobj_on_set cc_utilized_but nsubj_used_they conj_utilized_used det_parser_a amod_parser_two-stage nsubj_lexicalized_parser partmod_parser_comprised prep_comprised_of pobj_of_Charniaks ccomp_used_lexicalized amod_parser_probabilistic dobj_lexicalized_parser prep_lexicalized_with amod_parsing_n-best pobj_with_parsing cc_parsing_and det__a amod__discriminative amod__reranking nn__parser conj_parsing_ cc_utilized_and advmod_categorized_thus nsubjpass_categorized_it aux_categorized_would auxpass_categorized_be advmod_categorized_better conj_utilized_categorized prep_categorized_as pobj_as_co-training ' +o,4337,'713 Similarity via pagerank Pagerank is the celebrated citation ranking algorithm that has been applied to several natural language problems from summarization to opinion mining to our task of lexical relatedness ',Hughes,'nn_Similarity_713 nsubj_algorithm_Similarity prep_Similarity_via nn__pagerank nn__Pagerank pobj_via_ cop_algorithm_is det_algorithm_the amod_algorithm_celebrated nn_algorithm_citation nn_algorithm_ranking nsubjpass_applied_that aux_applied_has auxpass_applied_been rcmod_algorithm_applied prep_applied_to amod_problems_several amod_problems_natural nn_problems_language pobj_to_problems prep_problems_from nn__summarization pobj_from_ prep_applied_to nn__opinion nn__mining pobj_to_ prep_applied_to poss_task_our pobj_to_task prep_task_of amod_relatedness_lexical pobj_of_relatedness ' +o,4338,'Furthermore , the underlying decoding strategies are too time consuming for our application We therefore use a translation model based on the simple linear interpolation given in equation 2 which combines predictions of two translation models - Ms and M ~ - both based on IBM-like model ',Brown,'advmod_consuming_Furthermore det_strategies_the amod_strategies_underlying nn_strategies_decoding nsubj_consuming_strategies aux_consuming_are advmod_consuming_too dep_too_time prep_consuming_for poss_application_our pobj_for_application nsubj_use_We advmod_use_therefore dep_consuming_use det_model_a nn_model_translation dobj_use_model prep_use_based dep_based_on det_interpolation_the amod_interpolation_simple amod_interpolation_linear pobj_on_interpolation prep_interpolation_given dep_given_in pobj_in_equation num_equation_2 nsubj_combines_which rcmod_equation_combines dobj_combines_predictions prep_predictions_of num_models_two nn_models_translation pobj_of_models dep_predictions_Ms cc_Ms_and nn_~_M conj_Ms_~ dep_predictions_both partmod_both_based prep_based_on amod_model_IBM-like pobj_on_model ' +o,4339,'We use the following features in our induced English-to-English grammar :3 3Hiero also uses lexical weights in both 122 ? ? The joint probability of the two English hierarchical paraphrases , conditioned on the nonterminal symbol , as defined by this formula : p -LRB- e1 , e2 x -RRB- = c -LRB- X ? ? ? e1 , e2 ? ? summationtext e1prime , e2prime c -LRB- X ? ? ? e1prime , e2prime ? ? ',Koehn,'nsubj_use_We det_features_the amod_features_following dobj_uses_features prep_features_in poss_:3_our amod_:3_induced amod_:3_English-to-English nn_:3_grammar pobj_in_:3 nsubj_uses_3Hiero advmod_uses_also dep_use_uses amod_weights_lexical dobj_uses_weights advmod_uses_ prep_uses_in dep_122_both pobj_in_122 det_joint_The nsubj_probability_joint dep_uses_probability prep_probability_of det_paraphrases_the num_paraphrases_two amod_paraphrases_English amod_paraphrases_hierarchical pobj_of_paraphrases partmod_paraphrases_conditioned prep_conditioned_on det_symbol_the amod_symbol_nonterminal pobj_on_symbol advmod_defined_as amod_symbol_defined prep_probability_by det_formula_this pobj_by_formula dep_=_p appos_p_e1 num_x_e2 dep_e1_x advmod_formula_= nn_X_c dep_formula_X dep_uses_e1 appos_e1_e2 nsubj_e1prime_summationtext dep_uses_e1prime advmod_c_e2prime parataxis_e1prime_c dobj_c_X dep_uses_e1prime appos_e1prime_e2prime ' +o,4340,' 866 867 119 Klein and Manning 869 857 863 309 110 Charniak 874 875 100 886 881 091 Table 3 : Comparison with other parsers -LRB- sentences of length 40 -RRB- as head information ',Collins,'number_866_ num_119_866 number_119_867 measure_Klein_119 cc_Klein_and conj_Klein_Manning number_091_ cc_091_869 conj_091_857 number_091_863 number_091_309 number_091_110 number_091_Charniak number_091_ number_091_874 number_091_875 number_091_100 number_091_ number_091_ number_091_886 number_091_881 dobj_Manning_091 cc_091_Table conj_091_3 dep_091_Comparison prep_Comparison_with amod_parsers_other pobj_with_parsers dep_091_sentences prep_sentences_of pobj_of_length num_length_40 prep_Manning_as nn_information_head pobj_as_information ' +o,4341,'The superiority of discriminative models has been shown on many tasks when the discriminative and generative models use exactly the same model structure ',Klein,'det_superiority_The nsubjpass_shown_superiority prep_superiority_of amod_models_discriminative pobj_of_models aux_shown_has auxpass_shown_been prep_shown_on amod_tasks_many pobj_on_tasks advmod_use_when det_models_the nn_models_discriminative cc_discriminative_and conj_discriminative_generative nsubj_use_models advcl_shown_use advmod_structure_exactly det_structure_the amod_structure_same nn_structure_model dobj_use_structure ' +o,4342,'ENGLISH GERMAN CHINESE TrainSet Section 2-21 Sentences 1-18 ,602 Articles 26-270 DevSet Section 22 18,603-19 ,602 Articles 1-25 TestSet Section 23 19,603-20 ,602 Articles 271-300 Table 3 : Experimental setup ',Marcus,'nn_CHINESE_ENGLISH nn_CHINESE_GERMAN dep_Section_CHINESE nn_Section_ nn_Section_ nn_Section_ nn_Section_TrainSet dep_CHINESE_Section num_Sentences_2-21 dep_Section_Sentences num_Section_1-18 num_Section_,602 num_Section_Articles num_Section_26-270 nn_Section_DevSet num_Section_22 num_Section_18,603-19 num_Section_,602 num_Section_Articles num_Section_1-25 nn_Section_TestSet dep_Section_Section num_3_23 num_3_19,603-20 num_3_,602 num_3_Articles num_3_271-300 nn_3_Table dep_Section_3 amod_setup_Experimental dep_Section_setup ' +o,4343,'In the following experiments , we run two machine learning classifiers : Bayes Point Machines -LRB- BPM -RRB- , and the maximum entropy model -LRB- ME -RRB- ',Berger,'prep_run_In det_experiments_the amod_experiments_following pobj_In_experiments nsubj_run_we num_machine_two dobj_run_machine partmod_machine_learning dobj_learning_classifiers nn__Bayes nn__Point nn__Machines abbrev__BPM dep_classifiers_ cc__and det_model_the amod_model_maximum nn_model_entropy conj__model abbrev_model_ME ' +o,4344,'Semi-supervised conditional random fields -LRB- CRFs -RRB- based on a minimum entropy regularizer -LRB- SS-CRF-MER -RRB- have been proposed in ',Jiao,'amod_fields_Semi-supervised amod_fields_conditional amod_fields_random nsubjpass_proposed_fields abbrev_fields_CRFs prep_fields_based dep_based_on det_regularizer_a amod_regularizer_minimum nn_regularizer_entropy pobj_on_regularizer abbrev_regularizer_SS-CRF-MER aux_proposed_have auxpass_proposed_been prep_proposed_in ' +o,4345,' and Gao & Johnson -LRB- 2008 -RRB- assume that words are generated by a hidden Markov model and find that the resulting states strongly correlate with POS tags ',Johnson,'nsubj_assume_ cc__and conj__Gao cc_Gao_& conj_Gao_Johnson appos_Gao_2008 complm_generated_that nsubjpass_generated_words auxpass_generated_are ccomp_assume_generated prep_generated_by det_model_a amod_model_hidden nn_model_Markov pobj_by_model cc_generated_and conj_generated_find complm_correlate_that det_states_the amod_states_resulting nsubj_correlate_states advmod_correlate_strongly ccomp_find_correlate prep_correlate_with dep_tags_POS pobj_with_tags ' +o,4346,'optimization approaches which aim at selecting those examples that optimize some -LRB- algorithm-dependent -RRB- objective function , such as prediction variance , and heuristic methods with uncertainty sampling and query-by-committee -LRB- QBC -RRB- just to name the most prominent ones ',Berger,'nn_approaches_optimization nsubj_aim_which rcmod_approaches_aim prep_aim_at pcomp_at_selecting det_examples_those dobj_selecting_examples nsubj_optimize_that rcmod_examples_optimize det_function_some appos_function_algorithm-dependent nn_function_objective dobj_optimize_function dep_as_such prep_function_as nn__prediction nn__variance pobj_as_ cc_examples_and amod_methods_heuristic conj_examples_methods prep_methods_with nn__uncertainty amod__sampling pobj_with_ cc__and amod__query-by-committee abbrev__QBC conj__ advmod_name_just aux_name_to infmod_approaches_name det_ones_the advmod_prominent_most amod_ones_prominent dobj_name_ones ' +o,4347,'Lexical Weighting : -LRB- e -RRB- the lexical weight a27 a14a12a91 a29 a92a93a21 of the block a9 a72 a14a12a91 a19a86a92a93a21 is computed similarly to , details are given in Section 34 ',Koehn,'nn_Weighting_Lexical dep_computed_e det_a92a93a21_the amod_a92a93a21_lexical nn_a92a93a21_weight nn_a92a93a21_a27 nn_a92a93a21_a14a12a91 nn_a92a93a21_a29 nsubj_computed_a92a93a21 prep_a92a93a21_of det_a19a86a92a93a21_the nn_a19a86a92a93a21_block nn_a19a86a92a93a21_a9 nn_a19a86a92a93a21_a72 nn_a19a86a92a93a21_a14a12a91 pobj_of_a19a86a92a93a21 cop_computed_is ccomp_given_computed advmod_computed_similarly aux__to xcomp_computed_ nsubjpass_given_details auxpass_given_are dep_Weighting_given prep_given_in pobj_in_Section num_Section_34 ' +o,4348,'Predicate argument structures , which consist of complements -LRB- case filler nouns and case markers -RRB- and verbs , have also been used in the task of noun classification ',Hindle,'amod_structures_Predicate nn_structures_argument nsubjpass_used_structures nsubj_consist_which rcmod_structures_consist prep_consist_of pobj_of_complements nn_nouns_case nn_nouns_filler dep_complements_nouns cc_nouns_and nn_markers_case conj_nouns_markers cc_consist_and conj_consist_verbs aux_used_have advmod_used_also auxpass_used_been prep_used_in det_task_the pobj_in_task prep_task_of nn_classification_noun pobj_of_classification ' +o,4349,'62 Experimental Settings We utilize a maximum entropy -LRB- ME -RRB- model to design the basic classifier for WSD and TC tasks ',Berger,'num_Settings_62 nn_Settings_Experimental nsubj_utilize_We dep_Settings_utilize det_entropy_a amod_entropy_maximum dobj_utilize_entropy abbrev_entropy_ME nsubj__model dep_Settings_ aux_design_to xcomp__design det_classifier_the amod_classifier_basic dobj_design_classifier prep_classifier_for nn_tasks_WSD cc_WSD_and conj_WSD_TC pobj_for_tasks ' +o,4350,'It has been used for a variety of tasks , such as wide-coverage parsing , sentence realization , learning semantic parsers , dialog systems , grammar engineering , and modeling syntactic priming ',Zettlemoyer,'nsubjpass_used_It aux_used_has auxpass_used_been prep_used_for det_variety_a pobj_for_variety prep_variety_of pobj_of_tasks dep_as_such prep_variety_as amod__wide-coverage amod__parsing pobj_as_ nn__sentence nn__realization conj__ conj__learning amod_parsers_semantic dobj_learning_parsers advmod_learning_ amod_systems_dialog measure__systems amod__ dep__grammar dep__engineering amod__ cc__and amod__modeling amod__syntactic amod__priming conj__ ' +o,4351,'3 Network Evaluation We present an evaluation which has been carried out on an initial set of annotations of English articles from The Wall Street Journal -LRB- covering those annotated at the syntactic level in the Penn Treebank -RRB- ',Marcus,'num_Evaluation_3 nn_Evaluation_Network nsubj_present_We rcmod_Evaluation_present det_evaluation_an dobj_present_evaluation nsubjpass_carried_which aux_carried_has auxpass_carried_been rcmod_evaluation_carried prt_carried_out prep_carried_on det_set_an amod_set_initial pobj_on_set prep_set_of pobj_of_annotations prep_annotations_of amod_articles_English pobj_of_articles prep_carried_from det_Journal_The nn_Journal_Wall nn_Journal_Street pobj_from_Journal dep_Evaluation_covering dobj_covering_those partmod_those_annotated prep_annotated_at det_level_the amod_level_syntactic pobj_at_level prep_level_in det__the nn__Penn nn__Treebank pobj_in_ ' +n,4352,'The process of phrase extraction is difficult to optimize in a non-discriminative setting : many heuristics have been proposed , but it is not obvious which one should be chosen for a given language pair ',Koehn,'det_process_The nsubj_difficult_process prep_process_of nn_extraction_phrase pobj_of_extraction cop_difficult_is aux_optimize_to xcomp_difficult_optimize prep_optimize_in det_setting_a amod_setting_non-discriminative pobj_in_setting amod_heuristics_many nsubjpass_proposed_heuristics aux_proposed_have auxpass_proposed_been parataxis_difficult_proposed dep_proposed_ cc_proposed_but nsubj_obvious_it cop_obvious_is neg_obvious_not conj_proposed_obvious dobj_chosen_which nsubjpass_chosen_one aux_chosen_should auxpass_chosen_be ccomp_obvious_chosen prep_chosen_for det_pair_a amod_pair_given nn_pair_language pobj_for_pair ' +o,4353,'The labeling agreement was 84 \% -LRB- n = 80 ; -RRB- ',Carletta,'det_agreement_The nn_agreement_labeling nsubj_\%_agreement cop_\%_was num_\%_84 nsubj_=_n dep_\%_= dobj_=_80 dep_80_ ' +n,4354,'While several methods have been proposed to automatically extract compounds , we know of no successful attempt to automatically make classes of compounds ',Smadja,'mark_proposed_While amod_methods_several nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been advcl_know_proposed aux_extract_to advmod_extract_automatically xcomp_proposed_extract amod__compounds dobj_extract_ nsubj_know_we prep_know_of det_attempt_no amod_attempt_successful pobj_of_attempt aux_make_to advmod_make_automatically xcomp_know_make dobj_make_classes prep_classes_of pobj_of_compounds ' +o,4355,'These tasks are generally treated as sequential labeling problems incorporating the IOB tagging scheme ',Ramshaw,'det_tasks_These nsubjpass_treated_tasks auxpass_treated_are advmod_treated_generally prep_treated_as amod_problems_sequential nn_problems_labeling pobj_as_problems xcomp_treated_incorporating det_scheme_the nn_scheme_IOB nn_scheme_tagging dobj_incorporating_scheme ' +o,4356,'We measure translation performance by the BLEU score with one reference for each hypothesis ',Papineni,'nsubj_measure_We nn_performance_translation dobj_measure_performance prep_measure_by det__the amod__BLEU nn__score pobj_by_ prep__with num_reference_one pobj_with_reference prep_reference_for det_hypothesis_each pobj_for_hypothesis ' +o,4357,'However , the Naive Bayes classifier has been found to perform well for word-sense disambiguation both here and in a variety of other works -LRB- eg , , , , and -RRB- ',Berger,'advmod_found_However det_classifier_the nn_classifier_Naive nn_classifier_Bayes nsubjpass_found_classifier aux_found_has auxpass_found_been aux_perform_to xcomp_found_perform advmod_perform_well prep_perform_for amod_disambiguation_word-sense pobj_for_disambiguation dep_disambiguation_both dep_both_here cc_both_and conj_both_in det_variety_a pobj_in_variety prep_variety_of amod_works_other pobj_of_works dep_works_eg nn__ conj__ conj__ cc__and appos_eg_ ' +o,4358,' describe an error-driven transformation-based learning -LRB- TBL -RRB- method for finding NP chunks in texts ',Ramshaw,'advmod_describe_ det_learning_an amod_learning_error-driven amod_learning_transformation-based nsubj_describe_learning abbrev_learning_TBL dep_learning_method prep_method_for pcomp_for_finding nn_chunks_NP dobj_finding_chunks prep_finding_in pobj_in_texts ' +p,4359,'Introduction The creation of the Penn Treebank and the word sense-annotated SEMCOR have shown how even limited amounts of annotated data can result in major improvements in complex natural language understanding systems ',Marcus,'det_creation_The nsubj_sense-annotated_creation prep_creation_of det__the nn__Penn nn__Treebank pobj_of_ cc__and det_word_the conj__word dep_Introduction_sense-annotated nn__SEMCOR nsubj_shown_ aux_shown_have ccomp_sense-annotated_shown advmod_result_how advmod_how_even amod_amounts_limited nsubj_result_amounts prep_amounts_of amod_data_annotated pobj_of_data aux_result_can ccomp_shown_result prep_result_in amod_improvements_major pobj_in_improvements prep_improvements_in amod_systems_complex amod_systems_natural nn_systems_language nn_systems_understanding pobj_in_systems ' +o,4360,'For practical reasons , the maximum size of a token was set at three for Chinese , andfor forKorean2 Minimum error rate training was run on each system afterwardsand BLEU score was calculated on the test sets ',Och,'prep_set_For amod_reasons_practical pobj_For_reasons det_size_the amod_size_maximum nsubjpass_set_size prep_size_of det_token_a pobj_of_token auxpass_set_was prep_set_at pobj_at_three prep_set_for amod_rate_Chinese amod_rate_andfor amod_rate_forKorean2 nn_rate_Minimum nn_rate_error pobj_for_rate nn__training nsubjpass_run_ auxpass_run_was dep_set_run prep_run_on det_system_each pobj_on_system partmod_system_afterwardsand amod__BLEU nn__score nsubjpass_calculated_ auxpass_calculated_was ccomp_afterwardsand_calculated prep_calculated_on det_sets_the nn_sets_test pobj_on_sets ' +n,4361,'Many approaches for POS tagging have been developed in the past , including rule-based tagging , HMM taggers , maximum-entropy models , cyclic dependency networks , memory-based learning , etc All of these approaches require either a large amount of annotated training data -LRB- for supervised tagging -RRB- or a lexicon listing all possible tags for each word -LRB- for unsupervised tagging -RRB- ',Cutting,'amod_approaches_Many nsubjpass_developed_approaches prep_approaches_for dep_tagging_POS pobj_for_tagging aux_developed_have auxpass_developed_been prep_developed_in det_past_the pobj_in_past prep_past_including amod__rule-based nn__tagging pobj_including_ nn__HMM nn__taggers appos__ amod__maximum-entropy nn__models appos__ amod__cyclic amod__dependency nn__networks appos__ amod__memory-based nn__learning appos__ appos__etc nsubj_require_All prep_All_of det_approaches_these pobj_of_approaches dep_developed_require advmod_amount_either det_amount_a amod_amount_large dobj_require_amount prep_amount_of amod_data_annotated nn_data_training pobj_of_data dep_data_for amod_tagging_supervised pobj_for_tagging cc_data_or det_lexicon_a conj_data_lexicon partmod_lexicon_listing det_tags_all amod_tags_possible dobj_listing_tags prep_tags_for det_word_each pobj_for_word dep_lexicon_for amod_tagging_unsupervised pobj_for_tagging ' +o,4362,'The concept of baseNP has undergone a number of revisions but has previously always been tied to extraction from a more completely annotated treebank , whose annotations are subject to other pressures than just initial material up to the head To our knowledge , our gures for inter-annotator agreement on the baseNP task itself 169 -LRB- ie not derived from a larger annotation task -RRB- are the rst to be reported ',Ramshaw,'det_concept_The nsubjpass_undergone_concept prep_concept_of pobj_of_baseNP aux_undergone_has det_number_a dobj_undergone_number prep_number_of nn__revisions pobj_of_ cc_undergone_but aux_tied_has advmod_always_previously advmod_tied_always auxpass_tied_been conj_undergone_tied prep_tied_to pobj_to_extraction prep_tied_from det_treebank_a dep_annotated_more advmod_annotated_completely amod_treebank_annotated pobj_from_treebank poss_subject_whose nsubj_subject_annotations cop_subject_are dep_treebank_subject prep_subject_to amod_pressures_other pobj_to_pressures mark_rst_than advmod_material_just amod_material_initial nsubj_rst_material rcmod_material_up dep_up_to det_head_the pobj_to_head prep_head_To poss_knowledge_our pobj_To_knowledge poss_gures_our appos_knowledge_gures prep_gures_for amod_agreement_inter-annotator pobj_for_agreement prep_agreement_on det_task_the nn_task_baseNP pobj_on_task measure_169_itself dep_up_169 dep_material_ie neg_derived_not partmod_ie_derived prep_derived_from det_task_a amod_task_larger nn_task_annotation pobj_from_task cop_rst_are det_rst_the advcl_subject_rst aux_reported_to auxpass_reported_be infmod_rst_reported ' +p,4363,'The Penn TreeBank -LRB- PTB -RRB- is an example of such a resource with worldwide impact on natural language processing ',Marcus,'det_TreeBank_The nn_TreeBank_Penn nsubj_example_TreeBank abbrev_TreeBank_PTB cop_example_is det_example_an prep_example_of predet_resource_such det_resource_a pobj_of_resource prep_resource_with amod_impact_worldwide pobj_with_impact prep_impact_on amod_processing_natural nn_processing_language pobj_on_processing ' +o,4364,'The noun phrase extraction module uses Brill \'s POS tagger -LRB- Brill -RRB- and a base NP chunker -LRB- -RRB- ',Ramshaw,'det_module_The amod_module_noun nn_module_phrase nn_module_extraction nsubj_uses_module poss_tagger_Brill possessive_Brill_\'s dep_tagger_POS dobj_uses_tagger nn__Brill appos_tagger_ cc_tagger_and det_chunker_a amod_chunker_base nn_chunker_NP conj_tagger_chunker appos_chunker_ ' +o,4365,' argues that many NLP tasks can be formulated in terms of analogical reasoning , and he applies his PairClass algorithm to a number of problems including SAT verbal analogy tests , synonym\\/antonym classification and distinction between semantically similar and semantically associated words ',Turney,'advmod_argues_ complm_formulated_that amod_tasks_many nn_tasks_NLP nsubjpass_formulated_tasks aux_formulated_can auxpass_formulated_be ccomp_argues_formulated prep_formulated_in pobj_in_terms prep_terms_of amod_reasoning_analogical pobj_of_reasoning cc_formulated_and nsubj_applies_he conj_formulated_applies poss_algorithm_his nn_algorithm_PairClass dobj_applies_algorithm prep_applies_to det_number_a pobj_to_number prep_number_of pobj_of_problems prep_problems_including pobj_including_SAT amod_tests_verbal nn_tests_analogy nsubj_argues_tests nn_classification_synonym\\/antonym appos_tests_classification cc_classification_and conj_classification_distinction prep_classification_between advmod_similar_semantically amod_words_similar cc_similar_and conj_similar_semantically amod_words_associated pobj_between_words ' +o,4366,'Five chunk tag sets , IOB1 , IOB2 , IOE1 , IOE2 and SE , are commonly used ',Ramshaw,'num_sets_Five nn_sets_chunk nn_sets_tag nsubjpass_used_sets appos_sets_IOB1 conj_IOB1_IOB2 conj_IOB1_IOE1 nn__IOE2 conj_IOB1_ cc_IOB1_and nn__SE conj_IOB1_ auxpass_used_are advmod_used_commonly ' +o,4367,'They have made semantic formalisms like those now usually associated with Davison attractive in artificial intelligence for many years ',Hobbs,'nsubj_made_They aux_made_have amod_formalisms_semantic dobj_made_formalisms mark_associated_like nsubj_associated_those advmod_associated_now advmod_associated_usually advcl_made_associated prep_associated_with nn__Davison pobj_with_ amod__attractive prep_attractive_in amod_intelligence_artificial pobj_in_intelligence prep_intelligence_for amod_years_many pobj_for_years ' +o,4368,'In fact , a limitation of the experiments described in this paper is that the loglinear weights for the glass-box techniques were optimized for BLEU using Ochs algorithm , while the linear weights for 55 black-box techniques were set heuristically ',Och,'prep_is_In pobj_In_fact det_limitation_a nsubj_is_limitation prep_limitation_of det_experiments_the pobj_of_experiments partmod_experiments_described prep_described_in det_paper_this pobj_in_paper complm_optimized_that det_weights_the nn_weights_loglinear nsubjpass_optimized_weights prep_weights_for det_techniques_the amod_techniques_glass-box pobj_for_techniques auxpass_optimized_were ccomp_is_optimized prep_optimized_for pobj_for_BLEU xcomp_optimized_using nn__Ochs nn__algorithm dobj_using_ mark_set_while det_weights_the amod_weights_linear nsubjpass_set_weights prep_weights_for num_techniques_55 amod_techniques_black-box pobj_for_techniques auxpass_set_were advcl_optimized_set advmod_set_heuristically ' +o,4369,'We computed the LCS and WLCS-based F-measure following using both the query pool and the sentence pool as in the previous section ',Lin,'nsubj_computed_We det_F-measure_the amod_F-measure_LCS cc_LCS_and conj_LCS_WLCS-based dobj_computed_F-measure prep_F-measure_following pobj_following_ partmod__using preconj_pool_both det_pool_the amod_pool_query dep_pool_pool cc_pool_and det_sentence_the conj_pool_sentence dobj_using_pool dep_in_as dep_pool_in det_section_the amod_section_previous pobj_in_section ' +o,4370,' and ',Turney,'amod__ cc__and conj__ ' +o,4371,'Systems which are able to acquire a small number of verbal subcategorisation classes automatically from corpus text have been described by Brent , and Ushioda et al ',Marcus,'nsubjpass_described_Systems nsubj_able_which cop_able_are rcmod_Systems_able aux_acquire_to xcomp_able_acquire det_number_a amod_number_small dobj_acquire_number prep_number_of amod_classes_verbal nn_classes_subcategorisation pobj_of_classes dep_from_automatically prep_acquire_from nn_text_corpus pobj_from_text aux_described_have auxpass_described_been prep_described_by nn__Brent pobj_by_ cc__and conj__Ushioda cc_Ushioda_et conj_Ushioda_al ' +o,4372,'Iterative cost reduction algorithm Input: An SCFG Output: An equivalent binary SCFG of 1: Function ITERATIVECOSTREDUCTION( ) 2: 0 3: for each 0do 4: ( ) = , 0 5: while ( ) does not converge do 6: for each do 7: [ ] ( ) 8: for each ( ) do 9: for each , do 10: 1 11: ( ) CKYBINARIZATION( , ) 12: [ ] ( ) 13: for each ( ) do 14: for each , do 15: + 1 16: return In the iterative cost reduction algorithm, we first obtain an initial binary SCFG 0 using the synchronous binarization method proposed in (Zhang et al., 2006).',Zhang,'' +o,4373,'The most relevant to our work are , Toral and Muoz -LRB- 2006 -RRB- , and Cucerzan -LRB- 2007 -RRB- ',Kazama,'dep_relevant_The advmod_relevant_most prep_relevant_to poss_work_our pobj_to_work cop__are dep_relevant_ conj__Toral cc__and conj__Muoz appos__2006 cc__and conj__Cucerzan appos_Cucerzan_2007 ' +o,4374,'A different approach in evaluating nonparametric Bayesian models for NLP is statesplitting ',Liang,'det_approach_A amod_approach_different nsubj_statesplitting_approach prep_approach_in pcomp_in_evaluating amod_models_nonparametric nn_models_Bayesian dobj_evaluating_models prep_evaluating_for pobj_for_NLP aux_statesplitting_is ' +o,4375,' Hindi is a verb final , flexible word order language and therefore , has frequent occurrences of non-projectivity in its dependency structures ',McDonald,'nn_Hindi_ nsubj_language_Hindi cop_language_is det_language_a amod_language_verb amod_language_final amod_language_flexible nn_language_word nn_language_order cc_language_and conj_language_therefore dep_therefore_has amod_occurrences_frequent dobj_has_occurrences prep_occurrences_of pobj_of_non-projectivity prep_non-projectivity_in poss_structures_its amod_structures_dependency pobj_in_structures ' +n,4376,'Our focus is on the sentence level , unlike and ; we employ a significantly larger set of seed words , and we explore as indicators of orientation words from syntactic classes other than adjectives -LRB- nouns , verbs , and adverbs -RRB- ',Turney,'poss_focus_Our nsubj_is_focus prep_is_on det_level_the nn_level_sentence pobj_on_level prep_is_unlike pobj_unlike_ cc__and conj__ nsubj_employ_we parataxis_is_employ det_set_a advmod_larger_significantly amod_set_larger dobj_employ_set prep_set_of nn_words_seed pobj_of_words cc_is_and nsubj_explore_we conj_is_explore prep_explore_as pobj_as_indicators prep_indicators_of nn_words_orientation pobj_of_words prep_explore_from amod_classes_syntactic pobj_from_classes amod_classes_other dep_classes_than dep_classes_adjectives dep_adjectives_nouns conj_nouns_verbs cc_nouns_and conj_nouns_adverbs ' +o,4377,'The goal of each selection stage is to select the feature f that maximizes the gain of the log likelihood, where the a and gain of f are derived through following steps: Let the log likelihood of the model be -= yx xZysump pL,, )(/|log()( ~ and the empirical expectation of feature f be E p (f)= p (x,y)f(x,y) x,y With the approximation assumption in Berger et al (1996)s paper, the un-normalized component and the normalization factor of the model have the following recursive forms: )|()|( aa exysumxysum SfS = | Z f + The approximate gain of the log likelihood is computed by G Sf (a)L(p Sf a )-L(p S ) =- p (x)(logZ Sf,a (x) x /Z S (x)) +aE p (f) (1) The maximum approximate gain and its corresponding a are represented as: )(max),(~ a fS GfSL =D maxarg f 3 A Fast Feature Selection Algorithm The inefficiency of the IFS algorithm is due to the following reasons.',Berger,'' +p,4378,'Exponential family models are a mainstay of modern statistical modeling and they are widely and successfully used for example in text classification ',Berger,'amod_models_Exponential nn_models_family nsubj_mainstay_models cop_mainstay_are det_mainstay_a prep_mainstay_of amod__modern amod__statistical nn__modeling pobj_of_ cc_mainstay_and nsubjpass_used_they auxpass_used_are advmod_used_widely cc_widely_and conj_widely_successfully conj_mainstay_used prep_used_for pobj_for_example prep_example_in nn_classification_text pobj_in_classification ' +o,4379,' described the use of a biased PageRank over the WordNet graph to compute word pair semantic relatedness using the divergence of the probability values over the graph created by each word ',Hughes,'nsubj_described_ det_use_the dobj_described_use prep_use_of det_PageRank_a nn_PageRank_biased pobj_of_PageRank prep_described_over det_graph_the nn_graph_WordNet pobj_over_graph aux_compute_to xcomp_described_compute nn_relatedness_word nn_relatedness_pair amod_relatedness_semantic dobj_compute_relatedness partmod_relatedness_using det_divergence_the dobj_using_divergence prep_divergence_of det_values_the nn_values_probability pobj_of_values prep_using_over det_graph_the pobj_over_graph partmod_graph_created prep_created_by det_word_each pobj_by_word ' +o,4380,' used a binary bracketing ITG to segment a sen19 tence while simultaneously word-aligning it to its translation , but the model was trained heuristically with a fixed segmentation ',Wu,'nsubj_used_ det_ITG_a advmod_bracketing_binary amod_ITG_bracketing dobj_used_ITG prep_used_to pobj_to_segment det_sen19_a nsubj_tence_sen19 dep_used_tence mark_word-aligning_while advmod_word-aligning_simultaneously advcl_tence_word-aligning dobj_word-aligning_it prep_word-aligning_to poss_translation_its pobj_to_translation cc_tence_but det_model_the nsubjpass_trained_model auxpass_trained_was conj_tence_trained advmod_trained_heuristically prep_trained_with det_segmentation_a amod_segmentation_fixed pobj_with_segmentation ' +o,4381,'Then , it models the correlations between the pivot features and all other features by training linear pivot predictors to predict occurrences of each pivot in the unlabeled data from both domains ',Blitzer,'advmod_models_Then nsubj_models_it det_correlations_the dobj_models_correlations prep_correlations_between det_features_the amod_features_pivot pobj_between_features cc_features_and det_features_all amod_features_other conj_features_features prep_models_by pcomp_by_training amod_predictors_linear nn_predictors_pivot dobj_training_predictors aux_predict_to xcomp_training_predict dobj_predict_occurrences prep_occurrences_of det_pivot_each pobj_of_pivot prep_pivot_in det_data_the amod_data_unlabeled pobj_in_data prep_predict_from det_domains_both pobj_from_domains ' +p,4382,'1 Introduction Recently linguistically-motivated syntax-based translation method has achieved great success in statistical machine translation -LRB- SMT -RRB- ',Galley,'nsubj_linguistically-motivated_Introduction advmod_linguistically-motivated_Recently dep_1_linguistically-motivated amod_method_syntax-based nn_method_translation nsubj_achieved_method aux_achieved_has ccomp_linguistically-motivated_achieved amod_success_great dobj_achieved_success prep_achieved_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +p,4383,'The abduction-based approach has provided a simple and elegant way to realize such a task ',Hobbs,'det__The amod__abduction-based nn__approach nsubj_provided_ aux_provided_has det_way_a amod_way_simple cc_simple_and conj_simple_elegant dobj_provided_way aux_realize_to xcomp_provided_realize predet_task_such det_task_a dobj_realize_task ' +o,4384,' proposed a joint model of text and aspect ratings which utilizes a modified LDA topic model to build topics that are representative of ratable aspects , and builds a set of sentiment predictors ',Titov,'nsubj_proposed_ det_model_a amod_model_joint dobj_proposed_model prep_model_of nn_ratings_text cc_text_and conj_text_aspect pobj_of_ratings nsubj_utilizes_which rcmod_model_utilizes det_model_a amod_model_modified nn_model_LDA nn_model_topic dobj_utilizes_model aux_build_to xcomp_utilizes_build dobj_build_topics nsubj_representative_that cop_representative_are rcmod_topics_representative prep_representative_of amod_aspects_ratable pobj_of_aspects cc_representative_and conj_representative_builds det_set_a dobj_builds_set prep_set_of nn_predictors_sentiment pobj_of_predictors ' +o,4385,'Furthermore , techniques such as iterative minimum errorrate training as well as web-based MT services require the decoder to translate a large number of source-language sentences per unit time ',Och,'advmod_require_Furthermore nsubj_require_techniques dep_as_such prep_techniques_as amod__iterative amod__minimum amod__errorrate nn__training pobj_as_ dep_well_as cc__well dep_well_as amod_services_web-based nn_services_MT conj__services det_decoder_the nsubj_translate_decoder aux_translate_to xcomp_require_translate det_number_a amod_number_large dobj_translate_number prep_number_of amod_sentences_source-language pobj_of_sentences prep_sentences_per nn_time_unit pobj_per_time ' +o,4386,'Its previous applications demonstrated that cooccurrence statistics on a target word is often sufficient for its automatical classification into one of numerous classes such as synsets of WordNet ',Grefenstette,'poss__Its amod__previous nn__applications nsubj_demonstrated_ complm_sufficient_that amod_statistics_cooccurrence nsubj_sufficient_statistics prep_statistics_on det_word_a nn_word_target pobj_on_word cop_sufficient_is advmod_sufficient_often ccomp_demonstrated_sufficient prep_sufficient_for poss_classification_its amod_classification_automatical pobj_for_classification prep_sufficient_into pobj_into_one prep_one_of amod_classes_numerous pobj_of_classes dep_as_such prep_classes_as pobj_as_synsets prep_synsets_of pobj_of_WordNet ' +o,4387,'No documentation of tile construction algorithm of the su -LRB- lix lexicon in was available ',Cutting,'det_documentation_No nsubj_available_documentation prep_documentation_of nn_algorithm_tile nn_algorithm_construction pobj_of_algorithm prep_algorithm_of det_su_the pobj_of_su dep_su_lix advmod_su_lexicon dep_lexicon_in pobj_in_ cop_available_was ' +o,4388,'It is also related to loglinear models for machine translation ',Och,'nsubjpass_related_It auxpass_related_is advmod_related_also prep_related_to nn_models_loglinear pobj_to_models prep_models_for nn_translation_machine pobj_for_translation ' +o,4389,'3 Baseline MT System The phrase-based SMT system used in our experiments is Moses , phrase translation pro ing probabilities , and languag ties are combined in the log-linear model to obtain the best translation best e of the source sentence f : = = M p -RRB- -LRB- maxarg fee ebest -LRB- 2 -RRB- m mm h 1 , -LRB- maxarg f -RRB- e e The weights are set by a discriminative training method using a held-out data set as describ in ',Och,'dep_combined_3 nn_System_Baseline nn_System_MT nsubjpass_combined_System det_system_The amod_system_phrase-based nn_system_SMT nsubj_Moses_system partmod_system_used prep_used_in poss_experiments_our pobj_in_experiments cop_Moses_is rcmod_System_Moses nn_translation_phrase conj_Moses_translation amod_probabilities_pro nn_probabilities_ing dep_translation_probabilities cc_Moses_and amod_ties_languag conj_Moses_ties auxpass_combined_are prep_combined_in det_model_the amod_model_log-linear pobj_in_model aux_obtain_to xcomp_combined_obtain det_translation_the amod_translation_best dobj_obtain_translation dep_e_best amod_translation_e prep_translation_of det_sentence_the nn_sentence_source pobj_of_sentence dep_translation_f dep_obtain_= nn_p_= nn_p_M dobj_=_p amod_ebest_maxarg nn_ebest_fee dep_p_ebest appos_ebest_2 nn_h_m nn_h_mm dep_ebest_h num_h_1 dep_p_maxarg dep_maxarg_f dep_set_e dep_set_e det_weights_The nsubjpass_set_weights auxpass_set_are dep_combined_set prep_set_by det_method_a amod_method_discriminative nn_method_training pobj_by_method partmod_method_using det_data_a amod_data_held-out dobj_using_data partmod_data_set prep_set_as pobj_as_describ prep_set_in ' +o,4390,'Introduction Verb subcategorizafion probabilities play an important role in both computational linguistic applications and psycholinguisfic models of language processing ',Marcus,'nn_probabilities_Introduction nn_probabilities_Verb nn_probabilities_subcategorizafion nsubj_play_probabilities det_role_an amod_role_important dobj_play_role prep_role_in det__both amod__computational amod__linguistic nn__applications pobj_in_ cc_role_and amod_models_psycholinguisfic conj_role_models prep_models_of nn__language nn__processing pobj_of_ ' +p,4391,'Recent comparisons of approaches that can be trained on corpora have shown that in most cases statistical aproaches yield better results than finite-state , rule-based , or memory-based taggers ',Cutting,'amod_comparisons_Recent nsubj_shown_comparisons prep_comparisons_of pobj_of_approaches nsubjpass_trained_that aux_trained_can auxpass_trained_be rcmod_approaches_trained prep_trained_on nn__corpora pobj_on_ aux_shown_have complm__that prep__in amod_cases_most pobj_in_cases amod_aproaches_statistical nsubj__aproaches ccomp_shown_ xcomp__yield amod_results_better dobj_yield_results prep_yield_than amod__finite-state conj_finite-state_rule-based cc_finite-state_or conj_finite-state_memory-based dep_finite-state_taggers pobj_than_ ' +o,4392,'For example , in John saw Mary yesterday at the station , only John and Mary are required arguments while the other constituents are optional -LRB- adjuncts -RRB- 3 The problem of SF identification using statistical methods has had a rich discussion in the literature -LRB- also see the refences cited in -RRB- ',Dunning,'prep_arguments_For pobj_For_example mark_saw_in nsubj_saw_John dep_arguments_saw nn_yesterday_Mary tmod_saw_yesterday prep_saw_at det_station_the pobj_at_station advmod_saw_only nsubjpass_arguments_John cc_John_and conj_John_Mary auxpass_arguments_are dep_arguments_required mark_optional_while det_constituents_the amod_constituents_other nsubj_optional_constituents aux_optional_are advcl_arguments_optional dobj_optional_adjuncts dep_had_3 det_problem_The nsubj_had_problem prep_problem_of nn_identification_SF pobj_of_identification partmod_identification_using amod_methods_statistical dobj_using_methods aux_had_has dep_arguments_had det_discussion_a amod_discussion_rich dobj_had_discussion prep_discussion_in det__the nn__literature pobj_in_ advmod_see_also dep__see det_refences_the nsubj_cited_refences ccomp_see_cited prep_cited_in pobj_in_ ' +o,4393,'Given sentence-aligned bi-lingual training data , we first use GIZA + + to generate word level alignment ',Och,'prep_use_Given amod_data_sentence-aligned amod_data_bi-lingual nn_data_training dep_Given_data nsubj_use_we advmod_use_first nsubj_generate_GIZA dep_generate_+ cc_+_+ conj_+_ aux_generate_to xcomp_use_generate nn_alignment_word nn_alignment_level dobj_generate_alignment ' +o,4394,'This strategy is commonly used in multi-document summarization , where the combination step eliminates the redundancy across selected excerpts ',Goldstein,'det_strategy_This nsubjpass_used_strategy auxpass_used_is advmod_used_commonly prep_used_in amod__multi-document nn__summarization pobj_in_ advmod_eliminates_where det_step_the nn_step_combination nsubj_eliminates_step rcmod__eliminates det_redundancy_the dobj_eliminates_redundancy prep_redundancy_across amod_excerpts_selected pobj_across_excerpts ' +p,4395,'For a full derivation of the modified updates and for quite technical convergence proofs , see ',Collins,'det_derivation_a amod_derivation_full pobj_For_derivation prep_derivation_of det_updates_the amod_updates_modified pobj_of_updates cc_For_and mark_see_for advmod_technical_quite amod_proofs_technical nn_proofs_convergence nsubj_see_proofs conj_For_see ' +o,4396,'Previous uses of this model include language modeling , machine translation , prepositional phrase attachment , and word morphology ',Marcus,'amod_uses_Previous nsubj_include_uses prep_uses_of det_model_this pobj_of_model nn__language nn__modeling dobj_include_ nn__machine nn__translation conj__ amod__prepositional nn__phrase nn__attachment conj__ cc__and nn__word nn__morphology conj__ ' +o,4397,'In his Xtract system , first extracted significant pairs of words that consistently co-occur within a single syntactic structure using statistical scores called distance , strength and spread , and then examined concordances of the bi-grams to find longer frequent multiword units ',Smadja,'prep_extracted_In poss_system_his nn_system_Xtract pobj_In_system nsubj_extracted_ advmod_extracted_first amod_pairs_significant dobj_extracted_pairs prep_pairs_of pobj_of_words complm_called_that advmod_co-occur_consistently csubj_called_co-occur prep_co-occur_within det_structure_a amod_structure_single amod_structure_syntactic pobj_within_structure partmod_structure_using amod_scores_statistical dobj_using_scores ccomp_extracted_called dobj_called_distance conj_distance_strength cc_distance_and conj_distance_spread cc_called_and advmod_called_then conj_called_examined dobj_examined_concordances prep_concordances_of det_bi-grams_the pobj_of_bi-grams aux_find_to xcomp_examined_find amod_units_longer amod_units_frequent nn_units_multiword dobj_find_units ' +o,4398,'22 Brown clustering algorithm In order to provide word clusters for our experiments , we used the Brown clustering algorithm ',Brown,'num_algorithm_22 nn_algorithm_Brown nn_algorithm_clustering mark_provide_In dep_provide_order aux_provide_to dep_used_provide nn_clusters_word dobj_provide_clusters prep_provide_for poss_experiments_our pobj_for_experiments nsubj_used_we dep_algorithm_used det_algorithm_the nn_algorithm_Brown nn_algorithm_clustering dobj_used_algorithm ' +o,4399,'a Hindle and Rooth and used partial parses generated by Fidditch to study word ~ urrtnc patterns m syntactic contexts ',Church,'det__a nn__Hindle cc_Hindle_and conj_Hindle_Rooth nsubj_used_ cc__and conj__ amod_parses_partial nsubj_generated_parses ccomp_used_generated prep_generated_by pobj_by_Fidditch aux_study_to xcomp_generated_study nn_~_word dobj_study_~ amod_patterns_urrtnc nsubj_m_patterns dep_used_m amod_contexts_syntactic dobj_m_contexts ' +p,4400,'State-of-the-art measures such as BLEU or NIST aim at measuring the translation quality rather on the document level1 than on the level of single sentences ',Papineni,'amod_measures_State-of-the-art nsubj_aim_measures dep_as_such prep_measures_as nn__BLEU pobj_as_ cc__or nn__NIST conj__ prep_aim_at pcomp_at_measuring det_quality_the nn_quality_translation dobj_measuring_quality dep_on_rather prep_measuring_on det_level1_the nn_level1_document pobj_on_level1 dep_on_than prep_measuring_on det_level_the pobj_on_level prep_level_of amod_sentences_single pobj_of_sentences ' +o,4401,' observed , however , that the piecewiseconstant property could be exploited to characterize the function exhaustively along any line in parameter space , and hence to minimize it globally along that line ',Och,'nsubj_observed_ advmod_observed_however complm_exploited_that det_property_the nn_property_piecewiseconstant nsubjpass_exploited_property aux_exploited_could auxpass_exploited_be ccomp_observed_exploited aux_characterize_to ccomp_exploited_characterize det_exhaustively_the nn_exhaustively_function dobj_characterize_exhaustively prep_characterize_along det_line_any pobj_along_line prep_line_in nn_space_parameter pobj_in_space cc_characterize_and advmod_characterize_hence aux_minimize_to conj_characterize_minimize dobj_minimize_it advmod_minimize_globally prep_minimize_along det_line_that pobj_along_line ' +o,4402,'To overcome the knowledge acquisition bottleneck problem suffered by supervised methods , these methods make use of a small annotated corpus as seed data in a bootstrapping process ',Yarowsky,'aux_overcome_To dep_make_overcome det_problem_the nn_problem_knowledge nn_problem_acquisition nn_problem_bottleneck dobj_overcome_problem partmod_problem_suffered prep_suffered_by amod_methods_supervised pobj_by_methods det_methods_these nsubj_make_methods dobj_make_use prep_use_of det_corpus_a amod_corpus_small amod_corpus_annotated pobj_of_corpus prep_make_as nn_data_seed pobj_as_data prep_data_in det__a amod__bootstrapping nn__process pobj_in_ ' +o,4403,'The tagging scheme is a variant of the IOB scheme originally put forward by ',Ramshaw,'det_scheme_The amod_scheme_tagging nsubj_variant_scheme cop_variant_is det_variant_a prep_variant_of det_scheme_the nn_scheme_IOB pobj_of_scheme advmod_put_originally partmod_scheme_put advmod_put_forward prep_put_by ' +o,4404,'The central question in learning is how to set the parameters a , given the training examples b x 1 , y 1 , x 2 , y 2 , : : : , x n , y n Logistic regression and boosting involve different algorithms and criteria for training the parameters a , but recent work has shown that the methods have strong similarities ',Collins,'det_question_The amod_question_central nsubj_is_question prep_question_in pobj_in_learning dep_set_how aux_set_to dep_is_set det_a_the amod_a_parameters dobj_set_a dep_a_given det_examples_the nn_examples_training dep_given_examples dep_x_b dep_1_x dep_examples_1 appos_1_y num_y_1 appos_1_x dep_x_2 appos_1_y num_y_2 amod_n_x nsubj_involve_n nn_regression_y nn_regression_n nn_regression_Logistic conj_n_regression cc_n_and conj_n_boosting parataxis_is_involve amod_algorithms_different dobj_involve_algorithms cc_algorithms_and conj_algorithms_criteria prep_involve_for pcomp_for_training det_parameters_the dobj_training_parameters dep_training_a cc_involve_but amod__recent nn__work nsubj_shown_ aux_shown_has conj_involve_shown complm_have_that det_methods_the nsubj_have_methods ccomp_shown_have amod_similarities_strong dobj_have_similarities ' +o,4405,' -RRB- , the tagger for grammatical functions works with lexical and contextual probability measures Pq -LRB- -RRB- ',Cutting,'dep_works_ det_tagger_the nsubj_works_tagger prep_tagger_for amod_functions_grammatical pobj_for_functions prep_works_with amod_measures_lexical cc_lexical_and conj_lexical_contextual nn_measures_probability pobj_with_measures tmod_works_Pq ' +o,4406,'4 Semantic Class Induction from Wikipedia Wikipedia has recently been used as a knowledge source for various language processing tasks , including taxonomy construction , coreference resolution , and English NER -LRB- eg , Bunescu and Pasca , Cucerzan , Kazama and Torisawa , et al ',Watanabe,'num_Induction_4 nn_Induction_Semantic nn_Induction_Class nsubjpass_used_Induction prep_Induction_from nn_Wikipedia_Wikipedia pobj_from_Wikipedia aux_used_has advmod_used_recently auxpass_used_been prep_used_as det_source_a nn_source_knowledge pobj_as_source prep_source_for amod_tasks_various nn_tasks_language nn_tasks_processing pobj_for_tasks prep_source_including amod__taxonomy nn__construction pobj_including_ amod__coreference nn__resolution conj__ cc__and nn_NER_English conj__NER dep_NER_eg nn__Bunescu cc_Bunescu_and conj_Bunescu_Pasca conj_eg_ nn__Cucerzan conj_eg_ conj_eg_Kazama cc_eg_and nn__Torisawa conj_eg_ conj__ cc__et conj__al ' +o,4407,'The translation models and lexical scores were estimated on the training corpus whichwasautomaticallyalignedusingGiza + + in both directions between source and target and symmetrised using the growing heuristic ',Koehn,'det_models_The nn_models_translation nsubjpass_estimated_models cc_models_and nn_scores_lexical conj_models_scores auxpass_estimated_were prep_estimated_on det_whichwasautomaticallyalignedusingGiza_the nn_whichwasautomaticallyalignedusingGiza_training nn_whichwasautomaticallyalignedusingGiza_corpus pobj_on_whichwasautomaticallyalignedusingGiza amod_whichwasautomaticallyalignedusingGiza_+ cc_+_+ conj_+_ prep_estimated_in det_directions_both pobj_in_directions prep_directions_between pobj_between_source cc_source_and conj_source_target cc_estimated_and conj_estimated_symmetrised xcomp_symmetrised_using det_heuristic_the amod_heuristic_growing dobj_using_heuristic ' +o,4408,'Inter-annotator agreement is typically measured by the kappa statistic , dekappa frequency 00 02 04 06 08 10 0 2 4 6 8 Figure 2 : Distribution of -LRB- inter-annotator agreement -RRB- across the 54 ICSI meetings tagged by two annotators ',Carletta,'nn_agreement_Inter-annotator nsubjpass_measured_agreement auxpass_measured_is advmod_measured_typically prep_measured_by det__the nn__kappa nn__statistic pobj_by_ number_8_dekappa cc_8_frequency conj_8_00 number_8_02 number_8_04 number_8_06 number_8_08 number_8_10 cc_8_0 conj_8_2 number_8_4 number_8_6 appos__8 cc_8_Figure conj_8_2 dep_8_Distribution prep_Distribution_of amod_agreement_inter-annotator pobj_of_agreement prep_measured_across det_meetings_the num_meetings_54 nn_meetings_ICSI pobj_across_meetings partmod_meetings_tagged prep_tagged_by num_annotators_two pobj_by_annotators ' +o,4409,'Table 6 shows 3An exception is Golding , who uses the entire Brown corpus for training -LRB- 1M words -RRB- and 3\\/4 of the Wall Street Journal corpus for testing ',Marcus,'nsubj_shows_Table num_Table_6 amod_exception_3An nsubj__exception cop__is nn__Golding ccomp_shows_ nsubj_uses_who rcmod__uses det_corpus_the amod_corpus_entire nn_corpus_Brown dobj_uses_corpus prep_corpus_for pobj_for_training num_words_1M appos_training_words cc_corpus_and conj_corpus_3\\/4 prep_3\\/4_of det__the nn__Wall nn__Street nn__Journal nn__corpus pobj_of_ prep__for pobj_for_testing ' +p,4410,'As a result , the good results of with large seed sets do not immediately imply success with small seed sets ',McClosky,'prep_imply_As det_result_a pobj_As_result det_results_the amod_results_good nsubj_imply_results prep_results_of pobj_of_ prep__with amod_sets_large nn_sets_seed pobj_with_sets aux_imply_do neg_imply_not advmod_imply_immediately dobj_imply_success prep_imply_with amod_sets_small nn_sets_seed pobj_with_sets ' +p,4411,'Its also worth noting that saw a LFMS improvement of 08 \% over their baseline discriminative parser after adding punctuation features , one of which encoded the sentence-final punctuation ',Collins,'poss_parser_Its amod_parser_also prep_also_worth pcomp_worth_noting complm_saw_that nsubj_saw_ ccomp_noting_saw det_improvement_a nn_improvement_LFMS dobj_saw_improvement prep_improvement_of num_\%_08 pobj_of_\% prep_saw_over poss_baseline_their pobj_over_baseline amod_parser_discriminative prep_parser_after pcomp_after_adding nn_features_punctuation dobj_adding_features nsubj_encoded_one prep_one_of pobj_of_which rcmod_features_encoded det_punctuation_the amod_punctuation_sentence-final dobj_encoded_punctuation ' +o,4412,'As in , confusion networks built around all skeletons are joined into a lattice which is expanded and rescored with language models ',Rosti,'prep_joined_As dep_As_in pobj_in_ nn_networks_confusion nsubjpass_joined_networks partmod_networks_built prep_built_around det_skeletons_all pobj_around_skeletons auxpass_joined_are prep_joined_into det_lattice_a pobj_into_lattice nsubjpass_expanded_which auxpass_expanded_is rcmod_lattice_expanded cc_expanded_and conj_expanded_rescored prep_expanded_with nn_models_language pobj_with_models ' +p,4413,'It is often straightforward to obtain large amounts of unlabeled data , making semi-supervised approaches appealing ; previous work on semisupervised methods for dependency parsing includes ',Koo,'nsubj_straightforward_It cop_straightforward_is advmod_straightforward_often aux_obtain_to xcomp_straightforward_obtain amod_amounts_large dobj_obtain_amounts prep_amounts_of amod_data_unlabeled pobj_of_data xcomp_obtain_making amod_approaches_semi-supervised nsubj_appealing_approaches xcomp_making_appealing amod_work_previous nsubj_includes_work prep_work_on amod_methods_semisupervised pobj_on_methods prep_methods_for amod_parsing_dependency pobj_for_parsing parataxis_straightforward_includes ' +o,4414,'For example , in machine translation evaluation , approaches such as BLEU use n-gram overlap comparisons with a model to judge overall goodness , with higher n-grams meant to capture fluency considerations ',Papineni,'prep_use_For pobj_For_example prep_use_in nn_evaluation_machine nn_evaluation_translation pobj_in_evaluation nsubj_use_approaches dep_as_such prep_approaches_as nn__BLEU pobj_as_ amod_comparisons_n-gram amod_comparisons_overlap dobj_use_comparisons prep_use_with det_model_a pobj_with_model aux_judge_to infmod_model_judge amod_goodness_overall dobj_judge_goodness prep_use_with amod_n-grams_higher pobj_with_n-grams partmod_n-grams_meant aux_capture_to xcomp_meant_capture amod_considerations_fluency dobj_capture_considerations ' +o,4415,'2 We used the to generate the constituency parse and a dependency converter to obtain the dependency parse of English sentences ',Collins,'nsubj_used_We dep_2_used det__the dobj_used_ aux_generate_to infmod__generate det_parse_the nn_parse_constituency nsubj_obtain_parse cc_parse_and det__a amod__dependency nn__converter conj_parse_ aux_obtain_to xcomp_generate_obtain det_parse_the amod_parse_dependency dobj_obtain_parse prep_parse_of amod_sentences_English pobj_of_sentences ' +o,4416,'Verbs and possible senses in our corpus Both corpora were lemmatized and part-of-speech -LRB- POS -RRB- tagged using Minipar and Mxpost , respectivelly ',Ratnaparkhi,'nsubj_lemmatized_Verbs cc_Verbs_and amod_senses_possible conj_Verbs_senses prep_senses_in poss_corpus_our pobj_in_corpus det_corpora_Both dep_senses_corpora cop_lemmatized_were cc_lemmatized_and conj_lemmatized_part-of-speech appos_part-of-speech_POS partmod_part-of-speech_tagged xcomp_tagged_using nn__Minipar dobj_using_ cc__and nn__Mxpost conj__ advmod_lemmatized_respectivelly ' +o,4417,'The method uses a translation model based on IBM Model 1 , in which translation candidates of a phrase are generated by combining translations and transliterations of the phrase components , and matching the result against a large corpus ',Brown,'det_method_The nsubj_uses_method det_model_a nn_model_translation dobj_uses_model prep_uses_based dep_based_on nn__IBM nn__Model num__1 pobj_on_ rel_generated_in pobj_in_which nn_candidates_translation nsubjpass_generated_candidates prep_candidates_of det_phrase_a pobj_of_phrase auxpass_generated_are rcmod__generated prep_generated_by pcomp_by_combining dobj_combining_translations cc_translations_and conj_translations_transliterations prep_translations_of det_components_the nn_components_phrase pobj_of_components cc_combining_and conj_combining_matching det_result_the dobj_matching_result prep_result_against det_corpus_a amod_corpus_large pobj_against_corpus ' +p,4418,'In an experiment on 16,800 sentences of Chinese-English newswire text with segment-level human evaluation from the Linguistic Data Consortium?s (LDC) Multiple Translation project, we compare the LFG-based evaluation method with other popular metrics like BLEU, NIST, General Text Matcher (GTM) (Turian et al. , 2003), Translation Error Rate (TER) (Snover et al. , 2006)1, and METEOR (Banerjee and Lavie, 2005), and we show that combining dependency representations with synonyms leads to a more accurate evaluation that correlates better with human judgment.',Banerjee,'' +o,4419,'Recently used machine learning methods including maximum entropy models and support vector machines provide grounds for this type of modeling , because it allows various dependent features to be incorporated into the model without the independence assumption ',Berger,'advmod_used_Recently dobj_used_machine xcomp_used_learning dobj_learning_methods prep_methods_including amod_models_maximum nn_models_entropy pobj_including_models nn_machines_ cc__and conj__support nn_machines_vector nsubj__machines rcmod_models_ xcomp__provide dobj_provide_grounds prep_provide_for det_type_this pobj_for_type prep_type_of pobj_of_modeling mark_allows_because nsubj_allows_it dep_used_allows amod_features_various amod_features_dependent dobj_allows_features aux_incorporated_to auxpass_incorporated_be xcomp_allows_incorporated prep_incorporated_into det_model_the pobj_into_model prep_incorporated_without det_assumption_the nn_assumption_independence pobj_without_assumption ' +o,4420,'Gibbs sampling is not new to the natural language processing community ',Teh,'nsubj_new_Gibbs partmod_Gibbs_sampling cop_new_is neg_new_not prep_new_to det_community_the amod_community_natural nn_community_language nn_community_processing pobj_to_community ' +o,4421,'In the field of parsing , compared parsing errors between graphbased and transition-based parsers ',McDonald,'prep__In det_field_the pobj_In_field prep_field_of pobj_of_parsing prep__compared amod_errors_parsing dep_compared_errors prep_errors_between amod_parsers_graphbased cc_graphbased_and conj_graphbased_transition-based pobj_between_parsers ' +o,4422,'CLL has then been applied to a corpus of declarative sentences from the Penn Treebank on which it has been shown to perform comparatively well with respect to much less psychologically plausible systems , which are significantly more supervised and are applied to somewhat simpler problems ',Marcus,'nsubjpass_applied_CLL aux_applied_has advmod_applied_then auxpass_applied_been prep_applied_to det_corpus_a pobj_to_corpus prep_corpus_of amod_sentences_declarative pobj_of_sentences prep_applied_from det__the nn__Penn nn__Treebank pobj_from_ rel_shown_on pobj_on_which nsubjpass_shown_it aux_shown_has auxpass_shown_been rcmod__shown aux_perform_to xcomp_shown_perform acomp_perform_comparatively advmod_comparatively_well prep_comparatively_with pobj_with_respect prep_perform_to amod_systems_much dep_plausible_less advmod_plausible_psychologically amod_systems_plausible pobj_to_systems nsubj_supervised_which cop_supervised_are advmod_supervised_significantly advmod_supervised_more rcmod_systems_supervised cc_applied_and auxpass_applied_are conj_applied_applied prep_applied_to advmod_simpler_somewhat amod_problems_simpler pobj_to_problems ' +p,4423,'Typical examples of linguistically sophisticated annotation include tagging words with their syntactic category -LRB- although this has not been found to be effective for 1R -RRB- , lemma of the word -LRB- eg ` corpus \' for ` corpora \' -RRB- , phrasal information -LRB- eg identifying noun groups and phrases -RRB- , and subject-predicate identification ',Hindle,'amod_examples_Typical nsubj_include_examples prep_examples_of advmod_sophisticated_linguistically amod_annotation_sophisticated pobj_of_annotation xcomp_include_tagging dobj_tagging_words prep_words_with poss_category_their amod_category_syntactic pobj_with_category mark_found_although nsubjpass_found_this aux_found_has neg_found_not auxpass_found_been dep_words_found aux_effective_to cop_effective_be xcomp_found_effective prep_effective_for pobj_for_1R conj_words_lemma prep_lemma_of det_word_the pobj_of_word nn_corpus_eg dep_word_corpus prep_corpus_for pobj_for_corpora amod_information_phrasal appos_word_information dep_words_eg xcomp_eg_identifying nn_groups_noun nsubj__groups cc_groups_and conj_groups_phrases xcomp_identifying_ cc_words_and amod_identification_subject-predicate conj_words_identification ' +o,4424,'41 Data We used Penn-Treebank data , presented in Table 1 ',Marcus,'num_Data_41 nsubj_used_We dep_Data_used nn_data_Penn-Treebank nn_data_ dobj_used_data partmod_data_presented prep_presented_in pobj_in_Table num_Table_1 ' +o,4425,'1 Introduction Current state-of-the-art statistical parsers are trained on large annotated corpora such as the Penn Treebank ',Marcus,'num_parsers_1 nn_parsers_Introduction nn_parsers_Current amod_parsers_state-of-the-art amod_parsers_statistical nsubjpass_trained_ auxpass_trained_are dep_parsers_trained prep_trained_on amod_corpora_large amod_corpora_annotated pobj_on_corpora dep_as_such prep_corpora_as det_Treebank_the nn_Treebank_Penn pobj_as_Treebank ' +o,4426,'There are two tasks for the domain adaptation problem ','Daume III','expl_are_There num_tasks_two measure__tasks acomp_are_ prep_are_for det_problem_the nn_problem_domain nn_problem_adaptation pobj_for_problem ' +p,4427,'Coming from the other direction , such observations about phrase reordering between different languages are precisely thekindsoffactsthatparsingapproachestomachine translation are designed to handle and do successfully handle ',Wu,'dep_thekindsoffactsthatparsingapproachestomachine_Coming prep_Coming_from det_direction_the amod_direction_other pobj_from_direction amod_observations_such nsubj_thekindsoffactsthatparsingapproachestomachine_observations prep_observations_about nn_reordering_phrase pobj_about_reordering prep_reordering_between amod_languages_different pobj_between_languages cop_thekindsoffactsthatparsingapproachestomachine_are advmod_thekindsoffactsthatparsingapproachestomachine_precisely nsubjpass_designed_translation auxpass_designed_are dep_thekindsoffactsthatparsingapproachestomachine_designed aux_handle_to xcomp_designed_handle cc_handle_and conj_handle_do advmod_do_successfully ccomp_do_handle ' +o,4428,'Fox , and Wellington et al ',Galley,'num_Fox_ num_Fox_ cc_Fox_and conj_Fox_Wellington cc_Wellington_et conj_Wellington_al ' +o,4429,'Many of these tasks have been addressed in other fields , for example , hypothesis verification in the field of machine translation , sense disambiguation in speech synthesis , and relation tagging in information retrieval ',Yarowsky,'nsubjpass_addressed_Many prep_Many_of det_tasks_these pobj_of_tasks aux_addressed_have auxpass_addressed_been prep_addressed_in amod_fields_other pobj_in_fields prep_addressed_for pobj_for_example nn_verification_hypothesis conj_example_verification prep_verification_in det_field_the pobj_in_field prep_field_of nn__machine nn__translation pobj_of_ nn_disambiguation_sense conj_example_disambiguation prep_disambiguation_in nn__speech nn__synthesis pobj_in_ cc_example_and conj_example_relation partmod_relation_tagging prep_tagging_in nn__information nn__retrieval pobj_in_ ' +o,4430,'Hierarchical rules were extracted from a subset which has about 35M\\/41M words5 , and the rest of the training data were used to extract phrasal rules as in ',Och,'amod_rules_Hierarchical nsubjpass_extracted_rules auxpass_extracted_were prep_extracted_from det_subset_a pobj_from_subset nsubj_has_which rcmod_subset_has quantmod_35M\\/41M_about num_words5_35M\\/41M dobj_has_words5 cc_extracted_and det_rest_the nsubjpass_used_rest prep_rest_of det_data_the nn_data_training pobj_of_data auxpass_used_were conj_extracted_used aux_extract_to xcomp_used_extract amod_rules_phrasal dobj_extract_rules prep_extract_as dep_as_in ' +o,4431,'In this paper we focus on the second issue , constraining the grammar to the binary-branching Inversion Transduction Grammar of ',Wu,'prep_focus_In det_paper_this pobj_In_paper nsubj_focus_we prep_focus_on det_issue_the amod_issue_second pobj_on_issue xcomp_focus_constraining det_grammar_the dobj_constraining_grammar prep_constraining_to det_Grammar_the amod_Grammar_binary-branching nn_Grammar_Inversion nn_Grammar_Transduction pobj_to_Grammar prep_Grammar_of ' +o,4432,'The tagger was tested on two corpora-the Brown corpus -LRB- from the Treebank II CDROM -RRB- and the Wall Street Journal corpus -LRB- from the same source -RRB- ',Marcus,'det_tagger_The nsubjpass_tested_tagger auxpass_tested_was prep_tested_on num_corpus_two amod_corpus_corpora-the nn_corpus_Brown pobj_on_corpus dep_corpus_from det__the nn__Treebank nn__II nn__CDROM pobj_from_ cc_corpus_and det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal conj_corpus_corpus dep_corpus_from det_source_the amod_source_same pobj_from_source ' +o,4433,'The second alternative used BerkeleyAligner , which shares information between the two alignment directions to improve alignment quality ',DeNero,'det_alternative_The amod_alternative_second nsubj_used_alternative nn__BerkeleyAligner dobj_used_ nsubj_shares_which rcmod__shares nsubj_improve_information prep_information_between det_directions_the num_directions_two nn_directions_alignment pobj_between_directions aux_improve_to xcomp_shares_improve amod_quality_alignment dobj_improve_quality ' +o,4434,'Further enhancement of these utilities include compiling collocation statistics and semi-automatic gloassary construction ',Smadja,'amod_enhancement_Further prep_enhancement_of det_utilities_these pobj_of_utilities rcmod_enhancement_include xcomp_include_compiling nn_statistics_collocation dobj_compiling_statistics amod__ cc__and conj__semi-automatic amod__gloassary nn__construction dep_enhancement_ ' +o,4435,'corpus (Dunning, 1993; Scott, 1997; Rayson et al., 2004).',Dunning,'' +o,4436,'Firstly , resorted to heuristics to extract the Stringto-Dependency trees , whereas our approach employs the well formalized CCG grammatical theory ',Shen,'advmod_resorted_Firstly nsubj_resorted_ prep_resorted_to pobj_to_heuristics aux_extract_to xcomp_resorted_extract det_trees_the amod_trees_Stringto-Dependency dobj_extract_trees mark_employs_whereas poss_approach_our nsubj_employs_approach advcl_resorted_employs det_theory_the advmod_formalized_well amod_theory_formalized nn_theory_CCG amod_theory_grammatical dobj_employs_theory ' +o,4437,'Minimum Error Rate training over BLEU was used to optimize the weights for each of these models over the development test data ',Och,'nn__Minimum nn__Error nn__Rate nn__training nsubjpass_used_ prep__over pobj_over_BLEU auxpass_used_was aux_optimize_to xcomp_used_optimize det_weights_the dobj_optimize_weights prep_optimize_for pobj_for_each prep_each_of det_models_these pobj_of_models prep_models_over det_data_the nn_data_development nn_data_test pobj_over_data ' +o,4438,'Kappa is a better measurement of agreement than raw percentage agreement because it factors out the level of agreement which would be reached by random annotators using the same distribution of categories as the real coders ',Carletta,'nsubj_measurement_Kappa cop_measurement_is det_measurement_a amod_measurement_better prep_measurement_of pobj_of_agreement prep_agreement_than amod_agreement_raw nn_agreement_percentage pobj_than_agreement partmod_agreement_ prep__because dep_factors_it pobj_because_factors prep__out det_level_the pobj_out_level prep_level_of pobj_of_agreement nsubjpass_reached_which aux_reached_would auxpass_reached_be rcmod_level_reached prep_reached_by amod_annotators_random pobj_by_annotators partmod_annotators_using det_distribution_the amod_distribution_same dobj_using_distribution prep_distribution_of pobj_of_categories prep_categories_as det_coders_the amod_coders_real pobj_as_coders ' +o,4439,'We use the log-likelihood X ~ statistic , rather than the Pearson \'s X 2 statistic , as this is thought to be more appropriate when the counts in the contingency table are low ',Dunning,'nsubj_use_We det_statistic_the amod_statistic_log-likelihood nn_statistic_X nn_statistic_~ dobj_use_statistic cc_statistic_rather dep_rather_than det_Pearson_the poss_statistic_Pearson possessive_Pearson_\'s nn_statistic_X num_statistic_2 conj_statistic_statistic mark_thought_as nsubjpass_thought_this auxpass_thought_is dep_statistic_thought aux_appropriate_to cop_appropriate_be advmod_appropriate_more xcomp_thought_appropriate advmod_low_when det_counts_the nsubj_low_counts prep_counts_in det_table_the nn_table_contingency pobj_in_table cop_low_are advcl_appropriate_low ' +o,4440,'3 Experimental Results and Discussion We test our parsing models on the CONLL-2007 data set on various languages including Arabic , Basque , Catalan , Chinese , English , Italian , Hungarian , and Turkish ',Marcus,'num_Results_3 nn_Results_Experimental cc_Results_and conj_Results_Discussion nsubj_test_We dep_Results_test poss_models_our amod_models_parsing dobj_test_models prep_test_on det_data_the nn_data_CONLL-2007 nn_data_ pobj_on_data partmod_data_set prep_set_on amod_languages_various pobj_on_languages prep_languages_including nn_Hungarian_Arabic appos_Hungarian_Basque appos_Hungarian_Catalan appos_Hungarian_Chinese appos_Hungarian_English amod_Hungarian_Italian pobj_including_Hungarian cc_Hungarian_and conj_Hungarian_Turkish ' +o,4441,'NP chunks and technical terms fall into this difficult-toassess category ',Ramshaw,'nn__NP nn__chunks nsubj_fall_ cc__and amod__technical nn__terms conj__ prep_fall_into det_category_this amod_category_difficult-toassess pobj_into_category ' +o,4442,'Under the maximum entropy framework , evidence from different features can be combined with no assumptions of feature independence ',Berger,'prep_combined_Under det__the amod__maximum amod__entropy nn__framework pobj_Under_ nsubjpass_combined_evidence prep_evidence_from amod_features_different pobj_from_features aux_combined_can auxpass_combined_be prep_combined_with det_assumptions_no pobj_with_assumptions prep_assumptions_of nn_independence_feature pobj_of_independence ' +o,4443,'31 Background introduced the quasisynchronous grammar formalism ',Smith,'num__31 nn__Background dep_introduced_ det_formalism_the amod_formalism_quasisynchronous nn_formalism_grammar dobj_introduced_formalism ' +o,4444,'Our technique is based on a novel Gibbs sampler that draws samples from the posterior distributionofaphrase-basedtranslationmodel but operates in linear time with respect to the number of input words -LRB- Section 2 -RRB- ',Koehn,'poss_technique_Our nsubjpass_based_technique auxpass_based_is prep_based_on det_sampler_a amod_sampler_novel nn_sampler_Gibbs pobj_on_sampler nsubj_draws_that rcmod_sampler_draws dobj_draws_samples prep_draws_from det__the amod__posterior amod__distributionofaphrase-basedtranslationmodel pobj_from_ cc_draws_but conj_draws_operates prep_operates_in amod_time_linear pobj_in_time prep_time_with pobj_with_respect prep_operates_to det_number_the pobj_to_number prep_number_of nn_words_input pobj_of_words appos_words_Section num_Section_2 ' +o,4445,'We analyze our results using syntactic features extracted from a parse tree generated by Collins parser and compare those to models built using features extracted from FrameNets human annotations ',Collins,'nsubj_analyze_We poss_results_our dobj_analyze_results xcomp_analyze_using amod_features_syntactic dobj_using_features partmod_features_extracted prep_extracted_from det_tree_a amod_tree_parse pobj_from_tree partmod_tree_generated prep_generated_by nn__Collins nn__parser pobj_by_ cc_analyze_and conj_analyze_compare dobj_compare_those prep_compare_to pobj_to_models partmod_models_built xcomp_built_using dobj_using_features partmod_features_extracted prep_extracted_from nn_annotations_FrameNets amod_annotations_human pobj_from_annotations ' +o,4446,'Weights for the log-linear model are set using the 500-sentence tuning set provided for the shared task with minimum error rate training as implemented by Venugopal and Vogel -LRB- 2005 -RRB- ',Och,'nsubjpass_set_Weights prep_Weights_for det_model_the amod_model_log-linear pobj_for_model auxpass_set_are xcomp_set_using det_tuning_the amod_tuning_500-sentence dobj_using_tuning partmod_tuning_set prep_set_provided dep_provided_for det_task_the amod_task_shared pobj_for_task prep_task_with amod__minimum nn__error nn__rate nn__training pobj_with_ mark_implemented_as advcl_set_implemented prep_implemented_by pobj_by_Venugopal cc_Venugopal_and conj_Venugopal_Vogel dep_implemented_2005 ' +o,4447,'both relevant and non-redundant , some recent work focuses on improved search ',Goldstein,'det__both amod__relevant cc_relevant_and conj_relevant_non-redundant nsubj_focuses_ det_work_some amod_work_recent appos__work prep_focuses_on amod__improved nn__search pobj_on_ ' +o,4448,'There are multiple studies showing that the agreement between two -LRB- untrained -RRB- native speakers is about upper a15 a12a14a7 to lower a0a4a12a14a7 ',Berger,'expl_are_There amod_studies_multiple nsubj_are_studies partmod_studies_ xcomp__showing complm_is_that det_agreement_the nsubj_is_agreement prep_agreement_between pobj_between_two dep_agreement_untrained amod_speakers_native dep_agreement_speakers ccomp_showing_is prep_is_about amod_a12a14a7_upper amod_a12a14a7_a15 pobj_about_a12a14a7 prep_a12a14a7_to amod_a0a4a12a14a7_lower pobj_to_a0a4a12a14a7 ' +o,4449,'More specifically , a statistical word alignment model is used to acquire a bilingual lexicon consisting of NL substrings coupled with their translations in the target MRL ',Brown,'advmod_specifically_More advmod_used_specifically det__a amod__statistical nn__word nn__alignment nn__model nsubjpass_used_ auxpass_used_is aux_acquire_to xcomp_used_acquire det_lexicon_a amod_lexicon_bilingual dobj_acquire_lexicon partmod_lexicon_consisting prep_consisting_of nn_substrings_NL pobj_of_substrings prep_consisting_coupled dep_coupled_with poss_translations_their pobj_with_translations prep_translations_in det_MRL_the nn_MRL_target pobj_in_MRL ' +o,4450,'The statistical methods are based on distributional analysis -LRB- we defined a measure called mutual conditioned plausibility , a derivation of the well known mutual information -RRB- , and cluster analysis ',Dunning,'det_methods_The amod_methods_statistical nsubjpass_based_methods auxpass_based_are prep_based_on amod_analysis_distributional pobj_on_analysis nsubj_defined_we dep_based_defined det_measure_a nsubj_called_measure ccomp_defined_called amod_plausibility_mutual amod_plausibility_conditioned dobj_called_plausibility det_derivation_a appos_plausibility_derivation prep_derivation_of det_information_the advmod_known_well amod_information_known amod_information_mutual pobj_of_information cc_based_and nn_analysis_cluster nsubj__analysis conj_based_ ' +o,4451,'5 Experiments We compare the performance of our forest reranker against n-best reranking on the Penn English Treebank ',Marcus,'num_Experiments_5 nsubj_compare_We dep_Experiments_compare det_performance_the dobj_compare_performance prep_performance_of poss_reranker_our nn_reranker_forest pobj_of_reranker prep_compare_against amod_reranking_n-best pobj_against_reranking prep_reranking_on det_Treebank_the nn_Treebank_Penn nn_Treebank_English pobj_on_Treebank ' +p,4452,'This translation model differs from the well known phrase-based translation approach in two basic issues : rst , training data is monotonously segmented into bilingual units ; and second , the model considers n-gram probabilities instead of relative frequencies ',Koehn,'det_model_This nn_model_translation nsubj_differs_model prep_differs_from det__the advmod_known_well amod__known amod__phrase-based nn__translation nn__approach pobj_from_ prep__in num_issues_two amod_issues_basic pobj_in_issues nsubj_segmented_rst amod_data_training nsubj_segmented_data cop_segmented_is advmod_segmented_monotonously parataxis_differs_segmented prep_segmented_into amod_units_bilingual pobj_into_units cc_differs_and tmod_considers_second det_model_the nsubj_considers_model conj_differs_considers amod_probabilities_n-gram dobj_considers_probabilities dep_of_instead prep_probabilities_of amod_frequencies_relative pobj_of_frequencies ' +o,4453,'Confusion network and re-decoding have been well studied in the combination of different MT systems ',Rosti,'nn_network_Confusion nsubjpass_studied_network cc_network_and conj_network_re-decoding aux_studied_have auxpass_studied_been advmod_studied_well prep_studied_in det_combination_the pobj_in_combination prep_combination_of amod_systems_different nn_systems_MT pobj_of_systems ' +p,4454,'A = adjoin , T = attach , C = conjoin , G = generate In this paper , we use the perceptron-like algorithm proposed in which does not suffer from the label bias problem , and is fast in training ',Collins,'det_adjoin_A amod_adjoin_= nsubj_attach_adjoin appos_adjoin_T dep_T_= ccomp_generate_attach nsubj_conjoin_C dep_conjoin_= ccomp_attach_conjoin nn_=_G nsubj_generate_= prep_use_In det_paper_this pobj_In_paper nsubj_use_we dep_generate_use det_algorithm_the amod_algorithm_perceptron-like dobj_use_algorithm amod_algorithm_proposed prep_algorithm_in pobj_in_ nsubj_suffer_which aux_suffer_does neg_suffer_not rcmod_algorithm_suffer prep_suffer_from det_problem_the nn_problem_label nn_problem_bias pobj_from_problem cc_suffer_and conj_suffer_is advmod_is_fast prep_is_in pobj_in_training ' +o,4455,'These probabilities are estimated with IBM model 1 on parallel corpora ',Brown,'det_probabilities_These nsubjpass_estimated_probabilities auxpass_estimated_are prep_estimated_with nn__IBM nn__model num__1 pobj_with_ prep_estimated_on amod_corpora_parallel pobj_on_corpora ' +p,4456,'Eigenvector centrality in particular has been successfully applied to many different types of networks , including hyperlinked web pages , lexical networks , and semantic networks ',Mihalcea,'amod_centrality_Eigenvector nsubjpass_applied_centrality prep_centrality_in pobj_in_particular aux_applied_has auxpass_applied_been advmod_applied_successfully prep_applied_to amod_types_many amod_types_different pobj_to_types prep_types_of pobj_of_networks prep_types_including amod__hyperlinked nn__web nn__pages pobj_including_ amod__lexical nn__networks conj__ cc__and amod_networks_semantic conj__networks ' +o,4457,'However , current sentence alignment models , ',Brown,'advmod_alignment_However nsubj_alignment_current advmod_alignment_sentence dobj_alignment_models ' +o,4458,'A number of bootstrapping methods have been proposed for NLP tasks -LRB- eg , Collins and Singer -LRB- 1999 -RRB- , Riloff and Jones -LRB- 1999 -RRB- -RRB- ',Yarowsky,'det_number_A nsubjpass_proposed_number prep_number_of amod_methods_bootstrapping pobj_of_methods aux_proposed_have auxpass_proposed_been prep_proposed_for nn_tasks_NLP pobj_for_tasks nn__eg dep_tasks_ conj__Collins cc__and conj__Singer appos_Singer_1999 appos_Singer_Riloff cc_Riloff_and conj_Riloff_Jones appos_Riloff_1999 ' +o,4459,'Dependency relations are produced using a version of the Collins parser that has been adapted for building dependencies ',Collins,'amod_relations_Dependency nsubjpass_produced_relations auxpass_produced_are xcomp_produced_using det_version_a dobj_using_version prep_version_of det__the nn__Collins nn__parser pobj_of_ nsubjpass_adapted_that aux_adapted_has auxpass_adapted_been rcmod_version_adapted prep_adapted_for nn_dependencies_building pobj_for_dependencies ' +o,4460,'For each , we give case-insensitive scores on version 06 of METEOR with all modules enabled , version 104 of IBMstyle BLEU , and version 5 of TER ',Papineni,'prep_give_For pobj_For_each nsubj_give_we amod_scores_case-insensitive dobj_give_scores prep_scores_on pobj_on_version num_version_06 prep_version_of nn__METEOR pobj_of_ prep__with det_modules_all pobj_with_modules partmod_modules_enabled conj_scores_version num_version_104 prep_version_of nn__IBMstyle nn__BLEU pobj_of_ cc_scores_and conj_scores_version num_version_5 prep_version_of nn__TER pobj_of_ ' +o,4461,'For an alignment model , most of these use the Aachen HMM approach , the implementation of IBM Model 4 in GIZA + + or , more recently , the semi-supervised EMD algorithm ',Fraser,'det_model_an amod_model_alignment pobj_For_model conj_model_most prep_most_of det_use_these pobj_of_use det__the nn__Aachen nn__HMM nn__approach dep_use_ det_implementation_the appos_use_implementation prep_implementation_of nn_Model_IBM pobj_of_Model num_Model_4 prep_Model_in pobj_in_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_model_or advmod_recently_more dep_model_recently det_algorithm_the amod_algorithm_semi-supervised nn_algorithm_EMD conj_model_algorithm ' +o,4462,'In the context of part-of-speech tagging , argue for the same distinctions made here between discriminative models and discriminative training criteria , and come to the same conclusions ',Klein,'prep_argue_In det_context_the pobj_In_context prep_context_of amod_tagging_part-of-speech pobj_of_tagging nsubj_argue_ prep_argue_for det_distinctions_the amod_distinctions_same pobj_for_distinctions dep_distinctions_made advmod_made_here prep_made_between amod_models_discriminative pobj_between_models cc_models_and amod_criteria_discriminative nn_criteria_training conj_models_criteria cc_made_and conj_made_come prep_come_to det_conclusions_the amod_conclusions_same pobj_to_conclusions ' +o,4463,'One of the applications is in automatic summarization in order to compress sentences extracted for the summary ',Jing,'nsubj_is_One prep_One_of det_applications_the pobj_of_applications prep_is_in amod_summarization_automatic pobj_in_summarization prep_summarization_in pobj_in_order prep_is_to amod_sentences_compress pobj_to_sentences partmod_sentences_extracted prep_extracted_for det_summary_the pobj_for_summary ' +o,4464,'Several researchers -LRB- eg , -RRB- work on reducing the granularity of sense inventories for WSD ',Navigli,'amod_researchers_Several nsubj_work_researchers appos_researchers_eg dep_eg_ prep_work_on pcomp_on_reducing det_granularity_the dobj_reducing_granularity prep_granularity_of nn_inventories_sense pobj_of_inventories prep_inventories_for pobj_for_WSD ' +o,4465,'Evaluation 61 Evaluation at the Token Level This section compares translation model estimation methods A , B , and C to each other and to Model 1 ',Brown,'num_Evaluation_61 dobj_Evaluation_Evaluation prep_Evaluation_at det_section_the nn_section_Token nn_section_Level nn_section_This pobj_at_section dep_Evaluation_compares nn_methods_translation nn_methods_model nn_methods_estimation dobj_compares_methods nn_C_A conj_C_B cc_C_and nsubj_Evaluation_C prep_C_to det_other_each pobj_to_other cc_to_and conj_to_to nn_Model_ pobj_to_Model dep_to_1 ' +o,4466,'Movie and product reviews have been the main focus of many of the recent studies in this area ',Turney,'nn_reviews_Movie cc_Movie_and conj_Movie_product nsubj_been_reviews aux_been_have det_focus_the amod_focus_main dobj_been_focus prep_focus_of pobj_of_many prep_many_of det_studies_the amod_studies_recent pobj_of_studies prep_been_in det_area_this pobj_in_area ' +p,4467,'Discriminative methods such as Conditional Random Fields -LRB- CRFs -RRB- , Semi-Markov Random Fields , and perceptrons have been popular approaches for sequence labeling because of their excellent performance , which is mainly due to their ability to incorporate many kinds of overlapping and non-independent features ',Collins,'amod_methods_Discriminative nsubj_been_methods dep_as_such prep_methods_as nn__Conditional nn__Random nn__Fields abbrev__CRFs pobj_as_ nn__Semi-Markov nn__Random nn__Fields conj__ cc__and amod__perceptrons conj__ aux_been_have amod_approaches_popular dobj_been_approaches prep_approaches_for nn_labeling_sequence pobj_for_labeling dep_of_because prep_labeling_of poss_performance_their amod_performance_excellent pobj_of_performance nsubj_due_which cop_due_is advmod_due_mainly rcmod_labeling_due prep_due_to poss_ability_their pobj_to_ability aux_incorporate_to xcomp_due_incorporate amod_kinds_many dobj_incorporate_kinds prep_kinds_of nn_features_overlapping cc_overlapping_and conj_overlapping_non-independent pobj_of_features ' +o,4468,'Note that our use of cepts differs slightly from that of , inasmuch cepts may not overlap , according to our definition ',Brown,'ccomp_overlap_Note complm_differs_that poss_use_our nsubj_differs_use prep_use_of pobj_of_cepts ccomp_Note_differs advmod_differs_slightly prep_differs_from pobj_from_that prep_that_of pobj_of_ amod_cepts_inasmuch nsubj_overlap_cepts aux_overlap_may neg_overlap_not prep_overlap_according dep_according_to poss_definition_our pobj_to_definition ' +o,4469,'B = ; M = ; O = our data ; R = ; W = ',Ratnaparkhi,'nn__B nn__= dep__M dep__= rcmod_M_ dep_M_O dep_data_= poss_data_our rcmod_O_data dep_M_R dep__= rcmod_R_ dep_M_W dep__= rcmod_W_ ' +o,4470,'1 Introduction The last few decades have seen the emergence of multiple treebanks annotated with different grammar formalisms , motivated by the diversity of languages and linguistic theories , which is crucial to the success of statistical parsing ',Marcus,'num_Introduction_1 nsubj_seen_Introduction det_decades_The amod_decades_last amod_decades_few dep_seen_decades aux_seen_have det_emergence_the dobj_seen_emergence prep_emergence_of amod_treebanks_multiple pobj_of_treebanks partmod_treebanks_annotated prep_annotated_with amod_formalisms_different nn_formalisms_grammar pobj_with_formalisms partmod_formalisms_motivated prep_motivated_by det_diversity_the pobj_by_diversity prep_diversity_of pobj_of_languages cc_languages_and amod_theories_linguistic conj_languages_theories nsubj_crucial_which cop_crucial_is rcmod_formalisms_crucial prep_crucial_to det_success_the pobj_to_success prep_success_of amod_parsing_statistical pobj_of_parsing ' +o,4471,'This normal form allows simpler algorithm descriptions than the normal forms used by and Melamed -LRB- 2003 -RRB- ',Wu,'det_form_This amod_form_normal nsubj_allows_form amod_descriptions_simpler nn_descriptions_algorithm dobj_allows_descriptions prep_allows_than det_forms_the amod_forms_normal pobj_than_forms partmod_forms_used prep_used_by pobj_by_ cc__and conj__Melamed dep_used_2003 ' +o,4472,'Studies on self-training have focused mainly on generative , constituent based parsing ',McClosky,'nsubj_focused_Studies prep_Studies_on pobj_on_self-training aux_focused_have advmod_focused_mainly prep_focused_on pobj_on_generative appos_generative_constituent partmod_constituent_based xcomp_based_parsing ' +o,4473,'~ F ~ c ~ R ~ cR -LRB- 2 -RRB- ~ -RRB- ~ -RRB- continue explanations , we begin by mentioning the ` Xtrgct \' tool by Smadja ',Smadja,'nn_~_~ nn_~_F nsubj_~_c dep_~_~ advmod_~_R amod_cR_~ dobj_~_cR dep_~_2 dep_~_~ dobj_~_~ ccomp_begin_continue dobj_continue_explanations nsubj_begin_we dep_~_begin prep_begin_by pcomp_by_mentioning det_tool_the amod_tool_Xtrgct dobj_mentioning_tool prep_mentioning_by pobj_by_Smadja ' +o,4474,'The line search is an extension of that described in -LRB- 2003 ; Quirk et al 2005 ',Och,'det_search_The nn_search_line nsubj_extension_search cop_extension_is det_extension_an prep_extension_of pobj_of_that partmod_that_described prep_described_in nn_2005_ nn_2005_2003 conj_2003_Quirk cc_2003_et conj_2003_al pobj_in_2005 ' +o,4475,'34 Learning algorithm Maximum entropy -LRB- ME -RRB- models , also known as log-linear and exponential learning models , has been adopted in the SC classification task ',Berger,'num_algorithm_34 nn_algorithm_Learning nsubj_Maximum_algorithm amod_models_entropy abbrev_models_ME dobj_Maximum_models nsubjpass_adopted_ advmod_known_also partmod__known prep_known_as nn_models_log-linear cc_log-linear_and conj_log-linear_exponential nn_models_learning pobj_as_models aux_adopted_has auxpass_adopted_been dep_Maximum_adopted prep_adopted_in det_task_the nn_task_SC nn_task_classification pobj_in_task ' +o,4476,'Its roots are the same as computational linguistics -LRB- CL -RRB- , but it has been largely ignored in CL until recently ',Carletta,'poss_roots_Its nsubj_linguistics_roots cop_linguistics_are det_linguistics_the amod_linguistics_same advmod_computational_as amod_linguistics_computational abbrev_linguistics_CL cc_linguistics_but nsubjpass_ignored_it aux_ignored_has auxpass_ignored_been advmod_ignored_largely conj_linguistics_ignored prep_ignored_in pobj_in_CL prep_ignored_until dep_until_recently ' +o,4477,'The recent emphasis on improving these components of a translation system is likely due in part to the widespread availability of NLP tools for the language that is most frequently the target : English ',Brants,'det_emphasis_The amod_emphasis_recent nsubj_likely_emphasis prep_emphasis_on pcomp_on_improving det_components_these dobj_improving_components prep_components_of det__a nn__translation nn__system pobj_of_ cop_likely_is advmod_likely_due prep_likely_in pobj_in_part prep_likely_to det_availability_the amod_availability_widespread pobj_to_availability prep_availability_of nn_tools_NLP pobj_of_tools prep_tools_for det_language_the pobj_for_language nsubj_target_that cop_target_is advmod_frequently_most advmod_target_frequently det_target_the rcmod_language_target amod_target_English ' +o,4478,'However more recent results have shown that it can indeed improve parser performance ',McClosky,'dep_results_However amod_results_more amod_results_recent aux_shown_have dep_results_shown complm_improve_that nsubj_improve_it aux_improve_can advmod_improve_indeed ccomp_shown_improve nn_performance_parser dobj_improve_performance ' +o,4479,'4 Experiments We evaluated our classifier-based best-first parser on the Wall Street Journal corpus of the Penn Treebank using the standard split : sections 2-21 were used for training , section 22 was used for development and tuning of parameters and features , and section 23 was used for testing ',Marcus,'num_Experiments_4 nsubjpass_used_Experiments nsubj_evaluated_We rcmod_Experiments_evaluated poss_parser_our amod_parser_classifier-based amod_parser_best-first dobj_evaluated_parser prep_evaluated_on det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nsubj_using_corpus prep_corpus_of det__the nn__Penn nn__Treebank pobj_of_ pcomp_on_using det_split_the amod_split_standard dobj_using_split dep_split_sections dep_sections_2-21 auxpass_used_were prep_used_for pobj_for_training nsubjpass_used_section num_section_22 auxpass_used_was ccomp_used_used prep_used_for pobj_for_development cc_development_and conj_development_tuning prep_development_of pobj_of_parameters cc_parameters_and conj_parameters_features cc_used_and nsubjpass_used_section num_section_23 auxpass_used_was conj_used_used prep_used_for pobj_for_testing ' +o,4480,'For example , the constrained optimization method of relies on approximations of sensitivity -LRB- which they call CA -RRB- and specificity2 -LRB- their CR -RRB- ; related techniques rely on approximations of true positives , false positives , and false negatives , and , indirectly , recall and precision ',Jansche,'prep_relies_For pobj_For_example det_method_the amod_method_constrained nn_method_optimization nsubj_relies_method prep_method_of pobj_of_ prep_relies_on pobj_on_approximations prep_approximations_of pobj_of_sensitivity dobj_call_which nsubj_call_they dep_approximations_call dobj_call_CA cc_approximations_and conj_approximations_specificity2 poss_CR_their appos_specificity2_CR amod_techniques_related nsubj_rely_techniques advmod_rely_ parataxis_relies_rely prt_rely_on dobj_rely_approximations prep_approximations_of amod_positives_true pobj_of_positives amod_positives_false conj_positives_positives cc_positives_and amod_negatives_false conj_positives_negatives dep_recall_and dep_recall_indirectly dep_rely_recall cc_recall_and conj_recall_precision ' +o,4481,'1 Introduction Aligning parallel texts has recently received considerable attention ',Brown,'num_texts_1 nn_texts_Introduction nn_texts_Aligning nn_texts_parallel nsubj_received_texts aux_received_has advmod_received_recently amod_attention_considerable dobj_received_attention ' +o,4482,'One aspect of VPCs that makes them dicult to extract -LRB- cited in , eg , -RRB- is that the verb and particle can be non-contiguous , eg hand the paper in and battle right on ',Smadja,'num_aspect_One nsubj_is_aspect prep_aspect_of pobj_of_VPCs nsubj_makes_that rcmod_VPCs_makes nsubj_dicult_them ccomp_makes_dicult aux_extract_to xcomp_dicult_extract dep_aspect_cited prep_cited_in dep__eg pobj_in_ complm_non-contiguous_that det_verb_the nsubj_non-contiguous_verb cc_verb_and conj_verb_particle aux_non-contiguous_can cop_non-contiguous_be ccomp_is_non-contiguous conj_non-contiguous_eg dobj_eg_hand det_paper_the measure_in_paper advmod_eg_in cc_non-contiguous_and conj_non-contiguous_battle advmod_on_right advmod_battle_on ' +o,4483,'the syntax-based system , we ran a reimplementation of the Collins parser on the English half of the bitext to produce parse trees , then restructured and relabeled them as described in Section 32 ',Collins,'det_system_the amod_system_syntax-based nsubj_restructured_system nsubj_ran_we dep_system_ran det_reimplementation_a dobj_ran_reimplementation prep_reimplementation_of det__the nn__Collins nn__parser pobj_of_ prep_ran_on det_half_the amod_half_English pobj_on_half prep_half_of det_bitext_the pobj_of_bitext aux_produce_to infmod_bitext_produce amod_trees_parse dobj_produce_trees advmod_restructured_then cc_restructured_and conj_restructured_relabeled nsubj_described_them advmod_described_as xcomp_relabeled_described prep_described_in pobj_in_Section num_Section_32 ' +o,4484,'23 Online Learning Again following , we have used the single best MIRA , which is a margin aware variant of perceptron for structured prediction ',Collins,'num_Learning_23 nn_Learning_Online dep_following_Again prep_used_following pobj_following_ nsubj_used_we aux_used_have dep_Learning_used det__the amod__single amod__best nn__MIRA dobj_used_ nsubj_margin_which cop_margin_is det_margin_a rcmod__margin amod_margin_aware dep_aware_variant prep_margin_of nn__perceptron pobj_of_ prep__for amod_prediction_structured pobj_for_prediction ' +o,4485,'For each span in the chart , we get a weight factor that is multiplied with the parameter-based expectations9 4 Experiments We applied GIZA + + to word-align parts of the Europarl corpus for English and all other 10 languages ',Koehn,'prep_get_For det_span_each pobj_For_span prep_span_in det_chart_the pobj_in_chart nsubj_get_we det_factor_a nn_factor_weight dobj_get_factor nsubjpass_multiplied_that auxpass_multiplied_is rcmod_factor_multiplied prep_multiplied_with det_expectations9_the amod_expectations9_parameter-based pobj_with_expectations9 num_Experiments_4 tmod_multiplied_Experiments nsubj_applied_We dep_Experiments_applied dobj_applied_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_applied_to amod_parts_word-align pobj_to_parts prep_parts_of det__the nn__Europarl nn__corpus pobj_of_ prep_applied_for pobj_for_English cc_English_and det_languages_all amod_languages_other num_languages_10 conj_English_languages ' +o,4486,'2 The WFST Reordering Model The Translation Template Model -LRB- TTM -RRB- is a generative model of phrase-based translation ',Brown,'nsubj_model_2 det_Model_The nn_Model_WFST nn_Model_Reordering dep_2_Model det_Translation_The dep_2_Translation nn_Model_Template dep_Translation_Model abbrev_Model_TTM cop_model_is det_model_a amod_model_generative prep_model_of amod_translation_phrase-based pobj_of_translation ' +o,4487,'al 2006 , , et al 2007a -RRB- ',Rosti,'num_al_2006 num_al_ nn_2007a_et nn_2007a_al appos_al_2007a ' +o,4488,'A tree sequence to string rule 174 A tree-sequence to string translation rule in a forest is a triple , where L is the tree sequence in source language , R is the string containing words and variables in target language , and A is the alignment between the leaf nodes of L and R This definition is similar to that of except our treesequence is defined in forest ',Zhang,'det_sequence_A nn_sequence_tree nsubj_174_sequence prep_sequence_to nn_rule_string pobj_to_rule det_rule_A amod_rule_tree-sequence prep_tree-sequence_to nn_translation_string pobj_to_translation nsubj__rule prep_rule_in det_forest_a pobj_in_forest cop__is det__a amod__triple ccomp_174_ advmod_sequence_where nsubj_sequence_L cop_sequence_is det_sequence_the nn_sequence_tree dep_string_sequence prep_sequence_in nn_language_source pobj_in_language nsubj_string_R cop_string_is det_string_the ccomp_174_string partmod_string_containing dobj_containing_words cc_words_and conj_words_variables prep_containing_in nn_language_target pobj_in_language cc_174_and nsubj_alignment_A cop_alignment_is det_alignment_the conj_174_alignment prep_alignment_between det_nodes_the nn_nodes_leaf pobj_between_nodes prep_nodes_of pobj_of_L cc_L_and conj_L_R det_definition_This nsubj_similar_definition cop_similar_is rcmod_alignment_similar prep_similar_to pobj_to_that prep_that_of pobj_of_ mark_defined_except poss_treesequence_our nsubjpass_defined_treesequence auxpass_defined_is advcl_similar_defined prep_defined_in pobj_in_forest ' +o,4489,'The weights of these models are determined using the max-BLEU method described in ',Och,'det_weights_The nsubjpass_determined_weights prep_weights_of det_models_these pobj_of_models auxpass_determined_are xcomp_determined_using det_method_the amod_method_max-BLEU dobj_using_method partmod_method_described prep_described_in ' +o,4490,'We adopted the stop condition suggested in the maximization of the likelihood on a cross-validation set of samples which is unseen at the parameter estimation ',Berger,'nsubj_adopted_We det_condition_the nn_condition_stop nsubj_suggested_condition ccomp_adopted_suggested prep_suggested_in pcomp_in_ det_maximization_the dobj__maximization prep_maximization_of det_likelihood_the pobj_of_likelihood prep__on det_set_a amod_set_cross-validation pobj_on_set prep_set_of pobj_of_samples nsubj_unseen_which cop_unseen_is rcmod_samples_unseen prep_unseen_at det_estimation_the nn_estimation_parameter pobj_at_estimation ' +p,4491,'First , several of the best-performing parsers on the WSJ treebank are cases of history-based models ',Collins,'advmod_cases_First nsubj_cases_several prep_several_of det_parsers_the amod_parsers_best-performing pobj_of_parsers prep_parsers_on det__the nn__WSJ nn__treebank pobj_on_ cop_cases_are prep_cases_of amod_models_history-based pobj_of_models ' +p,4492,'Recently , we can see an important development in natural language processing and computational linguistics towards the use of empirical learning methods -LRB- for instance , -RRB- ',Marcus,'advmod_see_Recently nsubj_see_we aux_see_can det_development_an amod_development_important dobj_see_development prep_see_in amod_processing_natural nn_processing_language pobj_in_processing cc_processing_and nn_linguistics_computational conj_processing_linguistics prep_processing_towards det_use_the pobj_towards_use prep_use_of amod_methods_empirical nn_methods_learning pobj_of_methods dep_see_for pobj_for_instance appos_instance_ ' +n,4493,'This additional conditioning has the effect of making the choice of generation rules sensitive to the history of the generation process , and , we argue , provides a simpler , more uniform , general , intuitive and natural probabilistic generation model obviating the need for CFG-grammar transforms in the original proposal of ',Cahill,'det_conditioning_This amod_conditioning_additional nsubj_has_conditioning det_effect_the dobj_has_effect prep_effect_of pcomp_of_making det_choice_the nsubj_sensitive_choice prep_choice_of nn_rules_generation pobj_of_rules xcomp_making_sensitive prep_sensitive_to det_history_the pobj_to_history prep_history_of det_process_the nn_process_generation pobj_of_process cc_has_and nsubj_argue_we parataxis_has_argue conj_has_provides det_model_a amod_model_simpler advmod_uniform_more amod_model_uniform conj_uniform_general conj_uniform_intuitive cc_uniform_and conj_uniform_natural amod_model_probabilistic nn_model_generation nsubj_transforms_model partmod_model_obviating det_need_the dobj_obviating_need prep_need_for pobj_for_CFG-grammar dep_has_transforms prep_transforms_in det_proposal_the amod_proposal_original pobj_in_proposal prep_proposal_of ' +o,4494,'Chinese word segmentation is a well-known problem that has been studied extensively and it is known that human agreement is relatively low ',Berger,'amod_segmentation_Chinese nn_segmentation_word nsubj_problem_segmentation cop_problem_is det_problem_a amod_problem_well-known nsubjpass_studied_that aux_studied_has auxpass_studied_been rcmod_problem_studied advmod_studied_extensively nsubjpass_known_ cc__and conj__it auxpass_known_is dep_problem_known complm_low_that amod_agreement_human nsubj_low_agreement cop_low_is advmod_low_relatively ccomp_known_low ' +o,4495,'565 es ',Wellington,'num_es_565 ' +o,4496,'Yarowsky proposed an unsupervised method that used heuristics to obtain seed classifications and expanded the results to the other parts of the corpus , thus avoided the need to hand-annotate any examples ',Yarowsky,'nn__Yarowsky nsubj_proposed_ det_method_an amod_method_unsupervised dobj_proposed_method nsubj_used_that rcmod_method_used dobj_used_heuristics aux_obtain_to xcomp_used_obtain nn_classifications_seed dobj_obtain_classifications cc_used_and conj_used_expanded det_results_the dobj_expanded_results prep_expanded_to det_parts_the amod_parts_other pobj_to_parts prep_parts_of det_corpus_the pobj_of_corpus advmod_used_thus dep_used_avoided det_need_the dobj_avoided_need aux_hand-annotate_to xcomp_avoided_hand-annotate det_examples_any dobj_hand-annotate_examples ' +o,4497,'Work focusses on analyzing subjective features of text or speech , such as sentiment , opinion , emotion or point of view ',Turney,'dobj_Work_focusses prep_Work_on pcomp_on_analyzing amod_features_subjective dobj_analyzing_features prep_features_of pobj_of_text cc_text_or conj_text_speech dep_as_such prep_features_as pobj_as_sentiment conj_sentiment_opinion conj_sentiment_emotion cc_sentiment_or conj_sentiment_point prep_sentiment_of pobj_of_view ' +o,4498,'There are also attempts at a more fine-grained analysis of accuracy , targeting specific linguistic constructions or grammatical functions ',McDonald,'expl_are_There advmod_are_also nsubj_are_attempts prep_are_at det_analysis_a advmod_fine-grained_more amod_analysis_fine-grained pobj_at_analysis prep_analysis_of pobj_of_accuracy dep_are_targeting amod_constructions_specific amod_constructions_linguistic dobj_targeting_constructions cc_constructions_or amod_functions_grammatical conj_constructions_functions ' +o,4499,'We train our feature weights using max-BLEU and decode with a CKY-based decoder that supports language model scoring directly integrated into the search ',Och,'nsubj_train_We poss_weights_our nn_weights_feature dobj_train_weights xcomp_train_using amod__max-BLEU dobj_using_ cc__and conj__decode prep_using_with det_decoder_a amod_decoder_CKY-based pobj_with_decoder nsubj_supports_that rcmod_decoder_supports nn_model_language dobj_supports_model partmod_model_scoring advmod_integrated_directly acomp_scoring_integrated prep_integrated_into det_search_the pobj_into_search ' +o,4500,'In previous work , we developed an unsupervised learning algorithm that automatically recognizes definite NPs that are existential without syntactic modification because their meaning is universally understood ',Bean,'prep_developed_In amod__previous nn__work pobj_In_ nsubj_developed_we det_algorithm_an amod_algorithm_unsupervised nn_algorithm_learning dobj_developed_algorithm nsubj_recognizes_that advmod_recognizes_automatically rcmod_algorithm_recognizes amod_NPs_definite dobj_recognizes_NPs nsubj_existential_that cop_existential_are rcmod_NPs_existential prep_existential_without amod_modification_syntactic pobj_without_modification mark_understood_because poss_meaning_their nsubjpass_understood_meaning auxpass_understood_is advmod_understood_universally advcl_existential_understood ' +o,4501,'For nonprojective parsing , the analogy to the inside algorithm is the O -LRB- n3 -RRB- matrix-tree algorithm , which is dominated asymptotically by a matrix determinant ',Smith,'prep_algorithm_For amod_parsing_nonprojective pobj_For_parsing det_analogy_the nsubj_algorithm_analogy prep_analogy_to det_algorithm_the amod_algorithm_inside pobj_to_algorithm cop_algorithm_is det_algorithm_the nn_algorithm_O appos_algorithm_n3 amod_algorithm_matrix-tree nsubjpass_dominated_which auxpass_dominated_is rcmod_algorithm_dominated advmod_dominated_asymptotically prep_dominated_by det_determinant_a nn_determinant_matrix pobj_by_determinant ' +o,4502,'We compare against several competing systems , the first of which is based on the original IBM Model 4 for machine translation and the HMM machine translation alignment model as implemented in the GIZA + + package ',Brown,'nsubj_compare_We prep_compare_against amod_systems_several amod_systems_competing pobj_against_systems det_first_the nsubjpass_based_first prep_first_of pobj_of_which auxpass_based_is rcmod_systems_based prep_based_on det_Model_the amod_Model_original nn_Model_IBM pobj_on_Model num_Model_4 prep_Model_for nn__machine nn__translation pobj_for_ cc_Model_and det_model_the nn_model_HMM nn_model_machine nn_model_translation nn_model_alignment conj_Model_model dep_as_ advmod_implemented_as partmod_model_implemented prep_implemented_in det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__package pobj_in_ ' +o,4503,'Probability estimates of the RHS given the LHS are often smoothed by making a Markov assumption regarding the conditional independence of a category on those more than k categories away : P -LRB- X Y1Yn -RRB- = P -LRB- Y1 X -RRB- nY i = 2 P -LRB- Yi X , Y1 Yi1 -RRB- P -LRB- Y1 X -RRB- nY i = 2 P -LRB- Yi X , Yik Yi1 -RRB- ',Collins,'amod_estimates_Probability nsubj_nY_estimates prep_estimates_of det_RHS_the pobj_of_RHS partmod_RHS_given det_LHS_the nsubjpass_smoothed_LHS auxpass_smoothed_are advmod_smoothed_often ccomp_given_smoothed prep_smoothed_by pcomp_by_making det_assumption_a nn_assumption_Markov dobj_making_assumption prep_assumption_regarding det_independence_the amod_independence_conditional pobj_regarding_independence prep_independence_of det_category_a pobj_of_category prep_category_on det_categories_those dep_than_more quantmod_k_than num_categories_k pobj_on_categories advmod__away amod_categories_ dep_category_P nn_Y1Yn_X nsubj_P_Y1Yn dep_P_= ccomp_P_P nn_X_Y1 appos_category_X nsubj_=_i ccomp_nY_= number_P_2 num_nY_P nn_X_Yi dep_nY_X nn_Yi1_Y1 appos_X_Yi1 nn_nY_P nn_X_Y1 appos_nY_X dobj_=_nY partmod_nY_i dep_P_= num_P_2 ccomp_i_P nn_X_Yi appos_nY_X nn_Yi1_Yik dep_X_Yi1 ' +o,4504,'Because treebank annotation for individual formalisms is prohibitively expensive , there have been a number of efforts to extract TAGs , LFGs , and , more recently , HPSGs , from the Penn Treebank ',Cahill,'mark_expensive_Because nn_annotation_treebank nsubj_expensive_annotation prep_annotation_for amod_formalisms_individual pobj_for_formalisms cop_expensive_is advmod_expensive_prohibitively advcl_been_expensive expl_been_there aux_been_have det_number_a dobj_been_number prep_number_of pobj_of_efforts aux_extract_to infmod_efforts_extract dobj_extract_TAGs appos_TAGs_LFGs dep_HPSGs_and advmod_recently_more dep_HPSGs_recently dep_extract_HPSGs prep_HPSGs_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,4505,'For experiment on English , we used the English Penn Treebank -LRB- PTB -RRB- and the constituency structures were converted to dependency trees using the same rules as ',Marcus,'prep_used_For pobj_For_experiment prep_experiment_on pobj_on_English nsubj_used_we det__the amod__English nn__Penn nn__Treebank abbrev__PTB nsubjpass_converted_ cc__and det_structures_the nn_structures_constituency conj__structures auxpass_converted_were ccomp_used_converted prep_converted_to amod_trees_dependency pobj_to_trees xcomp_converted_using det_rules_the amod_rules_same dobj_using_rules prep_using_as pobj_as_ ' +p,4506,'Whereas language generation has benefited from syntax , the performance of statistical phrase-based machine translation when relying solely on syntactic phrases has been reported to be poor ',Wu,'mark_benefited_Whereas nn_generation_language nsubj_benefited_generation aux_benefited_has advcl__benefited prep_benefited_from nn__syntax pobj_from_ det_performance_the nsubj__performance prep_performance_of amod_translation_statistical amod_translation_phrase-based nn_translation_machine pobj_of_translation advmod_reported_when csubjpass_reported_relying advmod_relying_solely prep_relying_on amod_phrases_syntactic pobj_on_phrases aux_reported_has auxpass_reported_been dep_translation_reported aux_poor_to cop_poor_be xcomp_reported_poor ' +o,4507,'In fact , it has already been established that sentence level classification can improve document level analysis ',Pang,'prep_established_In pobj_In_fact nsubjpass_established_it aux_established_has advmod_established_already auxpass_established_been complm_improve_that nn_classification_sentence nn_classification_level nsubj_improve_classification aux_improve_can ccomp_established_improve nn_analysis_document nn_analysis_level dobj_improve_analysis ' +o,4508,'In this paper , we investigate the effectiveness of structural correspondence learning -LRB- SCL -RRB- in the domain adaptation task given by the CoNLL 2007 ',Blitzer,'prep_investigate_In det_paper_this pobj_In_paper nsubj_investigate_we det_effectiveness_the dobj_investigate_effectiveness prep_effectiveness_of amod_correspondence_structural pobj_of_correspondence partmod_correspondence_learning abbrev__SCL dobj_learning_ prep_learning_in det_task_the nn_task_domain nn_task_adaptation pobj_in_task prep_task_given dep_given_by det_CoNLL_the pobj_by_CoNLL num_CoNLL_2007 ' +o,4509,'Starting from a N-Best list generated from a translation decoder , an optimizer , such as Minimum Error Rate -LRB- MER -RRB- training , proposes directions to search for a better weight-vector to combine feature functions ',Och,'prep_generated_Starting dep_Starting_from det_list_a amod_list_N-Best pobj_from_list prep_generated_from det_decoder_a nn_decoder_translation pobj_from_decoder det_optimizer_an appos_decoder_optimizer dep_as_such prep_decoder_as nn_Rate_Minimum nn_Rate_Error pobj_as_Rate abbrev_Rate_MER amod_training_ nsubj_generated_training amod_directions_proposes appos_training_directions aux_search_to infmod_directions_search prep_search_for det_weight-vector_a amod_weight-vector_better pobj_for_weight-vector aux_combine_to xcomp_search_combine nn_functions_feature dobj_combine_functions ' +o,4510,'It compares favorably 505 with conventional phrase-based translation on Chinese-English news translation ',Koehn,'nsubj_compares_It advmod_compares_favorably dep_favorably_505 dep_favorably_with amod__conventional amod__phrase-based nn__translation pobj_with_ prep_compares_on amod__Chinese-English nn__news nn__translation pobj_on_ ' +o,4511,'The only trainable approaches -LRB- known to the author -RRB- to surface generation are the purely statistical machine translation -LRB- MT -RRB- systems such as and the corpus-based generation system described in ',Berger,'det_approaches_The advmod_approaches_only amod_approaches_trainable nsubj_systems_approaches partmod_approaches_known prep_known_to det_author_the pobj_to_author prep_approaches_to nn_generation_surface pobj_to_generation cop_systems_are det_systems_the advmod_statistical_purely amod_systems_statistical nn_systems_machine nn_systems_translation abbrev_systems_MT dep_as_such prep_systems_as pobj_as_ cc_systems_and det_system_the amod_system_corpus-based nn_system_generation conj_systems_system partmod_system_described prep_described_in pobj_in_ ' +o,4512,'For our experiments we used the following features , analogous to Pharaohs default feature set : P -LRB- -RRB- and P -LRB- -RRB- , the latter of which is not found in the noisy-channel model , but has been previously found to be a helpful feature ; the lexical weights Pw -LRB- -RRB- and Pw -LRB- -RRB- , which estimate how well the words in translate the words in ; 2 a phrase penalty exp -LRB- 1 -RRB- , which allows the model to learn a preference for longer or shorter derivations , analogous to Koehns phrase penalty ',Och,'prep_used_For poss_experiments_our pobj_For_experiments nsubj_used_we det_features_the amod_features_following dobj_used_features amod_feature_analogous prep_analogous_to pobj_to_Pharaohs nn_feature_default appos_features_feature partmod_feature_set dep_feature_P cc_-RRB-_and conj_-RRB-_P det__the amod__latter prep_latter_of nsubjpass_found_which auxpass_found_is neg_found_not dep_weights_found prep_found_in det_model_the amod_model_noisy-channel pobj_in_model cc_found_but aux_found_has auxpass_found_been advmod_found_previously conj_found_found aux__to cop__be det__a amod__helpful nn__feature xcomp_found_ det_weights_the amod_weights_lexical pobj_of_weights nn__Pw cc_-RRB-_and conj_-RRB-_Pw appos_feature_ nsubj_estimate_which rcmod_feature_estimate advmod_translate_how advmod_how_well det_words_the nsubj_translate_words advmod_translate_in ccomp_estimate_translate det_words_the dobj_translate_words prep_translate_in dep_exp_2 det_exp_a amod_exp_phrase nn_exp_penalty dep_features_exp dep_features_1 nsubj_allows_which dep_1_allows det_model_the dobj_allows_model aux_learn_to infmod_model_learn det_preference_a dobj_learn_preference prep_preference_for amod_derivations_longer cc_longer_or conj_longer_shorter pobj_for_derivations amod_features_analogous prep_analogous_to nn__Koehns nn__phrase nn__penalty pobj_to_ ' +o,4513,'In the area of statistical machine translation -LRB- SMT -RRB- , recently a combination of the BLEU evaluation metric and the bootstrap method for statistical significance testing has become popular ',Koehn,'prep_become_In det_area_the pobj_In_area prep_area_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_SMT advmod_combination_recently det_combination_a nsubj_become_combination prep_combination_of det__the amod__BLEU nn__evaluation amod__metric pobj_of_ cc_combination_and det_method_the nn_method_bootstrap conj_combination_method prep_method_for amod__statistical nn__significance nn__testing pobj_for_ aux_become_has dep_become_popular ' +o,4514,' , and in some cases , to factor the translation problem so that the baseline MT system can take advantage of the reduction in sparsity by being able to work on word stems ',Finkel,'cc__and conj__in det_cases_some pobj_in_cases aux_factor_to infmod_cases_factor det_problem_the nn_problem_translation dobj_factor_problem advmod_take_so dep_take_that det_system_the amod_system_baseline nn_system_MT nsubj_take_system aux_take_can dep_factor_take dobj_take_advantage prep_advantage_of det_reduction_the pobj_of_reduction prep_reduction_in pobj_in_sparsity prep_take_by cop_able_being pcomp_by_able aux_work_to xcomp_able_work prep_work_on pobj_on_word dep_word_stems ' +o,4515,'We use a tagger based on Adwait Ratnaparkhi \'s method ',Ratnaparkhi,'nsubj_use_We det_tagger_a dobj_use_tagger prep_use_based dep_based_on nn_Ratnaparkhi_Adwait poss_method_Ratnaparkhi possessive_Ratnaparkhi_\'s pobj_on_method ' +o,4516,'(Turney, 2002; Pang et al., 2002; Dave at al., 2003).',Turney,'' +o,4517,' proposes Inversion Transduction Grammars , treating translation as a process of parallel parsing of the source and target language via a synchronized grammar ',Wu,'advmod_proposes_ nn_Grammars_Inversion nn_Grammars_Transduction nsubj_proposes_Grammars partmod_Grammars_treating dobj_treating_translation prep_treating_as det_process_a pobj_as_process prep_process_of amod_parsing_parallel pobj_of_parsing prep_parsing_of det_language_the nn_language_source cc_source_and conj_source_target pobj_of_language prep_language_via det_grammar_a amod_grammar_synchronized pobj_via_grammar ' +o,4518,'Some previous approaches handle unknown words explicitly using ambiguity class components conditioned on various morphological features , and this has shown to produce good tagging results , especially when dealing with incomplete dictionaries ',Goldberg,'det__Some amod__previous nn__approaches nsubj_handle_ amod_words_unknown dobj_handle_words advmod_using_explicitly xcomp_handle_using amod_components_ambiguity nn_components_class dobj_using_components partmod_components_conditioned prep_conditioned_on amod_features_various amod_features_morphological pobj_on_features cc_handle_and nsubj_shown_this aux_shown_has conj_handle_shown aux_produce_to xcomp_shown_produce amod_results_good nn_results_tagging dobj_produce_results advmod_when_especially dep_produce_when xcomp_produce_dealing prep_dealing_with amod_dictionaries_incomplete pobj_with_dictionaries ' +o,4519,'Various corpus-based approaches to word sense disambiguation have been proposed ',Yarowsky,'amod_approaches_Various amod_approaches_corpus-based nsubjpass_proposed_approaches prep_approaches_to nn_disambiguation_word nn_disambiguation_sense pobj_to_disambiguation aux_proposed_have auxpass_proposed_been ' +o,4520,'5http : \\/ \\/ opennlpsourceforgenet \\/ We use the standard four-reference NIST MTEval data sets for the years 2003 , 2004 and 2005 -LRB- henceforth MT03 , MT04 and MT05 , respectively -RRB- for testing and the 2002 data set for tuning6 BLEU4 , METEOR and multiple-reference Word Error Rate scores are reported ',Papineni,'amod_\\/_\\/ amod_\\/_\\/ nn_\\/_opennlpsourceforgenet nsubjpass_reported_\\/ nsubj_use_We rcmod_\\/_use det_data_the amod_data_standard amod_data_four-reference nn_data_NIST nn_data_MTEval nsubj_sets_data ccomp_use_sets prep_sets_for dep_years_the pobj_for_years tmod_years_2003 num_2003_2004 dep_2003_and dep_2003_2005 amod_MT03_henceforth dep_years_MT03 conj_MT03_MT04 cc_MT03_and conj_MT03_MT05 advmod_MT03_respectively prep_sets_for pobj_for_testing cc_\\/_and det_scores_the num_data_2002 dep_set_data amod_scores_set prep_set_for amod__tuning6 nn__BLEU4 pobj_for_ nn__METEOR conj__ cc__and amod_Rate_multiple-reference nn_Rate_Word nn_Rate_Error conj__Rate conj_\\/_scores auxpass_reported_are dep_5http_reported ' +o,4521,' uses the observed frequencies within a specific syntactic pattern -LRB- subject\\/verb , and verb\\/object -RRB- to derive a cooccu , -RRB- rence score which is an estimate of mutual information ',Hindle,'advmod_uses_ det_frequencies_the amod_frequencies_observed dobj_uses_frequencies prep_uses_within det_score_a amod_score_specific dep_specific_syntactic nn_score_pattern dep_cooccu_subject\\/verb cc_subject\\/verb_and conj_subject\\/verb_verb\\/object prep_subject\\/verb_to pobj_to_derive det_cooccu_a dep_score_cooccu amod_score_rence pobj_within_score nsubj_estimate_which cop_estimate_is det_estimate_an dep_uses_estimate prep_estimate_of amod__mutual nn__information pobj_of_ ' +o,4522,'1 Introduction Machine translation systems based on probabilistic translation models are generally trained using sentence-aligned parallel corpora ',Brown,'num_systems_1 nn_systems_Introduction nn_systems_Machine nn_systems_translation nsubj_based_systems prep_based_on amod_models_probabilistic nn_models_translation pobj_on_models nsubjpass_trained_ auxpass_trained_are advmod_trained_generally dep_based_trained xcomp_trained_using amod_corpora_sentence-aligned amod_corpora_parallel dobj_using_corpora ' +o,4523,'To summarize , we can describe our system as follows : it is based on s implementation of , which has been fed at each iteration by a different dataset consisting of the supervised and unsupervised part : precisely , by a concatenation of the manually tagged training data -LRB- WSJ portion of the PTB 3 for English , morphologically disambiguated data from PDT 20 for Czech -RRB- and a chunk of automatically tagged unsupervised data ',Collins,'aux_summarize_To dep_describe_summarize nsubj_describe_we aux_describe_can poss_system_our dobj_describe_system mark_follows_as advcl_describe_follows nsubjpass_based_it auxpass_based_is parataxis_describe_based prep_based_on amod_s_ pobj_on_s dobj_based_implementation prep_implementation_of pobj_of_ nsubjpass_fed_which aux_fed_has auxpass_fed_been rcmod__fed prep_fed_at det_iteration_each pobj_at_iteration prep_based_by det_dataset_a amod_dataset_different pobj_by_dataset partmod_dataset_consisting prep_consisting_of det_part_the amod_part_supervised cc_supervised_and conj_supervised_unsupervised pobj_of_part advmod_based_precisely prep_based_by det_concatenation_a pobj_by_concatenation prep_concatenation_of det_manually_the pobj_of_manually partmod_manually_tagged amod_data_training dobj_tagged_data nn_portion_WSJ dep_data_portion prep_portion_of det_PTB_the pobj_of_PTB num_PTB_3 prep_PTB_for pobj_for_English advmod_portion_morphologically amod_data_disambiguated dep_portion_data prep_data_from pobj_from_PDT num_PDT_20 prep_data_for pobj_for_Czech cc_data_and det_chunk_a conj_data_chunk prep_chunk_of advmod_tagged_automatically pcomp_of_tagged amod_data_unsupervised dobj_tagged_data ' +o,4524,'The time complexity of the CKY-based binarization algorithm is -LRB- n3 -RRB- , which is higher than that of the linear binarization such as the synchronous binarization ',Zhang,'det_complexity_The nn_complexity_time nsubj_n3_complexity prep_complexity_of det_algorithm_the amod_algorithm_CKY-based nn_algorithm_binarization pobj_of_algorithm cop_n3_is nsubj_higher_which cop_higher_is rcmod_n3_higher prep_higher_than pobj_than_that prep_that_of det_binarization_the amod_binarization_linear pobj_of_binarization dep_as_such prep_binarization_as det_binarization_the amod_binarization_synchronous pobj_as_binarization ' +o,4525,'presented in ',Collins,'prep_presented_in ' +o,4526,'Until now , translation models have been evaluated either subjectively or using relative metrics , such as perplexity with respect to other models ',Brown,'prep_evaluated_Until pobj_Until_now nn_models_translation nsubjpass_evaluated_models aux_evaluated_have auxpass_evaluated_been advmod_evaluated_either advmod__subjectively acomp_evaluated_ cc__or conj__using amod_metrics_relative dobj_using_metrics dep_as_such prep_metrics_as pobj_as_perplexity prep_perplexity_with pobj_with_respect prep_using_to amod_models_other pobj_to_models ' +o,4527,'For Penn Treebank II style annotation , in which a nonterminal symbol is a category together with zero or more functional tags , we adopt the following scheme : the atomic pattern a matches any label with category a or functional tag a ; moreover , we define Boolean operators ^ , _ , and : ',Marcus,'prep_adopt_For nn__Penn nn__Treebank nn__II nn__style nn__annotation pobj_For_ rel_is_in pobj_in_which det_symbol_a amod_symbol_nonterminal nsubj_is_symbol rcmod__is det_category_a dep_together_category advmod_is_together prep_is_with nn_tags_zero cc_zero_or advmod_functional_more conj_zero_functional pobj_with_tags nsubj_adopt_we det_scheme_the amod_scheme_following nsubj_matches_scheme det_pattern_the amod_pattern_atomic dep_scheme_pattern dep_scheme_a dep_adopt_matches det_label_any dobj_matches_label prep_label_with pobj_with_category dep_category_a cc_category_or amod_tag_functional conj_category_tag dep_label_a advmod_define_moreover nsubj_define_we parataxis_matches_define nn_operators_Boolean nsubj_and_operators nn_and_^ appos_and__ xcomp_define_and ' +o,4528,'5 The task : Base NP chunking The task is base NP chunking on section 20 of the Wall Street Journal corpus , using sections 15 to 18 of the corpus as training data as in ',Ramshaw,'det_task_The nsubj_NP_task nn_NP_Base dep_task_NP partmod_NP_chunking det_task_The dobj_chunking_task cop_NP_is nn_NP_base dep_5_NP partmod_NP_chunking prep_chunking_on pobj_on_section num_section_20 prep_section_of det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_of_corpus xcomp_chunking_using dobj_using_sections number_18_15 dep_18_to measure_as_18 prep_18_of det_corpus_the pobj_of_corpus prep_corpus_as nn_data_training pobj_as_data advmod_using_as prep_using_in ' +p,4529,'This type of direct optimization is known as Minimum Error Rate Training in the MT community , and is an essential component in building the stateof-art MT systems ',Och,'det_type_This nsubjpass_known_type prep_type_of amod_optimization_direct pobj_of_optimization auxpass_known_is prep_known_as nn__Minimum nn__Error nn__Rate nn__Training pobj_as_ prep__in det_community_the nn_community_MT pobj_in_community cc_known_and cop_component_is det_component_an amod_component_essential conj_known_component prep_component_in pcomp_in_building det_systems_the amod_systems_stateof-art nn_systems_MT dobj_building_systems ' +o,4530,'4 Experiments and evaluation We carried out an evaluation on the local rephrasing of French sentences , using English as the pivot language2 We extracted phrase alignments of up to 7 word forms using the Giza + + alignment tool and the grow-diag-final-and heuristics described in on 948,507 sentences of the French-English part of the Europarl corpus and obtained some 42 million phrase pairs for which probabilities were estimated using maximum likelihood estimation ',Koehn,'num_Experiments_4 tmod_carried_Experiments dep_Experiments_and dep_Experiments_evaluation nsubj_carried_We prt_carried_out det_evaluation_an dobj_carried_evaluation prep_evaluation_on det_rephrasing_the amod_rephrasing_local pobj_on_rephrasing prep_rephrasing_of amod_sentences_French pobj_of_sentences dep_extracted_using dobj_using_English prep_English_as det_language2_the amod_language2_pivot pobj_as_language2 nsubj_extracted_We conj_carried_extracted nn_alignments_phrase dobj_extracted_alignments prep_alignments_of advmod_using_up dep_up_to num_forms_7 nn_forms_word pobj_to_forms pcomp_of_using det__the amod__Giza cc_Giza_+ conj_Giza_+ amod__alignment nn__tool dobj_using_ cc_carried_and det_heuristics_the amod_heuristics_grow-diag-final-and nsubj_described_heuristics conj_carried_described prep_described_in pobj_in_ prep_described_on num_sentences_948,507 pobj_on_sentences prep_sentences_of det_part_the amod_part_French-English pobj_of_part prep_part_of det__the nn__Europarl nn__corpus pobj_of_ cc_described_and conj_described_obtained det_pairs_some number_million_42 num_pairs_million nn_pairs_phrase dobj_obtained_pairs rel_estimated_for pobj_for_which nsubjpass_estimated_probabilities auxpass_estimated_were rcmod_pairs_estimated xcomp_estimated_using amod_estimation_maximum nn_estimation_likelihood dobj_using_estimation ' +o,4531,'This paper explores an alternative approach to parsing , based on the perceptron training algorithm introduced in ',Collins,'det_paper_This nsubj_explores_paper det_approach_an amod_approach_alternative dobj_explores_approach prep_explores_to pobj_to_parsing prep_explores_based dep_based_on det_algorithm_the nn_algorithm_perceptron nn_algorithm_training pobj_on_algorithm partmod_algorithm_introduced prep_introduced_in ' +o,4532,'A simpler , related idea of penalizing distortion from some ideal matching pattern can be found in the statistical translation and word alignment models ',Brown,'det_idea_A amod_idea_simpler amod_idea_related nsubjpass_found_idea prep_idea_of pcomp_of_penalizing dobj_penalizing_distortion prep_penalizing_from det_pattern_some amod_pattern_ideal amod_pattern_matching pobj_from_pattern aux_found_can auxpass_found_be prep_found_in det__the amod__statistical nn__translation pobj_in_ cc__and nn_models_word nn_models_alignment nn_models_ conj__models ' +o,4533,'Examples have been class-based D2-gram models , smoothing techniques for structural disambiguation and word sense disambiguation ',Brown,'nsubjpass_class-based_Examples aux_class-based_have auxpass_class-based_been amod_models_D2-gram dobj_class-based_models advmod_class-based_ xcomp_class-based_smoothing dobj_smoothing_techniques prep_smoothing_for amod__structural nn__disambiguation pobj_for_ cc__and nn__word nn__sense nn__disambiguation conj__ ' +o,4534,'21 Model 2 of Both parsing models discussed in this paper inherit a great deal from this model , so we briefly describe its ` progenitive \' features here , describing only how each of the two models of this paper differ in the subsequent two sections ',Collins,'nsubj_inherit_Model num_Model_2 prep_Model_of pcomp_of_ det_models_Both amod_models_parsing dobj__models partmod_models_discussed prep_discussed_in det_paper_this pobj_in_paper dep_21_inherit det_deal_a amod_deal_great dobj_inherit_deal prep_inherit_from det_model_this pobj_from_model dep_inherit_so nsubj_describe_we advmod_describe_briefly dep_inherit_describe poss_features_its amod_features_progenitive dobj_describe_features advmod_describe_here xcomp_describe_describing advmod_describing_only advmod_differ_how nsubj_differ_each prep_each_of det_models_the num_models_two pobj_of_models prep_models_of det_paper_this pobj_of_paper ccomp_describing_differ prep_differ_in det_sections_the amod_sections_subsequent num_sections_two pobj_in_sections ' +o,4535,'Study in collocation extraction using lexical statistics has gained some insights to the issues faced in collocation extraction ',Church,'nsubj_gained_Study prep_Study_in nn_extraction_collocation pobj_in_extraction partmod_extraction_using amod_statistics_lexical dobj_using_statistics aux_gained_has det_insights_some dobj_gained_insights prep_gained_to det_issues_the pobj_to_issues partmod_issues_faced prep_faced_in nn_extraction_collocation pobj_in_extraction ' +p,4536,'In particular , the use of SVMs in initially sparked interest in using machine learning methods for sentiment classi cation ',Pang,'prep_sparked_In pobj_In_particular det_use_the nsubj_sparked_use prep_use_of pobj_of_SVMs prep_SVMs_in pobj_in_ advmod_sparked_initially dobj_sparked_interest prep_sparked_in pcomp_in_using dobj_using_machine xcomp_using_learning dobj_learning_methods prep_learning_for nn_cation_sentiment nn_cation_classi pobj_for_cation ' +o,4537,'1 Introduction Several recent syntax-based models for machine translation can be seen as instances of the general framework of synchronous grammars and tree transducers ',Galley,'num_models_1 amod_models_Introduction amod_models_Several amod_models_recent amod_models_syntax-based nsubjpass_seen_models prep_models_for nn__machine nn__translation pobj_for_ aux_seen_can auxpass_seen_be prep_seen_as pobj_as_instances prep_instances_of det_framework_the amod_framework_general pobj_of_framework prep_framework_of amod_grammars_synchronous pobj_of_grammars cc_grammars_and nn_transducers_tree conj_grammars_transducers ' +o,4538,'Solving this first methodological issue , has led to solutions dubbed hereafter as unlexicalized statistical parsing ',Matsuzaki,'dep_led_Solving det_issue_this amod_issue_first amod_issue_methodological dobj_Solving_issue aux_led_has prep_led_to pobj_to_solutions partmod_solutions_dubbed advmod_dubbed_hereafter prep_dubbed_as pobj_as_unlexicalized amod_parsing_statistical nsubj_led_parsing ' +p,4539,'Our training and test corpora , for instance , are lessthan-gargantuan compared to such collections as the Penn Treebank ',Marcus,'poss_training_Our nsubj_lessthan-gargantuan_training cc_training_and nn_corpora_test conj_training_corpora prep_training_for pobj_for_instance cop_lessthan-gargantuan_are prep_lessthan-gargantuan_compared dep_compared_to amod_collections_such pobj_to_collections prep_collections_as det_Treebank_the nn_Treebank_Penn pobj_as_Treebank ' +o,4540,'1 Introduction Translational equivalence is a mathematical relation that holds between linguistic expressions with the same meaning ',Wellington,'num_equivalence_1 nn_equivalence_Introduction nn_equivalence_Translational nsubj_relation_equivalence cop_relation_is det_relation_a amod_relation_mathematical nsubj_holds_that rcmod_relation_holds prep_holds_between amod_expressions_linguistic pobj_between_expressions prep_expressions_with det_meaning_the amod_meaning_same pobj_with_meaning ' +o,4541,'Most existing work to capture labelconsistency , has attempted to create all parenleftbign2parenrightbig pairwise dependencies between the different occurrences of an entity , , where n is the number of occurrences of the given entity ',Finkel,'amod_work_Most amod_work_existing nsubj__work aux_capture_to infmod_work_capture advmod_capture_labelconsistency aux__has dep__attempted aux_create_to xcomp__create det_dependencies_all amod_dependencies_parenleftbign2parenrightbig nn_dependencies_pairwise dobj_create_dependencies prep_dependencies_between det_occurrences_the amod_occurrences_different pobj_between_occurrences prep_occurrences_of det_entity_an pobj_of_entity advmod_number_where nsubj_number_n cop_number_is det_number_the advcl__number prep_number_of pobj_of_occurrences prep_occurrences_of det_entity_the amod_entity_given pobj_of_entity ' +o,4542,'While the research in statistical machine translation -LRB- SMT -RRB- has made significant progress , most SMT systems relyonparallel corpora toextract translation entries ',Koehn,'mark_made_While det_research_the nsubj_made_research prep_research_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT aux_made_has advcl__made amod_progress_significant dobj_made_progress amod_systems_most nn_systems_SMT nsubj__systems amod_entries_relyonparallel nn_entries_corpora nn_entries_toextract nn_entries_translation dobj__entries ' +o,4543,'In our context , bootstrapping has a similar motivation to the annealing approach of Smith and Eisner , which also tries to alter the space of hidden outputs in the E-step over time to facilitate learning in the M-step , though of course the use of bootstrapping in general is quite widespread ',Yarowsky,'prep_has_In poss_context_our pobj_In_context nsubj_has_bootstrapping det_motivation_a amod_motivation_similar dobj_has_motivation prep_motivation_to det_approach_the amod_approach_annealing pobj_to_approach prep_approach_of nn__Smith cc_Smith_and conj_Smith_Eisner pobj_of_ nsubj_tries_which advmod_tries_also rcmod__tries aux_alter_to xcomp_tries_alter det_space_the dobj_alter_space prep_space_of amod_outputs_hidden pobj_of_outputs prep_alter_in det_E-step_the pobj_in_E-step prep_E-step_over pobj_over_time aux_facilitate_to xcomp_alter_facilitate xcomp_facilitate_learning prep_learning_in det_M-step_the pobj_in_M-step mark_widespread_though prep_widespread_of pobj_of_course det_use_the nsubj_widespread_use prep_use_of pcomp_of_bootstrapping prep_bootstrapping_in pobj_in_general cop_widespread_is advmod_widespread_quite advcl_tries_widespread ' +o,4544,'Training Data Our source for syntactically annotated training data was the Penn Treebank ',Marcus,'nn_Data_Training poss_source_Our nsubj_Treebank_source prep_source_for advmod_annotated_syntactically amod_data_annotated nn_data_training pobj_for_data cop_Treebank_was det_Treebank_the nn_Treebank_Penn dep_Data_Treebank ' +o,4545,'To deal with the difficulties in parse-to-parse matching , utilizes inversion transduction grammar -LRB- ITG -RRB- for bilingual parsing ',Wu,'aux_deal_To dep_utilizes_deal prep_deal_with det_difficulties_the pobj_with_difficulties prep_difficulties_in amod_matching_parse-to-parse pobj_in_matching nsubj_utilizes_ nn_grammar_inversion nn_grammar_transduction dobj_utilizes_grammar abbrev_grammar_ITG prep_grammar_for amod_parsing_bilingual pobj_for_parsing ' +p,4546,'The IBM models have shown good performance in machine translation , and especially so within certain families of languages , for example in translating between French and English or between Sinhalese and Tamil ',Brown,'det_models_The nn_models_IBM nsubj_shown_models aux_shown_have amod_performance_good dobj_shown_performance prep_performance_in nn_translation_machine pobj_in_translation cc_performance_and advmod_performance_especially advmod_performance_so dep_performance_within amod_families_certain dep_performance_families prep_families_of pobj_of_languages prep_shown_for pobj_for_example prep_example_in pcomp_in_translating prep_translating_between pobj_between_French cc_French_and conj_French_English cc_between_or conj_between_between pobj_between_Sinhalese cc_Sinhalese_and conj_Sinhalese_Tamil ' +o,4547,'177 Proceedings of EACL \'99 IOB1 IOB2 IOE1 IOE2 [+] [+ IO IO +] (Ramshaw and Marcus, 1995) (Veenstra, 1998) (Argamon et al. , 1998) (Cardie and Pierce, 1998) accuracy 97.58\% 96.50\% 97.58\% 96.77\% 97.37\% 97.2\% precision 92.50\% 91.24\% 92.41\% 91.93\% 93.66\% 91.47\% 91.25\% 91.80\% 89.0\% 91.6 \% 90.7\% recall F~=I 92.25\% 92.37 92.32\% 91.78 92.04\% 92.23 92.46\% 92.20 90.81\% 92.22 92.61\% 92.04 92.54\% 91.89 92.27\% 92.03 94.3\% 91.6 91.6\% 91.6 91.1\% 90.9 Table 6: The F~=I scores for the (Ramshaw and Marcus, 1995) test set after training with their training data set.',Ramshaw,'' +o,4548,'The bracketed portions of Figure 1 , for example , show the base NPs in one sentence from the Penn Treebank Wall Street Journal -LRB- WSJ -RRB- corpus ',Marcus,'det_portions_The amod_portions_bracketed nsubj_show_portions prep_portions_of pobj_of_Figure num_Figure_1 prep_portions_for pobj_for_example det_NPs_the nn_NPs_base dobj_show_NPs prep_show_in num_sentence_one pobj_in_sentence prep_sentence_from det_corpus_the nn_Journal_Penn nn_Journal_Treebank nn_Journal_Wall nn_Journal_Street nn_corpus_Journal abbrev_Journal_WSJ pobj_from_corpus ' +o,4549,'Vilain and Day identify -LRB- and classify -RRB- name phrases such as company names , locations , etc detect noun phrases , by classifying each word as being inside a phrase , outside or on the boundary between phrases ',Ramshaw,'nsubj__Vilain cc_Vilain_and conj_Vilain_Day xcomp__identify dep_-LRB-_and dep_-LRB-_classify nn_phrases_name dobj_identify_phrases dep_as_such prep_phrases_as nn_names_company pobj_as_names appos_names_locations dep_names_etc partmod_phrases_ ccomp__detect nn_phrases_noun dobj_detect_phrases prep_identify_by pcomp_by_classifying det_word_each dobj_classifying_word prep_classifying_as pcomp_as_being prep_being_inside det_phrase_a pobj_inside_phrase advmod_phrase_outside cc_inside_or conj_inside_on det_boundary_the pobj_on_boundary prep_boundary_between pobj_between_phrases ' +o,4550,'(Koehn et al. , 2003) used the following distortion model, which simply penalizes nonmonotonic phrase alignments based on the word distance of successively translated source phrases with an appropriate value for the parameter a71, a36a51a4a39a38 a33 a40a52a42 a33a53a45 a32 a8 a10 a71a26a72a73a25a74 a45a62a75 a74a77a76a24a78 a45 a32 a72 (3) a79a17a80a82a81a84a83a85a15a86a88a87a70a89a91a90 languageis a means communication of MG RA RA b1 b2 b3 b4 Figure 1: Phrase alignment and reordering bi-1 bi fi-1 fi ei-1 ei bi-1 bi fi-1 fi ei-1 ei bi-1 bi fi-1 fi ei-1 ei bi-1 bi fi-1 fi ei-1 ei source target target source target target source source d=MA d=MG d=RA d=RG Figure 2: Four types of reordering patterns 3 The Global Phrase Reordering Model Figure 1 shows an example of Japanese-English phrase alignment that consists of four phrase pairs.',Koehn,'' +o,4551,'We show that link 1For a complete discussion of alignment symmetrization heuristics , including union , intersection , and refined , refer to ',Och,'nsubj_show_We complm_refer_that nn_1For_link nsubj_refer_1For det_discussion_a amod_discussion_complete dep_1For_discussion prep_discussion_of amod_heuristics_alignment nn_heuristics_symmetrization pobj_of_heuristics prep_refer_including pobj_including_union conj_union_intersection cc_union_and conj_union_refined ccomp_show_refer prep_refer_to ' +o,4552,'Specifically , stochastic translation lexicons estimated using the IBM method from a fairly large sentence-aligned Chinese-English parallel corpus are used in their approach a considerable demand for a resourcedeficient language ',Brown,'advmod_used_Specifically amod_lexicons_stochastic nn_lexicons_translation nsubjpass_used_lexicons partmod_lexicons_estimated xcomp_estimated_using det__the nn__IBM nn__method dobj_using_ prep_using_from det_corpus_a advmod_large_fairly amod_corpus_large amod_corpus_sentence-aligned amod_corpus_Chinese-English amod_corpus_parallel pobj_from_corpus auxpass_used_are prep_used_in poss_approach_their pobj_in_approach det_demand_a amod_demand_considerable dobj_used_demand prep_demand_for det_language_a amod_language_resourcedeficient pobj_for_language ' +p,4553,'Such word-based lexicalizations of probability models are used successfully in the statistical parsing models of , eg , , Charniak -LRB- 1997 -RRB- , or Ratnaparkhi -LRB- 1997 -RRB- ',Collins,'amod_lexicalizations_Such amod_lexicalizations_word-based nsubjpass_used_lexicalizations prep_lexicalizations_of nn_models_probability pobj_of_models auxpass_used_are advmod_used_successfully prep_used_in det_models_the amod_models_statistical amod_models_parsing pobj_in_models prep_models_of dep__eg pobj_of_ conj__Charniak appos_Charniak_1997 cc__or conj__Ratnaparkhi appos_Ratnaparkhi_1997 ' +o,4554,'1 Introduction The field of sentiment classification has received considerable attention from researchers in recent years ',Pang,'num_Introduction_1 det_field_The nsubj_received_field prep_field_of nn_classification_sentiment pobj_of_classification aux_received_has dep_Introduction_received amod_attention_considerable dobj_received_attention prep_received_from pobj_from_researchers prep_researchers_in amod_years_recent pobj_in_years ' +o,4555,'In paraphrase generation , a text unit that matches a pattern P can be rewritten using the paraphrase patterns of P Avarietyofmethodshavebeenproposedonparaphrase patterns extraction ',Barzilay,'prep_rewritten_In amod_generation_paraphrase pobj_In_generation det_unit_a nn_unit_text nsubjpass_rewritten_unit nsubj_matches_that rcmod_unit_matches det_P_a nn_P_pattern dobj_matches_P aux_rewritten_can auxpass_rewritten_be purpcl_rewritten_using det_patterns_the amod_patterns_paraphrase dobj_using_patterns prep_patterns_of amod_extraction_P nn_extraction_Avarietyofmethodshavebeenproposedonparaphrase nn_extraction_patterns pobj_of_extraction ' +o,4556,'Chunks as a separate level have also been used in Collins and ',Ratnaparkhi,'nsubjpass_used_Chunks prep_Chunks_as det_level_a amod_level_separate pobj_as_level aux_used_have advmod_used_also auxpass_used_been prep_used_in nn_and_Collins nn_and_ pobj_in_and ' +p,4557,'To scale LMs to larger corpora with higher-order dependencies , researchers Work completed while this author was at Google Inc have considered alternative parameterizations such as class-based models , model reduction techniques such as entropy-based pruning , novel represention schemes such as suffix arrays , Golomb Coding and distributed language models that scale more readily ',Brants,'aux_scale_To dep_completed_scale dobj_scale_LMs prep_scale_to amod_corpora_larger pobj_to_corpora prep_corpora_with amod_dependencies_higher-order pobj_with_dependencies nn_Work_researchers nsubj_completed_Work mark_was_while det_author_this nsubj_was_author advcl_completed_was prep_was_at nn_Inc_Google pobj_at_Inc aux_considered_have rcmod_Inc_considered amod_parameterizations_alternative dobj_considered_parameterizations dep_as_such prep_parameterizations_as amod__class-based nn__models pobj_as_ nn_techniques_model nn_techniques_reduction conj_Inc_techniques dep_as_such prep_techniques_as amod__entropy-based nn__pruning pobj_as_ amod_schemes_novel nn_schemes_represention conj_Inc_schemes dep_as_such prep_schemes_as nn__suffix nn__arrays pobj_as_ nn__Golomb nn__Coding conj_Inc_ cc_Inc_and amod_models_distributed nn_models_language conj_Inc_models nsubj_scale_that rcmod_models_scale advmod_readily_more advmod_scale_readily ' +o,4558,'We rst recast the problem of estimating the IBM models in a discriminative framework , which leads to an initial increase in word-alignment accuracy ',Brown,'nsubj_rst_We xcomp_rst_recast det_problem_the dobj_recast_problem prep_problem_of pcomp_of_estimating det__the nn__IBM nn__models dobj_estimating_ prep__in det_framework_a amod_framework_discriminative pobj_in_framework nsubj_leads_which rcmod_framework_leads prep_leads_to det_increase_an amod_increase_initial pobj_to_increase prep_increase_in amod_accuracy_word-alignment pobj_in_accuracy ' +o,4559,'In practice , when training the parameters of an SMT system , for example using the discriminative methods of , the cost for skips of this kind is typically set to a very high value ',Och,'prep_set_In pobj_In_practice advmod_training_when dep_set_training det_parameters_the dobj_training_parameters prep_parameters_of det_system_an nn_system_SMT pobj_of_system prep_parameters_for pobj_for_example partmod_example_using det_methods_the amod_methods_discriminative dobj_using_methods prep_methods_of pobj_of_ det_cost_the nsubjpass_set_cost prep_cost_for pobj_for_skips prep_skips_of det_kind_this pobj_of_kind auxpass_set_is advmod_set_typically prep_set_to det_value_a advmod_high_very amod_value_high pobj_to_value ' +o,4560,'Section 4 compares our results to Itindle \'s ones ',Hindle,'nsubj_compares_Section num_Section_4 poss_results_our dobj_compares_results prep_compares_to poss_ones_Itindle possessive_Itindle_\'s pobj_to_ones ' +o,4561,'Without removing them , extracted rules can not be triggered until when completely the same strings appear in a text4 6 Performance Evaluation We measured the performance of our robust parsing algorithm by measuring coverage and degree of overgeneration for the Wall Street Journal in the Penn Treebank ',Marcus,'prep_triggered_Without pcomp_Without_removing dobj_removing_them amod_rules_extracted nsubjpass_triggered_rules aux_triggered_can neg_triggered_not auxpass_triggered_be mark_measured_until advmod_appear_when advmod_appear_completely det_strings_the amod_strings_same nsubj_appear_strings dep_measured_appear prep_appear_in det_Evaluation_a amod_Evaluation_text4 num_Evaluation_6 nn_Evaluation_Performance pobj_in_Evaluation nsubj_measured_We advcl_triggered_measured det_performance_the dobj_measured_performance prep_performance_of poss_algorithm_our amod_algorithm_robust nn_algorithm_parsing pobj_of_algorithm prep_measured_by pcomp_by_measuring dobj_measuring_coverage cc_coverage_and conj_coverage_degree prep_coverage_of pobj_of_overgeneration prep_measuring_for det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_for_Journal prep_Journal_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +o,4562,'4 Features Features used in our experiments are inspired by previous work on corpus-based approaches for discourse analysis ',Marcu,'num_Features_4 nn_Features_Features nsubjpass_inspired_Features partmod_Features_used prep_used_in poss_experiments_our pobj_in_experiments auxpass_inspired_are prep_inspired_by amod_work_previous pobj_by_work prep_inspired_on amod_approaches_corpus-based pobj_on_approaches prep_approaches_for nn_analysis_discourse pobj_for_analysis ' +o,4563,'Snow etal use known hypernym\\/hyponym pairs to generate training data for a machine-learning system , which then learns many lexico-syntactic patterns ',Snow,'nn_use_Snow amod_use_etal amod_use_ dep_known_use nn_pairs_hypernym\\/hyponym dobj_known_pairs aux_generate_to xcomp_known_generate nn_data_training dobj_generate_data prep_generate_for det_system_a amod_system_machine-learning pobj_for_system nsubj_learns_which advmod_learns_then rcmod_system_learns amod_patterns_many amod_patterns_lexico-syntactic dobj_learns_patterns ' +o,4564,'CIT -RRB- ',Koo,'nn_-RRB-_CIT ' +o,4565,'We run the decoder with its default settings -LRB- maximum phrase length 7 -RRB- and then use Koehn \'s implementation of minimum error rate training to tune the feature weights on the de2 The full name of HTRDP is National High Technology Research and Development Program of China , also named as 863 Program ',Och,'nsubj_run_We det_decoder_the dobj_run_decoder prep_run_with poss_settings_its nn_settings_default pobj_with_settings amod_length_maximum nn_length_phrase appos_settings_length num_length_7 cc_run_and advmod_use_then conj_run_use poss_implementation_Koehn possessive_Koehn_\'s dobj_use_implementation prep_implementation_of amod__minimum nn__error nn__rate nn__training pobj_of_ prep__to pobj_to_tune det_weights_the nn_weights_feature dobj_run_weights prep_run_on det_de2_the pobj_on_de2 det_name_The amod_name_full nsubj_Research_name prep_name_of pobj_of_HTRDP cop_Research_is nn_Research_National nn_Research_High nn_Research_Technology dep_run_Research cc_Research_and nn_Program_Development conj_Research_Program prep_Research_of pobj_of_China advmod_named_also partmod_Research_named prep_named_as num_Program_863 pobj_as_Program ' +p,4566,'Furthermore , use of the self-training techniques described in raise this to 878 \% -LRB- an error reduction of 28 \% -RRB- again without any use of labeled Brown data ',McClosky,'advmod_raise_Furthermore nsubj_raise_use prep_use_of det_techniques_the amod_techniques_self-training pobj_of_techniques partmod_techniques_described prep_described_in pobj_in_ dobj_raise_this prep_raise_to num_\%_878 pobj_to_\% det_reduction_an nn_reduction_error dep_\%_reduction prep_reduction_of num_\%_28 pobj_of_\% dep_without_again prep_raise_without det_use_any pobj_without_use prep_use_of nn_data_labeled nn_data_Brown pobj_of_data ' +o,4567,' represent chunking as tagging problem and the CoNLL2000 shared task is now the standard evaluation task for chunking English ',Ramshaw,'nsubj_represent_ xcomp_represent_chunking mark_shared_as amod_problem_tagging nsubj_shared_problem cc_problem_and det_CoNLL2000_the conj_problem_CoNLL2000 advcl_chunking_shared nn__task nsubj_task_ cop_task_is advmod_task_now det_task_the amod_task_standard nn_task_evaluation ccomp_shared_task prep_task_for pcomp_for_chunking dobj_chunking_English ' +o,4568,'h1 -LRB- eI1 , fJ1 -RRB- = log Kproductdisplay k = 1 N -LRB- z -RRB- -LRB- T -LRB- z -RRB- , Tk -RRB- N -LRB- T -LRB- z -RRB- -RRB- h2 -LRB- eI1 , fJ1 -RRB- = log Kproductdisplay k = 1 N -LRB- z -RRB- -LRB- T -LRB- z -RRB- , Tk -RRB- N -LRB- S -LRB- z -RRB- -RRB- h3 -LRB- eI1 , fJ1 -RRB- = log Kproductdisplay k = 1 lex -LRB- T -LRB- z -RRB- S -LRB- z -RRB- -RRB- -LRB- T -LRB- z -RRB- , Tk -RRB- h4 -LRB- eI1 , fJ1 -RRB- = log Kproductdisplay k = 1 lex -LRB- S -LRB- z -RRB- T -LRB- z -RRB- -RRB- -LRB- T -LRB- z -RRB- , Tk -RRB- h5 -LRB- eI1 , fJ1 -RRB- = K h6 -LRB- eI1 , fJ1 -RRB- = log Iproductdisplay i = 1 p -LRB- ei ei2 , ei1 -RRB- h7 -LRB- eI1 , fJ1 -RRB- = I 4When computing lexical weighting features , we take only terminals into account ',Koehn,'amod_=_h1 appos_=_eI1 dep_eI1_fJ1 nsubj_log_= nn_k_Kproductdisplay nsubj_=_k ccomp_log_= iobj_=_1 dobj_=_N dep_-LRB-_z dep_N_T dep_-LRB-_z appos_T_Tk nn_h2_N dep_N_T dep_-LRB-_z dobj_=_h2 appos_h2_eI1 dep_eI1_fJ1 nn_k_= nn_k_log nn_k_Kproductdisplay dobj_=_k dep_1_= rcmod_k_1 dobj_=_N dep_-LRB-_z dep_N_T dep_-LRB-_z appos_T_Tk nn_h3_N dep_N_S dep_-LRB-_z dobj_=_h3 appos_h3_eI1 dep_eI1_fJ1 nn_k_= nn_k_log nn_k_Kproductdisplay dobj_=_k dep_k_= num_lex_1 dobj_=_lex dep_lex_T dep_-LRB-_z dep_T_S dep_-LRB-_z dep_lex_T dep_-LRB-_z appos_T_Tk amod_=_h4 appos_=_eI1 dep_eI1_fJ1 dobj_=_= nn_k_log nn_k_Kproductdisplay dobj_=_k dep_k_= num_lex_1 dobj_=_lex nn_T_S nn_T_z dep_lex_T dep_-LRB-_z dep_lex_T dep_-LRB-_z appos_T_Tk amod_=_h5 appos_=_eI1 dep_eI1_fJ1 dobj_=_= nn_h6_K dobj_=_h6 appos_h6_eI1 dep_eI1_fJ1 dep_=_= ccomp_=_log nn_i_Iproductdisplay nsubj_p_i dep_p_= num_p_1 ccomp_log_p nn_ei2_ei appos_h7_ei2 dep_ei2_ei1 dep_log_h7 appos_h7_eI1 dep_eI1_fJ1 ccomp_take_= dep_computing_I amod_computing_4When iobj_=_computing amod__lexical nn__weighting nn__features dobj_=_ nsubj_take_we dep_log_take amod_terminals_only dobj_take_terminals prep_take_into pobj_into_account ' +o,4569,'Our results agree , at least at the level of morphology , with ',Marcus,'poss_results_Our nsubj_agree_results advmod_agree_at dep_at_least dep_at_at det_level_the pobj_at_level prep_level_of pobj_of_morphology prep_agree_with ' +o,4570,'Methods that use bigrams or trigrams cluster words considering as a word \'s context the one or two immediately adjacent words and employ as clustering criteria the minimal loss of average 836 nmtual information and the perplexity improvement respectively ',Brown,'complm__that nn_bigrams_use nsubj__bigrams ccomp_Methods_ cc__or conj__trigrams amod_words_ nn_words_cluster dobj__words partmod_words_considering prep_considering_as det_word_a poss_context_word possessive_word_\'s pobj_as_context det_one_the dep_context_one dep_one_or num_words_two advmod_adjacent_immediately amod_words_adjacent dep_one_words cc_Methods_and conj_Methods_employ prep_employ_as amod_criteria_clustering pobj_as_criteria det_loss_the amod_loss_minimal nsubj_Methods_loss prep_loss_of nn_information_average num_information_836 nn_information_nmtual pobj_of_information cc_loss_and det_improvement_the amod_improvement_perplexity conj_loss_improvement advmod_improvement_respectively ' +p,4571,'When efficient techniques have been proposed , they have been mostly evaluated on safe pairs of languages where the notion of word is rather clear ',Brown,'advmod_proposed_When amod_techniques_efficient nsubjpass_proposed_techniques aux_proposed_have auxpass_proposed_been dep_evaluated_proposed dep_proposed_ nsubjpass_evaluated_they aux_evaluated_have auxpass_evaluated_been advmod_evaluated_mostly prep_evaluated_on amod_pairs_safe pobj_on_pairs prep_pairs_of pobj_of_languages advmod_clear_where det_notion_the nsubj_clear_notion prep_notion_of pobj_of_word cop_clear_is advmod_clear_rather rcmod_languages_clear ' +o,4572,'We also experimented with a method suggested by Brent which applies the binomial test on frame frequency data ',Dunning,'nsubj_experimented_We advmod_experimented_also prep_experimented_with det_method_a pobj_with_method partmod_method_suggested prep_suggested_by nn__Brent pobj_by_ nsubj_applies_which rcmod__applies det_test_the amod_test_binomial dobj_applies_test prep_test_on nn_data_frame nn_data_frequency pobj_on_data ' +o,4573,'85 Recently some alignment evaluation metrics have been proposed which are more informative when the alignments are used to extract translation units ',Fraser,'num_metrics_85 advmod_alignment_Recently dep_alignment_some amod_metrics_alignment nn_metrics_evaluation nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been nsubj_informative_which cop_informative_are advmod_informative_more ccomp_proposed_informative advmod_used_when det_alignments_the nsubjpass_used_alignments auxpass_used_are ccomp_informative_used aux_extract_to xcomp_used_extract nn_units_translation dobj_extract_units ' +o,4574,'However , the achieved accuracy was not better than that of related work based on CRFs ',Finkel,'advmod_based_However det_accuracy_the amod_accuracy_achieved nsubjpass_based_accuracy auxpass_based_was neg_based_not advmod_based_better dep_better_than pobj_than_that prep_that_of amod__related nn__work pobj_of_ prep_based_on pobj_on_CRFs ' +o,4575,'Syntax-based MT approaches began with , who introduced the Inversion Transduction Grammars ',Wu,'amod_approaches_Syntax-based nn_approaches_MT nsubj_began_approaches dep_introduced_began prep_began_with pobj_with_ nsubj_introduced_who det_Grammars_the nn_Grammars_Inversion nn_Grammars_Transduction dobj_introduced_Grammars ' +o,4576,'For instance , the to-PP frame is poorly \' represented in the syntactically annotated version of the Penn Treebank ',Marcus,'prep_is_For pobj_For_instance det_frame_the amod_frame_to-PP nsubj_is_frame advmod_is_poorly ccomp_is_represented prep_represented_in det_version_the advmod_annotated_syntactically amod_version_annotated pobj_in_version prep_version_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,4577,' , or -LRB- S+T - -RRB- , where no labeled target domain data is available , eg ',Blitzer,'cc__or nn_-_S+T advmod_available_where det_data_no amod_data_labeled nn_data_target nn_data_domain nsubj_available_data cop_available_is dep_eg_available dep__eg ' +o,4578,'Using a vector-based topic identification process , these keywords are used to determine a set of likely values -LRB- including null -RRB- for that attribute ',Chu-Carroll,'dep_used_Using det__a amod__vector-based nn__topic nn__identification nn__process dobj_Using_ det_keywords_these nsubjpass_used_keywords auxpass_used_are aux_determine_to xcomp_used_determine det_set_a dobj_determine_set prep_set_of amod_values_likely pobj_of_values dep_used_including pobj_including_null prep_used_for nsubj_attribute_that pcomp_for_attribute ' +o,4579,'For example , suggested two different methods : using only the alignment with the maximum probability , the so-called Viterbi alignment , or generating a set of alignments by starting from the Viterbi alignment and making changes , which keep the alignment probability high ',Brown,'prep_suggested_For pobj_For_example nsubj_suggested_ num_methods_two amod_methods_different dobj_suggested_methods parataxis_suggested_using advmod_alignment_only det_alignment_the dobj_using_alignment prep_using_with det_probability_the amod_probability_maximum pobj_with_probability det_alignment_the amod_alignment_so-called nn_alignment_Viterbi appos_probability_alignment cc_using_or conj_using_generating det_set_a dobj_generating_set prep_set_of pobj_of_alignments prep_generating_by pcomp_by_starting prep_starting_from det_alignment_the nn_alignment_Viterbi pobj_from_alignment cc_starting_and conj_starting_making dobj_making_changes nsubj_keep_which rcmod_changes_keep det_probability_the nn_probability_alignment nsubj_high_probability xcomp_keep_high ' +o,4580,'Success is indicated by the proportion of the original sentence regenerated , as measured by any string comparison method : in our case , using the BLEU metric ',Papineni,'nsubjpass_indicated_Success auxpass_indicated_is prep_indicated_by det_proportion_the pobj_by_proportion prep_proportion_of det_sentence_the amod_sentence_original pobj_of_sentence partmod_sentence_regenerated mark_measured_as advcl_regenerated_measured prep_measured_by det_method_any nn_method_string nn_method_comparison pobj_by_method prep_regenerated_in poss_case_our pobj_in_case xcomp_regenerated_using det_metric_the amod_metric_BLEU dobj_using_metric ' +o,4581,' 8910 8914 8912 kitchen sink 8926 8955 8940 parser 8 , the only one that we were able to train and test under exactly the same experimental conditions -LRB- including the use of POS tags from the tagger of -RRB- ',Ratnaparkhi,'num__8910 number_8912_8914 num__8912 nn__kitchen nn__sink num__8926 number_8940_8955 dep_parser_8940 dep__parser dobj__ nsubj__8 det_one_the amod_one_only appos_8_one complm_able_that nsubj_able_we cop_able_were dep_one_able aux_train_to xcomp_able_train cc_train_and conj_train_test prep_test_under advmod_conditions_exactly det_conditions_the amod_conditions_same amod_conditions_experimental pobj_under_conditions dep_conditions_including det_use_the pobj_including_use prep_use_of dep_tags_POS pobj_of_tags prep_tags_from det_tagger_the pobj_from_tagger prep_tagger_of pobj_of_ ' +o,4582,'a larger number of labeled documents , its performance on this corpus is comparable to that of Support Vector Machines and Maximum Entropy models ',Pang,'det_number_a amod_number_larger nsubj_comparable_number prep_number_of amod_documents_labeled pobj_of_documents poss_performance_its nsubj_comparable_performance prep_performance_on det_corpus_this pobj_on_corpus cop_comparable_is prep_comparable_to pobj_to_that prep_that_of nn_Machines_Support nn_Machines_Vector pobj_of_Machines cc_Machines_and nn_models_Maximum nn_models_Entropy conj_Machines_models ' +o,4583,'In the original work (Brown et al. , 1993) the posterior probability p(eI1|fJ1 ) is decomposed following a noisy-channel approach, but current stateof-the-art systems model the translation probability directly using a log-linear model(Och and Ney, 2002): p(eI1|fJ1 ) = exp parenleftBigsummationtextM m=1 mhm(e I1,fJ1 ) parenrightBig summationdisplay ?eI1 exp parenleftBigsummationtextM m=1 mhm(?eI1,fJ1 ) parenrightBig, (2) with hm different models, m scaling factors and the denominator a normalization factor that can be ignored in the maximization process.',Brown,'' +o,4584,'They propose a two-level hierarchy , with 5 classes at the first level and 30 classes at the second one ; other researchers have used their class scheme and data set ',Nakov,'nsubj_propose_They det_hierarchy_a amod_hierarchy_two-level dobj_propose_hierarchy prep_propose_with num_classes_5 pobj_with_classes prep_classes_at det_level_the amod_level_first pobj_at_level cc_classes_and num_classes_30 conj_classes_classes prep_classes_at det_one_the amod_one_second pobj_at_one amod_researchers_other nsubj_used_researchers advmod_used_ aux_used_have parataxis_propose_used poss_scheme_their nn_scheme_class dobj_used_scheme cc_propose_and nsubj_set_data conj_propose_set ' +p,4585,'Note that our result on Dataset A is as strong as that obtained by via their subjectivity summarization algorithm , which retains only the subjective portions of a document ',Pang,'complm_strong_that poss_result_our nsubj_strong_result prep_result_on nn_A_Dataset pobj_on_A cop_strong_is advmod_strong_as ccomp_Note_strong mark_obtained_as nsubj_obtained_that advcl_strong_obtained prep_obtained_by pobj_by_ prep__via poss_algorithm_their amod_algorithm_subjectivity nn_algorithm_summarization pobj_via_algorithm nsubj_retains_which rcmod_algorithm_retains advmod_portions_only det_portions_the amod_portions_subjective dobj_retains_portions prep_portions_of det_document_a pobj_of_document ' +o,4586,'1 Introduction The last years have seen a boost of work devoted to the development of machine learning based coreference resolution systems ',Pedersen,'num_Introduction_1 det_years_The amod_years_last nsubj_seen_years aux_seen_have dep_Introduction_seen det_boost_a dobj_seen_boost prep_boost_of pobj_of_work partmod_work_devoted prep_devoted_to det_development_the pobj_to_development prep_development_of pobj_of_machine xcomp_devoted_learning prep_learning_based amod_systems_coreference nn_systems_resolution dep_based_systems partmod_systems_ ' +o,4587,'In natural language processing , label propagation has been used for document classification , word sense disambiguation , and sentiment categorization ',Alexandrescu,'prep_used_In amod_processing_natural nn_processing_language pobj_In_processing nn_propagation_label nsubjpass_used_propagation aux_used_has auxpass_used_been prep_used_for nn__document nn__classification pobj_for_ nn__word nn__sense nn__disambiguation conj__ cc__and nn__sentiment nn__categorization conj__ ' +o,4588,'This conclusion is supported by the fact that true IMT is not , to our knowledge , used in most modern translator \'s support environments , eg ',Brown,'det_conclusion_This nsubj_supported_conclusion aux_supported_is prep_supported_by det_fact_the pobj_by_fact complm_is_that amod_IMT_true nsubj_is_IMT ccomp_supported_is neg_is_not prep_supported_to poss_knowledge_our pobj_to_knowledge dep_supported_used prep_used_in amod_translator_most amod_translator_modern poss_environments_translator possessive_translator_\'s nn_environments_support pobj_in_environments nn__eg dobj_supported_ ' +o,4589,'We trained the parser on the Penn Treebank ',Marcus,'nsubj_trained_We det_parser_the dobj_trained_parser prep_parser_on det_Treebank_the nn_Treebank_Penn pobj_on_Treebank ' +o,4590,'We obtain aligned parallel sentences and the phrase table after the training of Moses , which includes running GIZA + + , grow-diagonal-final symmetrization and phrase extraction ',Och,'nsubj_obtain_We ccomp_obtain_aligned amod_sentences_parallel dobj_aligned_sentences cc_sentences_and det_table_the nn_table_phrase conj_sentences_table prep_aligned_after det_training_the pobj_after_training prep_training_of pobj_of_Moses nsubj_includes_which rcmod_training_includes xcomp_includes_running nn__GIZA nn__+ nn__+ dobj_running_ amod_symmetrization_grow-diagonal-final conj__symmetrization cc__and nn__phrase nn__extraction conj__ ' +o,4591,'The PT grammar 2 was extracted from the Penn Treebank ',Marcus,'det_grammar_The nn_grammar_PT nsubjpass_extracted_grammar num_grammar_2 auxpass_extracted_was prep_extracted_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +p,4592,'Several classification models can be adopted here , however , we choose the averaged perceptron algorithm because of its simplicity and high accuracy ',Collins,'amod_models_Several nn_models_classification nsubjpass_adopted_models aux_adopted_can auxpass_adopted_be advmod_choose_here advmod_choose_however nsubj_choose_we dep_adopted_choose det__the amod__averaged nn__perceptron nn__algorithm dobj_choose_ dep_of_because prep__of poss_simplicity_its pobj_of_simplicity cc_simplicity_and amod_accuracy_high conj_simplicity_accuracy ' +o,4593,'In order to resolve all Chinese NLDs represented in the CTB, we modify and substantially extend the (Cahill et al. , 2004) (henceforth C04 for short) algorithm as follows: Given the set of subcat frames s for the word w, and a set of paths p for the trace t, the algorithm traverses the f-structure f to: predict a dislocated argument t at a sub-fstructure h by comparing the local PRED:w to ws subcat frames s t can be inserted at h if h together with t is complete and coherent relative to subcat frame s traverse f starting from t along the path p link t to its antecedent a if ps ending GF a exists in a sub-f-structure within f; or leave t without an antecedent if an empty path for t exists In the modified algorithm, we condition the probability of NLD path p (including the empty path without an antecedent) on the GF associated of the trace t rather than the antecedent a as in C04.',Cahill,'' +o,4594,'Finally , following and Johnson -LRB- 2007 -RRB- we can instead insist that at most one HMM state can be mapped to any part-of-speech tag ',Haghighi,'advmod_mapped_Finally amod__following nsubjpass_mapped_ cc__and conj__Johnson appos_Johnson_2007 nsubj_insist_we aux_insist_can advmod_insist_instead rcmod_Johnson_insist det_state_that quantmod_one_at dep_at_most num_state_one nn_state_HMM dobj_insist_state aux_mapped_can auxpass_mapped_be prep_mapped_to det_tag_any amod_tag_part-of-speech pobj_to_tag ' +o,4595,'We measure semantic similarity using the shortest path length in WordNet as implemented in the WordNet Similarity package ',Pedersen,'nsubj_measure_We amod_similarity_semantic dobj_measure_similarity partmod_similarity_using det_length_the amod_length_shortest nn_length_path dobj_using_length prep_length_in nn__WordNet pobj_in_ mark_implemented_as advcl_using_implemented prep_implemented_in det_package_the nn_package_WordNet nn_package_Similarity pobj_in_package ' +o,4596,'After that , several million instances of people , locations , and other facts were added ',Fleischman,'prep_added_After pobj_After_that quantmod_million_several num_instances_million nsubjpass_added_instances prep_instances_of pobj_of_people conj_people_locations cc_people_and amod_facts_other conj_people_facts auxpass_added_were ' +o,4597,'While movie reviews have been the most studied domain , sentiment analysis has extended to a number of new domains , ranging from stock message boards to congressional floor debates ',Thomas,'mark_been_While nn_reviews_movie nsubj_been_reviews aux_been_have advcl_extended_been det_domain_the advmod_studied_most amod_domain_studied dobj_been_domain nn_analysis_sentiment nsubj_extended_analysis aux_extended_has prep_extended_to det_number_a pobj_to_number prep_number_of amod_domains_new pobj_of_domains xcomp_extended_ranging prep_ranging_from nn_boards_stock nn_boards_message pobj_from_boards prep_ranging_to amod_debates_congressional nn_debates_floor pobj_to_debates ' +o,4598,'c2009 Association for Computational Linguistics Structural Correspondence Learning for Parse Disambiguation Barbara Plank Alfa-informatica University of Groningen , The Netherlands bplank @ rugnl Abstract The paper presents an application of Structural Correspondence Learning -LRB- SCL -RRB- for domain adaptation of a stochastic attribute-value grammar -LRB- SAVG -RRB- ',Blitzer,'nn_Association_c2009 prep_Association_for nn_Learning_Computational nn_Learning_Linguistics nn_Learning_Structural nn_Learning_Correspondence pobj_for_Learning prep_Learning_for nn_University_Parse nn_University_Disambiguation nn_University_Barbara nn_University_Plank nn_University_Alfa-informatica pobj_for_University prep_University_of pobj_of_Groningen det_bplank_The nn_bplank_Netherlands appos_Groningen_bplank dep_Abstract_@ amod_Abstract_rugnl dep_presents_Abstract det_paper_The nsubj_presents_paper dep_Association_presents det_application_an dobj_presents_application prep_application_of pcomp_of_Structural nn__Correspondence nn__Learning abbrev__SCL dobj_Structural_ prep__for nn_adaptation_domain pobj_for_adaptation prep_adaptation_of det_grammar_a amod_grammar_stochastic amod_grammar_attribute-value pobj_of_grammar abbrev_grammar_SAVG ' +o,4599,'We utilize a maximum entropy -LRB- ME -RRB- model to design the basic classifier used in active learning for WSD ',Berger,'nsubj_utilize_We det__a amod__maximum nn__entropy nn__ME nn__model dobj_utilize_ aux_design_to xcomp_utilize_design det_classifier_the amod_classifier_basic dobj_design_classifier partmod_classifier_used prep_used_in amod_learning_active pobj_in_learning prep_learning_for pobj_for_WSD ' +o,4600,'Some work identifies inflammatory texts -LRB- eg , -RRB- or classifies reviews as positive or negative -LRB- -RRB- ',Turney,'det_work_Some nsubj_identifies_work amod_texts_inflammatory dobj_identifies_texts appos_texts_eg dep_eg_ cc_identifies_or conj_identifies_classifies dobj_classifies_reviews advmod_positive_as amod_reviews_positive cc_positive_or conj_positive_negative appos_reviews_ ' +o,4601,'Our statistical tagging model is modified from the standard bigrams using Viterbi search plus onthe-fly extra computing of lexical probabilities for unknown morphemes ',Cutting,'poss_model_Our amod_model_statistical amod_model_tagging nsubjpass_modified_model auxpass_modified_is prep_modified_from det__the amod__standard nn__bigrams pobj_from_ partmod__using nn_search_Viterbi dobj_using_search prep_search_plus amod_computing_onthe-fly amod_computing_extra dep_plus_computing prep_computing_of amod_probabilities_lexical pobj_of_probabilities prep_probabilities_for amod_morphemes_unknown pobj_for_morphemes ' +o,4602,'In , an undirected graphical model is used for parse reranking ',Koo,'prep_used_In pobj_In_ det_model_an amod_model_undirected amod_model_graphical nsubjpass_used_model auxpass_used_is prep_used_for amod_reranking_parse pobj_for_reranking ' +o,4603,'4 Experiments Our experiments involve data from two treebanks : the Wall Street Journal Penn treebank and the Chinese treebank ',Marcus,'num_Experiments_4 poss_experiments_Our nsubj_involve_experiments dep_Experiments_involve dobj_involve_data prep_involve_from num_treebanks_two pobj_from_treebanks det__the nn__Wall nn__Street nn__Journal nn__Penn nn__treebank dep_treebanks_ cc__and det__the amod__Chinese nn__treebank conj__ ' +p,4604,'In a later study , present a loglinear combination of the HMM and IBM Model 4 that produces better alignments than either of those ',Och,'prep_present_In det_study_a amod_study_later pobj_In_study nsubj_present_ det_combination_a nn_combination_loglinear dobj_present_combination prep_combination_of det_HMM_the pobj_of_HMM cc_HMM_and nn_Model_IBM conj_HMM_Model num_Model_4 nsubj_produces_that rcmod_HMM_produces amod_alignments_better dobj_produces_alignments prep_produces_than pobj_than_either prep_either_of pobj_of_those ' +p,4605,'354 supervised induction techniques that have been successfully developed for English -LRB- eg , Schutze , Clark -RRB- , including the recentlyproposed prototype-driven approach and Bayesian approach ',Haghighi,'nsubj_supervised_354 nn_techniques_induction dobj_supervised_techniques nsubjpass_developed_that aux_developed_have auxpass_developed_been advmod_developed_successfully rcmod_techniques_developed prep_developed_for pobj_for_English appos_English_eg nn__Schutze dep_eg_ nn__Clark dep_eg_ prep_English_including det__the amod__recentlyproposed amod__prototype-driven nn__approach pobj_including_ cc__and nn__Bayesian nn__approach conj__ ' +o,4606,'12 Decoding in Statistical Machine Translation and have discussed the first two of the three problems in statistical machine translation ',Brown,'num_Decoding_12 nsubj_discussed_Decoding prep_Decoding_in nn__Statistical nn__Machine nn__Translation pobj_in_ cc__and conj__ aux_discussed_have det_two_the amod_two_first dobj_discussed_two prep_two_of det_problems_the num_problems_three pobj_of_problems prep_discussed_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,4607,'eBonsai first performs syntactic analysis of a sentence using a parser based on GLR algorithm -LRB- MSLR parser -RRB- , and provides candidates of its syntactic structure ',Marcus,'nsubj_performs_eBonsai advmod_performs_first amod_analysis_syntactic dobj_performs_analysis prep_analysis_of det_sentence_a pobj_of_sentence partmod_sentence_using det_parser_a dobj_using_parser prep_using_based dep_based_on nn__GLR nn__algorithm nn_parser_MSLR abbrev__parser pobj_on_ cc_performs_and conj_performs_provides dobj_provides_candidates prep_candidates_of poss_structure_its amod_structure_syntactic pobj_of_structure ' +o,4608,'It reconfirms that only allowing sibling nodes reordering as done in SCFG may be inadequate for translational equivalence modeling 4 3 -RRB- All the three models on the FBIS corpus show much lower performance than that on the other two corpora ',Galley,'nsubj_reconfirms_It complm_inadequate_that advmod_allowing_only csubj_inadequate_allowing amod_reordering_sibling amod_reordering_nodes dobj_allowing_reordering advmod_done_as amod_reordering_done prep_allowing_in pobj_in_SCFG aux_inadequate_may cop_inadequate_be ccomp_reconfirms_inadequate prep_inadequate_for amod__translational nn__equivalence nn__modeling pobj_for_ num__4 dep_show_3 dep_models_All det_models_the num_models_three nsubj_show_models prep_models_on det_corpus_the nn_corpus_FBIS pobj_on_corpus dep_reconfirms_show advmod_lower_much amod_performance_lower dobj_show_performance prep_show_than pobj_than_that prep_that_on det_corpora_the amod_corpora_other num_corpora_two pobj_on_corpora ' +o,4609,'have been used in statistical machine translation , terminology research and translation aids , bilingual lexicography , word-sense disambiguation and information retrieval in a multilingual environment ',Smadja,'aux_used_have auxpass_used_been prep_used_in amod_research_statistical nn_research_machine nn_research_translation nn_research_ dep_research_terminology pobj_in_research cc_research_and nn_aids_translation conj_research_aids amod__ amod__bilingual nn__lexicography nsubjpass_used_ amod__word-sense nn__disambiguation appos__ cc__and nn_retrieval_information conj__retrieval prep__in det__a amod__multilingual nn__environment pobj_in_ ' +o,4610,'44 Text chunking Next , a rule-based text chunker is applied on the tagged sentences to further identify phrasal units , such as base noun phrases NP and verbal units VB ',Ramshaw,'num_chunking_44 nn_chunking_Text dep_Next_chunking advmod_applied_Next det__a amod__rule-based nn__text nn__chunker nsubjpass_applied_ auxpass_applied_is prep_applied_on det_sentences_the amod_sentences_tagged pobj_on_sentences aux_identify_to advmod_identify_further xcomp_applied_identify amod_units_phrasal dobj_identify_units dep_as_such prep_units_as amod_NP_base nn_NP_noun nn_NP_phrases pobj_as_NP cc_NP_and amod_VB_verbal nn_VB_units conj_NP_VB ' +o,4611,'The unlabeled data for English we use is the union of the Penn Treebank tagged WSJ data and the BLLIP corpus5 For the rest of the languages we use only the text of George Orwells novel 1984 , which is provided in morphologically disambiguated form as part of MultextEast -LRB- but we dont use the annotations -RRB- ',Marcus,'det_data_The amod_data_unlabeled prep_data_for amod_use_English dep_use_we nsubj_union_use cop_union_is det_union_the pcomp_for_union prep_union_of det__the nn__Penn nn__Treebank nn__tagged nn__WSJ nn__data pobj_of_ cc_union_and det_corpus5_the nn_corpus5_BLLIP conj_union_corpus5 dep_corpus5_For det_rest_the pobj_For_rest prep_data_of det_languages_the pobj_of_languages nsubj_use_we rcmod_data_use advmod_text_only det_text_the dobj_use_text prep_text_of nn_Orwells_George pobj_of_Orwells amod_Orwells_novel tmod_novel_1984 nsubjpass_provided_which auxpass_provided_is rcmod_Orwells_provided prep_provided_in advmod_disambiguated_morphologically amod_form_disambiguated pobj_in_form prep_provided_as pobj_as_part prep_part_of pobj_of_MultextEast dep_dont_but dep_but_we dep_data_dont rcmod_dont_use det_annotations_the dobj_use_annotations ' +o,4612,'The results are consistent with the idea in ',Yarowsky,'det_results_The nsubj_consistent_results cop_consistent_are prep_consistent_with det_idea_the pobj_with_idea prep_consistent_in ' +o,4613,'There have been many studies of zero-pronoun identification ',Iida,'expl_been_There aux_been_have amod_studies_many dobj_been_studies prep_studies_of amod__zero-pronoun nn__identification nn__ pobj_of_ ' +p,4614,'For example , shows that training a learning algorithm on the weighted union of different data sets -LRB- which is basically what we did -RRB- performs almost as well as more involved domain adaptation approaches ','Daume III','prep_shows_For pobj_For_example nsubj_shows_ complm_performs_that csubj_performs_training det_algorithm_a nn_algorithm_learning dobj_training_algorithm prep_training_on det_union_the amod_union_weighted pobj_on_union prep_union_of amod_sets_different nn_sets_data pobj_of_sets nsubj_is_which dep_union_is advmod_is_basically dobj_did_what nsubj_did_we ccomp_is_did ccomp_shows_performs amod_approaches_almost advmod_almost_as cc_almost_well dep_well_as advmod_involved_more conj_almost_involved nn_approaches_domain nn_approaches_adaptation dobj_performs_approaches ' +p,4615,'Similar models have been successfully applied in the past to other tasks including parsing , chunking , and machine translation ',Collins,'amod_models_Similar nsubjpass_applied_models aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_in det_past_the pobj_in_past prep_applied_to amod_tasks_other pobj_to_tasks prep_tasks_including amod__parsing pobj_including_ amod__chunking conj__ cc__and nn__machine nn__translation conj__ ' +o,4616,'recent advances in parsing technology are due to the explicit stochastic modeling of dependency information ',Collins,'amod_advances_recent nsubj_due_advances prep_advances_in amod_technology_parsing pobj_in_technology cop_due_are prep_due_to det_modeling_the amod_modeling_explicit amod_modeling_stochastic pobj_to_modeling prep_modeling_of amod_information_dependency pobj_of_information ' +o,4617,'We use a bootstrap approach in which we first extract 1-to-n word alignments using an existing word aligner , and then estimate the confidence of those alignments to decide whether or not the n words have to be grouped ; if so , this group is conwould thus be completely driven by the bilingual alignment process -LRB- see also for related considerations -RRB- ',Wu,'nsubj_use_We det_approach_a nn_approach_bootstrap dobj_use_approach rel_extract_in pobj_in_which nsubj_extract_we advmod_extract_first rcmod_approach_extract amod_alignments_1-to-n nn_alignments_word dobj_extract_alignments xcomp_extract_using det_aligner_an amod_aligner_existing nn_aligner_word dobj_using_aligner cc_extract_and nsubj_estimate_then conj_extract_estimate det_confidence_the nsubj_decide_confidence prep_confidence_of det_alignments_those pobj_of_alignments aux_decide_to xcomp_estimate_decide complm_have_whether cc_have_or conj_have_not det_words_the nn_words_n nsubj_have_words ccomp_decide_have aux_grouped_to auxpass_grouped_be xcomp_have_grouped mark_conwould_if advmod_conwould_so det_group_this nsubjpass_conwould_group auxpass_conwould_is parataxis_extract_conwould advmod_conwould_thus auxpass_driven_be advmod_driven_completely ccomp_conwould_driven prep_driven_by det_process_the amod_process_bilingual nn_process_alignment pobj_by_process parataxis_use_see advmod_see_also dobj_see_ prep__for amod_considerations_related pobj_for_considerations ' +o,4618,'For example , non-local features such as same phrases in a document do not have different entity classes were shown to be useful in named entity recognition ',Finkel,'prep_have_For pobj_For_example amod_features_non-local nsubj_have_features dep_as_such prep_features_as amod_phrases_same pobj_as_phrases prep_phrases_in det_document_a pobj_in_document aux_have_do neg_have_not amod_classes_different nn_classes_entity nsubjpass_shown_classes auxpass_shown_were ccomp_have_shown aux_useful_to cop_useful_be xcomp_shown_useful prep_useful_in amod_recognition_named nn_recognition_entity pobj_in_recognition ' +o,4619,'IBM Model 4 parameters are then estimated over this partial search space as an approximation to EM ',Brown,'nn_Model_IBM nsubjpass_estimated_Model num_parameters_4 dep_Model_parameters auxpass_estimated_are advmod_estimated_then prep_estimated_over det_space_this amod_space_partial nn_space_search pobj_over_space prep_space_as det_approximation_an pobj_as_approximation prep_estimated_to pobj_to_EM ' +o,4620,'Early experiments with syntactically-informed phrases , and syntactic reranking of K-best lists produced mostly negative results ',Koehn,'advmod_experiments_Early nsubj__experiments prep_experiments_with amod_phrases_syntactically-informed pobj_with_phrases cc__and amod_reranking_syntactic nsubj_produced_reranking prep_reranking_of amod_lists_K-best pobj_of_lists advmod_produced_ conj__produced advmod_negative_mostly amod_results_negative dobj_produced_results ' +o,4621,'This has been shown both in supervised settings and unsupervised settings in which constraints are used to bootstrap the model ',Haghighi,'nsubjpass_shown_This aux_shown_has auxpass_shown_been dep_in_both prep_shown_in amod__supervised nn__settings pobj_in_ cc_shown_and conj_shown_unsupervised nn__settings dobj_unsupervised_ rel_used_in pobj_in_which nsubjpass_used_constraints auxpass_used_are rcmod__used aux_bootstrap_to xcomp_used_bootstrap det_model_the dobj_bootstrap_model ' +o,4622,'EMD training combines generative and discriminative elements ',Fraser,'nn__EMD nn__training nsubj_combines_ amod_elements_generative cc_generative_and conj_generative_discriminative dobj_combines_elements ' +o,4623,'We tune all feature weights automatically to maximize the BLEU score on the dev set ',Papineni,'nsubj_tune_We det_weights_all nn_weights_feature nsubj__weights advmod__automatically ccomp_tune_ aux_maximize_to xcomp__maximize det_score_the amod_score_BLEU amod_score_ dobj_maximize_score prep_maximize_on det_set_the nn_set_dev pobj_on_set ' +o,4624,'Another line of research closely related to our work is the recognition of semantic orientation and sentiment analysis ',Turney,'det_line_Another nsubj_recognition_line prep_line_of pobj_of_research advmod_related_closely amod_research_related prep_related_to poss_work_our pobj_to_work cop_recognition_is det_recognition_the prep_recognition_of amod_orientation_semantic pobj_of_orientation cc_orientation_and nn_analysis_sentiment conj_orientation_analysis ' +o,4625,' , and others ',Yarowsky,'conj__ cc__and conj__others ' +o,4626,'Models of translational equivalence that are ignorant of indirect associations have \' a tendency to be confused by collocates \' ',Brown,'nsubj_have_Models prep_Models_of amod_equivalence_translational pobj_of_equivalence nsubj_ignorant_that cop_ignorant_are rcmod_equivalence_ignorant prep_ignorant_of amod_associations_indirect pobj_of_associations det_tendency_a dobj_have_tendency aux_confused_to auxpass_confused_be infmod_tendency_confused prep_confused_by poss__collocates pobj_by_ ' +o,4627,'Networks (Toutanova et al., 2003) 97.24 SVM (Gimenez and M`arquez, 2003) 97.05 ME based a bidirectional inference (Tsuruoka and Tsujii, 2005) 97.15 Guided learning for bidirectional sequence classification (Shen et al., 2007) 97.33 AdaBoost.SDF with candidate features (=2,=1,=100, W-dist) 97.32 AdaBoost.SDF with candidate features (=2,=10,=10, F-dist) 97.32 SVM with candidate features (C=0.1, d=2) 97.32 Text Chunking F=1 Regularized Winnow + full parser output (Zhang et al., 2001) 94.17 SVM-voting (Kudo and Matsumoto, 2001) 93.91 ASO + unlabeled data (Ando and Zhang, 2005) 94.39 CRF+Reranking(Kudo et al., 2005) 94.12 ME based a bidirectional inference (Tsuruoka and Tsujii, 2005) 93.70 LaSo (Approximate Large Margin Update) (Daume III and Marcu, 2005) 94.4 HySOL (Suzuki et al., 2007) 94.36 AdaBoost.SDF with candidate featuers (=2,=1,=, W-dist) 94.32 AdaBoost.SDF with candidate featuers (=2,=10,=10,W-dist) 94.30 SVM with candidate features (C=1, d=2) 94.31 One of the reasons that boosting-based classifiers realize faster classification speed is sparseness of rules.',Suzuki,'' +o,4628,' and Quirk et al ',Barzilay,'cc__and conj__Quirk cc_Quirk_et conj_Quirk_al ' +o,4629,'Clusters are created by means of distributional techniques in , while in low level synonim sets in WordNet are used ',Marcus,'nsubjpass_created_Clusters auxpass_created_are prep_created_by pobj_by_means prep_means_of amod_techniques_distributional pobj_of_techniques prep_created_in pobj_in_ mark_used_while prep_used_in amod_level_ amod_level_low pobj_in_level nn_sets_synonim nsubjpass_used_sets prep_sets_in pobj_in_WordNet auxpass_used_are advcl_created_used ' +o,4630,'22 Generalization pseudocode In order to identify the portions in common between the patterns , and to generalize them , we apply the following pseudocode -LRB- Ruiz-Casado et al , in press -RRB- : 1All the PoS examples in this paper are done with Penn Treebank labels ',Marcus,'num_pseudocode_22 amod_pseudocode_Generalization mark_identify_In dep_identify_order aux_identify_to dep_apply_identify det_portions_the dobj_identify_portions prep_identify_in pobj_in_common prep_common_between det_patterns_the pobj_between_patterns cc_identify_and aux_generalize_to conj_identify_generalize dobj_generalize_them nsubj_apply_we dep_pseudocode_apply det_pseudocode_the amod_pseudocode_following dobj_apply_pseudocode dep_pseudocode_Ruiz-Casado cc_Ruiz-Casado_et conj_Ruiz-Casado_al prep_Ruiz-Casado_in pobj_in_press dep_pseudocode_1All det_examples_the nn_examples_PoS nsubjpass_done_examples prep_examples_in det_paper_this pobj_in_paper auxpass_done_are dep_pseudocode_done prep_done_with nn_labels_Penn nn_labels_Treebank pobj_with_labels ' +o,4631,'Kappa coefficient is given in -LRB- 1 -RRB- -LRB- 1 -RRB- -RRB- -LRB- 1 -RRB- -LRB- -RRB- -LRB- EP EPAP Kappa = where P -LRB- A -RRB- is the proportion of times the annotators actually agree and P -LRB- E -RRB- is the proportion of times the annotators are expected to agree due to chance 3 ',Carletta,'nn_coefficient_Kappa nsubjpass_given_coefficient auxpass_given_is prep_given_in dep_expected_1 nsubjpass_expected_ appos__1 dep_agree_1 nn_=_EP nn_=_EPAP nn_=_Kappa nsubj_agree_= advmod_proportion_where nsubj_proportion_P appos_P_A cop_proportion_is det_proportion_the rcmod_=_proportion prep_proportion_of pobj_of_times det_annotators_the dep_times_annotators advmod_agree_actually rcmod__agree cc_agree_and nsubj_proportion_P appos_P_E cop_proportion_is det_proportion_the conj_agree_proportion prep_proportion_of pobj_of_times det_annotators_the dep_times_annotators auxpass_expected_are dep_given_expected aux_agree_to xcomp_expected_agree acomp_agree_due prep_due_to pobj_to_chance num_chance_3 ' +o,4632,'On the base of the chunk scheme proposed by Abney and the BIO tagging system proposed in , many machine learning techniques are used to deal with the problem ',Ramshaw,'prep_used_On det_base_the pobj_On_base prep_base_of det_scheme_the nn_scheme_chunk pobj_of_scheme partmod_scheme_proposed prep_proposed_by nn__Abney pobj_by_ cc__and det_system_the nn_system_BIO amod_system_tagging conj__system partmod_system_proposed prep_proposed_in pobj_in_ amod_techniques_many nn_techniques_machine nn_techniques_learning nsubjpass_used_techniques auxpass_used_are aux_deal_to xcomp_used_deal prep_deal_with det_problem_the pobj_with_problem ' +o,4633,'It has a lower bound of 0 , no upper bound , better scores indicate better translations , and it tends to be highly correlated with the adequacy of outputs ; mWER or Multiple Word Error Rate is the edit distance in words between the system output and the closest reference translation in a set ',Och,'nsubj_has_It det_bound_a amod_bound_lower dobj_has_bound prep_bound_of pobj_of_0 det_scores_no amod_scores_upper amod_scores_bound dep_bound_better nsubj_indicate_scores ccomp_has_indicate amod_translations_better dobj_indicate_translations cc_has_and nsubj_tends_it conj_has_tends aux_correlated_to auxpass_correlated_be advmod_correlated_highly xcomp_tends_correlated prep_correlated_with det_adequacy_the pobj_with_adequacy prep_adequacy_of pobj_of_outputs nn__mWER nsubj_distance_ cc__or amod_Rate_Multiple nn_Rate_Word nn_Rate_Error conj__Rate cop_distance_is det_distance_the amod_distance_edit parataxis_tends_distance prep_distance_in pobj_in_words prep_words_between det_output_the nn_output_system pobj_between_output cc_words_and det_translation_the amod_translation_closest nn_translation_reference conj_words_translation prep_translation_in det_set_a pobj_in_set ' +o,4634,'A new automatic metric METEOR uses stems and synonyms of the words ',Banerjee,'det_uses_A amod_uses_new amod_uses_automatic amod_uses_metric amod_uses_METEOR amod_uses_ nsubj_stems_uses cc_stems_and conj_stems_synonyms prep_stems_of det_words_the pobj_of_words ' +o,4635,'We compare the following model types : conventional -LRB- ie , non-exponential -RRB- word n-gram models ; conventional IBM class n-gram models interpolated with conventional word n-gram models ; and model M All conventional n-gram models are smoothed with modified Kneser-Ney smoothing , except we also evaluate word n-gram models with Katz smoothing ',Brown,'nsubj_compare_We det_types_the amod_types_following nn_types_model dobj_compare_types amod_word_conventional appos_word_ie dep_ie_non-exponential dep_types_word amod_models_n-gram dep_word_models amod_models_conventional nn_models_IBM nn_models_class nn_models_n-gram nsubj_interpolated_models parataxis_compare_interpolated prep_interpolated_with amod__conventional nn__word nn__n-gram nn__models pobj_with_ cc_compare_and nn_M_model nsubjpass_smoothed_M det_models_All amod_models_conventional amod_models_n-gram dep_M_models auxpass_smoothed_are conj_compare_smoothed prep_smoothed_with amod__modified amod__Kneser-Ney amod__smoothing pobj_with_ mark_evaluate_except nsubj_evaluate_we advmod_evaluate_also advcl_smoothed_evaluate nn_models_word nn_models_n-gram dobj_evaluate_models prep_evaluate_with nn__Katz nn__smoothing pobj_with_ ' +o,4636,'6 Comparison With Previous Work The two parsers which have previously reported the best accuracies on the Penn Treebank Wall St Journal are the bigram parser described in and the SPATTER parser described in ',Berger,'num_Comparison_6 dep_Comparison_With amod_Work_Previous pobj_With_Work det_parsers_The num_parsers_two nsubj_parser_parsers nsubj_reported_which aux_reported_have advmod_reported_previously rcmod_parsers_reported det_accuracies_the amod_accuracies_best dobj_reported_accuracies prep_accuracies_on det_Journal_the nn_Journal_Penn nn_Journal_Treebank nn_Journal_Wall nn_Journal_St pobj_on_Journal cop_parser_are det_parser_the nn_parser_bigram rcmod_Work_parser partmod_parser_described prep_described_in pobj_in_ cc_parser_and det_parser_the amod_parser_SPATTER conj_parser_parser partmod_parser_described prep_described_in pobj_in_ ' +o,4637,'P -LRB- d -RRB- P L -LRB- d -RRB- -LRB- 4 -RRB- Statistical approaches to language modeling have been used in much NLP research , such as machine translation and speech recognition ',Brown,'nsubj_L_P dep_L_d amod_L_P appos_L_d dep_L_4 nn_approaches_Statistical nsubjpass_used_approaches prep_approaches_to nn_modeling_language pobj_to_modeling aux_used_have auxpass_used_been dep_L_used prep_used_in amod_research_much nn_research_NLP pobj_in_research dep_as_such prep_research_as nn__machine nn__translation pobj_as_ cc__and nn__speech nn__recognition conj__ ' +o,4638,'We computed precision , recall and error rate on the entire set for each data set6 For an initial alignment , we used GIZA + + in both directions -LRB- E-to-F and F-to-E , where F is either Chinese -LRB- C -RRB- or Spanish -LRB- S -RRB- -RRB- , and also two different combined alignments : intersection of E-to-F and F-to-E ; and RA using a heuristic combination approach called grow-diag-final ',Koehn,'nsubj_computed_We nn_rate_precision conj_precision_recall cc_precision_and conj_precision_error dobj_computed_rate prep_computed_on det_set_the amod_set_entire pobj_on_set prep_computed_for det_set6_each nn_set6_data pobj_for_set6 prep_used_For det_alignment_an amod_alignment_initial pobj_For_alignment nsubj_used_we dep_computed_used nn_+_GIZA nn_+_+ dobj_used_+ prep_+_in det_directions_both pobj_in_directions dep_directions_E-to-F cc_E-to-F_and conj_E-to-F_F-to-E advmod_C_where nsubj_C_F cop_C_is advmod_C_either amod_C_Chinese rcmod_E-to-F_C cc_C_or conj_C_Spanish appos_Spanish_S cc_directions_and advmod_different_also dep_different_two amod_alignments_different amod_alignments_combined conj_directions_alignments dep_+_intersection prep_intersection_of pobj_of_E-to-F cc_E-to-F_and conj_E-to-F_F-to-E cc_+_and conj_+_RA partmod_RA_using det_approach_a amod_approach_heuristic nn_approach_combination dobj_using_approach partmod_approach_called advmod_called_grow-diag-final ' +o,4639,'In the following sections , we will use 2 statistics to measure the the mutual translation likelihood ',Church,'prep_use_In det_sections_the amod_sections_following pobj_In_sections nsubj_use_we aux_use_will num_statistics_2 dobj_use_statistics aux_measure_to xcomp_use_measure det_likelihood_the det_translation_the amod_translation_mutual nn_likelihood_translation dobj_measure_likelihood ' +o,4640,'Oncetraininghastakenplace , minimumerrorrate training is used to tune the parameters i Finally , decoding in Hiero takes place using a CKY synchronous parser with beam search , augmented to permit efficient incorporation of language model scores ',Och,'advmod_used_Oncetraininghastakenplace amod__minimumerrorrate nn__training nsubjpass_used_ auxpass_used_is prep_used_to pobj_to_tune nsubj_augmented_the nn_the_parameters nn_the_i dep_the_Finally partmod_the_decoding mark_takes_in nsubj_takes_Hiero advcl_decoding_takes dobj_takes_place partmod_place_using det_parser_a nn_parser_CKY amod_parser_synchronous dobj_using_parser prep_using_with nn_search_beam pobj_with_search dep_used_augmented aux_permit_to xcomp_augmented_permit amod_incorporation_efficient dobj_permit_incorporation prep_incorporation_of nn_scores_language nn_scores_model pobj_of_scores advmod_permit_ ' +o,4641,'Decomposing the translational equivalence relations in the training data into smaller units of knowledge can improve a models ability to generalize ',Zhang,'det_relations_the amod_relations_translational nn_relations_equivalence dobj_Decomposing_relations prep_relations_in det_data_the nn_data_training pobj_in_data prep_Decomposing_into amod_units_smaller pobj_into_units prep_units_of pobj_of_knowledge aux_improve_can dep_Decomposing_improve det_models_a dobj_improve_models nsubj_Decomposing_ability aux_generalize_to infmod_ability_generalize ' +o,4642,'Narrative retellings provide a natural , conversational speech sample that can be analyzed for many of the characteristics of speech and language that have been shown to discriminate between healthy and impaired subjects , including syntactic complexity and mean pause duration ',Marcus,'amod_retellings_Narrative nsubj_provide_retellings det_sample_a amod_sample_natural amod_sample_conversational nn_sample_speech dobj_provide_sample nsubjpass_analyzed_that aux_analyzed_can auxpass_analyzed_be rcmod_sample_analyzed prep_analyzed_for pobj_for_many prep_many_of det_characteristics_the pobj_of_characteristics prep_characteristics_of pobj_of_speech cc_speech_and conj_speech_language nsubjpass_shown_that aux_shown_have auxpass_shown_been rcmod_many_shown prep_shown_to pobj_to_discriminate prep_discriminate_between amod_subjects_healthy cc_healthy_and conj_healthy_impaired pobj_between_subjects prep_discriminate_including amod__syntactic nn__complexity pobj_including_ cc_analyzed_and conj_analyzed_mean nn__pause nn__duration dobj_mean_ ' +o,4643,'291 31 Level of Analysis Research on sentiment annotation is usually conducted at the text or at the sentence levels ',Pang,'num_Level_291 num_Level_31 nsubjpass_conducted_Level prep_Level_of nn_Research_Analysis pobj_of_Research prep_Research_on nn_annotation_sentiment pobj_on_annotation auxpass_conducted_is advmod_conducted_usually prep_conducted_at det__the nn__text pobj_at_ cc_at_or conj_at_at det__the nn__sentence nn__levels pobj_at_ ' +o,4644,'For testing purposes , we used the Wall Street Journal part of the Penn Treebank corpus ',Marcus,'prep_used_For nn_purposes_testing pobj_For_purposes nsubj_used_we det_part_the nn_part_Wall nn_part_Street nn_part_Journal dobj_used_part prep_part_of det_corpus_the nn_corpus_Penn nn_corpus_Treebank pobj_of_corpus ' +o,4645,'2 Task Description 21 Data Representation gave mainly two kinds of base NPs representation the open\\/close bracketing and IOB tagging ',Ramshaw,'num__2 nn__Task nn__Description num__21 nn__Data nn__Representation nsubj_gave_ advmod_kinds_mainly num_kinds_two iobj_gave_kinds prep_kinds_of nn_representation_base nn_representation_NPs pobj_of_representation det_bracketing_the amod_bracketing_open\\/close dobj_gave_bracketing cc_bracketing_and conj_bracketing_IOB amod_IOB_tagging ' +p,4646,'To perform translation , state-of-the-art MT systems use a statistical phrase-based approach by treating phrases as the basic units of translation ',Och,'aux_perform_To dep_use_perform dobj_perform_translation amod_systems_state-of-the-art nn_systems_MT nsubj_use_systems det__a amod__statistical amod__phrase-based nn__approach dobj_use_ prep_use_by pcomp_by_treating dobj_treating_phrases prep_treating_as det_units_the amod_units_basic pobj_as_units prep_units_of pobj_of_translation ' +o,4647,'26 Tuning procedure The Moses-based systems were tuned using the implementation of minimum error rate training -LRB- MERT -RRB- distributed with the Moses decoder , using the development corpus -LRB- dev2009a -RRB- ',Och,'num_procedure_26 nn_procedure_Tuning nsubj_systems_procedure det_systems_The amod_systems_Moses-based auxpass_tuned_were rcmod_systems_tuned dep_tuned_using det_implementation_the dobj_using_implementation prep_implementation_of amod__minimum nn__error nn__rate nn__training abbrev__MERT pobj_of_ partmod__distributed prep_distributed_with det_decoder_the nn_decoder_Moses pobj_with_decoder dep_using_using det_corpus_the nn_corpus_development dobj_using_corpus dep_systems_dev2009a ' +o,4648,'In many cases , improving semi-supervised models was done by seeding these models with domain information taken from dictionaries or ontology ',Collins,'prep_done_In amod_cases_many pobj_In_cases amod_models_improving amod_models_semi-supervised nsubjpass_done_models auxpass_done_was prep_done_by pcomp_by_seeding det_models_these dobj_seeding_models prep_seeding_with nn_information_domain pobj_with_information partmod_information_taken prep_taken_from pobj_from_dictionaries cc_dictionaries_or conj_dictionaries_ontology ' +o,4649,'This is applied to maximize coverage , which is similar as the final in ',Koehn,'nsubjpass_applied_This auxpass_applied_is aux_maximize_to xcomp_applied_maximize dobj_maximize_coverage nsubj_similar_which cop_similar_is rcmod_coverage_similar prep_similar_as det_final_the pobj_as_final prep_similar_in ' +o,4650,'(Lin, 2004b).',Lin,'' +o,4651,'Trained and tested using the same technique as ','Daume III','cc_Trained_and conj_Trained_tested xcomp_tested_using det_technique_the amod_technique_same dobj_using_technique prep_using_as ' +o,4652,'By introducing the hidden word alignment variable a, the following approximate optimization criterion can be applied for that purpose: e = argmaxe Pr(e | f) = argmaxe summationdisplay a Pr(e,a | f) argmaxe,a Pr(e,a | f) Exploiting the maximum entropy (Berger et al. , 1996) framework, the conditional distribution Pr(e,a | f) can be determined through suitable real valued functions (called features) hr(e,f,a),r = 1R, and takes the parametric form: p(e,a | f) exp Rsummationdisplay r=1 rhr(e,f,a)} The ITC-irst system (Chen et al. , 2005) is based on a log-linear model which extends the original IBM Model 4 (Brown et al. , 1993) to phrases (Koehn et al. , 2003; Federico and Bertoldi, 2005).',Berger,'' +o,4653,'The phrase bilexicon is derived from the intersection of bidirectional IBM Model 4 alignments , obtained with GIZA + + , augmented to improve recall using the grow-diag-final heuristic ',Och,'det_bilexicon_The nn_bilexicon_phrase nsubjpass_derived_bilexicon auxpass_derived_is prep_derived_from det_intersection_the pobj_from_intersection prep_intersection_of amod_Model_bidirectional nn_Model_IBM pobj_of_Model num_alignments_4 nsubj_augmented_alignments partmod_alignments_obtained prep_obtained_with pobj_with_GIZA amod_GIZA_+ cc_+_+ conj_+_ dep_derived_augmented aux_improve_to xcomp_augmented_improve dobj_improve_recall partmod_recall_using det_heuristic_the amod_heuristic_grow-diag-final dobj_using_heuristic ' +o,4654,'Work at the University of Dundee has shown that the extensive use of fixed text for sequences such as greetings and prestored narratives is beneficial in AAC ',Brown,'nsubj_shown_Work prep_Work_at det_University_the pobj_at_University prep_University_of nn__Dundee pobj_of_ aux_shown_has complm_beneficial_that det_use_the amod_use_extensive nsubj_beneficial_use prep_use_of amod_text_fixed pobj_of_text prep_text_for pobj_for_sequences dep_as_such prep_sequences_as pobj_as_greetings cc_greetings_and amod_narratives_prestored conj_greetings_narratives cop_beneficial_is ccomp_shown_beneficial prep_beneficial_in pobj_in_AAC ' +o,4655,'For many languages , large-scale syntactically annotated corpora have been built -LRB- eg the Penn Treebank -RRB- , and many parsing algorithms using CFGs have been proposed ',Marcus,'prep_built_For amod_languages_many pobj_For_languages amod_corpora_large-scale advmod_annotated_syntactically amod_corpora_annotated nsubjpass_built_corpora aux_built_have auxpass_built_been parataxis_built_eg det__the nn__Penn nn__Treebank dobj_eg_ cc_built_and amod_algorithms_many amod_algorithms_parsing nsubjpass_proposed_algorithms partmod_algorithms_using dobj_using_CFGs aux_proposed_have auxpass_proposed_been conj_built_proposed ' +o,4656,'More specifically , the work on optimizing preference factors and semantic collocations was done as part of a project on spoken language translation in which the CLE was used for analysis and generation of both English and Swedish ',Marcus,'advmod_specifically_More advmod_done_specifically det_work_the nsubjpass_done_work prep_work_on amod_factors_optimizing nn_factors_preference pobj_on_factors cc_factors_and amod_collocations_semantic conj_factors_collocations auxpass_done_was prep_done_as pobj_as_part prep_part_of det_project_a pobj_of_project prep_done_on amod_translation_spoken nn_translation_language pobj_on_translation rel_used_in pobj_in_which det_CLE_the nsubjpass_used_CLE auxpass_used_was rcmod_translation_used prep_used_for pobj_for_analysis cc_analysis_and conj_analysis_generation prep_analysis_of preconj__both nn__English cc_English_and conj_English_Swedish pobj_of_ ' +o,4657,'The data for all our experiments was extracted from the Penn Treebank II Wall Street Journal -LRB- WSJ -RRB- corpus ',Marcus,'det_data_The dep_extracted_data prep_data_for predet_experiments_all poss_experiments_our pobj_for_experiments auxpass_extracted_was prep_extracted_from det_Wall_the nn_Wall_Penn nn_Wall_Treebank nn_Wall_II pobj_from_Wall nn_corpus_Street nn_corpus_Journal abbrev_corpus_WSJ nsubjpass_extracted_corpus ' +o,4658,'The largest corpus that Goldwater and Griffiths studied contained 96,000 words , while used all of the 1,173,766 words in the full Penn WSJ treebank ',Johnson,'det_corpus_The amod_corpus_largest nsubj_contained_corpus complm_studied_that nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths nsubj_studied_ dep_corpus_studied num_words_96,000 dobj_contained_words mark_used_while nn__ nsubj_used_ advcl_contained_used dobj_used_all prep_all_of det_words_the num_words_1,173,766 pobj_of_words prep_used_in det_treebank_the amod_treebank_full nn_treebank_Penn nn_treebank_WSJ pobj_in_treebank ' +o,4659,'There are several works that try to learn paraphrase pairs from parallel or comparable corpora ',Barzilay,'expl_are_There amod_works_several nsubj_are_works nsubj_try_that rcmod_works_try aux_learn_to xcomp_try_learn amod_pairs_paraphrase dobj_learn_pairs prep_learn_from amod_corpora_parallel cc_parallel_or conj_parallel_comparable pobj_from_corpora ' +p,4660,'We also note that there are a number of bootstrapping methods successfully applied to text eg , word sense disambiguation , named entity instance classification , and the extraction of parts word given the whole word ',Yarowsky,'nsubj_note_We advmod_note_also complm_are_that expl_are_there ccomp_note_are det_number_a nsubj_are_number prep_number_of pcomp_of_bootstrapping dobj_bootstrapping_methods advmod_applied_successfully partmod_methods_applied prep_applied_to nn_eg_text pobj_to_eg nn__word nn__sense nn__disambiguation appos_eg_ partmod_methods_named nn__entity nn__instance nn__classification dobj_named_ cc_methods_and det_extraction_the conj_methods_extraction prep_extraction_of nn_word_parts pobj_of_word partmod_word_given det__the amod__whole nn__word dobj_given_ ' +o,4661,'Examples of such knowledge sources include stemming and TF-IDF weighting ',Banerjee,'nsubj_include_Examples prep_Examples_of amod_sources_such nn_sources_knowledge pobj_of_sources xcomp_include_stemming cc_stemming_and conj_stemming_TF-IDF dobj_stemming_weighting ' +o,4662,'1 Introduction Text chunking has been one of the most interesting problems in natural language learning community since the first work of using a machine learning method ',Ramshaw,'num_chunking_1 nn_chunking_Introduction nn_chunking_Text nsubj_been_chunking aux_been_has dobj_been_one prep_one_of det_problems_the advmod_interesting_most amod_problems_interesting pobj_of_problems prep_problems_in amod_language_natural pobj_in_language partmod_language_learning dobj_learning_community prep_learning_since det_work_the amod_work_first nsubj_using_work prep_work_of pobj_of_ pcomp_since_using det_machine_a dobj_using_machine partmod_machine_learning dobj_learning_method ' +o,4663,'A few unsupervised metrics have been applied to automatic paraphrase identification and extraction ',Barzilay,'det_metrics_A amod_metrics_few amod_metrics_unsupervised nsubjpass_applied_metrics aux_applied_have auxpass_applied_been prep_applied_to amod_identification_automatic nn_identification_paraphrase pobj_to_identification cc_identification_and conj_identification_extraction ' +o,4664,' showed that the learning strategy of bootstrapping from small tagged data led to results rivaling supervised training methods ',Yarowsky,'nsubj_showed_ complm_led_that det_strategy_the nn_strategy_learning nsubj_led_strategy prep_strategy_of pcomp_of_bootstrapping prep_bootstrapping_from amod_data_small amod_data_tagged pobj_from_data ccomp_showed_led prep_led_to pobj_to_results xcomp_led_rivaling amod_methods_supervised nn_methods_training dobj_rivaling_methods ' +p,4665,'2 Motivation and Prior Work While several authors have looked at the supervised adaptation case , there are less -LRB- and especially less successful -RRB- studies on semi-supervised domain adaptation ',McClosky,'dep_Motivation_2 cc_Motivation_and amod_Work_Prior conj_Motivation_Work mark_looked_While amod_authors_several nsubj_looked_authors aux_looked_have advcl_are_looked prep_looked_at det_case_the amod_case_supervised nn_case_adaptation pobj_at_case expl_are_there dep_Motivation_are acomp_are_less dep_-LRB-_and advmod_less_especially dep_-LRB-_less dep_less_successful dep_on_studies prep_are_on amod_adaptation_semi-supervised nn_adaptation_domain pobj_on_adaptation ' +o,4666,'Therefore , whenever we have access to a large amount of labeled data from some source -LRB- out-of-domain -RRB- , but we would like a model that performs well on some new target domain , we face the problem of domain adaptation ','Daume III','advmod_face_Therefore advmod_have_whenever nsubj_have_we dep_face_have dobj_have_access prep_have_to det_amount_a amod_amount_large pobj_to_amount prep_amount_of amod_data_labeled pobj_of_data prep_data_from det_source_some pobj_from_source appos_source_out-of-domain cc_have_but nsubj_like_we aux_like_would conj_have_like det_model_a dobj_like_model nsubj_performs_that rcmod_model_performs advmod_performs_well prep_performs_on det__some amod__new nn__target nn__domain pobj_on_ nsubj_face_we det_problem_the dobj_face_problem prep_problem_of nn_adaptation_domain pobj_of_adaptation ' +o,4667,'In the LFG-based generation algorithm presented by complex named entities -LRB- ie those consisting of more than one word token -RRB- and other multi-word units can be fragmented in the surface realization ',Cahill,'prep_named_In det_algorithm_the amod_algorithm_LFG-based nn_algorithm_generation pobj_In_algorithm partmod_algorithm_presented prep_presented_by amod_complex_ pobj_by_complex nsubj_named_entities dep_entities_ie dobj_ie_those partmod_those_consisting prep_consisting_of dep_than_more quantmod_one_than num_token_one nn_token_word pobj_of_token cc_entities_and amod_units_other amod_units_multi-word conj_entities_units aux_fragmented_can auxpass_fragmented_be rcmod_units_fragmented prep_fragmented_in det_realization_the nn_realization_surface pobj_in_realization ' +o,4668,'All submitted runs were evaluated with the automatic metrics : ROUGE , which calculates the proportion of n-grams shared between the candidate summary and the reference summaries , and Basic Elements , which compares the candidate to the models in terms of head-modifier pairs ',Lin,'nsubj_submitted_All nsubjpass_evaluated_runs auxpass_evaluated_were ccomp_submitted_evaluated prep_evaluated_with det_metrics_the amod_metrics_automatic pobj_with_metrics nn__ROUGE dep_metrics_ nsubj_calculates_which rcmod__calculates det_proportion_the dobj_calculates_proportion prep_proportion_of pobj_of_n-grams partmod_n-grams_shared prep_shared_between det_summary_the nn_summary_candidate pobj_between_summary cc_summary_and det_summaries_the nn_summaries_reference conj_summary_summaries cc__and nn__Basic nn__Elements conj__ nsubj_compares_which rcmod__compares det_candidate_the dobj_compares_candidate prep_compares_to det_models_the pobj_to_models prep_models_in pobj_in_terms prep_terms_of amod_pairs_head-modifier pobj_of_pairs ' +o,4669,'2 Latent Variable Parsing In latent variable parsing , we learn rule probabilities on latent annotations that , when marginalized out , maximize the likelihood of the unannotated training trees ',Matsuzaki,'dep_learn_2 nn_Parsing_Latent nn_Parsing_Variable dep_In_Parsing prep_learn_In amod__latent amod__variable nn__parsing pobj_In_ nsubj_learn_we nn_probabilities_rule dobj_learn_probabilities prep_learn_on amod_annotations_latent pobj_on_annotations complm_maximize_that advmod_marginalized_when dep_maximize_marginalized prt_marginalized_out ccomp_learn_maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_of det_trees_the amod_trees_unannotated nn_trees_training pobj_of_trees ' +o,4670,'Table 3 shows the differences between the treebank ~ utilized in on the one hand , and in the work reported here , on the other , is Table 4 shows relevant lSFigures for Average Sentence Length -LRB- ` l ~ raLuing Corpus -RRB- and Training Set Size , for the IBM ManuaLs Corpus , are approximate , and cz ~ e fzom ',Marcus,'nsubj_shows_Table num_Table_3 det_differences_the dobj_shows_differences prep_differences_between det_~_the nn_~_treebank pobj_between_~ partmod_~_utilized prep_utilized_in pobj_in_ prep_utilized_on det_hand_the num_hand_one pobj_on_hand cc_shows_and prep_approximate_in det_work_the pobj_in_work partmod_work_reported advmod_reported_here prep_reported_on det_other_the pobj_on_other aux_Table_is ccomp_approximate_Table num_shows_4 measure_relevant_shows amod_lSFigures_relevant dobj_Table_lSFigures prep_lSFigures_for nn_Length_Average nn_Length_Sentence pobj_for_Length amod_Corpus_l nn_Corpus_~ nn_Corpus_raLuing dep_Length_Corpus cc_Length_and nn_Size_Training nn_Size_Set conj_Length_Size prep_Table_for det_Corpus_the nn_Corpus_IBM nn_Corpus_ManuaLs pobj_for_Corpus aux_approximate_are conj_shows_approximate cc_shows_and nn_~_cz nsubj_e_~ conj_shows_e nn__fzom dobj_e_ ' +o,4671,'We removed all but the first two characters of each POS tag , resulting in a set of 57 tags which more closely resembles that of the Penn TreeBank ',Marcus,'nsubj_removed_We dep_but_all advmod_removed_but det_characters_the amod_characters_first num_characters_two dobj_removed_characters prep_characters_of det_tag_each dep_tag_POS pobj_of_tag xcomp_removed_resulting prep_resulting_in det_set_a pobj_in_set prep_set_of num_tags_57 pobj_of_tags nsubj_that_which advmod_closely_more advmod_that_closely cop_that_resembles rcmod_tags_that prep_that_of det_TreeBank_the nn_TreeBank_Penn pobj_of_TreeBank ' +o,4672,'More specialized methods also exist , for example for support vector machines and for conditional random fields ',Suzuki,'amod_methods_More amod_methods_specialized nsubj_exist_methods advmod_exist_also prep_exist_for pobj_for_example prep_example_for nn_machines_support nn_machines_vector pobj_for_machines dep_example_ cc__and conj__for amod_fields_conditional amod_fields_random pobj_for_fields ' +o,4673,' used bootstrapping to extend their semantic compatibility model , which they called contextual-role knowledge , by identifying certain cases of easily-resolved anaphors and antecedents ',Bean,'advmod_used_ xcomp_used_bootstrapping aux_extend_to xcomp_bootstrapping_extend poss_compatibility_their amod_compatibility_semantic dobj_extend_compatibility nsubj_used_model rel_called_which nsubj_called_they rcmod_model_called amod_knowledge_contextual-role dobj_called_knowledge prep_called_by pcomp_by_identifying amod_cases_certain dobj_identifying_cases prep_cases_of amod_anaphors_easily-resolved pobj_of_anaphors cc_anaphors_and conj_anaphors_antecedents ' +o,4674,'For comparison purposes , we consider two different algorithms for our AnswerExtraction module : one that does not bridge the lexical chasm , based on N-gram cooccurrences between the question terms and the answer terms ; and one that attempts to bridge the lexical chasm using Statistical Machine Translation inspired techniques in order to find the best answer for a given question ',Brown,'prep_consider_For nn_purposes_comparison pobj_For_purposes nsubj_consider_we num_algorithms_two amod_algorithms_different dobj_consider_algorithms prep_algorithms_for poss_module_our nn_module_AnswerExtraction pobj_for_module dep_algorithms_one nsubj_bridge_that aux_bridge_does neg_bridge_not rcmod_one_bridge det_chasm_the amod_chasm_lexical dobj_bridge_chasm prep_bridge_based dep_based_on amod_cooccurrences_N-gram pobj_on_cooccurrences prep_cooccurrences_between det_terms_the nn_terms_question pobj_between_terms cc_terms_and det_terms_the nn_terms_answer conj_terms_terms cc_algorithms_and conj_algorithms_one nsubj_attempts_that rcmod_one_attempts aux_bridge_to xcomp_attempts_bridge det_chasm_the amod_chasm_lexical dobj_bridge_chasm xcomp_bridge_using nn_Translation_Statistical nn_Translation_Machine nsubj_inspired_Translation ccomp_using_inspired nsubj__techniques ccomp_inspired_ mark_find_in dep_find_order aux_find_to purpcl__find det_answer_the amod_answer_best dobj_find_answer prep_answer_for det_question_a amod_question_given pobj_for_question ' +o,4675,'a1 Graduated in March 2006 Standard phrase-based translation systems use a word distance-based reordering model in which non-monotonic phrase alignment is penalized based on the word distance between successively translated source phrases without considering the orientation of the phrase alignment or the identities of the source and target phrases ',Koehn,'nn_Graduated_a1 nsubj_use_Graduated prep_Graduated_in nn_systems_March nn_systems_2006 nn_systems_Standard amod_systems_phrase-based nn_systems_translation pobj_in_systems det_model_a nn_model_word amod_model_distance-based nn_model_reordering dobj_use_model rel_penalized_in pobj_in_which amod_alignment_non-monotonic nn_alignment_phrase nsubjpass_penalized_alignment auxpass_penalized_is rcmod_model_penalized prep_penalized_based dep_based_on det_distance_the nn_distance_word pobj_on_distance prep_distance_between advmod_translated_successively amod_phrases_translated nn_phrases_source pobj_between_phrases prep_use_without pcomp_without_considering det_orientation_the dobj_considering_orientation prep_orientation_of det_alignment_the nn_alignment_phrase pobj_of_alignment cc_orientation_or det_identities_the conj_orientation_identities prep_identities_of det_phrases_the nn_phrases_source cc_source_and conj_source_target pobj_of_phrases ' +o,4676,'For the simple bag-of-word bilingual LSA as describedinSection221 , afterSVDonthesparsematrix using the toolkit SVDPACK , all source and target words are projected into a lowdimensional -LRB- R = 88 -RRB- LSA-space ',Brown,'prep_projected_For det_LSA_the amod_LSA_simple amod_LSA_bag-of-word amod_LSA_bilingual pobj_For_LSA prep_LSA_as pobj_as_describedinSection221 nsubjpass_projected_afterSVDonthesparsematrix partmod_afterSVDonthesparsematrix_using det__the amod__toolkit nn__SVDPACK dobj_using_ det_source_all conj__source cc__and nn_words_target conj__words auxpass_projected_are prep_projected_into det_88_a amod_88_lowdimensional nsubj_88_R dep_88_= pobj_into_88 amod_88_LSA-space ' +o,4677,'Sentiment analysis includes a variety of different problems, including: sentiment classification techniques to classify reviews as positive or negative, based on bag of words (Pang et al. , 2002) or positive and negative words (Turney, 2002; Mullen and Collier, 2004); classifying sentences in a document as either subjective or objective (Riloff and Wiebe, 2003; Pang and Lee, 2004); identifying or classifying appraisal targets (Nigam and Hurst, 2004); identifying the source of an opinion in a text (Choi et al. , 2005), whether the author is expressing the opinion, or whether he is attributing the opinion to someone else; and developing interactive and visual opinion mining methods (Gamon et al. , 2005; Popescu and Etzioni, 2005).',Pang,'' +o,4678,'MT output is evaluated using the standard MT evaluation metric BLEU ',Papineni,'nn_output_MT nsubjpass_evaluated_output auxpass_evaluated_is xcomp_evaluated_using det_BLEU_the amod_BLEU_standard nn_BLEU_MT nn_BLEU_evaluation amod_BLEU_metric dobj_using_BLEU ' +o,4679,'Word association norms based on co-occurrence information have been proposed by ',Church,'amod_norms_Word nn_norms_association nsubjpass_proposed_norms prep_norms_based dep_based_on amod_information_co-occurrence pobj_on_information aux_proposed_have auxpass_proposed_been prep_proposed_by ' +o,4680,'Some examples of language reuse include collocation analysis , the use of entire factual sentences extracted from corpora -LRB- eg , \' ` Toy Story \' is the Academy Award winning animated film developed by Pixar ~ \' -RRB- , and summarization using sentence extraction ',Smadja,'det_examples_Some nsubj_include_examples prep_examples_of nn_reuse_language pobj_of_reuse nn__collocation nn__analysis dobj_include_ det_use_the conj__use prep_use_of amod_sentences_entire amod_sentences_factual pobj_of_sentences partmod_sentences_extracted prep_extracted_from pobj_from_corpora nsubj_Award_eg nn_Story_Toy nsubj_Award_Story cop_Award_is det_Award_the nn_Award_Academy dep_corpora_Award partmod_Award_winning amod_film_animated dobj_winning_film partmod_film_developed prep_developed_by nn_~_Pixar pobj_by_~ cc__and conj__summarization partmod_summarization_using nn__sentence nn__extraction dobj_using_ ' +o,4681,'Since a handmade thesaurus is not slfitahle for machine use , and expensive to compile , automatical construction of ~ a thesaurus has been attempted using corpora ',Hindle,'mark_slfitahle_Since det_thesaurus_a amod_thesaurus_handmade nsubj_slfitahle_thesaurus cop_slfitahle_is neg_slfitahle_not advcl_attempted_slfitahle prep_slfitahle_for nn_use_machine pobj_for_use cc_slfitahle_and conj_slfitahle_expensive aux_compile_to xcomp_expensive_compile amod_construction_automatical nsubjpass_attempted_construction prep_construction_of pcomp_of_~ det_thesaurus_a dobj_~_thesaurus aux_attempted_has auxpass_attempted_been xcomp_attempted_using dobj_using_corpora ' +o,4682,'4 Analysis of Experimental Data Most of the existing research in computational linguistics that uses human annotators is within the framework of classification , where an annotator decides , for every test item , on an appropriate tag out of the pre-specified set of tags ',Marcus,'num_Analysis_4 nsubj_is_Analysis prep_Analysis_of nn_Most_Experimental nn_Most_Data pobj_of_Most prep_Most_of det_research_the amod_research_existing pobj_of_research prep_research_in amod_linguistics_computational pobj_in_linguistics nsubj_uses_that rcmod_research_uses amod_annotators_human dobj_uses_annotators prep_is_within det_framework_the pobj_within_framework prep_framework_of pobj_of_classification advmod_decides_where det_annotator_an nsubj_decides_annotator advcl_is_decides prep_decides_for det_item_every nn_item_test pobj_for_item prep_decides_on det_tag_an amod_tag_appropriate pobj_on_tag dep_of_out prep_tag_of det_set_the amod_set_pre-specified pobj_of_set prep_set_of pobj_of_tags ' +o,4683,'Following Hatzivassiloglou and McKeown and , we decided to observe how often the words from the headline co-occur with each one of the six emotions ',Turney,'prep_decided_Following nn__Hatzivassiloglou cc_Hatzivassiloglou_and conj_Hatzivassiloglou_McKeown pobj_Following_ cc__and conj__ nsubj_decided_we aux_observe_to xcomp_decided_observe advmod_co-occur_how advmod_how_often det_words_the nsubj_co-occur_words prep_words_from det_headline_the pobj_from_headline ccomp_observe_co-occur prep_co-occur_with det_one_each pobj_with_one prep_one_of det_emotions_the num_emotions_six pobj_of_emotions ' +o,4684,' calls this trade-off specificity ; equivalent observations were made by Church & Hanks -LRB- 1989 -RRB- and Church et al , who refer to the tendency for large windows to wash out , smear or defocus those associations exhibited at smaller scales ',Rapp,'nsubj_calls_ det_specificity_this nn_specificity_trade-off dobj_calls_specificity nn_observations_equivalent nsubjpass_made_observations auxpass_made_were parataxis_calls_made prep_made_by pobj_by_Church cc_Church_& conj_Church_Hanks appos_Church_1989 cc_calls_and nn__Church cc_Church_et conj_Church_al nsubj_smear_ nsubj_refer_who rcmod__refer prep_refer_to det_tendency_the pobj_to_tendency prep_tendency_for amod_windows_large pobj_for_windows aux_wash_to xcomp_refer_wash prt_wash_out conj_calls_smear cc_smear_or conj_smear_defocus det_associations_those nsubj_exhibited_associations ccomp_defocus_exhibited prep_exhibited_at amod_scales_smaller pobj_at_scales ' +o,4685,'Morphologicaltoolssuch as lemmatizers andPOStaggersarebeingcommonlyusedin extractionsystems ; they areemployedbothfordealingwithtext variationandfor validatingthe candidatepairs : combinationsof functionwordsare typicallyruledout , as are the ungrammaticalcombinationsin the systemsthatmake useofparsers -LRB- ChurchandHanks , 1990 ; ,1993 ; Basilietal ',Smadja,'prep_Morphologicaltoolssuch_as amod_extractionsystems_lemmatizers nn_extractionsystems_andPOStaggersarebeingcommonlyusedin pobj_as_extractionsystems nsubj_areemployedbothfordealingwithtext_they parataxis_Morphologicaltoolssuch_areemployedbothfordealingwithtext amod_candidatepairs_variationandfor nn_candidatepairs_validatingthe dobj_areemployedbothfordealingwithtext_candidatepairs parataxis_Morphologicaltoolssuch_combinationsof amod__functionwordsare nn__typicallyruledout dobj_combinationsof_ dep_ungrammaticalcombinationsin_as cop_ungrammaticalcombinationsin_are det_ungrammaticalcombinationsin_the dep__ungrammaticalcombinationsin det_useofparsers_the amod_useofparsers_systemsthatmake nsubj_ungrammaticalcombinationsin_useofparsers dep_useofparsers_ChurchandHanks dep_ChurchandHanks_1990 parataxis_Morphologicaltoolssuch_ dobj__,1993 parataxis_Morphologicaltoolssuch_Basilietal ' +o,4686,' report an accuracy of 9733 \% on the same data set using a perceptron-based bidirectional tagging model ',Shen,'nsubj_report_ det_accuracy_an dobj_report_accuracy prep_accuracy_of num_\%_9733 pobj_of_\% prep_report_on det_data_the amod_data_same pobj_on_data partmod_data_set xcomp_set_using det_model_a amod_model_perceptron-based amod_model_bidirectional nn_model_tagging dobj_using_model ' +o,4687,'They are generated from the training corpus via the ? diag-and ? ? method and smoothed using Kneser-Ney smoothing , ? ? one or several n-gram language model -LRB- s -RRB- trained with the SRILM toolkit ; in the baseline experiments reported here , we used a trigram model , ? ? a distortion model which assigns a penalty based on the number of source words which are skipped when generating a new target phrase , ? ? a word penalty ',Koehn,'nsubjpass_generated_They auxpass_generated_are prep_generated_from det_corpus_the nn_corpus_training pobj_from_corpus prep_corpus_via pobj_via_the partmod_the_diag-and nsubj__method dep_generated_ cc__and conj__smoothed xcomp_smoothed_using amod_smoothing_Kneser-Ney dobj_using_smoothing advmod_smoothed_ dep_in_one cc_one_or amod_model_several amod_model_n-gram nn_model_language conj_one_model dep_-LRB-_s amod_model_trained prep_trained_with det__the nn__SRILM nn__toolkit pobj_with_ dep_generated_in det_baseline_the pobj_in_baseline nsubj_reported_experiments dep_generated_reported advmod_used_here nsubj_used_we ccomp_reported_used det_model_a nn_model_trigram dobj_used_model det_distortion_a nsubj_model_distortion dep_reported_model nsubj_assigns_which ccomp_model_assigns det_penalty_a dobj_assigns_penalty prep_assigns_based dep_based_on det_number_the pobj_on_number prep_number_of nn_words_source pobj_of_words nsubj_skipped_which aux_skipped_are rcmod_number_skipped advmod_generating_when advcl_skipped_generating det_phrase_a amod_phrase_new nn_phrase_target dobj_generating_phrase det_word_a dep_reported_word dep_generated_penalty ' +o,4688,'This is the strategy that is usually adopted in other phrase-based MT approaches ',Och,'nsubj_strategy_This cop_strategy_is det_strategy_the nsubjpass_adopted_that auxpass_adopted_is advmod_adopted_usually rcmod_strategy_adopted prep_adopted_in amod_approaches_other amod_approaches_phrase-based nn_approaches_MT pobj_in_approaches ' +o,4689,'For example , used an English-Chinese bilingual parser based on stochastic transduction grammars to identify terms , including multiword expressions ',Wu,'prep_used_For pobj_For_example nsubj_used_ det_parser_an amod_parser_English-Chinese amod_parser_bilingual dobj_used_parser partmod_parser_based prep_based_on amod_grammars_stochastic nn_grammars_transduction pobj_on_grammars aux_identify_to xcomp_based_identify dobj_identify_terms prep_terms_including amod_expressions_multiword pobj_including_expressions ' +o,4690,' and Magerman used the clustering algorithm of et al ',Brown,'nn__ cc__and conj__Magerman nsubj_used_ det_algorithm_the amod_algorithm_clustering dobj_used_algorithm prep_algorithm_of pobj_of_ cc__et conj__al ' +o,4691,'According to the Bayes Rule , the problem is transformed into the noisy channel model paradigm , where the translation is the maximum a posteriori solution of a distribution for a channel target text given a channel source text and a prior distribution for the channel source text ',Brown,'prep_transformed_According dep_According_to det_Rule_the nn_Rule_Bayes pobj_to_Rule det_problem_the nsubjpass_transformed_problem auxpass_transformed_is prep_transformed_into det_paradigm_the amod_paradigm_noisy nn_paradigm_channel nn_paradigm_model pobj_into_paradigm advmod_solution_where det_translation_the nsubj_solution_translation cop_solution_is det_solution_the amod_solution_maximum det_solution_a nn_solution_posteriori rcmod_paradigm_solution prep_solution_of det_distribution_a pobj_of_distribution prep_distribution_for det_text_a nn_text_channel nn_text_target pobj_for_text partmod_text_given det_source_a nn_source_channel iobj_given_source dobj_given_text cc_text_and det_distribution_a amod_distribution_prior conj_text_distribution prep_given_for det_text_the nn_text_channel nn_text_source pobj_for_text ' +p,4692,'The maximum entropy model provides us with a well-founded framework for this purpose , which has been extensively used in natural lan guage processing tasks ranging from part-ofspeech tagging to machine translation ',Berger,'det__The amod__maximum amod__entropy nn__model nsubj_provides_ dobj_provides_us prep_provides_with det_framework_a amod_framework_well-founded pobj_with_framework prep_framework_for det_purpose_this pobj_for_purpose nsubjpass_used_which aux_used_has auxpass_used_been advmod_used_extensively rcmod_purpose_used prep_used_in amod_tasks_natural nn_tasks_lan nn_tasks_guage nn_tasks_processing pobj_in_tasks partmod_tasks_ranging prep_ranging_from amod_tagging_part-ofspeech pobj_from_tagging prep_ranging_to nn_translation_machine pobj_to_translation ' +o,4693,'Sometimes , due to data sparseness and\\/or limitations in the machine learning paradigm used , we need to extract features from the available representation in a manner that profoundly changes the representation -LRB- as is done in bilexical parsing -RRB- ',Collins,'advmod_need_Sometimes dep_to_due prep_need_to pobj_to_data rcmod_data_sparseness amod_limitations_and\\/or dobj_sparseness_limitations prep_sparseness_in det_machine_the pobj_in_machine partmod_machine_learning dobj_learning_paradigm partmod_paradigm_used nsubj_need_we aux_extract_to xcomp_need_extract dobj_extract_features prep_extract_from det_representation_the amod_representation_available pobj_from_representation prep_representation_in det_manner_a pobj_in_manner nsubj_changes_that advmod_changes_profoundly rcmod_manner_changes det_representation_the dobj_changes_representation mark_done_as auxpass_done_is dep_representation_done prep_done_in amod__bilexical amod__parsing pobj_in_ ' +n,4694,'1 Introduction The most widely applied training procedure for statistical machine translation IBM model 4 unsupervised training followed by post-processing with symmetrization heuristics yields low quality word alignments ',Och,'num_Introduction_1 det_most_The nsubj_applied_most advmod_applied_widely dep_Introduction_applied nn_procedure_training dobj_applied_procedure prep_applied_for amod_machine_statistical pobj_for_machine nn_training_translation nn_training_IBM nn_training_model num_training_4 amod_training_ nn_training_unsupervised nsubj_followed_training dep_Introduction_followed prep_followed_by pcomp_by_post-processing prep_post-processing_with nn_heuristics_symmetrization pobj_with_heuristics num_yields_ measure_low_yields amod_alignments_low nn_alignments_quality nn_alignments_word dobj_followed_alignments ' +p,4695,'Here , we use the more established ROUGE-W measure instead ',Lin,'advmod_use_Here nsubj_use_we det_measure_the amod_measure_more amod_measure_established amod_measure_ROUGE-W dobj_use_measure partmod_measure_ advmod__instead ' +o,4696,'Accurate automatic analysis of these aspects of language will augment existing research in the fields of sentiment andsubjectivityanalysis -LRB- Wiebeetal ',Pang,'amod_analysis_Accurate amod_analysis_automatic nsubj_augment_analysis prep_analysis_of det_aspects_these pobj_of_aspects prep_aspects_of pobj_of_language aux_augment_will amod_research_existing nsubj_andsubjectivityanalysis_research prep_research_in det_fields_the pobj_in_fields prep_fields_of pobj_of_sentiment advmod_andsubjectivityanalysis_ ccomp_augment_andsubjectivityanalysis dobj_andsubjectivityanalysis_Wiebeetal ' +p,4697,' discussed efficient implementation ',Och,'nsubj_discussed_ amod_implementation_efficient dobj_discussed_implementation ' +o,4698,'In order to extract the linguistic features necessary for the models , all sentences containing the target word were automatically part-of-speech-tagged using a maximum entropy tagger and parsed using the Collins parser ',Collins,'mark_extract_In dep_extract_order aux_extract_to dep_part-of-speech-tagged_extract det_features_the amod_features_linguistic nsubj_necessary_features xcomp_extract_necessary prep_necessary_for det_models_the pobj_for_models det_sentences_all nsubjpass_part-of-speech-tagged_sentences partmod_sentences_containing det_word_the nn_word_target dobj_containing_word auxpass_part-of-speech-tagged_were advmod_part-of-speech-tagged_automatically xcomp_part-of-speech-tagged_using det__a amod__maximum amod__entropy nn__tagger dobj_using_ cc_part-of-speech-tagged_and conj_part-of-speech-tagged_parsed xcomp_parsed_using det_parser_the nn_parser_Collins dobj_using_parser ' +o,4699,'1999 -RRB- , OpenCCG and XLE , or created semi-automatically , or fully automatically extracted from annotated corpora , like the HPSG , LFG and CCG resources derived from the Penn-II Treebank -LRB- PTB -RRB- ',Marcus,'dep__1999 nn__OpenCCG cc__and conj__XLE dep__ cc__or conj__created amod__semi-automatically dobj__ cc__or advmod__fully advmod_extracted_automatically conj__extracted prep_extracted_from amod_corpora_annotated pobj_from_corpora prep_extracted_like det__the nn__HPSG nn__ appos__LFG pobj_like_ cc__and nn_resources_CCG nn_resources_ conj__resources partmod__derived prep_derived_from det_Treebank_the nn_Treebank_Penn-II pobj_from_Treebank dep__PTB ' +o,4700,'The likelihood ratio is obtained by treating word and Ic as a bigram and computed with the formula in ',Dunning,'det_ratio_The nn_ratio_likelihood nsubjpass_obtained_ratio auxpass_obtained_is prep_obtained_by pcomp_by_treating dobj_treating_word cc_word_and conj_word_Ic prep_treating_as det_bigram_a pobj_as_bigram cc_obtained_and conj_obtained_computed prep_computed_with det_formula_the pobj_with_formula prep_computed_in ' +o,4701,'The text was split at the sentence level , tokenized and PoS tagged , in the style of the Wall Street Journal Penn TreeBank ',Marcus,'det_text_The nsubjpass_split_text auxpass_split_was prep_split_at det_level_the nn_level_sentence pobj_at_level conj_level_tokenized cc_level_and nn_tagged_PoS conj_level_tagged prep_split_in det_style_the pobj_in_style prep_style_of det_TreeBank_the nn_TreeBank_Wall nn_TreeBank_Street nn_TreeBank_Journal nn_TreeBank_Penn pobj_of_TreeBank ' +p,4702,'The current state-of-the-art is to optimize these parameters with respect to the final evaluation criterion ; this is the so-called minimum error rate training ',Och,'det_state-of-the-art_The amod_state-of-the-art_current nsubj_is_state-of-the-art aux_optimize_to xcomp_is_optimize det_parameters_these dobj_optimize_parameters prep_optimize_with pobj_with_respect prep_optimize_to det_criterion_the amod_criterion_final nn_criterion_evaluation pobj_to_criterion nsubj_training_this cop_training_is det_training_the amod_training_so-called amod_training_minimum nn_training_error nn_training_rate parataxis_is_training ' +o,4703,'The difference in accuracy between a SVM model applied to RRR dataset (RRR-basic experiment) and the same experiment applied to TB2 dataset (TB2278 Description Accuracy Data Extra Supervision Always noun 55.0 RRR Most likely for each P 72.19 RRR Most likely for each P 72.30 TB2 Most likely for each P 81.73 FN Average human, headwords (Ratnaparkhi et al. , 1994) 88.2 RRR Average human, whole sentence (Ratnaparkhi et al. , 1994) 93.2 RRR Maximum Likelihood-based (Hindle and Rooth, 1993) 79.7 AP Maximum entropy, words (Ratnaparkhi et al. , 1994) 77.7 RRR Maximum entropy, words & classes (Ratnaparkhi et al. , 1994) 81.6 RRR Decision trees (Ratnaparkhi et al. , 1994) 77.7 RRR Transformation-Based Learning (Brill and Resnik, 1994) 81.8 WordNet Maximum-Likelihood based (Collins and Brooks, 1995) 84.5 RRR Maximum-Likelihood based (Collins and Brooks, 1995) 86.1 TB2 Decision trees & WSD (Stetina and Nagao, 1997) 88.1 RRR WordNet Memory-based Learning (Zavrel et al. , 1997) 84.4 RRR LexSpace Maximum entropy, unsupervised (Ratnaparkhi, 1998) 81.9 Maximum entropy, supervised (Ratnaparkhi, 1998) 83.7 RRR Neural Nets (Alegre et al. , 1999) 86.0 RRR WordNet Boosting (Abney et al. , 1999) 84.4 RRR Semi-probabilistic (Pantel and Lin, 2000) 84.31 RRR Maximum entropy, ensemble (McLauchlan, 2001) 85.5 RRR LSA SVM (Vanschoenwinkel and Manderick, 2003) 84.8 RRR Nearest-neighbor (Zhao and Lin, 2004) 86.5 RRR DWS FN dataset, w/o semantic features (FN-best-no-sem) 91.79 FN PR-WWW FN dataset, w/ semantic features (FN-best-sem) 92.85 FN PR-WWW TB2 dataset, best feature set (TB2-best) 93.62 TB2 PR-WWW Table 5: Accuracy of PP-attachment ambiguity resolution (our results in bold) basic experiment) is 2.9\%.',Marcus,'' +o,4704,' has proposed automatically augmenting a small set of experimenter-supplied seed collocations -LRB- eg , manufacturing plant and plant life for two different senses of the noun plant -RRB- into a much larger set of training materials ',Yarowsky,'nsubj_proposed_ aux_proposed_has advmod_augmenting_automatically ccomp_proposed_augmenting det_set_a amod_set_small dobj_augmenting_set prep_set_of amod_collocations_experimenter-supplied nn_collocations_seed pobj_of_collocations nn_plant_eg amod_plant_manufacturing dep_set_plant cc_plant_and nn_life_plant conj_plant_life prep_plant_for num_senses_two amod_senses_different pobj_for_senses prep_senses_of det_plant_the nn_plant_noun pobj_of_plant prep_augmenting_into det_set_a advmod_larger_much amod_set_larger pobj_into_set prep_set_of nn_materials_training pobj_of_materials ' +o,4705,'We evaluate accuracy performance using two automatic metrics : an identity metric , ID , which measures the percent of sentences recreated exactly , and BLEU , which gives the geometric average of the number of uni - , bi - , tri - , and four-grams recreated exactly ',Papineni,'nsubj_evaluate_We nn_performance_accuracy dobj_evaluate_performance partmod_performance_using num_metrics_two amod_metrics_automatic dobj_using_metrics det_metric_an nn_metric_identity dep_metrics_metric appos_metrics_ID nsubj_measures_which rcmod_metrics_measures det_percent_the dobj_measures_percent prep_percent_of pobj_of_sentences partmod_sentences_recreated advmod_recreated_exactly cc_evaluate_and nn__BLEU nsubj_tri_ nsubj_gives_which rcmod__gives det_average_the amod_average_geometric dobj_gives_average dep_gives_of det_number_the pobj_of_number prep_number_of pobj_of_uni dep_of_bi conj_evaluate_tri cc_evaluate_and nsubj_recreated_four-grams conj_evaluate_recreated advmod_recreated_exactly ' +o,4706,'As shown by , the Single Malt parser tends to suffer from two problems : error propagation due to the deterministic parsing strategy , typicallyaffectinglongdependenciesmorethan short ones , and low precision on dependencies originating in the artificial root node due to fragmented parses9 The question is which of these problems is alleviatedbythemultipleviewsgivenbythecomponent parsers in the Blended system ',McDonald,'mark_shown_As advcl_tends_shown prep_shown_by pobj_by_ det_parser_the nn_parser_Single nn_parser_Malt nsubj_tends_parser aux_suffer_to xcomp_tends_suffer prep_suffer_from num_problems_two pobj_from_problems nn_propagation_error dep_due_propagation parataxis_tends_due prep_due_to det_strategy_the amod_strategy_deterministic nn_strategy_parsing pobj_to_strategy amod_ones_typicallyaffectinglongdependenciesmorethan amod_ones_short appos_strategy_ones cc_tends_and amod_precision_low nsubj_node_precision prep_precision_on pobj_on_dependencies partmod_dependencies_originating prep_originating_in det_root_the amod_root_artificial pobj_in_root conj_tends_node amod_parses9_due dep_due_to dep_due_fragmented dobj_node_parses9 det_question_The nsubj_is_question dep_tends_is nsubj_parsers_which prep_which_of det_problems_these pobj_of_problems cop_parsers_is amod_parsers_alleviatedbythemultipleviewsgivenbythecomponent ccomp_is_parsers prep_parsers_in det_system_the nn_system_Blended pobj_in_system ' +o,4707,'The definitions of the phrase and lexical translation probabilities are as follows ',Koehn,'det_definitions_The nsubj_follows_definitions prep_definitions_of det_phrase_the pobj_of_phrase cc_phrase_and amod_probabilities_lexical nn_probabilities_translation conj_phrase_probabilities cop_follows_are advmod_follows_as ' +o,4708,'The Pearson correlation is calculated over these ten pairs ',Papineni,'det_correlation_The nn_correlation_Pearson nsubjpass_calculated_correlation auxpass_calculated_is prep_calculated_over det_pairs_these amod_pairs_ten pobj_over_pairs ' +o,4709,'They mention that the resulting shallow parse tags are somewhat different than those used by , but that they found no significant accuracy differences in training on either set ',Ramshaw,'nsubj_mention_They dep_different_that det_tags_the amod_tags_resulting nn_tags_shallow nn_tags_parse nsubj_different_tags cop_different_are advmod_different_somewhat dep_mention_different prep_different_than pobj_than_those partmod_those_used prep_used_by pobj_by_ cc_different_but dep_found_that nsubj_found_they conj_different_found det_differences_no amod_differences_significant nn_differences_accuracy dobj_found_differences prep_found_in pobj_in_training prep_found_on det_set_either pobj_on_set ' +o,4710,'Firstly , rather than induce millions of xRS rules from parallel data , we extract phrase pairs in the standard way and associate with each phrase-pair a set of target language syntactic structures based on supertag sequences ',Och,'advmod_extract_Firstly dep_than_rather prep_extract_than pcomp_than_induce dobj_induce_millions prep_millions_of nn_rules_xRS pobj_of_rules prep_induce_from amod_data_parallel pobj_from_data nsubj_extract_we nn_pairs_phrase dobj_extract_pairs prep_extract_in det__the amod__standard nn__way pobj_in_ cc__and conj__associate prep__with det_phrase-pair_each pobj_with_phrase-pair det_set_a dep_phrase-pair_set prep__of nn_structures_target nn_structures_language amod_structures_syntactic pobj_of_structures prep_extract_based dep_based_on amod_sequences_supertag pobj_on_sequences ' +o,4711,'Previous publications on Meteor have described the details underlying the metric and have extensively compared its performance with Bleu and several other MT evaluation metrics ',Banerjee,'amod_publications_Previous nsubj_described_publications prep_publications_on nn__Meteor pobj_on_ aux_described_have det_details_the dobj_described_details xcomp_described_underlying det_metric_the dobj_underlying_metric cc_described_and aux_compared_have advmod_compared_extensively conj_described_compared poss_performance_its dobj_compared_performance prep_compared_with pobj_with_Bleu cc_Bleu_and amod_metrics_several amod_metrics_other nn_metrics_MT nn_metrics_evaluation conj_Bleu_metrics ' +o,4712,'The Gaussian prior -LRB- ie , the P k a 2 k = 7 2 k penalty -RRB- has been found in practice to be very effective in combating overfitting of the parameters to the training data ',Collins,'det_Gaussian_The advmod_found_prior dep_k_ie det_P_the nsubj_k_P dep_prior_k det_k_a num_k_2 nsubj_=_k ccomp_k_= num_2_7 iobj_=_2 nn_penalty_k dobj_=_penalty aux_found_has auxpass_found_been dep_Gaussian_found prep_found_in pobj_in_practice aux_effective_to cop_effective_be advmod_effective_very xcomp_found_effective prep_effective_in pcomp_in_combating dobj_combating_overfitting prep_overfitting_of det_parameters_the pobj_of_parameters prep_combating_to det__the nn__training nn__data pobj_to_ ' +o,4713,'Finally , the parameters i of the log-linear model -LRB- 18 -RRB- are learned by minimumerror-rate training , which tries to set the parameters so as to maximize the BLEU score of a development set ',Papineni,'advmod_learned_Finally nsubjpass_learned_the nn_the_parameters dep_the_i prep_the_of det_model_the amod_model_log-linear pobj_of_model appos_model_18 auxpass_learned_are prep_learned_by amod__minimumerror-rate nn__training pobj_by_ nsubj_tries_which rcmod__tries aux_set_to xcomp_tries_set det_parameters_the dobj_set_parameters advmod_as_so advmod_set_as aux_maximize_to dep_as_maximize det__the amod__BLEU nn__score dobj_maximize_ prep__of det_set_a nn_set_development pobj_of_set ' +o,4714,'43 Using Unlabeled Data for Parsing Recent studies on parsing indicate that the use of unlabeled data by self-training can help parsing on the WSJ data , even when labeled data is relatively large ',McClosky,'nsubj_indicate_43 partmod_43_Using nn_Data_Unlabeled dobj_Using_Data prep_Data_for pcomp_for_Parsing amod_studies_Recent dobj_Parsing_studies prep_Parsing_on pobj_on_parsing complm_help_that det_use_the nsubj_help_use prep_use_of amod_data_unlabeled pobj_of_data prep_data_by pobj_by_self-training aux_help_can ccomp_indicate_help xcomp_help_parsing prep_parsing_on det_data_the nn_data_WSJ pobj_on_data advmod_labeled_even advmod_labeled_when ccomp_help_labeled nsubj_large_data cop_large_is advmod_large_relatively ccomp_labeled_large ' +n,4715,'This is in contrast to purely statistical systems , which are difficult to inspect and modify ',Brown,'nsubj_is_This prep_is_in dep_in_contrast dep_in_to advmod_systems_purely amod_systems_statistical pobj_in_systems amod_systems_ nsubj_difficult_which cop_difficult_are rcmod_systems_difficult aux_inspect_to xcomp_difficult_inspect cc_inspect_and conj_inspect_modify ' +o,4716,'unlabeled R 100 \% 20\\/08\\/199605 \\/ 08\\/1997 -LRB- 351 days -RRB- 50 \% 20\\/08\\/199617 \\/ 02\\/1997 -LRB- 182 days -RRB- 10 \% 20\\/08\\/199624 \\/ 09\\/1996 -LRB- 36 days -RRB- labeled WSJ 50 \% sections 0012 25 \% lines 1 292960 5 \% lines 1 58284 1 \% lines 1 11720 -LRB- 500 sentences -RRB- 005 \% lines 1 611 -LRB- 23 sentences -RRB- Table 1 : Corpora used for the experiments : unlabeled Reuters -LRB- R -RRB- corpus for attachment statistics , labeled Penn treebank -LRB- WSJ -RRB- for training the parser ',Collins,'nsubj_R_unlabeled num_\%_100 iobj_R_\% num_\\/_20\\/08\\/199605 dobj_R_\\/ num_\\/_08\\/1997 num_days_351 appos_\\/_days dep_\%_50 amod_\\/_\% num_\\/_20\\/08\\/199617 dobj_R_\\/ num_\\/_02\\/1997 num_days_182 appos_\\/_days dep_\%_10 amod_\\/_\% number_\\/_20\\/08\\/199624 num_09\\/1996_\\/ dobj_R_09\\/1996 num_days_36 appos_WSJ_days nn_WSJ_labeled dobj_R_WSJ dep_\%_50 amod_WSJ_\% number_0012_sections num__0012 dobj_R_ dep_\%_25 amod_lines_\% dobj_R_lines number_292960_1 num__292960 dobj_R_ dep_\%_5 amod_lines_\% dobj_R_lines number_58284_1 num__58284 dobj_R_ dep_\%_1 amod_lines_\% dobj_R_lines number_11720_1 dobj_R_11720 num_sentences_500 appos_11720_sentences number_\%_005 num_lines_\% dobj_R_lines number_611_1 dobj_R_611 num_sentences_23 appos_611_sentences prep_R_Table pobj_Table_1 dobj_R_Corpora partmod_Corpora_used prep_used_for det_experiments_the pobj_for_experiments nn_corpus_unlabeled nn_corpus_Reuters appos_corpus_R dep_Corpora_corpus prep_corpus_for nn_statistics_attachment pobj_for_statistics partmod_statistics_labeled nn_treebank_Penn dobj_labeled_treebank abbrev_treebank_WSJ prep_labeled_for pcomp_for_training det_parser_the amod_parser_ dobj_training_parser ' +n,4717,'In addition , the semi-supervised Morce performs -LRB- on single CPU and development data set -RRB- 77 times faster than the combination and 23 times faster than ',Shen,'prep_performs_In pobj_In_addition det_Morce_the amod_Morce_semi-supervised nsubj_performs_Morce dep_performs_on amod_CPU_single pobj_on_CPU cc_CPU_and nn_data_development conj_CPU_data partmod_CPU_set num_times_77 dep_faster_times dep_than_faster prep_performs_than det_combination_the pobj_than_combination cc_combination_and num_times_23 conj_combination_times advmod_performs_faster dep_faster_than ' +o,4718,'The future score is based on the source-language words that are still to be translatedthis can be directly inferred from the items bit-stringthis is similar to the use of future scores in Pharoah , and in fact we use Pharoahs future scores in our model ',Koehn,'det_score_The amod_score_future nsubjpass_based_score auxpass_based_is prep_based_on det_words_the amod_words_source-language pobj_on_words nsubj_are_that rcmod_words_are advmod_are_still aux_translatedthis_to cop_translatedthis_be xcomp_are_translatedthis aux_inferred_can auxpass_inferred_be advmod_inferred_directly ccomp_translatedthis_inferred prep_inferred_from det_items_the pobj_from_items nsubj_similar_bit-stringthis cop_similar_is dep_based_similar prep_similar_to det_use_the pobj_to_use prep_use_of amod_scores_future pobj_of_scores prep_scores_in nn__Pharoah pobj_in_ cc_similar_and prep_use_in pobj_in_fact nsubj_use_we conj_similar_use nn_scores_Pharoahs amod_scores_future dobj_use_scores prep_scores_in poss_model_our pobj_in_model ' +o,4719,'For comparison purposes , we also computed the value of R 2 for fluency using the BLEU score formula given in , for the 7 systems using the same one reference , and we obtained a similar value , 7852 \% ; computing the value of R 2 for fluency using the BLEU scores computed with all 4 references available yielded a lower value for R 2 , 6496 \% , although BLEU scores obtained with multiple references are usually considered more reliable ',Papineni,'prep_computed_For nn_purposes_comparison pobj_For_purposes nsubj_computed_we advmod_computed_also det_value_the dobj_computed_value prep_value_of pobj_of_R num_R_2 prep_computed_for advmod_using_fluency pcomp_for_using det_formula_the amod_formula_BLEU nn_formula_score dobj_using_formula prep_using_given dep_given_in pobj_in_ prep_computed_for det_systems_the num_systems_7 pobj_for_systems partmod_systems_using det_reference_the amod_reference_same num_reference_one dobj_using_reference cc_computed_and nsubj_obtained_we conj_computed_obtained det_value_a amod_value_similar nsubj_value_value num_\%_7852 appos_value_\% partmod_value_computing det_value_the dobj_computing_value prep_value_of pobj_of_R num_R_2 prep_computing_for advmod_using_fluency pcomp_for_using det_scores_the amod_scores_BLEU dobj_using_scores partmod_scores_computed prep_computed_with pobj_with_all num_references_4 measure_available_references amod_all_available dep_value_yielded det_value_a amod_value_lower ccomp_obtained_value prep_value_for pobj_for_R num_R_2 num_\%_6496 appos_R_\% mark_considered_although amod_scores_BLEU nsubjpass_considered_scores partmod_scores_obtained prep_obtained_with amod_references_multiple pobj_with_references auxpass_considered_are advmod_considered_usually advcl_value_considered advmod_reliable_more dep_considered_reliable ' +o,4720,'The current version of the dataset gives semantic tags for the same sentencesas inthe PennTreebank , whichareexcerptsfromtheWallStreetJournal ',Marcus,'det_version_The amod_version_current nsubj_gives_version prep_version_of det_dataset_the pobj_of_dataset amod_tags_semantic dobj_gives_tags prep_gives_for det__the amod__same amod__sentencesas nn__inthe nn__PennTreebank pobj_for_ dep_gives_whichareexcerptsfromtheWallStreetJournal ' +o,4721,'This probability is computed using IBMs Model 1 : P -LRB- Q A -RRB- = productdisplay qQ P -LRB- q A -RRB- -LRB- 3 -RRB- P -LRB- q A -RRB- = -LRB- 1 -RRB- Pml -LRB- q A -RRB- + Pml -LRB- q C -RRB- -LRB- 4 -RRB- Pml -LRB- q A -RRB- = summationdisplay aA -LRB- T -LRB- q a -RRB- Pml -LRB- a A -RRB- -RRB- -LRB- 5 -RRB- where the probability that the question term q is generated from answer A , P -LRB- q A -RRB- , is smoothed using the prior probability that the term q is generated from the entire collection of answers C , Pml -LRB- q C -RRB- ',Marcus,'det_probability_This nsubjpass_computed_probability auxpass_computed_is xcomp_computed_using nn__IBMs nn__Model num__1 dobj_using_ nsubj_P_P nn_A_Q appos_P_A dep_P_= advmod_P_productdisplay nn_P_qQ parataxis_computed_P nn_A_q appos_P_A appos_P_3 dobj_P_P nn_A_q appos_P_A nn_Pml_= appos_Pml_1 dobj_P_Pml nn_A_q appos_Pml_A nn_Pml_+ dobj_P_Pml nn_C_q appos_Pml_C appos_Pml_4 dobj_P_Pml nn_A_q appos_Pml_A amod_Pml_= nn_Pml_summationdisplay nn_Pml_aA nn_Pml_T nn_Pml_q nn_Pml_a dobj_P_Pml det_A_a appos_Pml_A appos_Pml_5 advmod_smoothed_where det_probability_the nsubjpass_smoothed_probability complm_generated_that det_q_the nn_q_question nn_q_term nsubjpass_generated_q auxpass_generated_is dep_probability_generated prep_generated_from nn_A_answer pobj_from_A appos_A_P nn_A_q appos_P_A auxpass_smoothed_is advcl_P_smoothed dep_probability_using det_probability_the amod_probability_prior xcomp_smoothed_probability complm_generated_that det_q_the nn_q_term nsubjpass_generated_q auxpass_generated_is ccomp_probability_generated prep_generated_from det_collection_the amod_collection_entire pobj_from_collection prep_collection_of nn_C_answers pobj_of_C appos_C_Pml nn_C_q appos_Pml_C ' +o,4722,'Most SMT models try to model word-to-word corresl -RRB- ondences between source and target words using an alignment nmpl -RRB- ing from source l -RRB- osition j to target position i = aj ',Brown,'amod_models_Most nn_models_SMT nsubj_try_models advmod_try_ aux_model_to xcomp_try_model amod_ondences_word-to-word amod_ondences_corresl dobj_model_ondences prep_ondences_between nn_words_source cc_source_and conj_source_target pobj_between_words xcomp_model_using det_ing_an amod_ing_alignment nn_ing_nmpl dobj_using_ing prep_using_from pobj_from_source dep_j_l nsubj_j_osition dep_try_j aux_target_to xcomp_j_target nn_i_position nsubj_aj_i dep_aj_= ccomp_target_aj ' +o,4723,'One possible use for this technique is for parser adaptation initially training the parser on one type of data for which hand-labeled trees are available -LRB- eg , Wall Street Journal -RRB- and then self-training on a second type of data in order to adapt the parser to the second domain ',Marcus,'num_use_One amod_use_possible nsubj_is_use prep_use_for det_technique_this pobj_for_technique prep_is_for nn_adaptation_parser nsubj_training_adaptation advmod_training_initially pcomp_for_training det_parser_the dobj_training_parser prep_training_on num_type_one pobj_on_type prep_type_of pobj_of_data rel_available_for pobj_for_which amod_trees_hand-labeled nsubj_available_trees cop_available_are rcmod_data_available appos_data_eg nn__Wall nn__Street nn__Journal dep_eg_ cc_training_and advmod_training_then conj_training_self-training prep_self-training_on det_type_a amod_type_second pobj_on_type prep_type_of pobj_of_data prep_data_in pobj_in_order aux_adapt_to xcomp_self-training_adapt det_parser_the dobj_adapt_parser prep_adapt_to det_domain_the amod_domain_second pobj_to_domain ' +o,4724,'As the most concise definition we take the first sentence of each article , following ',Kazama,'prep_take_As det_definition_the advmod_concise_most amod_definition_concise pobj_As_definition nsubj_take_we det_sentence_the amod_sentence_first dobj_take_sentence prep_sentence_of det_article_each pobj_of_article partmod_sentence_following ' +o,4725,'It is true that various term extraction systems have been developed , such as Xtract , Termight , and TERMS among others -LRB- cf ',Smadja,'nsubj_true_It cop_true_is complm_developed_that amod_systems_various nn_systems_term nn_systems_extraction nsubjpass_developed_systems aux_developed_have auxpass_developed_been ccomp_true_developed prep_developed_such advmod_Xtract_as dep_such_Xtract dobj_Xtract_ nn__Termight conj__ cc__and nn__TERMS conj__ prep__among pobj_among_others prep_Xtract_cf ' +o,4726,' show that Bloomier filters can be used to create perfect hash functions for language models ',Talbot,'nsubj_show_ complm_used_that nn__Bloomier nn__filters nsubjpass_used_ aux_used_can auxpass_used_be ccomp_show_used aux_create_to purpcl_used_create amod_functions_perfect nn_functions_hash dobj_create_functions prep_create_for nn_models_language pobj_for_models ' +p,4727,'Compared to a basic treebank grammar , the grammars of highaccuracy parsers weaken independence assumptions by splitting grammar symbols and rules with either lexical or nonlexical conditioning information ',Matsuzaki,'prep_weaken_Compared dep_Compared_to det__a amod__basic nn__treebank nn__grammar pobj_to_ det_grammars_the appos__grammars prep_grammars_of amod_parsers_highaccuracy pobj_of_parsers nn_assumptions_independence nsubj_weaken_assumptions prep_assumptions_by pcomp_by_splitting nn_symbols_grammar dobj_splitting_symbols cc_symbols_and conj_symbols_rules prep_splitting_with det_information_either amod_information_lexical amod_information_ cc__or conj__nonlexical amod_information_ nn_information_conditioning pobj_with_information ' +o,4728,' and -RRB- , we generate training instances as follows : a positive instance is created for each anaphoric NP , NPj , and its closest antecedent , NPi ; and a negative instance is created for NPj paired with each of the intervening NPs , NPi +1 , NPi +2 , , NPj1 ',Ponzetto,'nsubjpass_created_ cc__and conj__ nsubj_generate_we dep__generate nn_instances_training dobj_generate_instances mark_follows_as advcl_generate_follows det_instance_a amod_instance_positive nsubjpass_created_instance auxpass_created_is parataxis_generate_created prep_created_for det_NP_each amod_NP_anaphoric pobj_for_NP conj_NP_NPj cc_NP_and poss_antecedent_its amod_antecedent_closest conj_NP_antecedent conj__NPi cc__and det_instance_a amod_instance_negative conj__instance auxpass_created_is prep_created_for pobj_for_NPj partmod_NPj_paired prep_paired_with pobj_with_each prep_each_of det_NPs_the amod_NPs_intervening pobj_of_NPs appos_NPs_NPi num_NPi_+1 appos_NPi_NPi num_NPi_+2 appos_NPi_NPj1 ' +o,4729,' also work with arguments ',Fujii,'nsubj_work_ advmod_work_also prep_work_with pobj_with_arguments ' +o,4730,'the Wall Street Journal -LRB- WSJ -RRB- sections of the Penn Treebank as training set , tests on BROWN Sections typically result in a 6-8 \% drop in labeled attachment scores , although the average sentence length is much shorter in BROWN than that in WSJ ',Marcus,'det_Journal_the nn_Journal_Wall nn_Journal_Street abbrev_Journal_WSJ nsubj_set_sections prep_sections_of det__the nn__Penn nn__Treebank pobj_of_ prep__as pobj_as_training dep_Journal_set nsubj_result_tests prep_tests_on nn_Sections_BROWN pobj_on_Sections advmod_result_typically parataxis_set_result prep_result_in det_drop_a num_drop_6-8 nn_drop_\% pobj_in_drop prep_drop_in amod_scores_labeled nn_scores_attachment pobj_in_scores mark_shorter_although det_length_the amod_length_average nn_length_sentence nsubj_shorter_length cop_shorter_is advmod_shorter_much advcl_set_shorter prep_shorter_in pobj_in_BROWN prep_shorter_than pobj_than_that prep_that_in pobj_in_WSJ ' +n,4731,'In a recent study by , nonlocal information is encoded using an independence model , and the inference is performed by Gibbs sampling , which enables us to use a stateof-the-art factored model and carry out training efficiently , but inference still incurs a considerable computational cost ',Finkel,'prep_encoded_In det_study_a amod_study_recent pobj_In_study prep_study_by pobj_by_ amod_information_nonlocal nsubjpass_encoded_information auxpass_encoded_is dep_incurs_encoded xcomp_encoded_using det_model_an nn_model_independence dobj_using_model cc_encoded_and det_inference_the nsubjpass_performed_inference auxpass_performed_is conj_encoded_performed prep_performed_by pobj_by_Gibbs amod_Gibbs_sampling nsubj_enables_which rcmod_Gibbs_enables nsubj_use_us aux_use_to xcomp_enables_use det_model_a amod_model_stateof-the-art amod_model_factored dobj_use_model cc_use_and conj_use_carry prt_carry_out dobj_carry_training advmod_carry_efficiently cc_enables_but conj_enables_inference advmod_inference_still det_cost_a amod_cost_considerable amod_cost_computational nsubj_incurs_cost ' +o,4732,'On the positive side , recent work exploring the automaticbinarizationofsynchronousgrammars has indicated that non-binarizable constructions seem to be relatively rare in practice ',Zhang,'prep_indicated_On det_side_the amod_side_positive pobj_On_side amod_work_recent nsubj_indicated_work partmod_work_exploring det__the amod__automaticbinarizationofsynchronousgrammars dobj_exploring_ aux_indicated_has complm_seem_that amod_constructions_non-binarizable nsubj_seem_constructions ccomp_indicated_seem aux_rare_to cop_rare_be advmod_rare_relatively xcomp_seem_rare prep_rare_in pobj_in_practice ' +o,4733,'In contrast , generative models are trained to maximize the joint probability of the training data , which is 1 used transformation-based learning , which for the present purposes can be tought of as a classi cation-based method ',Ramshaw,'prep_trained_In pobj_In_contrast amod_models_generative nsubj_trained_models cop_trained_are aux_maximize_to xcomp_trained_maximize det_probability_the amod_probability_joint dobj_maximize_probability prep_probability_of det_data_the nn_data_training pobj_of_data nsubj__which cop__is num__1 rcmod_data_ partmod__used amod__transformation-based nn__learning dobj_used_ nsubj_tought_which prep_tought_for det_purposes_the amod_purposes_present pobj_for_purposes aux_tought_can cop_tought_be rcmod__tought prep_tought_of dep_of_as det_method_a amod_method_classi amod_method_cation-based pobj_as_method ' +o,4734,'We then built separate English-to-Spanish and Spanish-to-English directed word alignments using IBM model 4 , combined them using the intersect + grow heuristic , and extracted phrase-level translation pairs of maximum length 7 using the alignment template approach ',Brown,'nsubj_built_We advmod_built_then dobj_built_separate dep_separate_English-to-Spanish cc_separate_and amod_alignments_Spanish-to-English amod_alignments_directed nn_alignments_word conj_separate_alignments xcomp_built_using nn_model_IBM nsubj__model num__4 xcomp_using_ partmod__combined nsubj_using_them dep_combined_using det__the amod__intersect cc_intersect_+ conj_intersect_grow amod__heuristic dobj_using_ cc_built_and conj_built_extracted amod_pairs_phrase-level nn_pairs_translation dobj_extracted_pairs prep_pairs_of nn_length_maximum pobj_of_length num_length_7 xcomp_built_using det__the amod__alignment amod__template nn__approach dobj_using_ ' +o,4735,'The second baseline is our implementation of the relevant part of the Wikipedia extraction in , taking the first noun after a be verb in the definition sentence , denoted as WikiBL ',Kazama,'det_baseline_The amod_baseline_second nsubj_implementation_baseline cop_implementation_is poss_implementation_our prep_implementation_of det_part_the amod_part_relevant pobj_of_part prep_part_of det_extraction_the nn_extraction_Wikipedia pobj_of_extraction prep_part_in pobj_in_ dep_noun_taking det_noun_the amod_noun_first partmod_part_noun mark_verb_after nsubj_verb_a cop_verb_be advcl_noun_verb prep_verb_in det_sentence_the nn_sentence_definition pobj_in_sentence partmod_part_denoted prep_denoted_as pobj_as_WikiBL ' +o,4736,'Most previous work on compositionality of MWEs either treat them as collocations , or examine the distributional similarity between the expression and its constituents ',Smadja,'amod_work_Most amod_work_previous nsubj_treat_work prep_work_on pobj_on_compositionality prep_compositionality_of pobj_of_MWEs advmod_treat_either dobj_treat_them mark__as nsubj__collocations advcl_treat_ cc__or conj__examine det_similarity_the amod_similarity_distributional dobj_examine_similarity prep_similarity_between det_expression_the pobj_between_expression cc_expression_and poss__its nn__constituents conj_expression_ ' +p,4737,'Weight averaging was also employed , which helped improve performance ',Collins,'nn_averaging_Weight nsubjpass_employed_averaging auxpass_employed_was advmod_employed_also dobj_employed_ nsubj_helped_which rcmod__helped xcomp_helped_improve dobj_improve_performance ' +o,4738,'In , a small set of sample results are presented ',Hindle,'prep_presented_In pobj_In_ det_set_a amod_set_small nsubjpass_presented_set prep_set_of nn_results_sample pobj_of_results auxpass_presented_are ' +o,4739,'It extracts all consistent phrase pairs from word-aligned bitext ',Koehn,'nsubj_extracts_It det_pairs_all amod_pairs_consistent nn_pairs_phrase dobj_extracts_pairs prep_extracts_from amod_bitext_word-aligned pobj_from_bitext ' +n,4740,' suggests use of an approximation summing over the training data , which does not sum over possible tags : \' h E f j = 2 P -LRB- ~ -RRB- p -LRB- ti l hi -RRB- f j -LRB- hi , ti -RRB- i = 1 However , we believe this passage is in error : such an estimate is ineffective in the iterative scaling algorithm ',Ratnaparkhi,'nsubj_suggests_ dobj_suggests_use prep_use_of det_approximation_an pobj_of_approximation partmod_approximation_summing prep_summing_over det_data_the nn_data_training pobj_over_data nsubj_sum_which aux_sum_does neg_sum_not rcmod_data_sum prep_sum_over amod_tags_possible pobj_over_tags nn_j_h nn_j_E nn_j_f dep_summing_j rcmod_j_= num_P_2 dobj_=_P appos_P_~ partmod_P_p amod_hi_ti amod_hi_l appos_P_hi dep_j_f dep_j_j appos_j_hi dep_hi_ti partmod_j_i dep_1_= ccomp_i_1 advmod_believe_However nsubj_believe_we dep_suggests_believe det_passage_this nsubj_is_passage ccomp_believe_is prep_is_in pobj_in_error predet_estimate_such det_estimate_an nsubj_ineffective_estimate cop_ineffective_is parataxis_believe_ineffective prep_ineffective_in det_algorithm_the nn_algorithm_iterative amod_algorithm_scaling pobj_in_algorithm ' +o,4741,'Automatic segmentation of spontaneous speech is an open research problem in its own right ',Berger,'nn_segmentation_Automatic nsubj_problem_segmentation prep_segmentation_of amod_speech_spontaneous pobj_of_speech cop_problem_is det_problem_an amod_problem_open nn_problem_research prep_problem_in poss__its amod__own nn__right pobj_in_ ' +n,4742,'IBM Model1 is a simplistic model which takes no account of the subtler aspects of language translation including the way word order tends to differ across languages ',Brown,'nn__IBM nn__Model1 nsubj_model_ cop_model_is det_model_a amod_model_simplistic nsubj_takes_which rcmod_model_takes det_account_no nsubj_tends_account prep_account_of det_aspects_the nn_aspects_subtler pobj_of_aspects prep_aspects_of nn_translation_language pobj_of_translation prep_translation_including det_order_the nn_order_way nn_order_word pobj_including_order ccomp_takes_tends aux_differ_to xcomp_tends_differ prep_differ_across pobj_across_languages ' +o,4743,'42 Classifier and Features For our AL framework we decided to employ a Maximum Entropy -LRB- ME -RRB- classifier ',Berger,'nsubj_decided_Classifier cc_Classifier_and conj_Classifier_Features prep_decided_For poss_framework_our nn_framework_AL pobj_For_framework nsubj_decided_we rcmod_42_decided aux_employ_to xcomp_decided_employ det_classifier_a amod_classifier_Maximum nn_classifier_Entropy abbrev_classifier_ME dep_42_classifier ' +o,4744,'The parameters of the models are estimated by iterative maximum-likelihood training on a large parallel corpus of natural language texts using the EM algorithm ',Brown,'det_parameters_The nsubjpass_estimated_parameters prep_parameters_of det_models_the pobj_of_models auxpass_estimated_are prep_estimated_by amod_training_iterative amod_training_maximum-likelihood pobj_by_training prep_training_on det_corpus_a amod_corpus_large amod_corpus_parallel pobj_on_corpus prep_corpus_of amod_texts_natural nn_texts_language pobj_of_texts xcomp_estimated_using det_algorithm_the nn_algorithm_EM dobj_using_algorithm ' +p,4745,'As a result of this tuning , our -LRB- fully supervised -RRB- version of the Morce tagger gives the best accuracy among all single taggers for Czech and also very good results for English , being beaten only by the tagger -LRB- by 010 \% absolute -RRB- and -LRB- not significantly -RRB- by ',Shen,'prep_gives_As det_result_a pobj_As_result prep_result_of det_tuning_this pobj_of_tuning poss_version_our advmod_supervised_fully dep_version_supervised nsubj_gives_version prep_version_of det_tagger_the nn_tagger_Morce pobj_of_tagger det_accuracy_the amod_accuracy_best iobj_gives_accuracy dep_results_among det_taggers_all amod_taggers_single pobj_among_taggers prep_taggers_for pobj_for_Czech cc_among_and conj_among_also advmod_good_very dep_among_good dobj_gives_results prep_results_for pobj_for_English auxpass_beaten_being partmod_results_beaten dep_by_only prep_beaten_by det__the nn__tagger pobj_by_ dep__by num_\%_010 pobj_by_\% partmod_\%_absolute cc_by_and neg_significantly_not dep_-LRB-_significantly conj_by_by pcomp_by_ ' +o,4746,'In particular , we adopt the approach of phrase-based statistical machine translation ',Koehn,'prep_adopt_In pobj_In_particular nsubj_adopt_we det_approach_the dobj_adopt_approach prep_approach_of amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,4747,'For instance , in , yd would be the polarity of the document and ysi would indicate whether sentence si is subjective or objective ',Pang,'prep_polarity_For pobj_For_instance prep_polarity_in pobj_in_ nsubj_polarity_yd aux_polarity_would cop_polarity_be det_polarity_the prep_polarity_of det_document_the pobj_of_document cc_document_and conj_document_ysi aux_indicate_would rcmod_document_indicate complm_subjective_whether nn_si_sentence nsubj_subjective_si cop_subjective_is ccomp_indicate_subjective cc_subjective_or conj_subjective_objective ' +o,4748,'73 EM algorithm The only other application of the EM algorithm to word-sense disambiguation is described in ',Yarowsky,'num_algorithm_73 nn_algorithm_EM det_application_The advmod_application_only amod_application_other nsubjpass_described_application prep_application_of det_algorithm_the nn_algorithm_EM pobj_of_algorithm prep_algorithm_to amod_disambiguation_word-sense pobj_to_disambiguation auxpass_described_is dep_algorithm_described prep_described_in ' +o,4749,'We expect that the mean field approximation should demonstrate better results than feed-forward approximation on this task as it is theoretically expected and confirmed on the constituent parsing task ',Titov,'nsubj_expect_We complm_results_that det_approximation_the amod_approximation_mean nn_approximation_field nsubj_results_approximation aux_results_should dep_results_demonstrate amod_results_better ccomp_expect_results prep_results_than amod_approximation_feed-forward pobj_than_approximation prep_approximation_on det_task_this pobj_on_task mark_expected_as nsubjpass_expected_it auxpass_expected_is advmod_expected_theoretically advcl_results_expected cc_expected_and conj_expected_confirmed prep_expected_on det_task_the nn_task_constituent amod_task_parsing pobj_on_task ' +o,4750,'The results of the comparison with ROUGE-N , ROUGE-S -LRB- U -RRB- and ROUGE-L show that our method correlates more closely with human evaluations and is more robust ',Lin,'det_results_The nsubj_show_results prep_results_of det_comparison_the pobj_of_comparison prep_comparison_with nn__ROUGE-N nn__ appos__ROUGE-S nn__U pobj_with_ cc__and nn__ROUGE-L conj__ complm_correlates_that poss_method_our nsubj_correlates_method ccomp_show_correlates advmod_closely_more advmod_correlates_closely prep_correlates_with amod_evaluations_human pobj_with_evaluations cc_correlates_and cop_robust_is advmod_robust_more conj_correlates_robust ' +o,4751,'To obtain these distances , Ratnaparkhis partof-speech -LRB- POS -RRB- tagger and Collins parser were used to obtain parse trees for the English side of the test corpus ',Ratnaparkhi,'aux_obtain_To dep_used_obtain det_distances_these dobj_obtain_distances amod__Ratnaparkhis amod__partof-speech appos__POS nn__tagger nsubjpass_used_ cc__and nn__Collins nn__parser conj__ auxpass_used_were aux_obtain_to xcomp_used_obtain amod_trees_parse dobj_obtain_trees prep_obtain_for det_side_the amod_side_English pobj_for_side prep_side_of det_corpus_the nn_corpus_test pobj_of_corpus ' +o,4752,'Model performance is evaluated using the standard BLEU metric which measures average n-gram precision , n 4 , and we use the NIST definition of the brevity penalty for multiple reference test sets ',Papineni,'nn_performance_Model nsubjpass_evaluated_performance auxpass_evaluated_is xcomp_evaluated_using det__the amod__standard amod__BLEU amod__metric dobj_using_ nsubj_measures_which rcmod__measures amod_precision_average amod_precision_n-gram dobj_measures_precision appos_precision_n num_n_4 cc_evaluated_and nsubj_use_we conj_evaluated_use det_definition_the amod_definition_NIST dobj_use_definition prep_definition_of det_penalty_the amod_penalty_brevity pobj_of_penalty prep_penalty_for amod_sets_multiple nn_sets_reference nn_sets_test pobj_for_sets ' +o,4753,'Bilexical CFG is at the heart of most modern statistical parsers , because the statistics associated with word-specific rules are more informative for disambiguation purposes ',Collins,'nn_CFG_Bilexical nsubj_is_CFG prep_is_at det_heart_the pobj_at_heart prep_heart_of advmod_modern_most amod__modern amod__statistical nn__parsers pobj_of_ mark_informative_because det_statistics_the nsubj_informative_statistics partmod_statistics_associated prep_associated_with amod_rules_word-specific pobj_with_rules cop_informative_are advmod_informative_more advcl_is_informative prep_informative_for nn_purposes_disambiguation pobj_for_purposes ' +o,4754,' describes a ` semi-unsupervised \' approach to the problem of sense disambiguation of words , also using a set of initial seeds , in this case a few high quality sense annotations ',Yarowsky,'advmod_describes_ det_approach_a amod_approach_semi-unsupervised dobj_describes_approach prep_describes_to det_problem_the pobj_to_problem prep_problem_of nn_disambiguation_sense pobj_of_disambiguation prep_problem_of pobj_of_words advmod_using_also partmod_problem_using det_set_a dobj_using_set prep_set_of amod_seeds_initial pobj_of_seeds prep_problem_in det_case_this pobj_in_case det_annotations_a amod_annotations_few amod_annotations_high nn_annotations_quality nn_annotations_sense nsubj_describes_annotations ' +p,4755,'One widely used model is the IBM model ',Brown,'num_model_One advmod_used_widely amod_model_used nsubj_model_model cop_model_is det_model_the nn_model_IBM ' +o,4756,'CIT -RRB- ',Cutting,'nn_-RRB-_CIT ' +o,4757,'It is analogous to the step in other translation model induction algorithms that sets all probabilities below a certain threshold to negligible values ',Brown,'nsubj_analogous_It cop_analogous_is prep_analogous_to det_step_the pobj_to_step prep_step_in amod_algorithms_other nn_algorithms_translation nn_algorithms_model nn_algorithms_induction pobj_in_algorithms nsubj_sets_that rcmod_algorithms_sets det_probabilities_all dobj_sets_probabilities prep_sets_below det_threshold_a amod_threshold_certain pobj_below_threshold prep_threshold_to amod_values_negligible pobj_to_values ' +o,4758,'The Brill tagger comes with an English default version also trained on general-purpose language corpora like the PENN TREEBANK ',Marcus,'det_tagger_The nn_tagger_Brill nsubj_comes_tagger prep_comes_with det_version_an amod_version_English nn_version_default pobj_with_version advmod_trained_also partmod_version_trained prep_trained_on amod_corpora_general-purpose nn_corpora_language pobj_on_corpora prep_corpora_like det_TREEBANK_the nn_TREEBANK_PENN pobj_like_TREEBANK ' +o,4759,'As to analysis of NPs , there have been a lot of work on statistical techniques for lexical dependency parsing of sentences , and these techniques potentially can be used for analysis of NPs if appropriate resources for NPs are available ',Collins,'prep_been_As dep_As_to pobj_to_analysis prep_analysis_of pobj_of_NPs expl_been_there aux_been_have det_lot_a dobj_been_lot prep_lot_of pobj_of_work prep_been_on amod_techniques_statistical pobj_on_techniques prep_techniques_for amod_parsing_lexical amod_parsing_dependency pobj_for_parsing prep_parsing_of nn__sentences pobj_of_ cc_been_and det_techniques_these nsubjpass_used_techniques advmod_used_potentially aux_used_can auxpass_used_be conj_been_used prep_used_for pobj_for_analysis prep_analysis_of pobj_of_NPs mark_available_if amod_resources_appropriate nsubj_available_resources prep_resources_for pobj_for_NPs cop_available_are advcl_used_available ' +o,4760,'His results may be improved if more sophisticated techniques and larger corpora are used to establish similarity between words -LRB- such as in -RRB- ',Hindle,'poss_results_His nsubjpass_improved_results aux_improved_may auxpass_improved_be mark_used_if advmod_sophisticated_more amod_techniques_sophisticated nsubjpass_used_techniques cc_techniques_and amod_corpora_larger conj_techniques_corpora auxpass_used_are advcl_improved_used aux_establish_to xcomp_used_establish dobj_establish_similarity prep_similarity_between pobj_between_words dep_as_such dep_words_as dep_as_in pobj_in_ ' +o,4761,'RIDF is like MI , but different References Word association norms , mutual information , and lexicography Computational Linguistics , 16:1 , pp ',Church,'nsubj_pp_RIDF cop_pp_is dep_pp_like pobj_like_MI cc_like_but amod_norms_different nn_norms_References nn_norms_ nn_norms_Word nn_norms_association conj_like_norms amod_information_mutual appos_norms_information cc_norms_and amod_Linguistics_lexicography nn_Linguistics_Computational conj_norms_Linguistics appos_Linguistics_16:1 ' +o,4762,'2We use a POS tagger trained on switchboard data with the additional tags of FP -LRB- filled pause -RRB- and FRAG -LRB- word fragment -RRB- ',Ratnaparkhi,'nsubj_use_2We det__a dep__POS nn__tagger dobj_use_ partmod__trained prep_trained_on amod_data_switchboard pobj_on_data prep_trained_with det_tags_the amod_tags_additional pobj_with_tags prep_tags_of pobj_of_FP dep__filled dobj_filled_pause cc__and conj__FRAG nn_fragment_word appos_FRAG_fragment ' +o,4763,'Fortunately , there is a straightforward parallel between our object recognition formulation and the statistical machine translation problem of building a lexicon from an aligned bitext ',Brown,'advmod_is_Fortunately expl_is_there det_parallel_a amod_parallel_straightforward nsubj_is_parallel prep_parallel_between poss_formulation_our nn_formulation_object nn_formulation_recognition pobj_between_formulation cc_parallel_and det_problem_the amod_problem_statistical nn_problem_machine nn_problem_translation conj_parallel_problem prep_problem_of pcomp_of_building det_lexicon_a dobj_building_lexicon prep_building_from det_bitext_an amod_bitext_aligned pobj_from_bitext ' +o,4764,'The kappa obtained on this feature was 093 ',Carletta,'det__The nn__kappa nsubj_093_ partmod__obtained prep_obtained_on det_feature_this pobj_on_feature cop_093_was ' +o,4765,'For example , the Penn Treebank was annotated with skeletal syntactic structure , and many syntactic parsers were evaluated and compared on the corpus ',Marcus,'prep_annotated_For pobj_For_example det__the nn__Penn nn__Treebank nsubjpass_annotated_ auxpass_annotated_was prep_annotated_with amod_structure_skeletal amod_structure_syntactic pobj_with_structure cc_annotated_and amod_parsers_many amod_parsers_syntactic nsubjpass_evaluated_parsers auxpass_evaluated_were conj_annotated_evaluated cc_evaluated_and conj_evaluated_compared prep_evaluated_on det_corpus_the pobj_on_corpus ' +o,4766,'By contrast , present three metrics that use syntactic and unlabelled dependency information ',Liu,'prep_present_By pobj_By_contrast nsubj_present_ num_metrics_three dobj_present_metrics nsubj_use_that rcmod_metrics_use amod_information_syntactic cc_syntactic_and conj_syntactic_unlabelled amod_information_dependency dobj_use_information ' +o,4767,'Some of them are based upon syntactic structure , with PropBank being one of the most relevant , building the annotation upon the syntactic representation of the TreeBank corpus ',Marcus,'nsubjpass_based_Some prep_Some_of pobj_of_them auxpass_based_are prep_based_upon amod_structure_syntactic pobj_upon_structure prep_based_with nn__PropBank nsubj_one_ cop_one_being pcomp_with_one prep_one_of dep_relevant_the advmod_relevant_most pobj_of_relevant xcomp_based_building det_annotation_the dobj_building_annotation prep_building_upon det_representation_the amod_representation_syntactic pobj_upon_representation prep_representation_of det_corpus_the nn_corpus_TreeBank pobj_of_corpus ' +o,4768,'The results were evaluated using the character\\/pinyin-based 4-gram BLEU score , word error rate -LRB- WER -RRB- , position independent word error rate -LRB- PER -RRB- , and exact match -LRB- EMatch -RRB- ',Papineni,'det_results_The nsubjpass_evaluated_results auxpass_evaluated_were xcomp_evaluated_using det__the amod__character\\/pinyin-based dep_character\\/pinyin-based_4-gram amod__BLEU nn__score dobj_using_ nn_rate_word nn_rate_error conj__rate abbrev_rate_WER conj__position amod_rate_independent nn_rate_word nn_rate_error dep_position_rate abbrev_rate_PER cc__and amod_match_exact conj__match abbrev__EMatch ' +o,4769,'4 are equivalent to a maximum entropy variant of the phrase sense disambiguation approach studied by ',Carpuat,'nsubj_equivalent_4 cop_equivalent_are prep_equivalent_to det_variant_a amod_variant_maximum nn_variant_entropy pobj_to_variant prep_variant_of det_approach_the nn_approach_phrase nn_approach_sense nn_approach_disambiguation pobj_of_approach partmod_approach_studied prep_studied_by ' +o,4770,' propose a tree sequence-based tree to tree translation model and et al ',Zhang,'advmod_propose_ det_tree_a nn_tree_tree amod_tree_sequence-based dobj_propose_tree prep_propose_to nn_model_tree nn_model_translation pobj_to_model cc_model_and conj_model_ cc__et conj__al ' +o,4771,'Class-based n-gram models have also been shown to benefit from their reduced number of parameters when scaling to higher-order n-grams , and even despite the increasing size and decreasing sparsity of language model training corpora , class-based n-gram models might lead to improvements when increasing the n-gram order ',Brants,'amod_models_Class-based amod_models_n-gram nsubjpass_shown_models aux_shown_have advmod_shown_also auxpass_shown_been aux_benefit_to xcomp_shown_benefit prep_benefit_from poss_number_their amod_number_reduced pobj_from_number prep_number_of pobj_of_parameters advmod_higher-order_when nsubj_higher-order_scaling aux_higher-order_to xcomp_benefit_higher-order amod__n-grams dobj_higher-order_ cc_shown_and dep_despite_even advcl_lead_despite det_size_the amod_size_increasing pobj_despite_size cc_size_and amod_sparsity_decreasing conj_size_sparsity prep_size_of nn__language nn__model nn__training nn__corpora pobj_of_ amod_models_class-based amod_models_n-gram nsubj_lead_models aux_lead_might conj_shown_lead prep_lead_to pobj_to_improvements advmod_increasing_when advcl_lead_increasing det_order_the amod_order_n-gram dobj_increasing_order ' +o,4772,'Automated evaluation metrics that rate system behavior based on automatically computable properties have been developed in a number of other fields : widely used measures include BLEU for machine translation and ROUGE for summarisation , for example ',Lin,'amod_metrics_Automated nn_metrics_evaluation nsubj_include_metrics complm_developed_that nn_behavior_rate nn_behavior_system nsubjpass_developed_behavior prep_behavior_based dep_based_on dep_on_automatically amod_properties_computable pobj_on_properties aux_developed_have auxpass_developed_been dep_metrics_developed prep_developed_in det_number_a pobj_in_number prep_number_of amod_fields_other pobj_of_fields advmod_used_widely amod_measures_used dep_number_measures nn__BLEU dobj_include_ prep__for nn_translation_machine pobj_for_translation cc__and nn__ROUGE conj__ prep__for pobj_for_summarisation prep_include_for pobj_for_example ' +o,4773,'It can also be considered as an extension from the monolingual to the bilingual case of the well-established methods for semantic or syntactic word clustering as proposed by Schtitze , , Ruge -LRB- 1995 -RRB- , Rapp -LRB- 1996 -RRB- , Lin -LRB- 1998 -RRB- , and others ',Grefenstette,'nsubjpass_considered_It aux_considered_can advmod_considered_also auxpass_considered_be prep_considered_as det_extension_an pobj_as_extension prep_considered_from dep_monolingual_the pobj_from_monolingual prep_monolingual_to det_case_the amod_case_bilingual pobj_to_case prep_case_of det_methods_the amod_methods_well-established pobj_of_methods prep_methods_for amod_word_semantic cc_semantic_or conj_semantic_syntactic pobj_for_word amod_monolingual_clustering advmod_proposed_as amod_monolingual_proposed prep_proposed_by pobj_by_Schtitze num_Schtitze_ num_Schtitze_ appos_monolingual_Ruge appos_Ruge_1995 conj_Ruge_Rapp appos_Rapp_1996 conj_Ruge_Lin appos_Lin_1998 cc_Ruge_and conj_Ruge_others ' +o,4774,'NeATS computes the likelihood ratio to identify key concepts in unigrams , bigrams , and trigrams and clusters these concepts in order to identify major subtopics within the main topic ',Dunning,'nsubj_computes_NeATS det__the nn__likelihood nn__ratio nsubj_identify_ aux_identify_to xcomp_computes_identify amod_concepts_key dobj_identify_concepts prep_identify_in nn_trigrams_unigrams conj_trigrams_bigrams cc_trigrams_and pobj_in_trigrams cc_identify_and conj_identify_clusters det_concepts_these dobj_clusters_concepts mark_identify_in dep_identify_order aux_identify_to purpcl_clusters_identify amod_subtopics_major dobj_identify_subtopics prep_identify_within det_topic_the amod_topic_main pobj_within_topic ' +p,4775,'Word segmentation and POS tagging in a joint process have received much attention in recent research and have shown improvements over a pipelined fashion ',Jiang,'amod_segmentation_Word nsubj_received_segmentation cc_segmentation_and conj_segmentation_POS partmod_segmentation_tagging prep_tagging_in det_process_a amod_process_joint pobj_in_process aux_received_have amod_attention_much dobj_received_attention prep_received_in amod_research_recent pobj_in_research cc_received_and aux_shown_have conj_received_shown dobj_shown_improvements prep_improvements_over det_fashion_a amod_fashion_pipelined pobj_over_fashion ' +p,4776,'Our method follows and substantially extends the earlier work of , who use syntactic features and unlabelled dependencies to evaluate MT quality , outperforming BLEU on segment-level correlation with human judgement ',Liu,'poss_method_Our nsubj_follows_method cc_follows_and advmod_extends_substantially conj_follows_extends det_work_the amod_work_earlier dobj_extends_work prep_work_of pobj_of_ nsubj_use_who rcmod__use amod_features_syntactic nsubj_evaluate_features cc_features_and nn_dependencies_unlabelled conj_features_dependencies aux_evaluate_to xcomp_use_evaluate nn_quality_MT dobj_evaluate_quality xcomp_use_outperforming dobj_outperforming_BLEU prep_outperforming_on amod_correlation_segment-level pobj_on_correlation prep_correlation_with amod_judgement_human pobj_with_judgement ' +o,4777,'In the last decade or so research on lexical semantics has focused more on sub-problems like word sense disambiguation , named entity recognition , and vocabulary construction for information extraction ',Yarowsky,'det_decade_the amod_decade_last pobj_In_decade cc_In_or mark_focused_so nsubj_focused_research prep_research_on amod_semantics_lexical pobj_on_semantics aux_focused_has conj_In_focused dep_on_more prep_focused_on pobj_on_sub-problems prep_sub-problems_like nn__word nn__sense nn__disambiguation pobj_like_ amod__named nn__entity nn__recognition conj__ cc__and nn_construction_vocabulary conj__construction prep_construction_for nn__information nn__extraction pobj_for_ ' +o,4778,' , , Dave et al ',Turney,'nn__ appos__ conj__Dave cc__et conj__al ' +o,4779,'We augment each labeled target instance xj with the label assigned by the source domain classifier ',Blitzer,'nsubj_augment_We det_xj_each amod_xj_labeled nn_xj_target nn_xj_instance dobj_augment_xj prep_augment_with det_label_the pobj_with_label partmod_label_assigned prep_assigned_by det_classifier_the nn_classifier_source nn_classifier_domain pobj_by_classifier ' +p,4780,'Among these advances , forest-based modeling and tree sequence-based modeling are two interesting modeling methods with promising results reported ',Zhang,'prep_methods_Among det_advances_these pobj_Among_advances amod__forest-based nn__modeling nsubj_methods_ cc__and nn__tree amod__sequence-based nn__modeling conj__ cop_methods_are num_methods_two amod_methods_interesting nn_methods_modeling prep_methods_with amod_results_promising pobj_with_results partmod_results_reported ' +o,4781,'Introduction Translation of two languages with highly different morphological structures as exemplified by Arabic and English poses a challenge to successful implementation of statistical machine translation models ',Brown,'nn_Translation_Introduction nsubj_poses_Translation prep_Translation_of num_languages_two pobj_of_languages prep_languages_with advmod_different_highly amod_structures_different amod_structures_morphological pobj_with_structures advmod_exemplified_as amod_structures_exemplified prep_exemplified_by pobj_by_Arabic cc_Arabic_and conj_Arabic_English det_challenge_a dobj_poses_challenge prep_poses_to amod_implementation_successful pobj_to_implementation prep_implementation_of amod_models_statistical nn_models_machine nn_models_translation pobj_of_models ' +o,4782,'5We use deterministic sampling , which is useful for reproducibility and for minimum error rate training ',Och,'nsubj_use_5We amod_sampling_deterministic dobj_use_sampling nsubj_useful_which cop_useful_is rcmod_sampling_useful prep_useful_for dep_training_reproducibility cc_reproducibility_and conj_reproducibility_for amod_error_minimum pobj_for_error nn_training_rate pobj_for_training ' +o,4783,'The second model is a maximum entropy model , since Klein and Manning found that this model yielded higher accuracy than nave Bayes in a subsequent comparison of WSD performance ',Klein,'det_model_The amod_model_second nsubj__model cop__is det__a amod__maximum amod__entropy nn__model mark_found_since nn__Klein cc_Klein_and conj_Klein_Manning nsubj_found_ advcl__found complm_yielded_that det_model_this nsubj_yielded_model ccomp_found_yielded amod_accuracy_higher dobj_yielded_accuracy prep_yielded_than amod_Bayes_nave pobj_than_Bayes prep_yielded_in det_comparison_a amod_comparison_subsequent pobj_in_comparison prep_comparison_of nn_performance_WSD pobj_of_performance ' +o,4784,'3 Incremental Parsing Method Based on Adjoining Operation In order to avoid the problem of infinite local ambiguity , the previous works have adopted the following approaches : -LRB- 1 -RRB- a beam search strategy , -LRB- 2 -RRB- limiting the allowable chains to those actually observed in the treebank , and -LRB- 3 -RRB- transforming the parse trees with a selective left-corner transformation before inducing the allowable chains and allowable triples ',Collins,'dep_Method_3 nn_Method_Incremental nn_Method_Parsing prep_Method_Based dep_Based_on nn_Operation_Adjoining pobj_on_Operation mark_avoid_In dep_avoid_order aux_avoid_to dep_adopted_avoid det_problem_the dobj_avoid_problem prep_problem_of amod_ambiguity_infinite amod_ambiguity_local pobj_of_ambiguity det_works_the amod_works_previous nsubj_adopted_works aux_adopted_have dep_Method_adopted det_approaches_the amod_approaches_following dobj_adopted_approaches dep__1 det__a amod__beam nn__search nn__strategy dep_approaches_ appos__2 partmod__limiting det_chains_the amod_chains_allowable dobj_limiting_chains prep_limiting_to pobj_to_those advmod_observed_actually partmod_those_observed prep_observed_in det__the nn__treebank pobj_in_ cc_approaches_and dep_trees_3 advmod_trees_transforming det_trees_the amod_trees_parse conj_approaches_trees prep_trees_with det_transformation_a amod_transformation_selective amod_transformation_left-corner pobj_with_transformation partmod_transformation_ prep__before pcomp_before_inducing det_chains_the amod_chains_allowable dobj_inducing_chains cc_chains_and amod_triples_allowable conj_chains_triples ' +o,4785,'52 Pseudo-Disambiguation Task Pseudo-disambiguation tasks have become a standard evaluation technique and , in the current setting , we may use a nouns neighbors to decide which of two co-occurrences is the most likely ',Dagan,'num_tasks_52 amod_tasks_Pseudo-Disambiguation nn_tasks_Task nn_tasks_Pseudo-disambiguation nsubj_become_tasks aux_become_have det_technique_a amod_technique_standard nn_technique_evaluation nsubj__technique dep_become_ cc__and dep__in det_setting_the amod_setting_current pobj_in_setting nsubj_use_we aux_use_may conj__use det_neighbors_a amod_neighbors_nouns nsubj_decide_neighbors aux_decide_to xcomp_use_decide nsubj_likely_which prep_which_of num_co-occurrences_two pobj_of_co-occurrences cop_likely_is det_most_the attr_likely_most ccomp_decide_likely ' +o,4786,'Following , the prevailing opinion in the research community has been that more complex patterns of word alignment in real bitexts are mostly attributable to alignment errors ',Wu,'nn__Following nsubj_been_ det_opinion_the amod_opinion_prevailing nsubj_been_opinion prep_opinion_in det_community_the nn_community_research pobj_in_community aux_been_has complm_attributable_that advmod_complex_more amod_patterns_complex nsubj_attributable_patterns prep_patterns_of nn_alignment_word pobj_of_alignment prep_alignment_in amod_bitexts_real pobj_in_bitexts cop_attributable_are advmod_attributable_mostly ccomp_been_attributable aux_alignment_to xcomp_attributable_alignment dobj_alignment_errors ' +o,4787,'For this reason , to our knowledge , all discriminative models proposed to date either side-step the problem by choosing simple model and feature structures , such that spurious ambiguity is lessened or removed entirely , or else ignore the problem and treat derivations as translations ',Liang,'prep_lessened_For det_reason_this pobj_For_reason prep_lessened_to poss_knowledge_our pobj_to_knowledge det_models_all amod_models_discriminative appos_knowledge_models dep_models_proposed prep_proposed_to pobj_to_date dep_proposed_either conj_proposed_side-step det_problem_the dobj_side-step_problem prep_side-step_by pcomp_by_choosing amod_model_simple dobj_choosing_model cc_model_and nn_structures_feature conj_model_structures predet_ambiguity_such det_ambiguity_that amod_ambiguity_spurious nsubjpass_lessened_ambiguity auxpass_lessened_is cc_lessened_or conj_lessened_removed advmod__entirely dep_removed_ cc_removed_or advmod_removed_else conj_removed_ignore det_problem_the dobj_ignore_problem cc_ignore_and conj_ignore_treat dobj_treat_derivations prep_treat_as pobj_as_translations ' +o,4788,'Here , we train word alignments in both directions with GIZA + + ',Och,'advmod_train_Here nsubj_train_we nn_alignments_word dobj_train_alignments prep_train_in det_directions_both pobj_in_directions prep_directions_with nn_+_GIZA nn_+_+ pobj_with_+ ' +o,4789,'The model employs a stochastic version of an inversion transduction grammar or ITG ',Wu,'det_model_The nsubj_employs_model det_version_a amod_version_stochastic dobj_employs_version prep_version_of det_grammar_an nn_grammar_inversion nn_grammar_transduction pobj_of_grammar cc_grammar_or conj_grammar_ITG ' +p,4790,'Some recent work on incremental parsing showed another way to handle this problem ',Collins,'det_work_Some amod_work_recent nsubj_showed_work prep_work_on amod__incremental amod__parsing pobj_on_ det_way_another dobj_showed_way aux_handle_to infmod_way_handle det_problem_this dobj_handle_problem ' +o,4791,'The statistical machine translation approach is based on the noisy channel paradigm and the Maximum-A-Posteriori decoding algorithm ',Brown,'det_approach_The amod_approach_statistical nn_approach_machine nn_approach_translation nsubjpass_based_approach auxpass_based_is prep_based_on det_paradigm_the amod_paradigm_noisy nn_paradigm_channel pobj_on_paradigm cc_paradigm_and det_algorithm_the amod_algorithm_Maximum-A-Posteriori amod_algorithm_decoding conj_paradigm_algorithm ' +o,4792,'We evaluate the string chosen by the log-linear model against the original treebank string in terms of exact match and BLEU score (Papineni et al., 821 Syntactic feature Type Definites Definite descriptions SIMPLE DEF simple definite descriptions POSS DEF simple definite descriptions with a possessive determiner (pronoun or possibly genitive name) DEF ATTR ADJ definite descriptions with adjectival modifier DEF GENARG definite descriptions with a genitive argument DEF PPADJ definite descriptions with a PP adjunct DEF RELARG definite descriptions including a relative clause DEF APP definite descriptions including a title or job description as well as a proper name (e.g. an apposition) Names PROPER combinations of position/title and proper name (without article) BARE PROPER bare proper names Demonstrative descriptions SIMPLE DEMON simple demonstrative descriptions MOD DEMON adjectivally modified demonstrative descriptions Pronouns PERS PRON personal pronouns EXPL PRON expletive pronoun REFL PRON reflexive pronoun DEMON PRON demonstrative pronouns (not: determiners) GENERIC PRON generic pronoun (man one) DA PRON da-pronouns (darauf, daruber, dazu, ) LOC ADV location-referring pronouns TEMP ADV,YEAR Dates and times Indefinites SIMPLE INDEF simple indefinites NEG INDEF negative indefinites INDEF ATTR indefinites with adjectival modifiers INDEF CONTRAST indefinites with contrastive modifiers (einige some, andere other, weitere further, ) INDEF PPADJ indefinites with PP adjuncts INDEF REL indefinites with relative clause adjunct INDEF GEN indefinites with genitive adjuncts INDEF NUM measure/number phrases INDEF QUANT quantified indefinites Table 5: An inventory of interesting syntactic characteristics in IS phrases Label 1 (+ features) Label 2 (+ features) B/A Total D-GIVEN-PRONOUN INDEF-REL 0 19 PERS PRON 39 INDEF ATTR 23 DA PRON 25 SIMPLE INDEF 17 DEMON PRON 19 GENERIC PRON 11 D-GIVEN-PRONOUN D-GIVEN-CATAPHOR 0.1 11 PERS PRON 39 SIMPLE DEF 13 DA PRON 25 DA PRON 10 DEMON PRON 19 GENERIC PRON 11 D-GIVEN-REFLEXIVE NEW 0.11 31 REFL PRON 54 SIMPLE INDEF 113 INDEF ATTR 53 INDEF NUM 32 INDEF PPADJ 26 INDEF GEN 25 Table 6: IS asymmetric pairs augmented with syntactic characteristics 822 2002).',Papineni,'' +o,4793,'The model was trained using minimum error rate training for Arabic and MIRA for Chinese ',Och,'det_model_The nsubjpass_trained_model auxpass_trained_was xcomp_trained_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training prep_training_for nn__Arabic pobj_for_ cc__and conj__MIRA prep_using_for amod__Chinese pobj_for_ ' +o,4794,'Given a set of features and a training corpus , the ME estimation process produces a model in which every feature fi has a weight i From , we can compute the conditional probability as : p -LRB- o h -RRB- = 1Z -LRB- h -RRB- productdisplay i fi -LRB- h , o -RRB- i -LRB- 2 -RRB- Z -LRB- h -RRB- = summationdisplay o productdisplay i fi -LRB- h , o -RRB- i -LRB- 3 -RRB- The probability is given by multiplying the weights of active features -LRB- ie , those fi -LRB- h , o -RRB- = 1 -RRB- ',Berger,'prep_produces_Given det_set_a dep_Given_set prep_set_of pobj_of_features cc_features_and det_corpus_a nn_corpus_training conj_features_corpus det_process_the nn_process_ME nn_process_estimation nsubj_produces_process det_model_a nsubj_i_model rel_has_in pobj_in_which det_fi_every nn_fi_feature nsubj_has_fi rcmod_model_has det_i_a nn_i_weight dobj_has_i prep_compute_From pobj_From_ nsubj_compute_we aux_compute_can rcmod_i_compute det_probability_the amod_probability_conditional dobj_compute_probability mark_=_as nsubj_=_p nn_h_o appos_p_h advcl_compute_= amod_fi_1Z appos_fi_h advmod_i_productdisplay amod_fi_i dobj_=_fi appos_fi_h dep_h_o ccomp_produces_i dep_i_2 nn_h_Z dobj_i_h dep_h_= amod_o_summationdisplay iobj_=_o advmod_i_productdisplay amod_i_i nn_i_fi appos_i_h dep_h_o dobj_=_i dep_given_3 det_probability_The nsubjpass_given_probability auxpass_given_is dep_produces_given prep_given_by pcomp_by_multiplying det_weights_the dobj_multiplying_weights prep_weights_of amod_features_active pobj_of_features dep_weights_ie nsubj_fi_those ccomp_ie_fi nsubj_1_h dep_h_o dep_1_= ccomp_fi_1 ' +p,4795,'Similar to WSD , used contextual information to solve the ambiguity problem for phrases ',Carpuat,'prep_Similar_to nn__WSD pobj_to_ partmod__used amod_information_contextual dobj_used_information aux_solve_to xcomp_used_solve det_problem_the amod_problem_ambiguity dobj_solve_problem prep_problem_for pobj_for_phrases ' +o,4796,' and with HPSG and LFG grammars ',Cahill,'cc__and conj__ prep__with nn_grammars_HPSG cc_HPSG_and conj_HPSG_LFG pobj_with_grammars ' +o,4797,'Meanwhile , some learning algorithms , like maximum likelihood for conditional log-linear models , unsupervised models , and models with hidden variables , require summing over the scores of many structures to calculate marginals ',Koo,'advmod__Meanwhile det_learning_some nsubj_algorithms_learning dep__algorithms prep__like amod_likelihood_maximum pobj_like_likelihood prep_likelihood_for amod__conditional amod__log-linear nn__models pobj_for_ amod_models_unsupervised nsubj__models cc__and nsubj__models prep_models_with amod_variables_hidden pobj_with_variables conj__ dep__require dobj__summing prep_summing_over det_scores_the pobj_over_scores prep_scores_of amod_structures_many pobj_of_structures aux_calculate_to xcomp__calculate dobj_calculate_marginals ' +o,4798,'In the past five years , important research on the automatic acquisition of word classes based on lexical distribution has been published ',Hindle,'prep_published_In det_years_the amod_years_past num_years_five pobj_In_years amod_research_important nsubjpass_published_research prep_research_on det_acquisition_the amod_acquisition_automatic pobj_on_acquisition prep_acquisition_of nn_classes_word pobj_of_classes partmod_classes_based prep_based_on amod_distribution_lexical pobj_on_distribution aux_published_has auxpass_published_been ' +o,4799,' use a pattern-based approach in mining instances of RSRs such as Contrast and Elaboration from large , unannotated corpora ',Marcu,'nsubj_use_ det_approach_a amod_approach_pattern-based dobj_use_approach prep_approach_in nn_instances_mining pobj_in_instances prep_instances_of pobj_of_RSRs dep_as_such prep_RSRs_as pobj_as_Contrast cc_Contrast_and conj_Contrast_Elaboration prep_use_from amod_corpora_large amod_corpora_unannotated pobj_from_corpora ' +o,4800,'These heuristics define a phrase pair to consist of a source and target ngrams of a word-aligned source-target sentence pair such that if one end of an alignment is in the one ngram , the other end is in the other ngram -LRB- and there is at least one such alignment -RRB- ',Och,'det_heuristics_These nsubj_define_heuristics det_pair_a nn_pair_phrase nsubj_consist_pair aux_consist_to xcomp_define_consist prep_consist_of det_ngrams_a nn_ngrams_source cc_source_and conj_source_target pobj_of_ngrams prep_ngrams_of det_pair_a amod_pair_word-aligned amod_pair_source-target nn_pair_sentence pobj_of_pair prep_consist_such complm_is_that mark_is_if num_end_one nsubj_is_end prep_end_of det_alignment_an pobj_of_alignment advcl_is_is prep_is_in det_ngram_the num_ngram_one pobj_in_ngram det_end_the amod_end_other nsubj_is_end ccomp_consist_is prep_is_in det_ngram_the amod_ngram_other pobj_in_ngram dep_is_and expl_is_there dep_ngram_is quantmod_one_at dep_at_least num_alignment_one amod_alignment_such nsubj_is_alignment ' +o,4801,'We employ loglinear models for the disambiguation ',Berger,'nsubj_employ_We nn_models_loglinear nsubj__models ccomp_employ_ prep__for det_disambiguation_the pobj_for_disambiguation ' +o,4802,'Whereas until recently the focus of research had been on sense disambiguation , papers like Pantel & Lin , Neill , and give evidence that sense induction now also attracts attention ',Rapp,'mark_been_Whereas dep_recently_until advmod_been_recently det_focus_the nsubj_been_focus prep_focus_of pobj_of_research aux_been_had advcl_give_been prep_been_on nn_disambiguation_sense pobj_on_disambiguation nsubj_give_papers prep_papers_like nn__Pantel cc_Pantel_& conj_Pantel_Lin pobj_like_ nn__Neill conj__ cc__and conj__ dobj_give_evidence complm_attracts_that nn_induction_sense nsubj_attracts_induction advmod_attracts_now advmod_attracts_also ccomp_give_attracts dobj_attracts_attention ' +n,4803,'A number of studies have investigated sentiment classification at document level , eg , , and at sentence level , eg , ; however , the accuracy is still less than desirable ',Pang,'det_number_A nsubj_investigated_number prep_number_of pobj_of_studies aux_investigated_have nn_classification_sentiment dobj_investigated_classification prep_investigated_at nn_level_document pobj_at_level appos_level_eg dep_at_ cc_at_and conj_at_at nn_level_sentence pobj_at_level appos_level_eg dep_at_ advmod_desirable_however det_accuracy_the nsubj_desirable_accuracy cop_desirable_is advmod_desirable_still advmod_desirable_less dep_less_than parataxis_investigated_desirable ' +o,4804,'(Fleischman et al. , 2003; Jijkoun et al. , 2003).',Fleischman,'' +o,4805,'Performance also degrades when the domain of the test set differs from the domain of the training set , in part because the test set includes more OOV words and words that appear only a few times in the training set -LRB- henceforth , rare words -RRB- ',Blitzer,'nsubj_degrades_Performance advmod_degrades_also advmod_differs_when det_domain_the nsubj_differs_domain prep_domain_of det_set_the nn_set_test pobj_of_set advcl_degrades_differs prep_differs_from det_domain_the pobj_from_domain prep_domain_of det_set_the nn_set_training pobj_of_set prep_domain_in pobj_in_part mark_includes_because det_set_the nn_set_test nsubj_includes_set advcl_differs_includes amod_words_more nn_words_OOV dobj_includes_words cc_words_and conj_words_words nsubj_appear_that rcmod_words_appear quantmod_few_only quantmod_few_a num_times_few dep_in_times prep_appear_in det_set_the nn_set_training pobj_in_set dep_set_henceforth amod_words_rare appos_henceforth_words ' +o,4806,'More details about the re-ranking algorithm are presented in ',Ji,'amod_details_More nsubjpass_presented_details prep_details_about det_algorithm_the amod_algorithm_re-ranking pobj_about_algorithm auxpass_presented_are prep_presented_in ' +o,4807,'This is similar to s and Charniak97s definition of a separate category for auxiliary verbs ',Collins,'nsubj_similar_This cop_similar_is aux__to xcomp_similar_ dobj__s cc_s_and nn_definition_Charniak97s conj_s_definition prep_s_of det_category_a amod_category_separate pobj_of_category prep_category_for amod_verbs_auxiliary pobj_for_verbs ' +o,4808,'The translation models were pharse-based created using the GIZA + + toolkit ',Och,'det_models_The nn_models_translation nsubj__models cop__were amod__pharse-based partmod__created xcomp_created_using det_GIZA_the nsubj_+_GIZA xcomp_using_+ cc_+_+ conj_+_toolkit ' +o,4809,'12 From Synchronous to Quasi-Synchronous Grammars Because our approach will let anything align to anything , it is reminiscent of IBM Models 15 ',Brown,'dep_12_From pobj_From_Synchronous prep_Synchronous_to nn_Grammars_Quasi-Synchronous pobj_to_Grammars mark_let_Because poss_approach_our nsubj_let_approach aux_let_will advcl_reminiscent_let nn_align_anything dobj_let_align prep_let_to pobj_to_anything nsubj_reminiscent_it cop_reminiscent_is rcmod_Grammars_reminiscent prep_Grammars_of nn_Models_IBM pobj_of_Models num_Models_15 ' +o,4810,'There are cases, though, where the labels consist of several related, but not entirely correlated, properties; examples include mention detectionthe task we are interested in, syntactic parsing with functional tag assignment (besides identifying the syntactic parse, also label the constituent nodes with their functional category, as defined in the Penn Treebank (Marcus et al. , 1993)), and, to a lesser extent, part-of-speech tagging in highly inflected languages.4 The particular type of mention detection that we are examining in this paper follows the ACE general definition: each mention in the text (a reference to a real-world entity) is assigned three types of information:5 An entity type, describing the type of the entity it points to (e.g. person, location, organization, etc) An entity subtype, further detailing the type (e.g. organizations can be commercial, governmental and non-profit, while locations can be a nation, population center, or an international region) A mention type, specifying the way the entity is realized a mention can be named (e.g. John Smith), nominal (e.g. professor), or pronominal (e.g. she).',Marcus,'' +p,4811,'1 Introduction State-of-the-art part of speech -LRB- POS -RRB- tagging accuracy is now above 97 \% for newspaper text ',Collins,'num_part_1 nn_part_Introduction nn_part_State-of-the-art nsubj_is_part prep_part_of pobj_of_speech appos_speech_POS dep_speech_tagging dobj_tagging_accuracy advmod_is_now prep_is_above num_\%_97 pobj_above_\% prep_\%_for nn_text_newspaper pobj_for_text ' +o,4812,'This algorithm can thus be viewed as a large-margin version of the perceptron algorithm for structured outputs ',Collins,'det_algorithm_This nsubjpass_viewed_algorithm aux_viewed_can advmod_viewed_thus auxpass_viewed_be prep_viewed_as det_version_a amod_version_large-margin pobj_as_version prep_version_of det_algorithm_the nn_algorithm_perceptron pobj_of_algorithm prep_algorithm_for amod_outputs_structured pobj_for_outputs ' +o,4813,'We -LRB- : an tin -LRB- l 1 ; 11 -LRB- : sam -LRB- ; l ; yl -RRB- olop ; y in other works , -LRB- Ca rdi -LRB- : and Pierc ',Ramshaw,'det_tin_an dep_We_tin dep_tin_l dep_l_1 appos_tin_11 dep_11_sam dep_sam_l dep_;_yl dep_l_olop appos_l_y prep_y_in amod__other nn__works pobj_in_ nn_-LRB-_Ca nn_-LRB-_rdi cc_-LRB-_and conj_-LRB-_Pierc dep_We_ ' +o,4814,'Self-training is a commonly used technique for semi-supervised learning that has been ap532 plied to several natural language processing tasks ',Yarowsky,'nsubj_technique_Self-training cop_technique_is det_technique_a advmod_used_commonly amod_technique_used prep_technique_for amod_learning_semi-supervised pobj_for_learning nsubjpass_plied_that aux_plied_has auxpass_plied_been advmod_plied_ap532 rcmod_learning_plied prep_plied_to amod_tasks_several amod_tasks_natural nn_tasks_language nn_tasks_processing pobj_to_tasks ' +o,4815,'These instances can be retagged with their countability by using the proposed method and some kind of bootstrapping ',Yarowsky,'det_instances_These nsubjpass_retagged_instances aux_retagged_can auxpass_retagged_be prep_retagged_with poss_countability_their pobj_with_countability prep_retagged_by pcomp_by_using det_method_the amod_method_proposed dobj_using_method cc_method_and det_kind_some conj_method_kind prep_kind_of pobj_of_bootstrapping ' +o,4816,'Methods such as , and employ a synchronous parsing procedure to constrain a statistical alignment ',Wu,'nsubj_employ_Methods dep_as_such prep_Methods_as pobj_as_ conj__ cc__and conj__ det_procedure_a amod_procedure_synchronous nn_procedure_parsing nsubj_constrain_procedure aux_constrain_to xcomp_employ_constrain det_alignment_a amod_alignment_statistical dobj_constrain_alignment ' +o,4817,'We compared our system with the concepts in WordNet and Fleischman et als instance\\/concept relations ',Fleischman,'nsubj_compared_We poss_system_our dobj_compared_system prep_compared_with det_concepts_the pobj_with_concepts prep_compared_in nn_relations_WordNet cc_WordNet_and conj_WordNet_Fleischman nn_relations_et amod_relations_als nn_relations_instance\\/concept pobj_in_relations ' +o,4818,'Relatedness scores are computed for each pair of senses of the grammatically linked pair of words -LRB- w1 ; w2 ; GR -RRB- , using the WordNet-Similarity-103 package and the lesk 759 option ',Pedersen,'amod_scores_Relatedness nsubjpass_computed_scores auxpass_computed_are prep_computed_for det_pair_each pobj_for_pair prep_pair_of pobj_of_senses prep_senses_of det_pair_the amod_pair_grammatically amod_pair_linked pobj_of_pair prep_pair_of pobj_of_words abbrev_pair_w1 dep_w1_w2 dep_w2_GR conj_pair_using det_package_the amod_package_WordNet-Similarity-103 dobj_using_package cc_pair_and det_option_the nn_option_lesk num_option_759 conj_pair_option ' +p,4819,'However , the only known work which automates part of a customer service center using natural language dialogue is the one by ',Chu-Carroll,'advmod_one_However det_work_the advmod_known_only amod_work_known nsubj_one_work nsubj_automates_which rcmod_work_automates dobj_automates_part prep_part_of det_center_a nn_center_customer nn_center_service pobj_of_center partmod_center_using amod_dialogue_natural nn_dialogue_language dobj_using_dialogue cop_one_is det_one_the prep_one_by ' +o,4820,'The release has implementations for BLEU , WER and PER error criteria and it has decoding interfaces for Phramer and Pharaoh ',Papineni,'det_release_The nsubj_has_release dobj_has_implementations prep_implementations_for nn_criteria_BLEU nn_criteria_ conj__WER cc__and conj__PER nn_criteria_error pobj_for_criteria cc_has_and nsubj_has_it conj_has_has amod_interfaces_decoding dobj_has_interfaces prep_interfaces_for pobj_for_Phramer cc_Phramer_and conj_Phramer_Pharaoh ' +o,4821,'In particular , we used this method with WordNet and using the same training data ',Marcus,'prep_used_In pobj_In_particular nsubj_used_we det_method_this dobj_used_method prep_used_with pobj_with_WordNet xcomp_used_ cc__and conj__using det_data_the amod_data_same nn_data_training dobj__data ' +o,4822,'-LRB- Case-insensitive -RRB- BLEU-4 is used as the evaluation metric ',Papineni,'nn__BLEU-4 nsubjpass_used_ auxpass_used_is dep_Case-insensitive_used prep_used_as det_metric_the nn_metric_evaluation pobj_as_metric ' +o,4823,'The corpus is aligned in the word level using IBM Model4 ',Brown,'det_corpus_The nsubjpass_aligned_corpus auxpass_aligned_is prep_aligned_in det_level_the nn_level_word pobj_in_level xcomp_aligned_using nn_Model4_IBM dobj_using_Model4 ' +o,4824,'Other techniques have tried to quantify the generalizability of certain features across domains , or tried to exploit the common structure of related problems ','Daume III','amod_techniques_Other nsubj_tried_techniques aux_tried_have aux_quantify_to xcomp_tried_quantify det_generalizability_the dobj_quantify_generalizability prep_generalizability_of amod_features_certain pobj_of_features prep_quantify_across nn__domains pobj_across_ cc_tried_or conj_tried_tried aux_exploit_to xcomp_tried_exploit det_structure_the amod_structure_common dobj_exploit_structure prep_structure_of amod__related nn__problems pobj_of_ ' +o,4825,'3 Analysis Results 31 Kappa Statistic Kappa coefficient is commonly used as a standard to reflect inter-annotator agreement ',Carletta,'num_Results_3 nn_Results_Analysis num__31 nn__Kappa nn__Statistic nn__Kappa nn__coefficient nsubjpass_used_ auxpass_used_is advmod_used_commonly dep_Results_used prep_used_as det_standard_a pobj_as_standard aux_reflect_to xcomp_used_reflect amod_agreement_inter-annotator dobj_reflect_agreement ' +o,4826,'In this paper , Stanford Named Entity Recognizer is used to classify noun phrases into four semantic categories : PERSON , LOCATION , ORGANIZARION and MISC ',Finkel,'prep_used_In det_paper_this pobj_In_paper nn__Stanford nn__Named nn__Entity nn__Recognizer nsubjpass_used_ auxpass_used_is aux_classify_to xcomp_used_classify nn_phrases_noun dobj_classify_phrases prep_classify_into num_categories_four amod_categories_semantic pobj_into_categories dep_categories_PERSON conj_PERSON_LOCATION conj_PERSON_ORGANIZARION cc_PERSON_and conj_PERSON_MISC ' +o,4827,'55 Applying F-score Optimization Technique In addition , we can simply apply the F-score optimization technique for the sequence labeling tasks proposed in to boost the HySOL performance since the base discriminative models pD -LRB- y x -RRB- and discriminative combination , namely Equation -LRB- 3 -RRB- , in our hybrid model basically uses the same optimization procedure as CRFs ',Suzuki,'nsubj_uses_55 partmod_55_Applying nn_Technique_F-score nn_Technique_Optimization dobj_Applying_Technique prep_apply_In pobj_In_addition nsubj_apply_we aux_apply_can advmod_apply_simply rcmod_Technique_apply det_technique_the amod_technique_F-score nn_technique_optimization dobj_apply_technique prep_apply_for det_sequence_the pobj_for_sequence partmod_sequence_labeling dobj_labeling_tasks partmod_tasks_proposed prep_proposed_in pobj_in_ aux_boost_to xcomp_proposed_boost det_performance_the nn_performance_HySOL dobj_boost_performance prep_boost_since det_pD_the amod_pD_base amod_pD_discriminative nn_pD_models pobj_since_pD number_x_y dep_pD_x cc_apply_and amod_combination_discriminative conj_apply_combination advmod_3_namely amod_3_Equation appos_combination_3 dep_combination_in poss_model_our amod_model_hybrid pobj_in_model advmod_uses_basically det_procedure_the amod_procedure_same nn_procedure_optimization dobj_uses_procedure prep_procedure_as pobj_as_CRFs ' +o,4828,'32 Classifying speech segments in isolation In our experiments , we employed the well-known classifier SVMlight to obtain individual-document classification scores , treating Y as the positive class and using plain unigrams as features5 Following standard practice in sentiment analysis , the input to SVMlight consisted of normalized presence-of-feature -LRB- rather than frequency-of-feature -RRB- vectors ',Pang,'num_segments_32 nn_segments_Classifying nn_segments_speech prep_segments_in pobj_in_isolation dep_segments_In poss_experiments_our pobj_In_experiments appos_experiments_we partmod_we_employed det_SVMlight_the amod_SVMlight_well-known nn_SVMlight_classifier dobj_employed_SVMlight aux_obtain_to xcomp_employed_obtain amod_scores_individual-document nn_scores_classification dobj_obtain_scores dep_employed_treating dobj_treating_Y prep_Y_as det_class_the amod_class_positive pobj_as_class cc_treating_and conj_treating_using amod_unigrams_plain dobj_using_unigrams prep_using_as pobj_as_features5 prep_features5_Following amod_practice_standard pobj_Following_practice prep_practice_in nn__sentiment nn__analysis pobj_in_ det_input_the appos__input prep_input_to nn_consisted_SVMlight pobj_to_consisted prep_consisted_of amod_vectors_normalized dep_normalized_presence-of-feature dep_than_rather dep_vectors_than pobj_than_frequency-of-feature pobj_of_vectors ' +o,4829,'2 Word Alignment Framework A statistical translation model describes the relationship between a pair of sentences in the source and target languages -LRB- f = fJ1 , e = eI1 -RRB- using a translation probability P -LRB- f e -RRB- ',Och,'nsubj_Framework_2 nn_Framework_Word nn_Framework_Alignment det__A amod__statistical nn__translation nn__model nsubj_describes_ rcmod_Framework_describes det_relationship_the dobj_describes_relationship prep_relationship_between det_pair_a pobj_between_pair prep_pair_of pobj_of_sentences prep_sentences_in det_languages_the nn_languages_source cc_source_and conj_source_target pobj_in_languages dep_eI1_f dep_fJ1_= ccomp_eI1_fJ1 dep_eI1_e nsubj_eI1_= parataxis_describes_eI1 xcomp_describes_using det_P_a nn_P_translation nn_P_probability dobj_using_P dep_-LRB-_f dep_f_e ' +o,4830,'Based on these grammars , a great number of SMT models have been recently proposed , including string-to-string model -LRB- Synchronous FSG -RRB- , tree-to-string model -LRB- TSG-string -RRB- , string-totree model -LRB- string-CFG\\/TSG -RRB- , tree-to-tree model -LRB- Synchronous CFG\\/TSG , Data-Oriented Translation -RRB- and so on ',Koehn,'prep_proposed_Based dep_Based_on det_grammars_these pobj_on_grammars det_number_a amod_number_great appos_grammars_number prep_number_of nn_models_SMT pobj_of_models aux_proposed_have auxpass_proposed_been advmod_proposed_recently prep_proposed_including amod_model_string-to-string pobj_including_model nn_FSG_Synchronous abbrev_model_FSG nsubjpass_proposed_ amod__tree-to-string nn__model abbrev__TSG-string appos__ amod__string-totree nn__model appos__string-CFG\\/TSG appos__ amod_model_tree-to-tree appos__model nn_CFG\\/TSG_Synchronous appos_model_CFG\\/TSG amod_Translation_Data-Oriented dep_CFG\\/TSG_Translation partmod_model_ cc_so_and advmod__so dep_so_on ' +o,4831,'4 Evaluation As our algorithm works in open domains , we were able to perform a corpus-based evaluation using the Penn WSJ Treebank ',Marcus,'num_Evaluation_4 mark_works_As poss_algorithm_our nsubj_works_algorithm advcl_able_works prep_works_in amod_domains_open pobj_in_domains nsubj_able_we cop_able_were dep_Evaluation_able aux_perform_to xcomp_able_perform det_evaluation_a amod_evaluation_corpus-based dobj_perform_evaluation partmod_evaluation_using det_Treebank_the nn_Treebank_Penn nn_Treebank_WSJ dobj_using_Treebank ' +o,4832,'Co-training can be informally described in the following manner : # 0F Pick two -LRB- or more -RRB- views of a classification problem ',Yarowsky,'amod__Co-training nsubjpass_described_ aux_described_can auxpass_described_be advmod_described_informally prep_described_in det_manner_the amod_manner_following pobj_in_manner dep_0F_# nsubj_views_0F dep_0F_Pick num_Pick_two cc_more_or dep_0F_more parataxis_described_views prep_views_of det_problem_a nn_problem_classification pobj_of_problem ' +o,4833,'Past work has synchronously binarized such rules for efficiency ',Zhang,'amod_work_Past nsubj_binarized_work aux_binarized_has advmod_binarized_synchronously amod_rules_such dobj_binarized_rules prep_binarized_for pobj_for_efficiency ' +o,4834,'The features are similar to the ones used in phrasal systems , and their weights are trained using max-BLEU training ',Och,'det_features_The nsubj_similar_features cop_similar_are prep_similar_to det_ones_the pobj_to_ones partmod_ones_used prep_used_in amod_systems_phrasal pobj_in_systems cc_similar_and poss_weights_their nsubjpass_trained_weights auxpass_trained_are conj_similar_trained xcomp_trained_using amod_training_max-BLEU dobj_using_training ' +o,4835,'Thus , also proposed an averaged perceptron , where the nal weight vector is 1Collins -LRB- 2002a -RRB- alsoprovidedproofthatguaranteedgood learning for the non-separable case ',Collins,'advmod_proposed_Thus nsubj_proposed_ advmod_proposed_also det_perceptron_an amod_perceptron_averaged dobj_proposed_perceptron advmod_1Collins_where det_vector_the nn_vector_nal nn_vector_weight nsubj_1Collins_vector cop_1Collins_is rcmod_perceptron_1Collins appos_1Collins_2002a partmod_1Collins_alsoprovidedproofthatguaranteedgood xcomp_alsoprovidedproofthatguaranteedgood_learning prep_learning_for det_case_the amod_case_non-separable pobj_for_case ' +o,4836,'However , as discussed in prior arts and this paper , linguistically-informed SCFG is an inadequate model for parallel corpora due to its nature that only allowing child-node reorderings ',Galley,'advmod_model_However mark_discussed_as dep_model_discussed prep_discussed_in amod__prior nn__arts pobj_in_ cc__and det_paper_this conj__paper amod_SCFG_linguistically-informed nsubj_model_SCFG cop_model_is det_model_an amod_model_inadequate prep_model_for amod_corpora_parallel pobj_for_corpora amod_corpora_due prep_due_to poss_nature_its pobj_to_nature nsubj_allowing_that advmod_allowing_only rcmod_nature_allowing amod_reorderings_child-node dobj_allowing_reorderings ' +o,4837,'In particular , Hockenmaier and Steedman report a generative model for CCG parsing roughly akin to the Collins parser specific to CCG ',Collins,'prep_report_In pobj_In_particular nn__Hockenmaier cc_Hockenmaier_and conj_Hockenmaier_Steedman nsubj_report_ det_model_a amod_model_generative dobj_report_model prep_model_for pobj_for_CCG partmod_CCG_parsing advmod_akin_roughly acomp_parsing_akin prep_akin_to det__the nn__Collins nn__parser pobj_to_ amod__specific prep_specific_to pobj_to_CCG ' +o,4838,'This weak supervision has been encoded using priors and initializations , specialized models , and implicit negative evidence ',Smith,'det_supervision_This amod_supervision_weak nsubjpass_encoded_supervision aux_encoded_has auxpass_encoded_been xcomp_encoded_using dobj_using_priors cc_priors_and nn__initializations conj_priors_ amod__specialized nn__models conj_priors_ cc_priors_and amod_evidence_implicit amod_evidence_negative conj_priors_evidence ' +o,4839,'The yield of this tree gives the target translation : the gunman was killed by police The Penn English Treebank -LRB- PTB -RRB- is our source of syntactic information , largely due to the availability of reliable parsers ',Marcus,'det_yield_The nsubj_gives_yield prep_yield_of det_tree_this pobj_of_tree det_translation_the nn_translation_target iobj_gives_translation det_gunman_the nsubjpass_killed_gunman auxpass_killed_was dep_translation_killed prep_killed_by pobj_by_police det_Treebank_The nn_Treebank_Penn nn_Treebank_English dobj_gives_Treebank abbrev_Treebank_PTB nsubj_source_ cop_source_is poss_source_our rcmod_Treebank_source prep_source_of amod_information_syntactic pobj_of_information advmod_due_largely amod_source_due prep_due_to det_availability_the pobj_to_availability prep_availability_of amod_parsers_reliable pobj_of_parsers ' +o,4840,'This is comparable to the accuracy of 9629 \% reported by on the newswire domain ','Daume III','nsubj_comparable_This cop_comparable_is prep_comparable_to det_accuracy_the pobj_to_accuracy prep_accuracy_of num_\%_9629 pobj_of_\% partmod_\%_reported prep_reported_by pobj_by_ prep_reported_on det_domain_the nn_domain_newswire pobj_on_domain ' +o,4841,'For example , incremental CFG parsing algorithms can be used with the CFGs produced by this transform , as can the Inside-Outside estimation algorithm and more exotic methods such as estimating adjoined hidden states ',Matsuzaki,'prep_used_For pobj_For_example amod_algorithms_incremental nn_algorithms_CFG nn_algorithms_parsing nsubjpass_used_algorithms aux_used_can auxpass_used_be prep_used_with det_CFGs_the pobj_with_CFGs partmod_CFGs_produced prep_produced_by det_transform_this pobj_by_transform dep__as aux__can det_algorithm_the amod_algorithm_Inside-Outside nn_algorithm_estimation nsubj__algorithm dep_transform_ cc_more_and num_methods_more amod_methods_exotic dobj__methods dep_as_such prep_methods_as pcomp_as_estimating amod_states_adjoined amod_states_hidden dobj_estimating_states ' +o,4842,'A few studies addressed this defect by selecting the appropriate translation rules for an input span based on its context in the input sentence ',Hasan,'det__A amod__few nn__studies nsubj_addressed_ det_defect_this dobj_addressed_defect prep_addressed_by pcomp_by_selecting det_rules_the amod_rules_appropriate nn_rules_translation dobj_selecting_rules prep_rules_for det_span_an nn_span_input pobj_for_span prep_selecting_based dep_based_on poss_context_its pobj_on_context prep_context_in det_sentence_the nn_sentence_input pobj_in_sentence ' +o,4843,'Tag sets for English are derived from the Penn Treebank ',Marcus,'nsubj_sets_Tag mark_derived_for nsubjpass_derived_English auxpass_derived_are advcl_sets_derived prep_derived_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,4844,'In , target trees were employed to improve the scoring of translation theories ',He,'prep_employed_In pobj_In_ nn_trees_target nsubjpass_employed_trees auxpass_employed_were aux_improve_to xcomp_employed_improve det_scoring_the dobj_improve_scoring prep_scoring_of nn_theories_translation pobj_of_theories ' +o,4845,'First , splitting and merging of sentences , which seems related to content planning and aggregation ',Jing,'conj_First_splitting cc_First_and conj_First_merging prep_First_of nn__sentences pobj_of_ nsubj_related_which cop_related_seems rcmod__related aux_content_to xcomp_related_content dobj_content_planning cc_planning_and conj_planning_aggregation ' +o,4846,'We referred to the studies of ',Berger,'nsubj_referred_We prep_referred_to det_studies_the pobj_to_studies prep_referred_of ' +o,4847,'For this experiment , we used sections 02 21 of the Penn Treebank -LRB- PTB -RRB- as the training data and section 23 -LRB- 2416 sentences -RRB- for evaluation , as is now standard ',Marcus,'prep_sections_For det_experiment_this pobj_For_experiment nsubj_sections_we dep_sections_used number_21_02 dep_of_21 prep_sections_of det_data_the nn__Penn nn__Treebank abbrev__PTB nn_data_ prep__as det_training_the pobj_as_training pobj_of_data cc_data_and conj_data_section num_section_23 num_sentences_2416 appos_section_sentences prep_sections_for pobj_for_evaluation mark_standard_as cop_standard_is advmod_standard_now advcl_sections_standard ' +o,4848,'The hypothesis scores and tuning are identical to the setup used in ',Rosti,'det_scores_The nn_scores_hypothesis nsubj_identical_scores cc_scores_and conj_scores_tuning cop_identical_are prep_identical_to det_setup_the pobj_to_setup partmod_setup_used prep_used_in ' +o,4849,'AL has already been applied to several NLP tasks , such as document classification , POS tagging , chunking , statistical parsing , and information extraction ',Berger,'nsubjpass_applied_AL aux_applied_has advmod_applied_already auxpass_applied_been prep_applied_to amod_tasks_several nn_tasks_NLP pobj_to_tasks dep_as_such prep_tasks_as nn__document nn__classification pobj_as_ dep__POS nn__tagging conj__ amod__chunking conj__ amod__statistical nn__parsing conj__ cc__and nn__information nn__extraction conj__ ' +o,4850,'Only the measures provided by LESK , HSO , VEC , , and provide a method for predicting adjective similarities ; of these , only LESK and VEC outperform the uninformed baseline on adjectives , while our learned measure achieves a 40 \% improvement over the LESK measure on adjectives ',Navigli,'advmod_measures_Only det_measures_the nsubj_provided_measures prep_provided_by nn__LESK appos__HSO appos__VEC pobj_by_ cc_provided_and nsubj_provide_ conj_provided_provide det_method_a dobj_provide_method prep_provide_for pcomp_for_predicting nn_similarities_adjective dobj_predicting_similarities prep_outperform_of pobj_of_these advmod_LESK_only nsubj_outperform_LESK cc_LESK_and conj_LESK_VEC parataxis_provided_outperform det_baseline_the amod_baseline_uninformed dobj_outperform_baseline prep_baseline_on pobj_on_adjectives mark_achieves_while poss_measure_our amod_measure_learned nsubj_achieves_measure advcl_outperform_achieves det_improvement_a dep_\%_40 amod_improvement_\% dobj_achieves_improvement prep_improvement_over det_measure_the nn_measure_LESK pobj_over_measure prep_achieves_on pobj_on_adjectives ' +o,4851,'Phrase tables were learned from the training corpus using the diag-and method , and using IBM model 2 to produce initial word alignments -LRB- these authors found this worked as well as IBM4 -RRB- ',Koehn,'amod_tables_Phrase nsubjpass_learned_tables auxpass_learned_were prep_learned_from det_corpus_the nn_corpus_training pobj_from_corpus dep_corpus_using det__the amod__diag-and nn__method dobj_using_ cc_using_and conj_using_using nn_model_IBM nsubj_produce_model num_model_2 aux_produce_to xcomp_using_produce amod_alignments_initial nn_alignments_word dobj_produce_alignments det_authors_these nsubj_found_authors dep_corpus_found dobj_found_this partmod_this_worked dep_well_as cc_this_well dep_well_as conj_this_IBM4 ' +o,4852,'The marginal relevance systems -LRB- MR and MR+IE -RRB- used a simple selection mechanism which does not involve search , inspired by the maximal marginal relevance -LRB- MMR -RRB- approach ',Goldstein,'det_systems_The amod_systems_marginal nn_systems_relevance nsubj_approach_systems appos_systems_MR cc_MR_and conj_MR_MR+IE partmod_systems_used det_mechanism_a amod_mechanism_simple nn_mechanism_selection dobj_used_mechanism nsubj_involve_which aux_involve_does neg_involve_not rcmod_mechanism_involve dobj_involve_search partmod_systems_inspired prep_inspired_by det_relevance_the amod_relevance_maximal amod_relevance_marginal pobj_by_relevance abbrev_relevance_MMR ' +o,4853,'Similar to bidirectional labeling in , there are two learning tasking in this model ',Shen,'ccomp_are_Similar aux_bidirectional_to xcomp_Similar_bidirectional xcomp_bidirectional_labeling prep_labeling_in pobj_in_ expl_are_there nsubj_are_two partmod_two_learning xcomp_learning_tasking prep_tasking_in det_model_this pobj_in_model ' +o,4854,'It is possible to use unsupervised learning to train stochastic taggers without the need for a manually annotated corpus by using the Baum-Welch algorithm ',Cutting,'nsubj_possible_It cop_possible_is aux_use_to xcomp_possible_use dep_use_unsupervised xcomp_unsupervised_learning aux_train_to xcomp_learning_train amod_taggers_stochastic dobj_train_taggers prep_train_without det_need_the pobj_without_need prep_need_for det_corpus_a amod_corpus_manually amod_corpus_annotated pobj_for_corpus prep_train_by pcomp_by_using det_algorithm_the amod_algorithm_Baum-Welch dobj_using_algorithm ' +p,4855,'1 Introduction Robust statistical syntactic parsers , made possible by new statistical techniques and by the availability of large , hand-annotated training corpora such as WSJ and Switchboard , have had a major impact on the field of natural language processing ',Marcus,'num_parsers_1 nn_parsers_Introduction nn_parsers_Robust amod_parsers_statistical amod_parsers_syntactic nsubj_had_parsers partmod_parsers_made acomp_made_possible prep_made_by amod__new amod__statistical nn__techniques pobj_by_ cc_by_and conj_by_by det_availability_the pobj_by_availability prep_availability_of amod_corpora_large amod_corpora_hand-annotated nn_corpora_training pobj_of_corpora dep_as_such prep_availability_as nn__WSJ pobj_as_ cc__and nn__Switchboard conj__ aux_had_have det_impact_a amod_impact_major dobj_had_impact prep_had_on det_field_the pobj_on_field prep_field_of amod_processing_natural nn_processing_language pobj_of_processing ' +o,4856,'For instance , work has been done in Chinese using the Penn Chinese Treebank , in Czech using the Prague Dependency Treebank , in French using the French Treebank , in German using the Negra Treebank , and in Spanish using the UAM Spanish Treebank ',Collins,'prep_done_For pobj_For_instance nsubjpass_done_work aux_done_has auxpass_done_been prep_done_in nsubj_using_Chinese pcomp_in_using det__the nn__Penn nn__Chinese nn__Treebank dobj_using_ prep_done_in nsubj_using_Czech pcomp_in_using det__the nn__Prague nn__Dependency nn__Treebank dobj_using_ prep_done_in pobj_in_French partmod_French_using det__the amod__French nn__Treebank dobj_using_ prep_using_in nsubj_using_German pcomp_in_using det__the nn__Negra nn__Treebank dobj_using_ cc_in_and conj_in_in pobj_in_Spanish partmod_Spanish_using det__the nn__UAM nn__Spanish nn__Treebank dobj_using_ ' +o,4857,'Statistical machine translation is based on the noisy channel model , where the translation hypothesis is searched over the space defined by a translation model and a target language ',Brown,'nn_translation_Statistical nn_translation_machine nsubjpass_based_translation auxpass_based_is prep_based_on det_model_the amod_model_noisy nn_model_channel pobj_on_model advmod_searched_where det_hypothesis_the nn_hypothesis_translation nsubj_searched_hypothesis cop_searched_is rcmod_model_searched prep_searched_over det_space_the pobj_over_space partmod_space_defined prep_defined_by det_model_a nn_model_translation pobj_by_model cc_model_and det_language_a nn_language_target conj_model_language ' +o,4858,'The terms graph-based and transition-based were used by to describe the difference between MSTParser , which is a graph-based parser with an exhaustive search decoder , and MaltParser , which is a transition-based parser with a greedy search decoder ',McDonald,'det_terms_The nsubjpass_used_terms amod_terms_graph-based cc_graph-based_and conj_graph-based_transition-based auxpass_used_were prep_used_by pobj_by_ aux_describe_to xcomp_used_describe det_difference_the dobj_describe_difference prep_difference_between nn__MSTParser pobj_between_ nsubj_parser_which cop_parser_is det_parser_a amod_parser_graph-based rcmod__parser prep_parser_with det_decoder_an amod_decoder_exhaustive nn_decoder_search pobj_with_decoder cc_difference_and nn__MaltParser conj_difference_ nsubj_parser_which cop_parser_is det_parser_a amod_parser_transition-based rcmod__parser prep_parser_with det_decoder_a amod_decoder_greedy nn_decoder_search pobj_with_decoder ' +o,4859,'One of the most relevant work is , which proposed to integrate various patterns in order to measure semantic similarity between words ',Bollegala,'nsubj__One prep_One_of det_work_the advmod_relevant_most amod_work_relevant pobj_of_work cop__is nsubj_proposed_which rcmod__proposed aux_integrate_to xcomp_proposed_integrate amod_patterns_various dobj_integrate_patterns mark_measure_in dep_measure_order aux_measure_to purpcl_integrate_measure amod_similarity_semantic dobj_measure_similarity prep_similarity_between pobj_between_words ' +o,4860,'Either pruning or lossy randomizing approaches may result in a compact representation for the application run-time ',Talbot,'dep_result_Either nn__pruning nsubj_result_ cc__or nn_approaches_lossy nn_approaches_randomizing conj__approaches advmod_result_ aux_result_may prep_result_in det_representation_a amod_representation_compact pobj_in_representation prep_representation_for det_run-time_the nn_run-time_application pobj_for_run-time ' +p,4861,'For the extraction problem , there have been various methods proposed to date , which are quite adequate ',Smadja,'prep_been_For det_problem_the nn_problem_extraction pobj_For_problem expl_been_there aux_been_have amod_methods_various dobj_been_methods partmod_methods_proposed prep_proposed_to pobj_to_date nsubj_adequate_which cop_adequate_are advmod_adequate_quite rcmod_date_adequate ' +o,4862,'Actually , now that SMT has reached some maturity , we see several attempts to integrate more structure into these systems , ranging from simple hierarchical alignment models to syntax-based statistical systems ',Wu,'advmod_see_Actually advmod_reached_now dep_reached_that nsubj_reached_SMT aux_reached_has dep_see_reached det_maturity_some dobj_reached_maturity nsubj_see_we amod_attempts_several nsubj_integrate_attempts aux_integrate_to xcomp_see_integrate amod_structure_more dobj_integrate_structure prep_integrate_into det_systems_these pobj_into_systems xcomp_integrate_ranging prep_ranging_from amod_models_simple amod_models_hierarchical nn_models_alignment pobj_from_models partmod_models_ prep__to amod__syntax-based amod__statistical nn__systems pobj_to_ ' +o,4863,'This is in sharp contrast to the smoothed fixed-word statistics in most lexicalized parsing models derived from sparse data -LRB- Magerman , , Charniak -LRB- 1997 -RRB- , etc -RRB- ',Collins,'nsubj_is_This prep_is_in amod_contrast_sharp pobj_in_contrast prep_contrast_to det_statistics_the amod_statistics_smoothed amod_statistics_fixed-word pobj_to_statistics prep_is_in advmod_lexicalized_most amod_models_lexicalized nn_models_parsing pobj_in_models partmod_models_derived prep_derived_from amod_data_sparse pobj_from_data nn__Magerman dep_data_ appos__ nn_1997_Charniak appos__1997 dep__etc ' +o,4864,' has induced clusters by mapping WordNet senses to a more coarse-grained lexical resource ',Navigli,'nsubj_induced_ aux_induced_has dobj_induced_clusters prep_induced_by pcomp_by_mapping nn_senses_WordNet dobj_mapping_senses prep_mapping_to det_resource_a advmod_coarse-grained_more amod_resource_coarse-grained amod_resource_lexical pobj_to_resource ' +o,4865,'Thus , given a hyponym definition -LRB- O -RRB- and a set of candidate hypernym definitions , this method selects the candidate hypernym definition -LRB- E -RRB- which returns the maximum score given by formula -LRB- 1 -RRB- : SC -LRB- O , E -RRB- : E cw -LRB- wi , wj -RRB- -LRB- I -RRB- ` wIEOAwj6E The cooccurrence weight -LRB- cw -RRB- between two words can be given by Cooccurrence Frequency , Mutual Information or Association Ratio ',Church,'advmod_selects_Thus prep_selects_given det_definition_a nn_definition_hyponym dep_given_definition appos_definition_O cc_definition_and det_set_a conj_definition_set prep_set_of nn_definitions_candidate nn_definitions_hypernym pobj_of_definitions det_method_this nsubj_selects_method det_definition_the nn_definition_candidate nn_definition_hypernym dobj_selects_definition appos_definition_E nsubj_returns_which rcmod_definition_returns det_score_the amod_score_maximum dobj_returns_score prep_returns_given dep_given_by pobj_by_formula appos_formula_1 dep_definition_SC appos_SC_O dep_O_E nn_cw_E dep_definition_cw appos_cw_wi dep_wi_wj dep_cw_I dep_definition_wIEOAwj6E det_weight_The amod_weight_cooccurrence nsubjpass_given_weight appos_weight_cw prep_weight_between num_words_two pobj_between_words aux_given_can auxpass_given_be dep_selects_given prep_given_by nn_Frequency_Cooccurrence pobj_by_Frequency nn__Mutual nn__Information conj_Frequency_ cc_Frequency_or nn__Association nn__Ratio conj_Frequency_ ' +o,4866,'The weights of the models are computed automatically using a variant of the Maximum Bleu training procedure proposed by ',Och,'det_weights_The nsubjpass_computed_weights prep_weights_of det_models_the pobj_of_models auxpass_computed_are advmod_computed_automatically xcomp_computed_using det_variant_a dobj_using_variant prep_variant_of det_procedure_the nn_procedure_Maximum nn_procedure_Bleu nn_procedure_training pobj_of_procedure partmod_procedure_proposed prep_proposed_by ' +n,4867,'Automatic evaluation methods such as BLEU , RED , or the weighted N-gram model proposed here may be more consistent in judging quality as compared to human evaluators , but human judgments remain the only criteria for metaevaluating the automatic methods ',Papineni,'nn_methods_Automatic nn_methods_evaluation nsubj_proposed_methods dep_as_such prep_methods_as nn__BLEU pobj_as_ nn__RED conj__ cc__or det_model_the amod_model_weighted amod_model_N-gram conj__model nsubj_consistent_here aux_consistent_may cop_consistent_be advmod_consistent_more dep_proposed_consistent prep_consistent_in pcomp_in_judging dobj_judging_quality advmod_quality_as prep_quality_compared dep_compared_to amod_evaluators_human pobj_to_evaluators cc_consistent_but amod_judgments_human nsubj_criteria_judgments cop_criteria_remain det_criteria_the amod_criteria_only conj_consistent_criteria prep_criteria_for pcomp_for_metaevaluating det_methods_the amod_methods_automatic dobj_metaevaluating_methods ' +o,4868,'In , we proposed a framework that automatically evaluated automatic MT evaluation metrics using only manual translations without further human involvement ',Lin,'prep_proposed_In pobj_In_ nsubj_proposed_we det_framework_a dobj_proposed_framework nsubj_evaluated_that advmod_evaluated_automatically rcmod_framework_evaluated amod_metrics_automatic nn_metrics_MT nn_metrics_evaluation dobj_evaluated_metrics partmod_metrics_using advmod_translations_only amod_translations_manual dobj_using_translations prep_using_without amod_involvement_further amod_involvement_human pobj_without_involvement ' +p,4869,'robust mforrmatlon extractlon, and readlly-avmlable on-hne NLP resources These techtuques and resources allow us to create a richer indexed source of Imgmstlc and domain knowledge than other frequency approaches Our approach attempts to apprommate text dlscourse structure through these multlple layers of mformatlon, ohtinned from automated methods m contrast to labor-lntenslve, discourse-based approaches Moreover, our planned training methodology will also allow us to explmt thin productlve infrastructure m ways whlch model human performance whde avoidmg hand-crafting domain-dependent rules of the knowledge-based approaches Our ultlmate goal m to make our summarlzatlon system scalable and portable by learning summarization rules from easily extractable text features 2 System Description Our summarization system DlmSum consmts of the Summarization Server and the Summarlzatzon Chent The Server extracts features (the Feature Extractor) from a document using various robust NLP techmques, described In Sectzon 2 1, and combines these features (the Feature Combiner) to basehne multiple combinations of features, as described m Section 2 2 Our work m progress to automattcally tram the Feature Combiner based upon user and apphcatlon needs m presented in Section 2 2 2 The Java-based Chent, which wdl be dmcnssed In Section 4, provides a graphical user interface (GUI) for the end user to cnstomlze the summamzatlon preferences and see multiple views of generated sumInarles 2.1 Extracting Stlmmarization Features In this section, we describe how we apply robust NLP technology to extract summarization features Our goal IS to add more mtelhgence to frequencybased approaches, to acqmre domain knowledge In a more automated fashion, and to apprommate text structure by recogmzing sources of dmcourse cohesion and coherence 2.1.1 Going Beyond a Word Frequency-based summarization systems typically use a single word stnng as a umt for counting frequencies Whde such a method IS very robust, it ignores the semantic content of words and their potential membership m multi-word phrases For example, zt does not dmtmgumh between \'bill\' m \'Bdl Table 1 Collocations with \'chlps\' {potato tortdla corn chocolate b~gle} chips {computer pentmm Intel macroprocessor memory} chips {wood oak plastlc} cchlps bsrgmmng clups blue clups mr chips Clmton\' and \'bill\' in \'reform bill\' This may introduce noise m frequency counting as the same strmgs are treated umformly no matter how the context may have dmamblguated the sense or regardless of membership in multl-word phrases For DlrnSum, we use term frequency based on tf*Idf (Salton and McGdl, 1983, Brandow, Mitze, and Rau, 1995) to derive ssgnature words as one of the summarization features If single words were the sole basra of countmg for our summarization application, nome would be introduced both m term frequency and reverse document frequency However, recent advances in statmtlcal NLP and information extraction make it possible to utilize features which go beyond the single word level Our approach is to extract multi-word phrases automatlcally with high accuracy and use them as the basic unit in the summarization process, including frequency calculation Ftrst, just as word association methods have proven effective m lemcal analysis, e g (Church and Hanks, 1990), we are exploring whether frequently occurring Collocatlonal reformation can improve on simple word-based approaches We have preprocessed about 800 MB of LA tlmes/Wastnngton Post newspaper articles nsmg a POS tagger (Bnll, 1993) and derived two-word noun collocations using mutual information The.',Church,'' +o,4870,'The training algorithm we used is the improved iterative scaling -LRB- IIS -RRB- described in 3 ',Berger,'det_algorithm_The nn_algorithm_training nsubj_scaling_algorithm nsubj_used_we rcmod_algorithm_used cop_scaling_is det_scaling_the amod_scaling_improved nn_scaling_iterative abbrev_scaling_IIS partmod_scaling_described prep_described_in number_3_ pobj_in_3 ' +o,4871,'Performance is measured by computing the BLEU scores of the systems translations , when compared against a single reference translation per sentence ',Papineni,'nsubjpass_measured_Performance auxpass_measured_is prep_measured_by pcomp_by_computing det__the amod__BLEU nn__scores dobj_computing_ prep__of det_translations_the nn_translations_systems pobj_of_translations advmod_compared_when advcl_measured_compared prep_compared_against det_translation_a amod_translation_single nn_translation_reference pobj_against_translation prep_translation_per pobj_per_sentence ' +o,4872,'In previous work , I described a Maximum Entropy\\/Minimum Divergence -LRB- MEMD -RRB- model for p -LRB- w -LRB- hi , s -RRB- which incorporates a trigram language model and a translation component which is an analog of the well-known IBM translation model 1 ',Berger,'prep_described_In amod__previous nn__work pobj_In_ nsubj_described_I det__a nn_Divergence_Maximum nn_Divergence_Entropy\\/Minimum nn__Divergence abbrev_Divergence_MEMD nn__model dobj_described_ prep__for pobj_for_p dep_p_w dep_p_hi dep_hi_s nsubj_incorporates_which rcmod__incorporates det_model_a nn_model_trigram nn_model_language dobj_incorporates_model cc_model_and det_component_a nn_component_translation conj_model_component nsubj_analog_which cop_analog_is det_analog_an rcmod_component_analog prep_analog_of det__the amod__well-known nn__IBM nn__translation nn__model num__1 pobj_of_ ' +o,4873,'Lexical cues of differing complexities have been used , including single words and Ngrams -LRB- eg , -RRB- , as well as phrases and lexico-syntactic patterns -LRB- eg , -RRB- ',Pang,'amod_cues_Lexical nsubjpass_used_cues prep_cues_of amod_complexities_differing pobj_of_complexities aux_used_have auxpass_used_been prep_used_including amod_words_single pobj_including_words cc_words_and conj_words_Ngrams appos_words_eg dep_eg_ dep_well_as cc_words_well dep_well_as conj_words_phrases cc_phrases_and amod_patterns_lexico-syntactic conj_phrases_patterns appos_patterns_eg dep_eg_ ' +n,4874,'For comparison purposes , we revisit fully-generative Bayesian model for unsupervised coreference resolution , discuss its potential weaknesses and consequently propose three modifications to their model ',Haghighi,'prep_revisit_For nn_purposes_comparison pobj_For_purposes nsubj_revisit_we dep_revisit_ amod_model_fully-generative amod_model_Bayesian dobj__model prep_model_for amod_resolution_unsupervised nn_resolution_coreference pobj_for_resolution conj__discuss poss_weaknesses_its amod_weaknesses_potential dobj_discuss_weaknesses cc__and advmod__consequently conj__propose num_modifications_three dobj_propose_modifications prep_propose_to poss_model_their pobj_to_model ' +o,4875,'-LRB- A similar intuition holds for the Machine Translation models generically known as the IBM models , which assume that certain words in a source language sentence tend to trigger the usage of certain words in a target language translation of that sentence -RRB- ',Brown,'det_intuition_A amod_intuition_similar nsubj_holds_intuition prep_holds_for det_models_the nn_models_Machine nn_models_Translation pobj_for_models advmod_known_generically partmod_models_known prep_known_as det__the nn__IBM nn__models pobj_as_ nsubj_assume_which rcmod__assume complm_tend_that amod_words_certain nsubj_tend_words prep_words_in det_sentence_a nn_sentence_source nn_sentence_language pobj_in_sentence ccomp_assume_tend aux_trigger_to xcomp_tend_trigger det_usage_the dobj_trigger_usage prep_usage_of amod_words_certain pobj_of_words prep_trigger_in det_translation_a nn_translation_target nn_translation_language pobj_in_translation prep_translation_of det_sentence_that pobj_of_sentence ' +o,4876,'been put forward by ',Ramshaw,'auxpass_put_been advmod_put_forward prep_put_by ' +o,4877,'Substring-based transliteration with a generative hybrid model is very similar to existing solutions for phrasal SMT , operating on characters rather than words ',Koehn,'amod_transliteration_Substring-based nsubj_operating_transliteration prep_transliteration_with det_model_a amod_model_generative amod_model_hybrid pobj_with_model cop_operating_is advmod_similar_very acomp_operating_similar prep_similar_to amod_solutions_existing pobj_to_solutions prep_operating_for amod__phrasal nn__SMT pobj_for_ prep_operating_on pobj_on_characters cc_characters_rather dep_rather_than conj_characters_words ' +o,4878,'In the post-editing step , a prediction engine helps to decrease the amount of human interaction ',Och,'prep_helps_In det_step_the amod_step_post-editing pobj_In_step det_engine_a nn_engine_prediction nsubj_helps_engine aux_decrease_to xcomp_helps_decrease det_amount_the dobj_decrease_amount prep_amount_of amod_interaction_human pobj_of_interaction ' +o,4879,'1 is a set of assumptions sufficient to support the inI , ` rl -RRB- n ` lation given S and R In other words , this is h , ~ crl -RRB- rctal , ion as abduction \' , since ~ ! -RRB- -LRB- i , -LRB- ` lion , not deduction , is needed to arrive at the : ~ -RRB- \'d H II I ~ tiOIIS ,4 ',Hobbs,'nsubj_set_1 cop_set_is det_set_a prep_set_of pobj_of_assumptions amod_S_sufficient aux_support_to xcomp_sufficient_support det_inI_the dobj_support_inI nn_n_rl appos_inI_n amod_S_lation amod_S_given nsubj_\'d_S cc_S_and conj_S_R prep_rctal_In amod_words_other pobj_In_words nsubj_rctal_this cop_rctal_is nn_rctal_h amod_rctal_~ nn_rctal_crl rcmod_R_rctal appos_R_ion prep_ion_as poss__abduction pobj_as_ dep_~_since dep__~ dep__i nn_deduction_lion neg_deduction_not nsubjpass_needed_deduction auxpass_needed_is dep_R_needed aux_arrive_to xcomp_needed_arrive advmod_arrive_at dep_at_the advmod_arrive_~ dep_set_\'d nn_II_H dobj_\'d_II nsubj_tiOIIS_I advmod_tiOIIS_~ rcmod_II_tiOIIS dobj_tiOIIS_,4 ' +o,4880,'Algorithms for the computation of first-order associations have been used in lexicography for the extraction of collocations and in cognitive psychology for the simulation of associative learning ',Smadja,'nsubjpass_used_Algorithms prep_Algorithms_for det_computation_the pobj_for_computation prep_computation_of amod_associations_first-order pobj_of_associations aux_used_have auxpass_used_been prep_used_in pobj_in_lexicography prep_lexicography_for det_extraction_the pobj_for_extraction prep_extraction_of amod__collocations pobj_of_ cc_in_and conj_in_in amod_psychology_cognitive pobj_in_psychology prep_psychology_for det_simulation_the pobj_for_simulation prep_simulation_of amod__associative nn__learning pobj_of_ ' +o,4881,'While traditional approaches to syntax based MT were dependent on availability of manual grammar , more recent approaches operate within the resources of PB-SMT and induce hierarchical or linguistic grammars from existing phrasal units , to provide better generality and structure for reordering ',Wu,'mark_syntax_While amod_approaches_traditional nsubj_syntax_approaches aux_syntax_to advcl_operate_syntax prep_syntax_based nsubj_dependent_MT cop_dependent_were dep_based_dependent prep_dependent_on pobj_on_availability prep_availability_of amod_grammar_manual pobj_of_grammar advmod_approaches_more amod_approaches_recent nsubj_operate_approaches prep_operate_within det_resources_the pobj_within_resources prep_resources_of pobj_of_PB-SMT cc_operate_and conj_operate_induce amod_grammars_hierarchical cc_hierarchical_or conj_hierarchical_linguistic dobj_induce_grammars prep_induce_from amod_units_existing amod_units_phrasal pobj_from_units aux_provide_to xcomp_induce_provide amod_generality_better dobj_provide_generality cc_generality_and conj_generality_structure prep_generality_for pobj_for_reordering ' +o,4882,'The use of such relations -LRB- mainly relations between verbs or nouns and their arguments and modifiers -RRB- for various purposes has received growing attention in recent research ',Hindle,'det_use_The nsubj_received_use prep_use_of amod_relations_such pobj_of_relations advmod_relations_mainly dep_relations_relations prep_relations_between pobj_between_verbs cc_verbs_or conj_verbs_nouns cc_verbs_and poss_arguments_their conj_verbs_arguments cc_arguments_and conj_arguments_modifiers prep_relations_for amod_purposes_various pobj_for_purposes aux_received_has amod_attention_growing dobj_received_attention prep_attention_in amod_research_recent pobj_in_research ' +o,4883,'Nevertheless , as and others have argued , semantic representations for natural language need not be higher-order in that ontological promiscuity can solve the problem ',Hobbs,'advmod_higher-order_Nevertheless mark_argued_as nsubj_argued_ cc__and conj__others aux_argued_have dep_higher-order_argued amod_representations_semantic nsubj_higher-order_representations prep_representations_for amod_language_natural pobj_for_language aux_higher-order_need neg_higher-order_not cop_higher-order_be prep_higher-order_in complm_solve_that amod_promiscuity_ontological nsubj_solve_promiscuity aux_solve_can pcomp_in_solve det_problem_the dobj_solve_problem ' +p,4884,' , whose training corpus for the noun drug was 9 times bigger than that of Karov and Edelman , reports 914 \% correct performance improved to impressive 939 \% when using the ` one sense per discourse \' constraint ',Yarowsky,'nsubj_reports_ poss_corpus_whose nn_corpus_training nsubj_bigger_corpus prep_corpus_for det_drug_the nn_drug_noun pobj_for_drug cop_bigger_was num_times_9 measure_bigger_times rcmod__bigger prep_bigger_than pobj_than_that prep_that_of pobj_of_Karov cc_Karov_and conj_Karov_Edelman num_\%_914 dobj_reports_\% amod_performance_correct nsubj_improved_performance dep_reports_improved aux_impressive_to xcomp_improved_impressive num_\%_939 dobj_impressive_\% advmod_using_when advcl_impressive_using det_sense_the nn_sense_one dobj_using_sense prep_sense_per poss_constraint_discourse pobj_per_constraint ' +o,4885,'WordNet sense information has been criticized to be too fine grained ',Navigli,'amod_information_WordNet nn_information_sense nsubjpass_criticized_information aux_criticized_has auxpass_criticized_been aux_fine_to cop_fine_be advmod_fine_too xcomp_criticized_fine advmod_fine_grained ' +o,4886,'The intercoder reliability is a constant concern of everyone working with corpora to test linguistic hypotheses , and the more so when one is coding for semanto-pragmatic interpretations , as in the case of the analysis of connectives ',Carletta,'det_reliability_The nn_reliability_intercoder nsubj_concern_reliability cop_concern_is det_concern_a amod_concern_constant prep_concern_of pobj_of_everyone partmod_everyone_working prep_working_with pobj_with_corpora aux_test_to xcomp_working_test amod_hypotheses_linguistic dobj_test_hypotheses advmod_test_ cc_concern_and det_more_the conj_concern_more advmod_when_so advmod_coding_when nsubj_coding_one aux_coding_is dep_more_coding prep_coding_for amod_interpretations_semanto-pragmatic pobj_for_interpretations dep_for_as dep_for_in det_case_the pobj_in_case prep_case_of det_analysis_the pobj_of_analysis prep_analysis_of pobj_of_connectives ' +o,4887,'The next section briefly reviews the word alignment based statistical machine translation ',Brown,'det_section_The amod_section_next nsubj_reviews_section advmod_reviews_briefly det_alignment_the nn_alignment_word dobj_reviews_alignment partmod_alignment_based amod_translation_statistical nn_translation_machine dobj_based_translation ' +o,4888,' describe a method for learning a probabilistic model that maps LFG parse structures in German into LFG parse structures in English ',Riezler,'advmod_describe_ det_method_a dobj_describe_method prep_describe_for pcomp_for_learning det_model_a amod_model_probabilistic dobj_learning_model dep_LFG_that nsubj_LFG_maps dep_describe_LFG amod_structures_parse dobj_LFG_structures prep_LFG_in amod_structures_German prep_German_into pobj_into_LFG nn_structures_parse pobj_in_structures prep_structures_in pobj_in_English ' +o,4889,'While previous researchers have used agglomerative nesting clustering -LRB- eg , Futrelle and Gauch -LRB- 1993 -RRB- -RRB- , comparisons with our work are difficult to draw , due to their use of the 1,000 commonest words from their respective corpora ',Brown,'mark_used_While amod_researchers_previous nsubj_used_researchers aux_used_have advcl_difficult_used amod_clustering_agglomerative amod_clustering_nesting dobj_used_clustering parataxis_used_eg dobj_eg_ conj__Futrelle cc__and conj__Gauch appos__1993 nsubj_difficult_comparisons prep_comparisons_with poss_work_our pobj_with_work cop_difficult_are aux_draw_to xcomp_difficult_draw dep_to_due prep_draw_to poss_use_their pobj_to_use prep_use_of det_words_the num_words_1,000 nn_words_commonest pobj_of_words prep_draw_from poss_corpora_their amod_corpora_respective pobj_from_corpora ' +o,4890,'The algorithm proposed by is labeled as Turney-PairClass ',Turney,'det_algorithm_The nsubjpass_labeled_algorithm partmod_algorithm_proposed prep_proposed_by pobj_by_ auxpass_labeled_is prep_labeled_as pobj_as_Turney-PairClass ' +o,4891,'The cohesion between words has been evaluated with the mutual information measure , as in ',Church,'det_cohesion_The nsubjpass_evaluated_cohesion prep_cohesion_between pobj_between_words aux_evaluated_has auxpass_evaluated_been prep_evaluated_with det_measure_the amod_measure_mutual nn_measure_information pobj_with_measure dep_with_as dep_with_in ' +o,4892,'Using this alignment strategy , we follow and compute one alignment for each translation direction -LRB- f e and e f -RRB- , and then combine them ',Och,'dep_follow_Using det_strategy_this amod_strategy_alignment dobj_Using_strategy nsubj_follow_we dep_follow_ cc__and conj__compute num_alignment_one dobj__alignment prep_alignment_for det_direction_each nn_direction_translation pobj_for_direction dep_f_f nn_f_e cc_e_and conj_e_e dep_-LRB-_f cc__and advmod__then conj__combine dobj_combine_them ' +o,4893,'2 Combining Classifiers for Chinesewordsegmentation Thetwomachine-learningmodelsweuseinthis work are the maximum entropy model and the error-driven transformation-based learning model Weusetheformerasthemainworkhorse and the latter to correct some of the errors producedbytheformer ',Ratnaparkhi,'dep__2 nn_Classifiers_Combining nsubj__Classifiers prep_Classifiers_for nn_work_Chinesewordsegmentation nn_work_Thetwomachine-learningmodelsweuseinthis pobj_for_work cop__are det__the amod__maximum amod__entropy nn__model cc__and det_model_the amod_model_error-driven amod_model_transformation-based nn_model_learning conj__model nn_Weusetheformerasthemainworkhorse_ nsubj_correct_Weusetheformerasthemainworkhorse cc_Weusetheformerasthemainworkhorse_and det_latter_the conj_Weusetheformerasthemainworkhorse_latter aux_correct_to dep__correct dobj_correct_some prep_some_of det_producedbytheformer_the nn_producedbytheformer_errors pobj_of_producedbytheformer ' +o,4894,' studied two context delineation methods of English nouns : the window-based and the syntactic , whereby all the different types of syntactic dependencies of the nouns were used in the same feature space ',Grefenstette,'nsubj_methods_ dep_methods_studied num_methods_two nn_methods_context nn_methods_delineation prep_methods_of amod_nouns_English pobj_of_nouns det_window-based_the dep_methods_window-based cc_window-based_and det_syntactic_the conj_window-based_syntactic advmod_used_whereby predet_types_all det_types_the amod_types_different nsubjpass_used_types prep_types_of amod_dependencies_syntactic pobj_of_dependencies prep_dependencies_of det_nouns_the pobj_of_nouns auxpass_used_were advcl_methods_used prep_used_in det_space_the amod_space_same nn_space_feature pobj_in_space ' +o,4895,'Wall-Street Journal -LRB- WSJ -RRB- Sections 15-18 and 20 were used by as training and test data respectively for evaluating their base-NP chunker ',Ramshaw,'nn_Sections_Wall-Street nn_Sections_Journal abbrev_Sections_WSJ nsubjpass_used_15-18 cc_15-18_and conj_15-18_20 auxpass_used_were dep_Sections_used prep_used_by pobj_by_ prep__as nn_data_training cc_training_and conj_training_test pobj_as_data dep_for_respectively prep_used_for pcomp_for_evaluating poss_chunker_their amod_chunker_base-NP dobj_evaluating_chunker ' +o,4896,'We then used the kappa statistic to assess the level of agreement between the three coders with respect to the 2 An agent holds the task initiative during a turn as long as some utterance during the turn directly proposes how the agents should accomplish their goal , as in utterance -LRB- 3c -RRB- ',Carletta,'nsubj_used_We advmod_used_then det_statistic_the nn_statistic_kappa nsubj__statistic ccomp_used_ aux_assess_to xcomp__assess det_level_the dobj_assess_level prep_level_of pobj_of_agreement prep_agreement_between det_coders_the num_coders_three pobj_between_coders prep_assess_with pobj_with_respect prep_assess_to dep_2_the pobj_to_2 det_agent_An nsubj_holds_agent dep_used_holds det_initiative_the nn_initiative_task dobj_holds_initiative prep_initiative_during det_turn_a pobj_during_turn advmod_long_as advmod_holds_long mark_proposes_as det_utterance_some nsubj_proposes_utterance prep_utterance_during det_turn_the pobj_during_turn advmod_proposes_directly dep_long_proposes advmod_accomplish_how det_agents_the nsubj_accomplish_agents aux_accomplish_should ccomp_proposes_accomplish poss_goal_their dobj_accomplish_goal dep_long_as dep_as_in pobj_in_utterance appos_utterance_3c ' +o,4897,'We could also use the value of semantic similarity and relatedness measures or the existence of hypernym or hyponym relations as features ',Pedersen,'nsubj_use_We aux_use_could advmod_use_also det_value_the dobj_use_value prep_value_of amod_similarity_semantic pobj_of_similarity cc_similarity_and nn__relatedness nn__measures conj_similarity_ cc_value_or det_existence_the conj_value_existence prep_existence_of nn_relations_hypernym cc_hypernym_or conj_hypernym_hyponym pobj_of_relations prep_use_as pobj_as_features ' +o,4898,'While the tag features , containing WSJ paxt-ofspeech tags , have about 45 values , the word features have more than 10,000 values ',Marcus,'mark_have_While det_features_the nn_features_tag nsubj_have_features partmod_have_containing nn_tags_WSJ amod_tags_paxt-ofspeech nsubj__tags xcomp_containing_ advcl_have_have quantmod_45_about num_values_45 dobj_have_values det_features_the nn_features_word nsubj_have_features dep_than_more quantmod_10,000_than num_values_10,000 dobj_have_values ' +o,4899,'41 Evaluation of Different Features and Models In pilot experiments on a subset of the features , we provide a comparison of HM-SVM with other two learning models , maximum entropy -LRB- MaxEnt -RRB- model and SVM model , to test the effectiveness of HMSVM on function labeling task , as well as the generality of our hypothesis on different learning 58 Table 3 : Features used in each experiment round ',Berger,'num_Evaluation_41 nsubj_provide_Evaluation prep_Evaluation_of amod_Features_Different pobj_of_Features cc_Features_and conj_Features_Models prep_provide_In nn_experiments_pilot pobj_In_experiments prep_experiments_on det_subset_a pobj_on_subset prep_subset_of det_features_the pobj_of_features nsubj_provide_we det_comparison_a nsubj_learning_comparison prep_comparison_of pobj_of_HM-SVM prep_HM-SVM_with amod_two_other pobj_with_two dep_provide_learning nsubj_test_models amod__maximum amod__entropy appos__MaxEnt nn__model appos_models_ cc__and nn__SVM nn__model conj__ aux_test_to xcomp_learning_test det_effectiveness_the dobj_test_effectiveness prep_effectiveness_of pobj_of_HMSVM prep_HMSVM_on nn_task_function nn_task_labeling pobj_on_task dep_well_as cc_HMSVM_well dep_well_as det_generality_the conj_HMSVM_generality prep_generality_of poss_hypothesis_our pobj_of_hypothesis prep_test_on pobj_on_different partmod_different_learning num_Table_58 dobj_learning_Table num_Table_3 dep_Table_Features partmod_Features_used prep_used_in det_round_each nn_round_experiment pobj_in_round ' +o,4900,'Their weights are optimized wrt BLEU score using the algorithm described in ',Och,'poss_weights_Their nsubjpass_optimized_weights auxpass_optimized_are nn_score_wrt nn_score_BLEU dobj_optimized_score partmod_score_using det_algorithm_the dobj_using_algorithm partmod_algorithm_described prep_described_in ' +p,4901,'This increase of probabilities is defined as multiplicative change -LRB- N -RRB- as follows : -LRB- N -RRB- = P -LRB- E Tprime -RRB- \\/ P -LRB- E T -RRB- -LRB- 2 -RRB- The main innovation of the model in is the possibility of adding at each step the best relation N = -LCB- Ri , j -RCB- as well as N = I -LRB- Ri , j -RRB- that is Ri , j with all the relations by the existing taxonomy ',Snow,'det_increase_This nsubjpass_defined_increase prep_increase_of pobj_of_probabilities auxpass_defined_is prep_defined_as amod_change_multiplicative pobj_as_change appos_change_N mark_follows_as advcl_defined_follows nsubj_=_N ccomp_follows_= dobj_=_P nn_Tprime_E appos_P_Tprime partmod_P_\\/ dobj_\\/_P nn_T_E appos_P_T dep_possibility_2 det_innovation_The amod_innovation_main nsubj_possibility_innovation prep_innovation_of det_model_the pobj_of_model prep_model_in pobj_in_ cop_possibility_is det_possibility_the dep_defined_possibility prep_possibility_of pcomp_of_adding prep_adding_at det_step_each pobj_at_step det_relation_the amod_relation_best dobj_adding_relation nn_=_N dep_relation_= appos_=_Ri dep_Ri_j dep_well_as cc_relation_well dep_well_as conj_relation_N dep_I_= rcmod_N_I appos_N_Ri dep_Ri_j nsubj_Ri_that cop_Ri_is rcmod_N_Ri appos_Ri_j prep_j_with predet_relations_all det_relations_the pobj_with_relations prep_relations_by det_taxonomy_the amod_taxonomy_existing pobj_by_taxonomy ' +p,4902,'However , as also pointed out by , this observation does not hold uniformly over all possible co-occurrences of two words ',Yarowsky,'advmod_hold_However mark_pointed_as advmod_pointed_also dep_hold_pointed prt_pointed_out prep_pointed_by pobj_by_ det_observation_this nsubj_hold_observation aux_hold_does neg_hold_not advmod_hold_uniformly dep_uniformly_over det_co-occurrences_all amod_co-occurrences_possible pobj_over_co-occurrences prep_co-occurrences_of num_words_two pobj_of_words ' +o,4903,'GIZA + + , an implementation of the IBM and HMM -LRB- ? -RRB- ',Brown,'nn_+_GIZA nn__+ dep_+_ det_implementation_an conj__implementation prep_implementation_of det__the nn__IBM pobj_of_ cc__and conj__HMM ' +o,4904,'As for parser , we train three off-shelf maximum-entropy parsers using the Arabic , Chinese and English Penn treebank ',Marcus,'prep_train_As dep_As_for pobj_for_parser nsubj_train_we num_parsers_three amod_parsers_off-shelf amod_parsers_maximum-entropy nsubj_using_parsers dep_using_ xcomp_train_using det_treebank_the amod_treebank_Arabic conj_Arabic_Chinese cc_Arabic_and conj_Arabic_English nn_treebank_Penn dobj_using_treebank ' +o,4905,'Our decoder is a phrase-based multi-stack implementation of the log-linear model similar to Pharaoh ',Koehn,'poss_decoder_Our nsubj_implementation_decoder cop_implementation_is det_implementation_a amod_implementation_phrase-based amod_implementation_multi-stack prep_implementation_of det_model_the amod_model_log-linear pobj_of_model amod_model_similar prep_similar_to pobj_to_Pharaoh ' +o,4906,'Words in test data that have not been seen in training are deterministically assigned the POS tag that is assigned by the tagger described in ',Ratnaparkhi,'nsubjpass_assigned_Words prep_Words_in nn_data_test pobj_in_data nsubjpass_seen_that aux_seen_have neg_seen_not auxpass_seen_been rcmod_data_seen prep_seen_in pobj_in_training auxpass_assigned_are advmod_assigned_deterministically det_tag_the dep_tag_POS dobj_assigned_tag nsubjpass_assigned_that auxpass_assigned_is rcmod_tag_assigned prep_assigned_by det_tagger_the pobj_by_tagger partmod_tagger_described prep_described_in ' +o,4907,' describes various strategies for the decomposition of the decoding into multiple translation models using the Moses decoder ',Koehn,'nsubj_describes_ amod_strategies_various dobj_describes_strategies prep_strategies_for det_decomposition_the pobj_for_decomposition prep_decomposition_of det_decoding_the pobj_of_decoding prep_describes_into amod_models_multiple nn_models_translation pobj_into_models partmod_models_using det_decoder_the nn_decoder_Moses dobj_using_decoder ' +o,4908,'We created a dependency training corpus based on the Penn Treebank , or more specifically on the HPSG Treebank generated from the Penn Treebank -LRB- see section 22 -RRB- ',Marcus,'nsubj_created_We det_corpus_a amod_corpus_dependency nn_corpus_training dobj_created_corpus dep_corpus_based prep_based_on det__the nn__Penn nn__Treebank pobj_on_ cc_based_or advmod_specifically_more advmod_based_specifically dep_specifically_on det_Treebank_the nn_Treebank_HPSG pobj_on_Treebank conj_based_generated prep_generated_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank parataxis_generated_see dobj_see_section num_section_22 ' +p,4909,'The MERT module is a highly modular , efficient and customizable implementation of the algorithm described in ',Och,'det_module_The nn_module_MERT nsubj_implementation_module cop_implementation_is det_implementation_a advmod_modular_highly amod_implementation_modular conj_modular_efficient cc_modular_and conj_modular_customizable prep_implementation_of det_algorithm_the pobj_of_algorithm partmod_algorithm_described prep_described_in ' +o,4910,'34 Feature Representation Ranking Models Following previous work on sentiment classi cation , we represent each review as a vector of lexical features ',Pang,'num_Models_34 nn_Models_Feature nn_Models_Representation nn_Models_Ranking prep_represent_Following amod_work_previous pobj_Following_work prep_work_on nn__sentiment nn__classi nn__cation pobj_on_ nsubj_represent_we dep_Models_represent det_review_each dobj_represent_review prep_represent_as det_vector_a pobj_as_vector prep_vector_of amod_features_lexical pobj_of_features ' +o,4911,'Perceptron Learning a discriminative structure prediction model with a perceptron update was first proposed by ',Collins,'nn_Learning_Perceptron det_model_a amod_model_discriminative nn_model_structure nn_model_prediction nsubjpass_proposed_model prep_model_with det_update_a nn_update_perceptron pobj_with_update auxpass_proposed_was advmod_proposed_first dep_Learning_proposed prep_proposed_by ' +o,4912,'English POS tags were assigned by MXPOST , which was trained on the training data described in Section 41 ',Ratnaparkhi,'nn_tags_English dep_tags_POS nsubjpass_assigned_tags auxpass_assigned_were prep_assigned_by nn__MXPOST pobj_by_ nsubjpass_trained_which auxpass_trained_was rcmod__trained prep_trained_on det_data_the nn_data_training pobj_on_data partmod_data_described prep_described_in pobj_in_Section num_Section_41 ' +o,4913,'ACM Transactions on Computer-Human Interaction (TOCHI), 11(3). M E Pollack Intelligent technology for an aging population: The use of AI to assist elders with cognitive impairment 2005 AI Magazine 26--2 rch on developing SDS for home-care and tele-care applications, Examples include scheduling appointments over the phone (Zajicek et al. 2004, Wolters et al., submitted), interactive reminder systems (Pollack, 2005), symptom management systems (Black et al. 2005) or environmental control systems (Clarke et al. 2005).',Lin,'' +o,4914,'For example , extractive text summarization generates a summary by selecting a few good sentences from one or more articles on the same topic ',Goldstein,'prep_generates_For pobj_For_example amod_summarization_extractive nn_summarization_text nsubj_generates_summarization det_summary_a dobj_generates_summary prep_generates_by pcomp_by_selecting det_sentences_a amod_sentences_few amod_sentences_good dobj_selecting_sentences prep_selecting_from num_articles_one cc_one_or conj_one_more pobj_from_articles prep_articles_on det_topic_the amod_topic_same pobj_on_topic ' +o,4915,' showed that the MSTParser and MaltParser produce different errors ',McDonald,'nsubj_showed_ complm_produce_that det_MSTParser_the nsubj_produce_MSTParser cc_MSTParser_and conj_MSTParser_MaltParser ccomp_showed_produce amod_errors_different dobj_produce_errors ' +o,4916,' proposed sentence alignment techniques based on dynamic programming , using sentence length and lexical mapping information ',Brown,'nsubj_proposed_ nn_techniques_sentence nn_techniques_alignment dobj_proposed_techniques prep_techniques_based dep_based_on amod_programming_dynamic pobj_on_programming xcomp_proposed_using nn_length_sentence dobj_using_length cc_length_and amod_information_lexical nn_information_mapping conj_length_information ' +o,4917,'For instance , BLEU and ROUGE are based on n-gram precisions , METEOR and STM use word-class or structural information , Kauchak leverages on paraphrases , and TER uses edit-distances ',Banerjee,'dep_leverages_For pobj_For_instance nn__BLEU cc_BLEU_and conj_BLEU_ROUGE nsubjpass_based_ auxpass_based_are dep_leverages_based prep_based_on amod_precisions_n-gram pobj_on_precisions nn__METEOR conj_precisions_ cc__and nn_word-class_STM nn_word-class_ nn_word-class_use conj__word-class cc_precisions_or amod_information_structural conj_precisions_information nn__Kauchak nsubj_leverages_ prep_leverages_on pobj_on_paraphrases cc_leverages_and nn__TER nsubj_uses_ conj_leverages_uses dobj_uses_edit-distances ' +o,4918,'One is distortion model which penalizes translations according to their jump distance instead of their content ',Koehn,'nsubj__One cop__is nn__distortion nn__model nsubj_penalizes_which rcmod__penalizes dobj_penalizes_translations prep_penalizes_according dep_according_to poss_distance_their nn_distance_jump pobj_to_distance dep_of_instead prep_distance_of poss_content_their pobj_of_content ' +o,4919,'CIT -RRB- ',Nakov,'nn_-RRB-_CIT ' +o,4920,'Results for chunking Penn Treebank data were previously presented by several authors ',Ramshaw,'nsubjpass_presented_Results prep_Results_for amod_data_chunking nn_data_Penn nn_data_Treebank pobj_for_data auxpass_presented_were advmod_presented_previously prep_presented_by amod_authors_several pobj_by_authors ' +o,4921,'1 Introduction The recent advances in statistical machine translation have been achieved by discriminatively training a small number of real-valued features based either on -LRB- hierarchical -RRB- phrase-based translation or syntax-based translation ',Och,'num_advances_1 nn_advances_Introduction nn_advances_The amod_advances_recent nsubjpass_achieved_advances prep_advances_in amod_translation_statistical nn_translation_machine pobj_in_translation aux_achieved_have auxpass_achieved_been prep_achieved_by advmod_training_discriminatively pcomp_by_training det_number_a amod_number_small dobj_training_number prep_number_of amod_features_real-valued pobj_of_features partmod_features_based advmod_based_either prep_based_on dep_on_hierarchical amod__phrase-based nn__translation pobj_on_ cc__or amod__syntax-based nn__translation conj__ ' +o,4922,'Manual processes , such as lexicon development could be automated in the future using standard contextbased , word distribution methods , or other corpus-based techniques ',Smadja,'amod_processes_Manual nsubjpass_automated_processes dep_as_such prep_processes_as nn_development_lexicon pobj_as_development aux_automated_could auxpass_automated_be prep_automated_in det_future_the pobj_in_future partmod_future_using amod_contextbased_standard dobj_using_contextbased nn__word nn__distribution nn__methods conj_contextbased_ cc_contextbased_or amod_techniques_other amod_techniques_corpus-based conj_contextbased_techniques ' +o,4923,'Separating the scoring from the source language reordering also has the advantage that the approach in essence is compatible with other approaches such as a traditional PSMT system or a hierarchical phrase system ',Koehn,'det_scoring_the dobj_Separating_scoring prep_Separating_from det_reordering_the nn_reordering_source nn_reordering_language pobj_from_reordering advmod_has_also dep_Separating_has det_advantage_the dobj_has_advantage complm_compatible_that det_approach_the nsubj_compatible_approach prep_approach_in pobj_in_essence cop_compatible_is dep_advantage_compatible prep_compatible_with amod_approaches_other pobj_with_approaches dep_as_such prep_approaches_as det_PSMT_a amod_PSMT_traditional pobj_as_PSMT nn__system nsubj_Separating_ cc__or det__a amod__hierarchical nn__phrase nn__system conj__ ' +o,4924,'(Koehn et al., 2003).',Koehn,'' +o,4925,'In order to create the necessary SMT language and translation models, they used: Giza++ (Och & Ney, 2003);2 the CMU-Cambridge statistical toolkit;3 the ISI ReWrite Decoder.4 Translation was performed from EnglishFrench and FrenchEnglish, and the resulting translations were evaluated using a range of automatic metrics: BLEU (Papineni et al. , 2002), Precision and Recall 2http://www.isi.edu/och/Giza++.html 3http://mi.eng.cam.ac.uk/prc14/toolkit.html 4http://www.isi.edu/licensed-sw/rewrite-decoder/ 185 (Turian et al. , 2003), and Wordand Sentence Error Rates.',Och,'' +o,4926,' , sometimes augmented by an HMM-based model or Och and Neys Model 6 ',Dunning,'advmod_augmented_ advmod_augmented_sometimes prep_augmented_by det_model_an amod_model_HMM-based pobj_by_model cc_model_or nn_Model_Och cc_Och_and conj_Och_Neys conj_model_Model num__6 nsubj_augmented_ ' +o,4927,' , is not very useful for applications like statistical machine translation , , for which an accurate word-to-word alignment between the source and the target languages is critical for high quality translations ',Brown,'advmod_useful_ cop_useful_is neg_useful_not advmod_useful_very prep_useful_for pobj_for_applications prep_applications_like amod_translation_statistical nn_translation_machine pobj_like_translation appos_translation_ rel_critical_for pobj_for_which det_alignment_an amod_alignment_accurate dep_accurate_word-to-word nsubj_critical_alignment prep_alignment_between det_source_the pobj_between_source cc_source_and det_languages_the nn_languages_target conj_source_languages cop_critical_is rcmod_translation_critical prep_critical_for pobj_for_high nn_translations_quality nsubj_useful_translations ' +o,4928,'Candidate term Segment result of GPWS for one sentence , in which term appears \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ \\/ Table 2 : Examples of candidates eliminated by GPWS 5 Relative frequency ratio against background corpus Relative frequency ratio -LRB- RFR -RRB- is a useful method to be used to discover characteristic linguistic phenomena of a corpus when compared with another ',Dunning,'amod_result_Candidate nn_result_term nn_result_Segment prep_result_of pobj_of_GPWS prep_GPWS_for num_sentence_one pobj_for_sentence rel_Examples_in pobj_in_which nsubj_Examples_term cop_Examples_appears nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ nn_Examples_\\/ dep_Examples_Table dep_Examples_2 rcmod_sentence_Examples prep_Examples_of pobj_of_candidates partmod_Examples_eliminated prep_eliminated_by pobj_by_GPWS num_GPWS_5 amod_ratio_Relative nn_ratio_frequency nsubj_method_ratio prep_ratio_against nn_ratio_background nn_ratio_corpus nn_ratio_Relative nn_ratio_frequency pobj_against_ratio abbrev_ratio_RFR cop_method_is det_method_a amod_method_useful dep_result_method aux_used_to auxpass_used_be infmod_method_used aux_discover_to purpcl_used_discover amod_phenomena_characteristic amod_phenomena_linguistic dobj_discover_phenomena prep_phenomena_of det_corpus_a pobj_of_corpus advmod_compared_when advcl_discover_compared prep_compared_with det__another pobj_with_ ' +o,4929,'Many methods have been proposed to compute distributional similarity between words , eg , , , and ',Hindle,'amod_methods_Many nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been aux_compute_to xcomp_proposed_compute amod_similarity_distributional dobj_compute_similarity prep_similarity_between pobj_between_words conj_words_eg conj_words_ conj_words_ conj_words_ cc_words_and conj_words_ ' +o,4930,'1 Introduction Decoding is one of the three fundamental problems in classical SMT -LRB- translation model and language model being the other two -RRB- as proposed by IBM in the early 1990s ',Brown,'num_Decoding_1 nn_Decoding_Introduction nsubj_one_Decoding cop_one_is prep_one_of det_problems_the num_problems_three amod_problems_fundamental pobj_of_problems prep_problems_in amod_SMT_classical pobj_in_SMT nn_model_translation dep_SMT_model cc_model_and nn_model_language conj_model_model cop_two_being det_two_the amod_two_other dep_model_two mark_proposed_as dep_SMT_proposed prep_proposed_by pobj_by_IBM prep_proposed_in det_1990s_the amod_1990s_early pobj_in_1990s ' +o,4931,'For example , minimum entropy regularization , aims to maximize the conditional likelihood of labeled data while minimizing the conditional entropy of unlabeled data : summationdisplay i logp -LRB- y -LRB- i -RRB- x -LRB- i -RRB- -RRB- 122bardblbardbl2H -LRB- y x -RRB- -LRB- 3 -RRB- This approach generally would result in sharper models which can be data-sensitive in practice ',Jiao,'prep_aims_For pobj_For_example amod_regularization_minimum nn_regularization_entropy dep__regularization ccomp_aims_ aux_maximize_to xcomp_aims_maximize det_likelihood_the amod_likelihood_conditional dobj_maximize_likelihood prep_likelihood_of amod_data_labeled pobj_of_data prep_maximize_while pcomp_while_minimizing det_entropy_the amod_entropy_conditional dobj_minimizing_entropy prep_entropy_of amod_data_unlabeled pobj_of_data advmod_i_summationdisplay amod_122bardblbardbl2H_i nn_122bardblbardbl2H_logp dep_122bardblbardbl2H_y dep_y_i dep_i_x appos_x_i dep_data_122bardblbardbl2H number_x_y dep_122bardblbardbl2H_x dep_result_3 det_approach_This nsubj_result_approach advmod_approach_generally aux_result_would dep_aims_result prep_result_in amod_models_sharper pobj_in_models nsubj_data-sensitive_which aux_data-sensitive_can cop_data-sensitive_be rcmod_models_data-sensitive prep_data-sensitive_in pobj_in_practice ' +o,4932,'Introduction Many applications that process natural language can be enhanced by incorporating information about the probabilities of word strings ; that is , by using statistical language model information ',Brown,'amod_applications_Many dobj_Introduction_applications complm_enhanced_that nn_language_process amod_language_natural nsubjpass_enhanced_language aux_enhanced_can auxpass_enhanced_be dep_applications_enhanced prep_enhanced_by pcomp_by_incorporating dobj_incorporating_information prep_incorporating_about det_probabilities_the pobj_about_probabilities prep_probabilities_of nn_strings_word pobj_of_strings nsubj_is_that rcmod_applications_is prep_Introduction_by pcomp_by_using amod__statistical nn__language nn__model nn__information dobj_using_ ' +o,4933,'While in this paper we evaluated our framework on the discovery of concepts , we have recently proposed fully unsupervised frameworks for the discovery of different relationship types ',Davidov,'mark_evaluated_While prep_evaluated_in det_paper_this pobj_in_paper nsubj_evaluated_we advcl_proposed_evaluated poss_framework_our dobj_evaluated_framework prep_evaluated_on det_discovery_the pobj_on_discovery prep_discovery_of pobj_of_concepts nsubj_proposed_we aux_proposed_have advmod_proposed_recently advmod_unsupervised_fully amod_frameworks_unsupervised dobj_proposed_frameworks prep_proposed_for det_discovery_the pobj_for_discovery prep_discovery_of amod_types_different nn_types_relationship pobj_of_types ' +p,4934,'53 Comparison with SS-CRF-MER When we consider semi-supervised SOL methods , SS-CRF-MER is the most competitive with HySOL , since both methods are defined based on CRFs ',Jiao,'num_Comparison_53 prep_Comparison_with pobj_with_SS-CRF-MER advmod_methods_When nsubj_methods_we dep_methods_consider amod_methods_semi-supervised nn_methods_SOL dep_Comparison_methods nn__SS-CRF-MER nsubj_competitive_ cop_competitive_is dep_competitive_the advmod_competitive_most parataxis_methods_competitive prep_competitive_with pobj_with_HySOL mark_defined_since det_methods_both nsubjpass_defined_methods auxpass_defined_are advcl_methods_defined prep_defined_based dep_based_on pobj_on_CRFs ' +o,4935,'Such a technique has been used with TER to combine the output of multiple translation systems ',Rosti,'predet_technique_Such det_technique_a nsubjpass_used_technique aux_used_has auxpass_used_been prep_used_with pobj_with_TER aux_combine_to xcomp_used_combine det_output_the dobj_combine_output prep_output_of amod_systems_multiple nn_systems_translation pobj_of_systems ' +o,4936,' of running GIZA + + in both directions and then merging the alignments using the grow-diag-final heuristic ',Och,'prep__of pcomp_of_running nn__GIZA nn__+ nn__+ dobj_running_ prep__in det_directions_both pobj_in_directions cc_running_and advmod_merging_then conj_running_merging det_alignments_the dobj_merging_alignments xcomp_merging_using det_heuristic_the amod_heuristic_grow-diag-final dobj_using_heuristic ' +o,4937,'Recent lexicalized stochastic parsers such as , Charniak -LRB- 1997 -RRB- , and others add additional features to each constituent , the most important being the head word of the parse constituent ',Collins,'amod_parsers_Recent amod_parsers_lexicalized amod_parsers_stochastic nsubj_add_parsers dep_as_such prep_parsers_as nn_Charniak_ pobj_as_Charniak appos_Charniak_1997 cc_Charniak_and conj_Charniak_others amod_features_additional dobj_add_features prep_add_to det_constituent_each pobj_to_constituent dep_important_the dep_important_most nsubj_word_important cop_word_being det_word_the nn_word_head rcmod_constituent_word prep_word_of det_constituent_the amod_constituent_parse pobj_of_constituent ' +o,4938,'So far , most previous work on domain adaptation for parsing has focused on data-driven systems , ie systems employing -LRB- constituent or dependency based -RRB- treebank grammars ',McClosky,'dep_far_So advmod_focused_far amod_work_most amod_work_previous nsubj_focused_work prep_work_on nn_adaptation_domain pobj_on_adaptation prep_adaptation_for pobj_for_parsing aux_focused_has prep_focused_on amod__data-driven nn__systems pobj_on_ nn_systems_ie appos__systems xcomp_focused_employing dep_employing_constituent cc_constituent_or conj_constituent_dependency partmod_constituent_based nn_grammars_treebank nsubj__grammars ccomp_employing_ ' +o,4939,'34 Perspectives for automatic paraphrase extraction There is a growing amount of work on automatic extraction of paraphrases from text corpora ',Barzilay,'num_Perspectives_34 nsubj_extraction_Perspectives prep_Perspectives_for amod_paraphrase_automatic pobj_for_paraphrase expl_is_There ccomp_extraction_is det_amount_a amod_amount_growing nsubj_is_amount prep_amount_of pobj_of_work prep_work_on amod_extraction_automatic pobj_on_extraction prep_extraction_of pobj_of_paraphrases prep_paraphrases_from nn_corpora_text pobj_from_corpora ' +o,4940,'Others have introduced alternative discriminative training methods , in which a recurring challenge is scalability : to train many features , we need many train218 ing examples , and to train discriminatively , we need to search through all possible translations of each training example ',Liang,'nsubj_methods_Others aux_methods_have dep_methods_introduced amod_methods_alternative amod_methods_discriminative nn_methods_training advmod_methods_ rel_scalability_in pobj_in_which det_challenge_a amod_challenge_recurring nsubj_scalability_challenge cop_scalability_is dobj_methods_scalability aux_train_to dep_scalability_train amod_features_many dobj_train_features nsubj_need_we ccomp_methods_need amod_examples_many amod_examples_train218 nn_examples_ing dobj_need_examples cc_methods_and aux_train_to dep_need_train advmod_train_discriminatively nsubj_need_we conj_methods_need aux_search_to xcomp_need_search prep_search_through det_translations_all amod_translations_possible pobj_through_translations prep_translations_of det_example_each nn_example_training pobj_of_example ' +o,4941,'Our system is a re-implementation of the phrase-based system described in Koehn , and uses publicly available components for word alignment 1 , decoding 2 , language modeling 3 and finite-state processing 4 ',Och,'poss_system_Our nsubjpass_described_system auxpass_described_is det_re-implementation_a attr_described_re-implementation prep_re-implementation_of det_system_the amod_system_phrase-based pobj_of_system prep_described_in nn__Koehn pobj_in_ cc_described_and conj_described_uses advmod_available_publicly amod_components_available dobj_uses_components prep_components_for nn_alignment_word pobj_for_alignment amod_components_ tmod__1 partmod_components_decoding number_2_ dobj_decoding_2 nn_modeling_language appos_2_modeling number_3_ dep_modeling_3 cc_components_and amod_processing_finite-state conj_components_processing number_4_ dep_processing_4 ' +o,4942,' and the HMM alignment model of ',Brown,'cc__and det_model_the nn_model_HMM nn_model_alignment conj__model prep__of pobj_of_ ' +o,4943,'31 Word Sequence Classification Similar to English text chunking , the word sequence classification model aims to classify each word via encoding its context features ',Ramshaw,'num_Classification_31 nn_Classification_Word nn_Classification_Sequence nsubj_aims_Classification amod_Classification_Similar prep_Similar_to amod__English nn__text nn__chunking pobj_to_ det_model_the nn_model_word nn_model_sequence nn_model_classification appos__model aux_classify_to xcomp_aims_classify det_word_each dobj_classify_word prep_word_via pcomp_via_encoding poss_features_its nn_features_context dobj_encoding_features ' +o,4944,'The table also shows Cohen \'s to , an agreement measure that corrects for chance agreement ; the most important t value in the table is the value of 07 for the two human judges , which can be interpreted as sufficiently high to indicate that the task is reasonably well defined ',Carletta,'det_table_The nsubj_shows_table advmod_shows_also poss__Cohen possessive_Cohen_\'s aux_,_to det_measure_an nn_measure_agreement xcomp_,_measure nsubj_corrects_that rcmod_measure_corrects prep_corrects_for nn_agreement_chance pobj_for_agreement dobj_shows_ det_value_the advmod_important_most amod_value_important nn_value_t nsubj_value_value prep_value_in det_table_the pobj_in_table cop_value_is det_value_the parataxis_shows_value prep_value_of pobj_of_07 prep_value_for det_judges_the num_judges_two amod_judges_human pobj_for_judges nsubjpass_interpreted_which aux_interpreted_can auxpass_interpreted_be rcmod_judges_interpreted advmod_sufficiently_as advmod_interpreted_sufficiently advmod_sufficiently_high aux_indicate_to purpcl_interpreted_indicate complm_defined_that det_task_the nsubjpass_defined_task auxpass_defined_is advmod_defined_reasonably advmod_defined_well ccomp_indicate_defined ' +o,4945,'Note that the need to consider segmentation and alignment at the same time is also mentioned in , and related issues are reported in ',Wu,'complm_mentioned_that det_need_the nsubjpass_mentioned_need aux_consider_to infmod_need_consider dobj_consider_segmentation cc_segmentation_and conj_segmentation_alignment prep_consider_at det_time_the amod_time_same pobj_at_time auxpass_mentioned_is advmod_mentioned_also ccomp_Note_mentioned prep_mentioned_in pobj_in_ cc_mentioned_and amod_issues_related nsubjpass_reported_issues auxpass_reported_are conj_mentioned_reported prep_reported_in ' +o,4946,'Unconstrained CL corresponds exactly to a conditional maximum entropy model ',Berger,'nn_CL_Unconstrained nsubj_corresponds_CL advmod_corresponds_exactly prep_corresponds_to det_model_a amod_model_conditional amod_model_maximum nn_model_entropy pobj_to_model ' +o,4947,' 1993 -RRB- , make use of both positive and negative instances of performing a task ',Dunning,'dep_make_1993 dobj__make rcmod_make_use prep_make_of preconj_positive_both amod_instances_positive cc_positive_and conj_positive_negative pobj_of_instances prep_instances_of pcomp_of_performing det_task_a dobj_performing_task ' +o,4948,'204 422 Correlation between TREC nuggets and non-text features Analyzing the features used could let us understand summarization better ',Nenkova,'nn_Correlation_204 nn_Correlation_422 nsubj_let_Correlation prep_Correlation_between nn_nuggets_TREC pobj_between_nuggets cc_nuggets_and amod_features_non-text conj_nuggets_features partmod_features_Analyzing det_features_the dobj_Analyzing_features partmod_features_used aux_let_could nsubj_understand_us ccomp_let_understand dobj_understand_summarization advmod_understand_better ' +o,4949,'For this work , an off-the-shelf maximum entropy tagger 10 was used ',Ratnaparkhi,'prep_used_For det_work_this pobj_For_work det__an amod__off-the-shelf amod__maximum nn__entropy nn__tagger num__10 nsubjpass_used_ auxpass_used_was ' +o,4950,'Our evaluation metrics is casesensitive BLEU-4 ',Papineni,'poss_metrics_Our nn_metrics_evaluation nsubj_BLEU-4_metrics cop_BLEU-4_is amod_BLEU-4_casesensitive ' +o,4951,'For example , the distancebased reordering model allows a decoder to translate in non-monotonous order , under the constraint that the distance between two phrases translated consecutively does not exceed a limit known as distortion limit ',Koehn,'prep_allows_For pobj_For_example det__the amod__distancebased nn__reordering nn__model nsubj_allows_ det_decoder_a dobj_allows_decoder aux_translate_to infmod_decoder_translate prep_translate_in amod_order_non-monotonous pobj_in_order prep_allows_under det_constraint_the pobj_under_constraint complm_exceed_that det_distance_the nsubj_exceed_distance prep_distance_between num_phrases_two pobj_between_phrases partmod_phrases_translated dep_translated_consecutively aux_exceed_does neg_exceed_not dep_constraint_exceed det_limit_a dobj_exceed_limit partmod_limit_known prep_known_as nn_limit_distortion pobj_as_limit ' +o,4952,'These three parsers have given the best reported parsing results on the Penn Treebank Wall Street Journal corpus ',Marcus,'det_parsers_These num_parsers_three nsubj_given_parsers aux_given_have det_best_the iobj_given_best amod_results_reported amod_results_parsing dobj_given_results prep_results_on det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_on_corpus ' +o,4953,'Recently there have been some improvements to the Charniak parser , use n-best re-ranking as reported in and selftraining and re-ranking using data from the North American News corpus -LRB- NANC -RRB- and adapts much better to the Brown corpus ',McClosky,'advmod_been_Recently advmod_been_there aux_been_have det_improvements_some dobj_been_improvements prep_improvements_to det_parser_the nn_parser_Charniak pobj_to_parser conj_been_use amod_re-ranking_n-best dobj_use_re-ranking mark_reported_as advcl_use_reported prep_reported_in pcomp_in_ cc__and conj__selftraining cc__and conj__re-ranking xcomp_re-ranking_using dobj_using_data prep_using_from det_corpus_the nn_corpus_North nn_corpus_American nn_corpus_News pobj_from_corpus abbrev_corpus_NANC cc_been_and conj_been_adapts advmod_better_much nsubj_been_better prep_better_to det_corpus_the nn_corpus_Brown pobj_to_corpus ' +o,4954,'1 Introduction introduced minimum error rate training -LRB- MERT -RRB- as an alternative training regime to the conditional likelihood objective previously used with log-linear translation models ',Och,'num__1 nn__Introduction nsubj_introduced_ amod_training_minimum nn_training_error nn_training_rate dobj_introduced_training abbrev_training_MERT prep_training_as det_regime_an amod_regime_alternative nn_regime_training pobj_as_regime prep_introduced_to det_objective_the amod_objective_conditional nn_objective_likelihood pobj_to_objective advmod_used_previously partmod_objective_used prep_used_with amod_models_log-linear nn_models_translation pobj_with_models ' +o,4955,'We evaluate our results with case-sensitive BLEU-4 metric ',Papineni,'nsubj_evaluate_We poss_results_our dobj_evaluate_results prep_evaluate_with amod_metric_case-sensitive nn_metric_BLEU-4 pobj_with_metric ' +o,4956,'(2002) do not use a feature selection technique, employing instead an objective function which includes a Table 4 Values of Savings (a, b) for various values of a, b. ab Savings (a, b) 1100,000 2,692.7 110 48.6 11100 83.5 1011,000 280.0 1,00110,000 1,263.9 10,00150,000 2,920.2 50,001100,000 4,229.8 Collins and Koo Discriminative Reranking for NLP Gaussian prior on the parameter values, thereby penalizing parameter values which become too large: a C3 arg min a LogLossa X k0:::m a 2 k 7 2 k 28 Closed-form updates under iterative scaling are not possible with this objective function; instead, optimization algorithms such as gradient descent or conjugate gradient methods are used to estimate parameter values.',Collins,'' +o,4957,'4 Experiments We evaluated the ISBN parser on all the languages considered in the shared task ',Marcus,'num_Experiments_4 nsubj_evaluated_We rcmod_Experiments_evaluated det_parser_the nn_parser_ISBN dobj_evaluated_parser prep_evaluated_on predet_languages_all det_languages_the pobj_on_languages partmod_languages_considered prep_considered_in det_task_the amod_task_shared pobj_in_task ' +o,4958,'For each pivot feature k , we use a loss function L k , -LRB- -RRB- 2 1 -RRB- -LRB- wxwxpL i i T ikk + = -LRB- 1 -RRB- where the function p k -LRB- x i -RRB- indicates whether the pivot feature k occurs in the instance x i , otherwise xif xp ik ik 0 1 1 -RRB- -LRB- -RRB- = , where the weight vector w encodes the correspondence of the non-pivot features with the pivot feature k ',Blitzer,'prep_use_For det_k_each amod_k_pivot nn_k_feature pobj_For_k nsubj_use_we det_loss_a dobj_use_loss nn_k_function nn_k_L dep_loss_k number_-RRB-_2 number_-RRB-_1 dep_loss_wxwxpL nn_+_i nn_+_i nn_+_T nn_+_ikk dobj_wxwxpL_+ dep_+_= appos_+_1 advmod_indicates_where det_k_the nn_k_function nn_k_p nsubj_indicates_k nn_i_x appos_k_i rcmod_+_indicates complm_occurs_whether det_k_the amod_k_pivot nn_k_feature nsubj_occurs_k ccomp_indicates_occurs prep_occurs_in det_instance_the pobj_in_instance dep_instance_x dep_x_i advmod_instance_otherwise nn_0_xif nn_0_xp nn_0_ik nn_0_ik dep_instance_0 number_1_1 dep_0_1 dep_loss_= advmod_encodes_where det_w_the nn_w_weight nn_w_vector nsubj_encodes_w rcmod_loss_encodes det_correspondence_the dobj_encodes_correspondence prep_correspondence_of det_features_the amod_features_non-pivot pobj_of_features prep_features_with det_k_the amod_k_pivot nn_k_feature pobj_with_k ' +o,4959,'Bilingual configurations that condition on tprime , wprime -LRB- 2 -RRB- are incorporated into the generative process as in ',Smith,'nsubj_configurations_Bilingual complm_incorporated_that nsubjpass_incorporated_condition prep_condition_on pobj_on_tprime appos_tprime_wprime appos_wprime_2 auxpass_incorporated_are ccomp_configurations_incorporated prep_incorporated_into det_process_the amod_process_generative pobj_into_process prep_incorporated_as dep_as_in ' +o,4960,'Their weights are calculated by deleted interpolation ',Brown,'poss_weights_Their nsubjpass_calculated_weights auxpass_calculated_are prep_calculated_by amod_interpolation_deleted pobj_by_interpolation ' +o,4961,'To generate word alignments we use GIZA + + , which implements both the IBM Models of Brown et al ',Och,'aux_generate_To nn_alignments_word dobj_generate_alignments nsubj_use_we rcmod_alignments_use dobj_use_GIZA amod_GIZA_+ cc_+_+ conj_+_ nsubj_implements_which rcmod_GIZA_implements predet_Models_both det_Models_the nn_Models_IBM dobj_implements_Models prep_Models_of pobj_of_Brown cc_Brown_et conj_Brown_al ' +o,4962,'During training , the early update strategy of is used : when the correct state item falls out of the beam at any stage , parsing is stopped immediately , and the model is updated using the current best partial item ',Collins,'prep_used_During pobj_During_training det_strategy_the amod_strategy_early nn_strategy_update nsubjpass_used_strategy prep_strategy_of pobj_of_ auxpass_used_is advmod_falls_when det_item_the amod_item_correct nn_item_state nsubj_falls_item dep_stopped_falls prt_falls_out prep_falls_of det_beam_the pobj_of_beam prep_beam_at det_stage_any pobj_at_stage nsubjpass_stopped_parsing auxpass_stopped_is parataxis_used_stopped advmod_stopped_immediately cc_used_and det_model_the nsubj_updated_model aux_updated_is conj_used_updated xcomp_updated_using det_item_the amod_item_current advmod_item_best amod_item_partial dobj_using_item ' +o,4963,'The phoneme prediction and sequence modeling are considered as tagging problems and a Perceptron HMM is used to model it ',Collins,'det_prediction_The nn_prediction_phoneme nsubjpass_considered_prediction cc_prediction_and nn_modeling_sequence conj_prediction_modeling auxpass_considered_are mark_used_as amod_problems_tagging nsubjpass_used_problems cc_problems_and det__a nn__Perceptron nn__HMM conj_problems_ auxpass_used_is advcl_considered_used aux_model_to xcomp_used_model dobj_model_it ' +o,4964,'The system used for baseline experiments is two runs of IBM Model 4 in the GIZA + + implementation , which includes smoothing extensions to Model 4 ',Och,'det_system_The nsubj_runs_system partmod_system_used prep_used_for nn_experiments_baseline pobj_for_experiments cop_runs_is num_runs_two prep_runs_of nn__IBM nn__Model num__4 pobj_of_ prep__in det_implementation_the nn_implementation_GIZA amod_implementation_+ cc_+_+ conj_+_ pobj_in_implementation nsubj_includes_which rcmod_implementation_includes xcomp_includes_smoothing dobj_smoothing_extensions prep_smoothing_to pobj_to_Model num_Model_4 ' +o,4965,'In particular , mutual information and other statistical methods such as and frequency-based methods such as exclude infrequent phrases because they tend to introduce too much noise ',Church,'prep_exclude_In amod__particular amod__mutual nn__information pobj_In_ cc__and amod_methods_other amod_methods_statistical conj__methods dep_as_such prep_methods_as amod_methods_ cc__and conj__frequency-based pobj_as_methods dep_as_such prep_methods_as pobj_as_ acomp_exclude_infrequent nsubj_exclude_phrases mark_tend_because nsubj_tend_they ccomp_phrases_tend aux_introduce_to xcomp_tend_introduce advmod_much_too amod_noise_much dobj_introduce_noise ' +o,4966,'Inspired by the idea of graph based algorithms to collectively rank and select the best candidate , research efforts in the natural language community have applied graph-based approaches on keyword selection , text summarization , word sense disambiguation , sentiment analysis , and sentence retrieval for question answering ',Pang,'prep_Inspired_by det_idea_the pobj_by_idea prep_idea_of pobj_of_graph partmod_graph_based dep_Inspired_algorithms aux_rank_to advmod_rank_collectively xcomp_algorithms_rank cc_rank_and conj_rank_select det_candidate_the amod_candidate_best nsubj_Inspired_candidate nn_efforts_research appos_candidate_efforts prep_efforts_in det_community_the amod_community_natural nn_community_language pobj_in_community aux_applied_have rcmod_community_applied amod_approaches_graph-based dobj_applied_approaches prep_applied_on amod__keyword nn__selection pobj_on_ nn__text nn__summarization conj__ nn__word nn__sense nn__disambiguation conj__ nn__sentiment nn__analysis conj__ cc__and nn_retrieval_sentence conj__retrieval prep_retrieval_for nn__question nn__answering pobj_for_ ' +o,4967,'By comparing derivation trees for parallel sentences in two languages , instances of structural divergences can be automatically detected ',Marcus,'prep_detected_By amod_trees_comparing nn_trees_derivation pobj_By_trees prep_trees_for amod_sentences_parallel pobj_for_sentences prep_sentences_in num_languages_two pobj_in_languages nsubjpass_detected_instances prep_instances_of amod__structural nn__divergences pobj_of_ aux_detected_can auxpass_detected_be advmod_detected_automatically ' +o,4968,'Language models , such as N-gram class models and Ergodic Hidden Markov Models were proposed and used in applications such as syntactic class -LRB- POS -RRB- tagging for English , clustering and scoring of recognizer sentence hypotheses ',Cutting,'amod_models_Language dep_as_such prep_models_as amod_models_N-gram nn_models_class pobj_as_models nn__ cc__and conj__Ergodic nn__Hidden nn__Markov nn__Models nsubjpass_proposed_ auxpass_proposed_were rcmod_models_proposed cc_proposed_and conj_proposed_used prep_proposed_in nsubj_tagging_applications dep_as_such prep_applications_as amod_class_syntactic pobj_as_class appos_class_POS pcomp_in_tagging prep_tagging_for nn__English pobj_for_ conj_tagging_clustering cc_tagging_and conj_tagging_scoring prep_scoring_of nn_hypotheses_recognizer nn_hypotheses_sentence pobj_of_hypotheses ' +o,4969,'5 Discussion and Future Work The work in this paper substantially differs from previous work in SMT based on the noisy channel approach presented in ',Brown,'dep_Discussion_5 cc_Discussion_and amod_Work_Future conj_Discussion_Work det_work_The nsubj_differs_work prep_work_in det_paper_this pobj_in_paper advmod_differs_substantially dep_Discussion_differs prep_differs_from amod_work_previous pobj_from_work prep_work_in pobj_in_SMT prep_differs_based dep_based_on det_approach_the amod_approach_noisy nn_approach_channel pobj_on_approach partmod_approach_presented prep_presented_in ' +o,4970,'22 Co-occurrence-based approaches The second class of algorithms uses cooccurrence statistics ',Hindle,'num_approaches_22 amod_approaches_Co-occurrence-based det_class_The amod_class_second dep_approaches_class prep_class_of amod_uses_algorithms pobj_of_uses amod_statistics_cooccurrence dep_approaches_statistics ' +o,4971,'An early exception to this was itself , where Model 2 used function tags during the training process for heuristics to identify arguments -LRB- eg , the TMP tag on the NP in Figure 1 disqualifies the NP-TMP from being treated as an argument -RRB- ',Collins,'det_exception_An amod_exception_early nsubjpass__exception prep_exception_to pobj_to_this auxpass__was dobj__itself advmod_used_where nsubj_used_Model num_Model_2 advcl__used nn_tags_function dobj_used_tags prep_used_during det_process_the nn_process_training pobj_during_process prep_process_for pobj_for_heuristics aux_identify_to xcomp_used_identify dobj_identify_arguments nsubj_disqualifies_eg det_tag_the nn_tag_TMP appos_eg_tag prep_tag_on det_NP_the pobj_on_NP prep_NP_in pobj_in_Figure num_Figure_1 dep_arguments_disqualifies det_NP-TMP_the dobj_disqualifies_NP-TMP prep_disqualifies_from auxpass_treated_being pcomp_from_treated prep_treated_as det_argument_an pobj_as_argument ' +o,4972,'Other metrics assess the impact of alignments externally , eg , different alignments are tested by comparing the corresponding MT outputs using automated evaluation metrics -LRB- eg , BLEU or METEOR -RRB- ',Papineni,'amod_metrics_Other nsubj_assess_metrics det_impact_the dobj_assess_impact prep_impact_of pobj_of_alignments advmod_tested_externally nn_alignments_eg amod_alignments_different nsubjpass_tested_alignments auxpass_tested_are dep_assess_tested prep_tested_by pcomp_by_comparing det_outputs_the amod_outputs_corresponding nn_outputs_MT dobj_comparing_outputs xcomp_comparing_using amod_metrics_automated nn_metrics_evaluation dobj_using_metrics dep_metrics_eg nn__BLEU conj_eg_ cc_eg_or nn__METEOR conj_eg_ ' +p,4973,'WSD is one of the fundamental problems in natural language processing and is important for applications such as machine translation -LRB- MT -RRB- , information retrieval -LRB- IR -RRB- , etc WSD is typically viewed as a classification problem where each ambiguous word is assigned a sense label -LRB- from a pre-defined sense inventory -RRB- during the disambiguation process ',Carpuat,'nsubj_one_WSD cop_one_is prep_one_of det_problems_the amod_problems_fundamental pobj_of_problems prep_problems_in amod_processing_natural nn_processing_language pobj_in_processing cc_one_and cop_important_is conj_one_important prep_important_for pobj_for_applications dep_as_such prep_applications_as nn_translation_machine pobj_as_translation abbrev_translation_MT nsubj_etc_ nn_retrieval_information appos__retrieval abbrev_retrieval_IR dep_one_etc nsubjpass_viewed_WSD auxpass_viewed_is advmod_viewed_typically ccomp_etc_viewed prep_viewed_as det_problem_a nn_problem_classification pobj_as_problem advmod_assigned_where det_word_each amod_word_ambiguous nsubjpass_assigned_word auxpass_assigned_is advcl_viewed_assigned det_label_a nn_label_sense dobj_assigned_label dep_label_from det_inventory_a amod_inventory_pre-defined nn_inventory_sense pobj_from_inventory prep_assigned_during det_process_the nn_process_disambiguation pobj_during_process ' +o,4974,'This approach is also used in base-NP chunking and named entity recognition as well as word segmentation ',Ramshaw,'det_approach_This nsubjpass_used_approach auxpass_used_is advmod_used_also prep_used_in amod__base-NP amod__chunking pobj_in_ cc_used_and conj_used_named nn__entity nn__recognition dobj_named_ dep_well_as cc__well dep_well_as nn_segmentation_word conj__segmentation ' +o,4975,'Then , the method of can be used to compute the probability of every possible edge conditioned on the presence of ki , p -LRB- yiprime = kprime yi = k , x -RRB- , using K1ki Multiplying this probability by p -LRB- yi = k x -RRB- yields the desired two edge marginal ',Smith,'advmod_used_Then det_method_the nsubjpass_used_method prep_method_of pobj_of_ aux_used_can auxpass_used_be aux_compute_to purpcl_used_compute det_probability_the dobj_compute_probability prep_probability_of det_edge_every amod_edge_possible pobj_of_edge partmod_edge_conditioned prep_conditioned_on det_presence_the pobj_on_presence prep_presence_of pobj_of_ki appos_ki_p nsubj_=_yiprime dep_p_= amod_yi_kprime dobj_=_yi dep_k_= rcmod_yi_k appos_yi_x csubj_yields_using dobj_using_K1ki xcomp_using_Multiplying det_probability_this dobj_Multiplying_probability prep_Multiplying_by pobj_by_p nsubj_x_yi dep_x_= nn_x_k parataxis_Multiplying_x dep_used_yields dep_two_the amod_two_desired nsubj_edge_two ccomp_yields_edge acomp_edge_marginal ' +o,4976,'Standard MET iterative parameter estimation under IBM BLEU is performed on the corresponding development set ',Och,'nn_estimation_Standard nn_estimation_MET nn_estimation_ nn_estimation_iterative nn_estimation_parameter nsubjpass_performed_estimation prep_estimation_under nn__IBM nn__BLEU pobj_under_ auxpass_performed_is prep_performed_on det_set_the amod_set_corresponding nn_set_development pobj_on_set ' +o,4977,'Item Form: a32 a2 a49a51 a15 a52 a49 a51a16a33 Goal: a32a35a34 a49 a51 a15 a23a4a3 a12 a0a36a5 a24 a49 a51a37a33 Inference Rules Scan component d, a10a38a8 a7 a8 a0 : a39a41a40a43a42a44 a44a45 a23a25a24 a49 a5a47a46 a49 a2 a23a25a24 a5a49a48 a49 a51 a50 a23a25a24 a49 a5a47a46 a49 a20a43a5 a3a22 a23a25a24 a5a49a48 a49 a51 a51a14a52 a52 a53 a54a55 a55 a56 a23a25a24 a49 a5a47a46 a49 a2 a23a25a24 a5a49a48 a49 a51 a50 a23a25a24 a49 a5a47a46 a49a23 a19a57a24 a10a13a12 a19 a24 a23a25a24 a5a49a48 a49 a51 a58a59 a59 a60 Compose: a61a63a62a65a64 a66a68a67a69 a64 a66a71a70 a61a35a72a37a64 a66a68a67a73 a64 a66a71a70a36a74a76a75 a32a78a77 a64 a66a76a67a69 a64 a66a80a79a81a73 a64 a66 a14 a62a82a64 a66 a14 a72a37a64 a66 a33 a10 a77 a64 a66 a67a69 a64 a66a37a83 a73 a64 a66 a18 Figure 3: Logic C (C for CKY) These constraints are enforced by the d-span operators a84 and a85 . Parser C is conceptually simpler than the synchronous parsers of Wu (1997), Alshawi et al.',Wu,'' +o,4978,'Several authors have attempted to improve results by using supplementary fields of information in the electronic version of the Longman Dictionary of Contemporary English -LRB- LDOCE -RRB- , in particular , the box codes and subject codes provided for each sense ',Brown,'amod__Several nn__authors nsubj_attempted_ aux_attempted_have aux_improve_to xcomp_attempted_improve dobj_improve_results prep_improve_by pcomp_by_using amod_fields_supplementary dobj_using_fields prep_fields_of pobj_of_information prep_using_in det_version_the amod_version_electronic pobj_in_version prep_version_of det_Dictionary_the nn_Dictionary_Longman pobj_of_Dictionary prep_Dictionary_of nn_English_Contemporary pobj_of_English abbrev_English_LDOCE prep_attempted_in pobj_in_particular det_codes_the nn_codes_box appos_particular_codes cc_codes_and nn_codes_subject conj_codes_codes partmod_codes_provided prep_provided_for det_sense_each pobj_for_sense ' +o,4979,'Daume allows an extra degree of freedom among the features of his domains , implicitly creating a two-level feature hierarchy with one branch for general features , and another for domain specific ones , but does not extend his hierarchy further -RRB- ','Daume III','nsubj_allows_Daume det_degree_an amod_degree_extra dobj_allows_degree prep_degree_of pobj_of_freedom prep_freedom_among det_features_the pobj_among_features prep_features_of poss_domains_his pobj_of_domains advmod_creating_implicitly partmod_degree_creating det_hierarchy_a amod_hierarchy_two-level nn_hierarchy_feature dobj_creating_hierarchy prep_hierarchy_with num_branch_one pobj_with_branch prep_branch_for amod_features_general pobj_for_features cc_hierarchy_and conj_hierarchy_another prep_another_for nn_ones_domain amod_ones_specific pobj_for_ones cc_allows_but aux_extend_does neg_extend_not conj_allows_extend poss__his nn__hierarchy amod__further dobj_extend_ ' +o,4980,'Then , we apply a grow-diag-final algorithm which is widely used in bilingual phrase extraction to monolingual alignments ',Koehn,'advmod_apply_Then nsubj_apply_we det_algorithm_a amod_algorithm_grow-diag-final dobj_apply_algorithm nsubjpass_used_which auxpass_used_is advmod_used_widely rcmod_algorithm_used prep_used_in amod__bilingual nn__phrase nn__extraction pobj_in_ prep_used_to amod_alignments_monolingual pobj_to_alignments ' +n,4981,'The class-based kappa statistic of can not be applied here , as the classes vary depending on the number of ambiguities per entry in the lexicon ',Carletta,'det_statistic_The amod_statistic_class-based nn_statistic_kappa nsubjpass_applied_statistic prep_statistic_of pobj_of_ aux_applied_can neg_applied_not auxpass_applied_be advmod_applied_here mark_vary_as det_classes_the nsubj_vary_classes advcl_applied_vary prep_vary_depending dep_depending_on det_number_the pobj_on_number prep_number_of pobj_of_ambiguities prep_ambiguities_per pobj_per_entry prep_entry_in det_lexicon_the pobj_in_lexicon ' +p,4982,' has described an efficient exact onedimensional accuracy maximization technique for a similar search problem in machine translation ',Och,'nsubj_described_ aux_described_has det_technique_an amod_technique_efficient dep_efficient_exact amod_technique_onedimensional nn_technique_accuracy nn_technique_maximization dobj_described_technique prep_technique_for det_problem_a amod_problem_similar nn_problem_search pobj_for_problem prep_described_in nn_translation_machine pobj_in_translation ' +o,4983,'The translation quality is evaluated by BLEU metric , as calculated by mteval-v11bpl with case-insensitive matching of n-grams , where n = 4 ',Papineni,'det_quality_The nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_is prep_evaluated_by amod__BLEU amod__metric pobj_by_ mark_calculated_as advcl_evaluated_calculated prep_calculated_by pobj_by_mteval-v11bpl prep_calculated_with amod_matching_case-insensitive pobj_with_matching prep_matching_of pobj_of_n-grams advmod_4_where nsubj_4_n dep_4_= rcmod_matching_4 ' +o,4984,'The second voting model is a maximum entropy model , since found that this model yielded higher accuracy than naive Bayes in a subsequent comparison of WSD performance ',Klein,'det_model_The amod_model_second nn_model_voting nsubj__model cop__is det__a amod__maximum amod__entropy nn__model mark_found_since nsubj_found_ advcl__found complm_yielded_that det_model_this nsubj_yielded_model ccomp_found_yielded amod_accuracy_higher dobj_yielded_accuracy prep_yielded_than nn_Bayes_naive pobj_than_Bayes prep_yielded_in det_comparison_a amod_comparison_subsequent pobj_in_comparison prep_comparison_of nn_performance_WSD pobj_of_performance ' +o,4985,'While transfer learning was proposed more than a decade ago , its application in natural language processing is still a relatively new territory , and its application in relation extraction is still unexplored ',Jiang,'mark_proposed_While nn_learning_transfer nsubjpass_proposed_learning auxpass_proposed_was advcl__proposed dep_than_more quantmod_a_than num_decade_a measure_ago_decade advmod_proposed_ago dep_ago_ poss_application_its nsubj__application prep_application_in amod_processing_natural nn_processing_language pobj_in_processing cop__is advmod__still det__a advmod_new_relatively amod__new nn__territory cc__and poss_application_its nsubjpass_unexplored_application prep_application_in nn_extraction_relation pobj_in_extraction auxpass_unexplored_is advmod_unexplored_still conj__unexplored ' +o,4986,'Other classes , such as the ones below can be extracted using lexico-statistical tools , such as in , and then checked by a human ',Smadja,'amod_classes_Other amod_classes_such mark_extracted_as det_ones_the nsubjpass_extracted_ones advmod_extracted_below aux_extracted_can auxpass_extracted_be ccomp_such_extracted purpcl_extracted_using amod_tools_lexico-statistical dobj_using_tools dep_as_such dep_classes_as dep_as_in pobj_in_ cc_classes_and advmod_checked_then conj_classes_checked prep_checked_by det_human_a pobj_by_human ' +o,4987,'These include the perceptron and its large-margin variants ',Collins,'nsubj_include_These det__the nn__perceptron dobj_include_ cc__and poss__its amod__large-margin amod__variants conj__ ' +o,4988,'The other intriguing issue is how our anchor-based method for shared argument identification can benefit from recent advances in coreference and zero-anaphora resolution ',Iida,'det_issue_The amod_issue_other amod_issue_intriguing nsubj_is_issue advmod_benefit_how poss_method_our amod_method_anchor-based nsubj_benefit_method prep_method_for amod_identification_shared nn_identification_argument pobj_for_identification aux_benefit_can ccomp_is_benefit prep_benefit_from amod_advances_recent pobj_from_advances prep_advances_in amod_resolution_coreference cc_coreference_and conj_coreference_zero-anaphora pobj_in_resolution ' +o,4989,'Another important direction is classifying sentences as subjective or objective , and classifying subjective sentences or clauses as positive or negative ',McDonald,'det_direction_Another amod_direction_important nsubj_classifying_direction aux_classifying_is dobj_classifying_sentences prep_classifying_as pobj_as_subjective cc_subjective_or conj_subjective_objective cc_classifying_and conj_classifying_classifying amod_sentences_subjective nsubj_positive_sentences cc_sentences_or conj_sentences_clauses advmod_positive_as xcomp_classifying_positive cc_positive_or conj_positive_negative ' +o,4990,' proposed a method to retrieve collocations by combining bigrams whose cooccurrences are greater than a given threshold 3 ',Smadja,'advmod_proposed_ det_method_a dobj_proposed_method aux_retrieve_to xcomp_proposed_retrieve dobj_retrieve_collocations prep_retrieve_by pcomp_by_combining dobj_combining_bigrams poss_greater_whose nsubj_greater_cooccurrences cop_greater_are dep_bigrams_greater prep_greater_than det_threshold_a amod_threshold_given pobj_than_threshold nsubj_proposed_3 ' +o,4991,'Discriminative models do not only have theoretical advantages over generative models , as we discuss in Section 2 , but they are also shown to be empirically favorable over generative models when features and objective functions are fixed ',Klein,'amod_models_Discriminative nsubj_have_models aux_have_do neg_have_not advmod_have_only amod_advantages_theoretical dobj_have_advantages prep_advantages_over amod_models_generative pobj_over_models mark_discuss_as nsubj_discuss_we advcl_have_discuss prep_discuss_in pobj_in_Section num_Section_2 cc_discuss_but nsubjpass_shown_they auxpass_shown_are advmod_shown_also conj_discuss_shown aux_favorable_to cop_favorable_be advmod_favorable_empirically xcomp_shown_favorable prep_favorable_over amod_models_generative pobj_over_models advmod_fixed_when nsubjpass_fixed_features cc_features_and nn_functions_objective conj_features_functions auxpass_fixed_are advcl_favorable_fixed ' +o,4992,'3 Feature selection proposed an iterative procedure of adding news features to feature set driven by data ',Berger,'num__3 nn__Feature nn__selection dep_proposed_ det_procedure_an amod_procedure_iterative dobj_proposed_procedure prep_procedure_of pcomp_of_adding nn_features_news dobj_adding_features aux_feature_to xcomp_adding_feature dobj_feature_set partmod_set_driven prep_driven_by pobj_by_data ' +o,4993,'One of our goals was to use for this study only information that could be annotated reliably , as we believe this will make our results easier to replicate ',Carletta,'nsubj_was_One prep_One_of poss_goals_our pobj_of_goals aux_use_to xcomp_was_use prep_use_for det_information_this nn_information_study amod_information_only pobj_for_information nsubjpass_annotated_that aux_annotated_could auxpass_annotated_be rcmod_information_annotated advmod_annotated_reliably advmod_annotated_ mark_believe_as nsubj_believe_we advcl_annotated_believe nsubj_make_this aux_make_will ccomp_believe_make poss_results_our nsubj_easier_results xcomp_make_easier aux_replicate_to xcomp_easier_replicate ' +o,4994,'task , originally introduced in and also described in , brackets just base NP constituents5 ',Ramshaw,'nsubj_brackets_task advmod_introduced_originally dep_task_introduced prep_introduced_in pobj_in_ cc_introduced_and advmod_described_also conj_introduced_described prep_described_in pobj_in_ advmod_constituents5_just nn_constituents5_base nn_constituents5_NP dobj_brackets_constituents5 ' +p,4995,'While these are based on a relatively few number of items, and while we have not performed any tests to determine whether the differences in ? are statistically significant, the results 7The Czech-English conditions were excluded since there were so few systems 146 are nevertheless interesting, since three metrics have higher correlation than Bleu: ??Semantic role overlap (Gimenez and M`arquez, 2007), which makes its debut in the proceedings of this workshop ??ParaEval measuring recall (Zhou et al. , 2006), which has a model of allowable variation in translation that uses automatically generated paraphrases (Callison-Burch, 2007) ??Meteor (Banerjee and Lavie, 2005) which also allows variation by introducing synonyms and by flexibly matches words using stemming.',Banerjee,'' +o,4996,'Following , we used the version 11a NIST BLEU script with its default settings to calculate the BLEU scores based on case-insensitive ngram matching , where n is up to 4 ',Papineni,'prep_used_Following pobj_Following_ nsubj_used_we det_11a_the nn_11a_version nsubj_script_11a dep_script_NIST amod_script_BLEU ccomp_used_script prep_script_with poss_settings_its nn_settings_default pobj_with_settings aux_calculate_to xcomp_script_calculate det__the amod__BLEU nn__scores dobj_calculate_ prep_calculate_based dep_based_on amod_matching_case-insensitive nn_matching_ngram pobj_on_matching advmod_4_where nsubj_4_n cop_4_is quantmod_4_up dep_4_to advcl_script_4 ' +o,4997,' predicates the sentiment orientation of a review by the average semantic orientation of the phrases in the review that contain adjectives or adverbs , which is denoted as the semantic oriented method ',Turney,'advmod_predicates_ det_orientation_the nn_orientation_sentiment dobj_predicates_orientation prep_orientation_of det_review_a pobj_of_review prep_predicates_by det_orientation_the amod_orientation_average amod_orientation_semantic pobj_by_orientation prep_orientation_of det_phrases_the pobj_of_phrases prep_phrases_in det_review_the pobj_in_review nsubj_contain_that dep_predicates_contain dobj_contain_adjectives cc_adjectives_or conj_adjectives_adverbs nsubjpass_denoted_which auxpass_denoted_is rcmod_adjectives_denoted prep_denoted_as det_method_the amod_method_semantic dep_semantic_oriented pobj_as_method ' +o,4998,'The Logllkelihood Ratio , G 2 , is a mathematically well-grounded and accurate method for calculating how ` surprising \' an event is ',Dunning,'det_Ratio_The nn_Ratio_Logllkelihood nsubj_method_Ratio appos_Ratio_G num_G_2 cop_method_is det_method_a amod_method_mathematically amod_method_well-grounded cc_well-grounded_and conj_well-grounded_accurate prep_method_for pcomp_for_calculating advmod_is_how ccomp_is_surprising det_event_an nsubj_is_event ccomp_calculating_is ' +o,4999,'417 structure of semantic networks was proposed in , with a disambiguation accuracy of 509 \% measured on all the words in the SENSEVAL-2 data set ',Mihalcea,'amod_structure_417 nsubjpass_proposed_structure prep_structure_of amod_networks_semantic pobj_of_networks auxpass_proposed_was prep_proposed_in pobj_in_ prep_proposed_with det_accuracy_a nn_accuracy_disambiguation pobj_with_accuracy prep_accuracy_of num_\%_509 pobj_of_\% partmod_\%_measured prep_measured_on predet_words_all det_words_the pobj_on_words prep_words_in det_set_the nn_set_SENSEVAL-2 nn_set_data pobj_in_set ' +o,5000,'We describe the experiment in greater detail 2The particular verbs selected were looked up in and the class for each verb in the classification system defined in was selected with some discussion with linguists ',Dunning,'nsubj_describe_We det_experiment_the dobj_describe_experiment prep_experiment_in amod_2The_greater nn_2The_detail pobj_in_2The amod_verbs_particular nsubjpass_looked_verbs partmod_verbs_selected auxpass_looked_were dep_describe_looked prt_looked_up prep_looked_in pobj_in_ cc_looked_and det_class_the nsubjpass_selected_class prep_class_for det_verb_each pobj_for_verb prep_verb_in det_system_the nn_system_classification pobj_in_system partmod_system_defined prep_defined_in pobj_in_ auxpass_selected_was conj_looked_selected prep_selected_with det_discussion_some pobj_with_discussion prep_discussion_with pobj_with_linguists ' +o,5001,'So fitr , we have implemented the following , : sentence ~ dignment btLsed-on word correspondence information , word correspondence estimation by cooccnl ` rence-ffequency-based methods in GMe mid Church -LRB- 19 ~ H -RRB- and Kay and R6scheisen , structured Imttehlng of parallel sentences , and case Dame acquisition of Japanese verbs ',Brown,'advmod_fitr_So nsubj_implemented_we aux_implemented_have parataxis_fitr_implemented det_following_the dobj_implemented_following nn_estimation_sentence nn_estimation_~ nn_estimation_dignment nn_estimation_btLsed-on nn_estimation_word nn_estimation_correspondence nn_estimation_information dep_estimation_word nn_estimation_correspondence dobj_fitr_estimation prep_estimation_by pobj_by_cooccnl amod_methods_rence-ffequency-based nsubj_fitr_methods prep_methods_in nn_Church_GMe nn_Church_mid pobj_in_Church num_H_19 nn_H_~ appos_Church_H cc_methods_and conj_methods_Kay cc_Kay_and nn__R6scheisen conj_Kay_ partmod_methods_structured dobj_structured_Imttehlng prep_Imttehlng_of amod__parallel nn__sentences pobj_of_ cc_Imttehlng_and nn_acquisition_case nn_acquisition_Dame conj_Imttehlng_acquisition prep_acquisition_of amod__Japanese nn__verbs pobj_of_ ' +o,5002,'of ACL 1990 , F Smadja , Retrieving collocations fi ` cma text : XTRACT , -LRB- 1993 -RRB- ',Smadja,'nn__ACL num__1990 pobj_of_ nn_Smadja_F dep_of_Smadja nn_collocations_Retrieving appos_Smadja_collocations partmod_collocations_fi nn_text_cma dep_fi_text dep_text_XTRACT dep_fi_1993 ' +o,5003,'322 Features We used eight features and their weights for the translations ',Och,'nn_Features_322 nsubj_used_We dep_Features_used num_features_eight dobj_used_features amod_features_ cc_features_and poss_weights_their conj_features_weights prep_used_for det_translations_the pobj_for_translations ' +o,5004,'-LRB- discusses the recovery of one kind of empty node , viz ',Collins,'amod_discusses_ det_recovery_the dep_of_recovery prep_discusses_of num_kind_one pobj_of_kind prep_kind_of amod_node_empty pobj_of_node appos_node_viz ' +o,5005,'This idea is the same as ',Turney,'det_idea_This nsubj_is_idea det_same_the dep_as_same advmod_is_as ' +p,5006,'A number of part-of-speech taggers are readily available and widely used , all trained and retrainable on text corpora ',Cutting,'det_number_A nsubj_available_number prep_number_of amod_taggers_part-of-speech pobj_of_taggers cop_available_are advmod_available_readily cc_available_and advmod_used_widely conj_available_used advmod_trained_all conj_available_trained cc_available_and conj_available_retrainable prep_retrainable_on nn_corpora_text pobj_on_corpora ' +n,5007,'While the amount of parallel data required to build such systems is orders of magnitude smaller than corresponding phrase based statistical systems , the variety of linguistic annotation required is greater ',Koehn,'mark_smaller_While det_amount_the nsubj_smaller_amount prep_amount_of amod_data_parallel pobj_of_data partmod_data_required aux_build_to xcomp_required_build amod_systems_such dobj_build_systems cop_smaller_is measure_smaller_orders prep_orders_of pobj_of_magnitude advcl_greater_smaller prep_smaller_than amod_phrase_corresponding pobj_than_phrase partmod_phrase_based amod_systems_statistical dobj_based_systems advmod_based_ det_variety_the nsubj_greater_variety prep_variety_of amod_annotation_linguistic pobj_of_annotation partmod_annotation_required cop_greater_is ' +o,5008,' present a lexical similarity model based on random walks on graphs derived from WordNet ; Rao et al ',Hughes,'advmod_present_ det_model_a amod_model_lexical nn_model_similarity dobj_present_model prep_present_based dep_based_on amod_walks_random pobj_on_walks prep_walks_on pobj_on_graphs partmod_graphs_derived prep_derived_from pobj_from_WordNet conj_WordNet_Rao cc_WordNet_et conj_WordNet_al ' +o,5009,'For example , syntactic features can be computed this way and are used in our system ',Luo,'prep_computed_For pobj_For_example amod__syntactic nn__features nsubjpass_computed_ aux_computed_can auxpass_computed_be det_way_this dobj_computed_way cc_computed_and auxpass_used_are conj_computed_used prep_used_in poss_system_our pobj_in_system ' +o,5010,'Due to the parameter interdependencies introduced by the one-to-one assumption , we are unlikely to find a method for decomposing the assignments into parameters that can be estimated independently of each other as in Brown et al -RRB- ',Dunning,'dep_to_Due prep_unlikely_to det_interdependencies_the nn_interdependencies_parameter pobj_to_interdependencies partmod_interdependencies_introduced prep_introduced_by det_assumption_the amod_assumption_one-to-one pobj_by_assumption nsubj_unlikely_we cop_unlikely_are aux_find_to xcomp_unlikely_find det_method_a dobj_find_method prep_find_for pcomp_for_decomposing det_assignments_the dobj_decomposing_assignments prep_decomposing_into pobj_into_parameters nsubjpass_estimated_that aux_estimated_can auxpass_estimated_be rcmod_parameters_estimated advmod_estimated_independently prep_estimated_of det_other_each pobj_of_other prep_estimated_as dep_as_in nn__Brown cc_Brown_et conj_Brown_al pobj_in_ ' +o,5011,'2 Disperp and Distortion Corpora 21 Defining Disperp The ultimate reason for choosing one SCM over another will be the performance of an MT system containing it , as measured by a metric like BLEU ',Papineni,'num_Disperp_2 nsubj_performance_Disperp cc_Disperp_and nn_Corpora_Distortion conj_Disperp_Corpora num_Corpora_21 partmod_Disperp_Defining xcomp_Defining_Disperp det_reason_The amod_reason_ultimate dobj_Disperp_reason prep_Disperp_for pcomp_for_choosing num_SCM_one dobj_choosing_SCM prep_SCM_over pobj_over_another aux_performance_will cop_performance_be det_performance_the prep_performance_of det_system_an nn_system_MT pobj_of_system partmod_system_containing dobj_containing_it mark_measured_as advcl_performance_measured prep_measured_by det_BLEU_a amod_BLEU_metric amod_BLEU_like pobj_by_BLEU ' +o,5012,'22 Statistical Parsers Pioneered by the IBM natural language group and later pursued by , for example , Schabes , Roth , and Osborne , Jelinek et al ',Marcus,'num_Parsers_22 nn_Parsers_Statistical nsubj_Pioneered_Parsers prep_Pioneered_by det__the nn__IBM amod__natural nn__language nn__group pobj_by_ cc_Pioneered_and advmod_pursued_later conj_Pioneered_pursued prep_pursued_by dep_by_for pobj_for_example pobj_by_Schabes conj_Schabes_Roth cc_Schabes_and nn__Osborne conj_Schabes_ conj__Jelinek cc__et conj__al ' +o,5013,'3ThePOS taggers The two POS taggers used in the experiments are TNT , a publicly available Markov model tagger , and a reimplementation of the maximum entropy -LRB- ME -RRB- tagger MXPOST ',Ratnaparkhi,'amod_taggers_3ThePOS det_taggers_The num_taggers_two dep_taggers_POS nsubj_TNT_taggers partmod_taggers_used prep_used_in det_experiments_the pobj_in_experiments cop_TNT_are dep_taggers_TNT det__a advmod_available_publicly amod__available amod__Markov nn__model nn__tagger conj_TNT_ cc_TNT_and det_reimplementation_a conj_TNT_reimplementation prep_reimplementation_of det_MXPOST_the amod_MXPOST_maximum amod_MXPOST_entropy abbrev_MXPOST_ME amod_MXPOST_tagger pobj_of_MXPOST ' +o,5014,'In earlier IBM translation systems each English word would be generated by , or ` aligned to \' , exactly one formal language word ',Brown,'prep__In amod_systems_earlier nn_systems_IBM nn_systems_translation pobj_In_systems det_word_each amod_word_English dobj__word aux_generated_would auxpass_generated_be rcmod_word_generated prep_generated_by cc_word_or conj_word_aligned prep_aligned_to advmod__exactly num_word_one amod_word_formal nn_word_language nsubj__word ' +o,5015,'52 Experimental Results Following and other work on general-purpose generators , BLEU score , average NIST simple string accuracy -LRB- SSA -RRB- and percentage of exactly matched sentences are adopted as evaluation metrics ',Papineni,'num_Results_52 amod_Results_Experimental nsubjpass_adopted_Results prep_Results_Following amod_work_ cc__and conj__other pobj_Following_work prep_work_on amod_generators_general-purpose pobj_on_generators amod__BLEU nn__score conj_work_ amod_accuracy_average amod_accuracy_NIST amod_accuracy_simple nn_accuracy_string conj_work_accuracy abbrev_accuracy_SSA cc_work_and conj_work_percentage prep_percentage_of advmod_sentences_exactly amod_sentences_matched pobj_of_sentences auxpass_adopted_are prep_adopted_as nn_metrics_evaluation pobj_as_metrics ' +o,5016,'Then the two models and a search module are used to decode the best translation ',Brown,'dep_used_the num_models_two nsubjpass_used_models cc_models_and det_module_a nn_module_search conj_models_module auxpass_used_are dep_Then_used aux_decode_to xcomp_used_decode det_translation_the amod_translation_best dobj_decode_translation ' +o,5017,'For process -LRB- 3 -RRB- , machine-learning methods are usually used to classify subjective descriptions into bipolar categories or multipoint scale categories ',Pang,'prep_used_For pobj_For_process appos_process_3 amod_methods_machine-learning nsubjpass_used_methods auxpass_used_are advmod_used_usually aux_classify_to xcomp_used_classify amod_descriptions_subjective dobj_classify_descriptions prep_classify_into amod__bipolar nn__categories pobj_into_ cc__or nn_categories_multipoint nn_categories_scale conj__categories ' +o,5018,'Then , those structurally matched parallel sentences are used as a source for acquiring lexical knowledge snch as verbal case frames ',Brown,'advmod_matched_Then det_structurally_those nsubj_matched_structurally amod_sentences_parallel nsubjpass_used_sentences auxpass_used_are ccomp_matched_used prep_used_as det_source_a pobj_as_source prep_used_for pcomp_for_acquiring amod_snch_lexical nn_snch_knowledge dobj_acquiring_snch prep_acquiring_as amod__verbal nn__case nn__frames pobj_as_ ' +o,5019,'Above the phrase level , some models perform no reordering , some have a simple distortion model that reorders phrases independently of their content , and some , for example , the Alignment Template System , hereafter ATS , and the IBM phrase-based system , have phrase-reordering models that add some lexical sensitivity ',Koehn,'dep_have_Above det_level_the nn_level_phrase pobj_Above_level det_models_some nsubj_perform_models dep_have_perform det__no nn__reordering dobj_perform_ nsubj_have_some det_model_a amod_model_simple nn_model_distortion dobj_have_model nsubj_reorders_that rcmod_model_reorders dobj_reorders_phrases dep_of_independently prep_phrases_of poss__their nn__content pobj_of_ cc_have_and nsubj_have_some prep_have_for pobj_for_example det__the nn__Alignment nn__Template nn__System conj_example_ nn_ATS_hereafter conj_example_ATS cc_example_and det__the nn__IBM amod__phrase-based nn__system conj_example_ conj_have_have amod_models_phrase-reordering dobj_have_models nsubj_add_that rcmod_models_add det_sensitivity_some amod_sensitivity_lexical dobj_add_sensitivity ' +o,5020,'In fact, when the perceptron update rule of (Dekel et al. , 2004) which modifies the weights of every divergent node along the predicted and true paths is used in the ranking framework, it becomes virtually identical with the standard, flat, ranking perceptron of Collins (2002).5 In contrast, our approach shares the idea of (Cesa-Bianchi et al. , 2006a) that if a parent class has been predicted wrongly, then errors in the children should not be taken into account. We also view this as one of the key ideas of the incremental perceptron algorithm of (Collins and Roark, 2004), which searches through a complex decision space step-by-step and is immediately updated at the first wrong move.',Collins,'' +p,5021,'To avoid this problem , we adopt cross-validation training as used in ',Collins,'aux_avoid_To dep_adopt_avoid det_problem_this dobj_avoid_problem nsubj_adopt_we amod_training_cross-validation dobj_adopt_training mark_used_as advcl_adopt_used prep_used_in ' +o,5022,'Recently , severalmethods have been proposed with similar motivation to ours ',Collins,'advmod_proposed_Recently amod__severalmethods nsubjpass_proposed_ aux_proposed_have auxpass_proposed_been prep_proposed_with amod_motivation_similar pobj_with_motivation prep_proposed_to pobj_to_ours ' +p,5023,'The default training set of Penn Treebank was used for the parser because the domain and style of those texts actually matches fairly well with the domain and style of the texts on which a reading level predictor for second language learners might be used ',Marcus,'det_set_The nn_set_default nn_set_training nsubjpass_used_set prep_set_of nn__Penn nn__Treebank pobj_of_ auxpass_used_was prep_used_for det_parser_the pobj_for_parser mark_matches_because det_domain_the nsubj_matches_domain cc_domain_and conj_domain_style prep_domain_of det_texts_those pobj_of_texts advmod_matches_actually advcl_used_matches advmod_well_fairly advmod_matches_well prep_matches_with det_domain_the pobj_with_domain cc_domain_and conj_domain_style prep_domain_of det_texts_the pobj_of_texts rel_used_on pobj_on_which det_predictor_a nn_predictor_reading nn_predictor_level nsubjpass_used_predictor prep_predictor_for amod_learners_second nn_learners_language pobj_for_learners aux_used_might auxpass_used_be rcmod_texts_used ' +o,5024,'It has been used for diverse problems such as machine translation and sense disambiguation ',Brown,'nsubjpass_used_It aux_used_has auxpass_used_been prep_used_for amod_problems_diverse pobj_for_problems dep_as_such prep_problems_as nn_translation_machine pobj_as_translation cc_translation_and nn__sense nn__disambiguation conj_translation_ ' +p,5025,'This source of overcounting is considered and fixed by and Zens and Ney -LRB- 2003 -RRB- , which we briefly review here ',Wu,'det_source_This nsubjpass_considered_source prep_source_of pobj_of_overcounting auxpass_considered_is cc_considered_and conj_considered_fixed prep_considered_by pobj_by_ cc__and conj__Zens cc__and conj__Ney appos_Ney_2003 dobj_review_which nsubj_review_we advmod_review_briefly rcmod_Ney_review advmod_review_here ' +o,5026,'In this study we have concentrated on the NPs ? ? term extraction , which comprises the focus of interest in several studies ',Dunning,'prep_concentrated_In det_study_this pobj_In_study nsubj_concentrated_we aux_concentrated_have prep_concentrated_on det_NPs_the pobj_on_NPs nn_extraction_term nsubj__extraction nsubj_comprises_which rcmod_extraction_comprises det_focus_the dobj_comprises_focus prep_focus_of pobj_of_interest prep_interest_in amod_studies_several pobj_in_studies dep_concentrated_ ' +o,5027,'An alternative training criterion therefore directly optimizes translation quality as measured by an automatic evaluation criterion ',Och,'det_criterion_An amod_criterion_alternative nn_criterion_training nsubj_optimizes_criterion advmod_criterion_therefore advmod_optimizes_directly nn_quality_translation dobj_optimizes_quality mark_measured_as advcl_optimizes_measured prep_measured_by det_criterion_an amod_criterion_automatic nn_criterion_evaluation pobj_by_criterion ' +o,5028,' proposed a simple feature augmentation method to achieve domain adaptation ','Daume III','advmod_proposed_ det_method_a amod_method_simple nn_method_feature nn_method_augmentation nsubj_proposed_method aux_achieve_to infmod_method_achieve nn_adaptation_domain dobj_achieve_adaptation ' +o,5029,'In open-domain opinion extraction , some approaches use syntactic features obtained from parsed input sentences , as is commonly done in semantic role labeling ',Kim,'prep_use_In amod_extraction_open-domain nn_extraction_opinion pobj_In_extraction det_approaches_some nsubj_use_approaches amod_features_syntactic nsubj__features partmod_features_obtained prep_obtained_from amod_sentences_parsed nn_sentences_input pobj_from_sentences xcomp_use_ mark_done_as auxpass_done_is advmod_done_commonly advcl_use_done prep_done_in amod_labeling_semantic nn_labeling_role pobj_in_labeling ' +o,5030,'We utilize a maximum entropy -LRB- ME -RRB- model to design the basic classifier used in active learning for WSD ',Berger,'nsubj_utilize_We det__a amod__maximum nn__entropy nn__ME nn__model dobj_utilize_ aux_design_to xcomp_utilize_design det_classifier_the amod_classifier_basic dobj_design_classifier partmod_classifier_used prep_used_in amod_learning_active pobj_in_learning prep_learning_for pobj_for_WSD ' +p,5031,'This further supports the claim by that loglikelihood ratio is much less sensitive than pmi to low counts ',Dunning,'nsubj_supports_This advmod_supports_further det_claim_the dobj_supports_claim prep_supports_by pcomp_by_ complm_sensitive_that amod_ratio_loglikelihood nsubj_sensitive_ratio cop_sensitive_is advmod_sensitive_much advmod_sensitive_less ccomp__sensitive prep_sensitive_than pobj_than_pmi prep_sensitive_to amod_counts_low pobj_to_counts ' +o,5032,'They can be used for discriminative training of reordering models ',Tillmann,'nsubjpass_used_They aux_used_can auxpass_used_be prep_used_for amod_training_discriminative pobj_for_training prep_training_of nn_models_reordering pobj_of_models ' +o,5033,'The metric we used is the kappa statistic , which factors out the agreement that is expected by chance : -RRB- -LRB- 1 -RRB- -LRB- -RRB- -LRB- EP EPAP = where P -LRB- A -RRB- is the observed agreement among the raters , and P -LRB- E -RRB- is the expected agreement , ie , the probability that the raters agree by chance ',Carletta,'det_metric_The nsubj_used_we dep_metric_used cop_kappa_is det_kappa_the ccomp_used_kappa nn__statistic dep_kappa_ dep_factors_which rcmod_kappa_factors dep_factors_out det_agreement_the pobj_out_agreement nsubjpass_expected_that auxpass_expected_is rcmod_agreement_expected prep_expected_by pobj_by_chance appos_kappa_1 nn_=_EP nn_=_EPAP pobj_-RRB-_= advmod_agreement_where nsubj_agreement_P appos_P_A cop_agreement_is det_agreement_the amod_agreement_observed advcl_kappa_agreement prep_agreement_among det_raters_the pobj_among_raters cc_used_and nsubj_agreement_P appos_P_E cop_agreement_is det_agreement_the amod_agreement_expected conj_used_agreement dep_probability_ie det_probability_the appos_agreement_probability complm_agree_that det_raters_the nsubj_agree_raters dep_probability_agree prep_agree_by pobj_by_chance ' +o,5034,'We trained a Chinese Treebank-style tokenizer and partof-speech tagger , both using a tagging model based on a perceptron learning algorithm ',Collins,'nsubj_trained_We det_tokenizer_a amod_tokenizer_Chinese amod_tokenizer_Treebank-style dobj_trained_tokenizer cc_tokenizer_and amod_tagger_partof-speech conj_tokenizer_tagger nsubj_using_both rcmod_tokenizer_using det_model_a amod_model_tagging dobj_using_model prep_using_based dep_based_on det_perceptron_a pobj_on_perceptron partmod_perceptron_learning dobj_learning_algorithm ' +o,5035,'We tuned our system on the development set devtest2006 for the EuroParl tasks and on nc-test2007 for CzechEnglish , using minimum error-rate training to optimise BLEU score ',Och,'nsubj_tuned_We poss_system_our dobj_tuned_system prep_tuned_on det_devtest2006_the nn_devtest2006_development nn_devtest2006_set pobj_on_devtest2006 prep_devtest2006_for det_tasks_the nn_tasks_EuroParl pobj_for_tasks cc_on_and conj_on_on pobj_on_nc-test2007 prep_tuned_for pobj_for_CzechEnglish xcomp_tuned_using amod__minimum amod__error-rate nn__training dobj_using_ aux_optimise_to xcomp_using_optimise amod_score_BLEU dobj_optimise_score ' +o,5036,'5 Related Work Evidence from the surrounding context has been used previously to determine if the current sentence should be subjective\\/objective -RRB- and adjacency pair information has been used to predict congressional votes ',Pang,'num_Evidence_5 amod_Evidence_Related nn_Evidence_Work nsubjpass_used_Evidence prep_Evidence_from det_context_the amod_context_surrounding pobj_from_context aux_used_has auxpass_used_been advmod_used_previously aux_determine_to xcomp_used_determine mark_and_if det_sentence_the amod_sentence_current nsubj_and_sentence aux_and_should cop_and_be nn_and_subjective\\/objective nn_and_ advcl_determine_and amod_information_adjacency nn_information_pair nsubjpass_used_information aux_used_has auxpass_used_been rcmod_and_used aux_predict_to xcomp_used_predict amod_votes_congressional dobj_predict_votes advmod_predict_ ' +o,5037,'41 Applications to phrase-based SMT Aphrase-basedtranslationmodelcanbeestimated in two stages : first a parallel corpus is aligned at the word-level and then phrase pairs are extracted ',Koehn,'num_Applications_41 nsubjpass_aligned_Applications prep_Applications_to amod_Aphrase-basedtranslationmodelcanbeestimated_phrase-based nn_Aphrase-basedtranslationmodelcanbeestimated_SMT pobj_to_Aphrase-basedtranslationmodelcanbeestimated prep_Aphrase-basedtranslationmodelcanbeestimated_in num_stages_two pobj_in_stages amod_Aphrase-basedtranslationmodelcanbeestimated_first det_corpus_a amod_corpus_parallel dep_Aphrase-basedtranslationmodelcanbeestimated_corpus auxpass_aligned_is prep_aligned_at det_word-level_the pobj_at_word-level cc_word-level_and conj_word-level_then nn_pairs_phrase nsubjpass_extracted_pairs auxpass_extracted_are dep_aligned_extracted ' +o,5038,'These blocks are used to compute the results in the fourth column : the BLEU score with a153 reference translation using a153 - grams along with 95 \% confidence interval is reported 4 ',Papineni,'det_blocks_These nsubjpass_used_blocks auxpass_used_are aux_compute_to xcomp_used_compute det_results_the dobj_compute_results prep_compute_in det_column_the amod_column_fourth pobj_in_column det__the amod__BLEU nn__score dep_column_ prep__with amod_translation_a153 nn_translation_reference pobj_with_translation partmod_translation_using dobj_using_a153 nsubjpass_reported_grams advmod_reported_along dep_along_with dep_\%_95 amod_interval_\% nn_interval_confidence pobj_with_interval auxpass_reported_is parataxis_used_reported dobj_reported_4 ' +o,5039,'22 Implementation of GIZA + + GIZA + + is an implementation of ML estimators for several statistical alignment models , including IBM Model 1 through 5 , HMM and Model 6 ',Brown,'num_Implementation_22 nsubj_implementation_Implementation prep_Implementation_of nn_+_GIZA pobj_of_+ cc_GIZA_+ amod_+_GIZA cc_GIZA_+ conj_GIZA_+ cop_implementation_is det_implementation_an prep_implementation_of nn_estimators_ML pobj_of_estimators prep_estimators_for amod_models_several amod_models_statistical nn_models_alignment pobj_for_models conj_models_including nn__IBM nn__Model num__1 dep__through num__5 pobj_including_ nn__HMM conj_models_ cc_models_and conj_models_Model num_Model_6 partmod_Model_ ' +n,5040,'Although this method is comparatively easy to be implemented , it just achieves the same performance as the synchronous binarization method for syntaxbased SMT systems ',Zhang,'mark_easy_Although det_method_this nsubj_easy_method cop_easy_is advmod_easy_comparatively advcl_achieves_easy aux_implemented_to auxpass_implemented_be xcomp_easy_implemented nsubj_achieves_it advmod_achieves_just det_performance_the amod_performance_same dobj_achieves_performance prep_achieves_as det__the amod__synchronous nn__binarization nn__method pobj_as_ prep__for amod_systems_syntaxbased nn_systems_SMT pobj_for_systems ' +p,5041,'Inversion transduction grammar , or ITG , is a wellstudied synchronous grammar formalism ',Wu,'nn__Inversion nn__transduction nn__grammar nsubj_formalism_ dep_ITG_or dep__ITG cop_formalism_is det_formalism_a amod_formalism_wellstudied dep_wellstudied_synchronous nn_formalism_grammar ' +p,5042,'For example , used cooccurrences between verbs and their subjects and objects , and proposed a similarity metric based on mutual information , but no exploration concerning the effectiveness of other kinds of word relationship is provided , although it is extendable to any kinds of contextual information ',Hindle,'prep_used_For pobj_For_example nsubj_used_ dobj_used_cooccurrences prep_cooccurrences_between pobj_between_verbs cc_verbs_and poss_subjects_their conj_verbs_subjects cc_subjects_and conj_subjects_objects cc_used_and conj_used_proposed det_metric_a nn_metric_similarity dobj_proposed_metric prep_proposed_based dep_based_on amod_information_mutual pobj_on_information cc_used_but det_exploration_no nsubjpass_provided_exploration prep_exploration_concerning det_effectiveness_the pobj_concerning_effectiveness prep_effectiveness_of amod_kinds_other pobj_of_kinds prep_kinds_of nn_relationship_word pobj_of_relationship auxpass_provided_is conj_used_provided mark_extendable_although nsubj_extendable_it cop_extendable_is advcl_provided_extendable prep_extendable_to det_kinds_any pobj_to_kinds prep_kinds_of amod_information_contextual pobj_of_information ' +o,5043,'An exception is the use of similarity for alleviating the sparse data problem in language modeling ',Brown,'det_exception_An nsubj_use_exception cop_use_is det_use_the prep_use_of pobj_of_similarity prep_similarity_for pcomp_for_alleviating det_problem_the amod_problem_sparse nn_problem_data dobj_alleviating_problem prep_alleviating_in nn_modeling_language pobj_in_modeling ' +o,5044,'The majority of these systems used models belonging to one of the twodominantapproachesindata-drivendependency parsinginrecentyears : In graph-based models , every possible dependency graph for a given input sentence is given a score that decomposes into scores for the arcs of the graph ',McDonald,'det_majority_The nsubj_used_majority prep_majority_of det_systems_these pobj_of_systems dobj_used_models partmod_models_belonging prep_belonging_to pobj_to_one prep_one_of det__the amod__twodominantapproachesindata-drivendependency nn__parsinginrecentyears pobj_of_ prep_given_In amod_models_graph-based pobj_In_models det_graph_every amod_graph_possible amod_graph_dependency nsubjpass_given_graph prep_graph_for det_sentence_a amod_sentence_given nn_sentence_input pobj_for_sentence auxpass_given_is parataxis_used_given det_score_a iobj_given_score det_decomposes_that dobj_given_decomposes prep_given_into pobj_into_scores prep_scores_for det_arcs_the pobj_for_arcs prep_arcs_of det_graph_the pobj_of_graph ' +o,5045,'579 The MaxEnt algorithm associates a set of weights -LRB- ij -RRB- i = 1nj = 1m with the features , which are estimated during the training phase to maximize the likelihood of the data ',Berger,'det_algorithm_The nn_algorithm_MaxEnt nsubj_associates_algorithm dep_579_associates det_set_a dobj_associates_set prep_set_of pobj_of_weights appos_set_ij nn_1m_i nn_1m_= nn_1m_1nj nn_1m_= dep_set_1m prep_1m_with det_features_the pobj_with_features nsubjpass_estimated_which auxpass_estimated_are rcmod_features_estimated prep_estimated_during det_phase_the nn_phase_training pobj_during_phase aux_maximize_to xcomp_estimated_maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_of det_data_the pobj_of_data ' +o,5046,'We have also applied our more general unification grammar acquisition methodology to the TIGER Treebank and Penn Chinese Treebank , extracting wide-coverage , probabilistic LFG grammar 361 Computational Linguistics Volume 31 , Number 3 approximations and lexical resources for German and Chinese ',Cahill,'nsubj_applied_We aux_applied_have advmod_applied_also poss_methodology_our advmod_general_more amod_methodology_general nn_methodology_unification nn_methodology_grammar nn_methodology_acquisition dobj_applied_methodology prep_methodology_to det__the nn__TIGER nn__Treebank pobj_to_ cc__and nn__Penn nn__Chinese nn__Treebank conj__ xcomp_applied_extracting dobj_extracting_wide-coverage amod_Volume_probabilistic nn_Volume_LFG nn_Volume_grammar num_Volume_361 nn_Volume_Computational nn_Volume_Linguistics conj_wide-coverage_Volume num_Volume_31 nn_approximations_Number num_approximations_3 conj_wide-coverage_approximations cc_wide-coverage_and amod_resources_lexical conj_wide-coverage_resources prep_extracting_for amod__German pobj_for_ cc__and amod__Chinese conj__ ' +o,5047,'The last issue is how our binarization performs on a lexicalized parser , like ',Collins,'det_issue_The amod_issue_last nsubj_is_issue advmod_performs_how poss_binarization_our nsubj_performs_binarization ccomp_is_performs prep_performs_on det_parser_a amod_parser_lexicalized pobj_on_parser prep_performs_like ' +o,5048,'33 Grid Line Search Our implementation of a grid search is a modified version of that proposed in ',Och,'num_Search_33 nn_Search_Grid nn_Search_Line poss_implementation_Our nsubj_version_implementation prep_implementation_of det_search_a nn_search_grid pobj_of_search cop_version_is det_version_a amod_version_modified dep_Search_version prep_version_of pobj_of_that partmod_that_proposed prep_proposed_in ' +o,5049,'Toremedythis situation , we can borrow the probabilistic model of PHARAOH , and define the parsing model as : Pr -LRB- d e -LRB- d -RRB- -RRB- = productdisplay dd w -LRB- r -LRB- d -RRB- -RRB- -LRB- 4 -RRB- which is the product of the weights of the rules used in a derivation d The rule weight , w -LRB- X , -RRB- , is in turn defined as : P -LRB- -RRB- 1P -LRB- -RRB- 2Pw -LRB- -RRB- 3Pw -LRB- -RRB- 4 exp -LRB- -RRB- 5 where P -LRB- -RRB- and P -LRB- -RRB- are the relative frequencies of and , and Pw -LRB- -RRB- and Pw -LRB- -RRB- are 176 the lexical weights ',Koehn,'prep_borrow_Toremedythis pobj_Toremedythis_situation nsubj_borrow_we aux_borrow_can det_model_the amod_model_probabilistic dobj_borrow_model prep_model_of pobj_of_PHARAOH cc_borrow_and conj_borrow_define det_model_the amod_model_parsing dobj_define_model prep_define_as nsubj_is_Pr nn_e_d dep_Pr_e appos_e_d rcmod_e_= amod_w_productdisplay nn_w_dd dobj_=_w dep_w_r appos_w_d appos_w_4 nsubj_product_which cop_product_is det_product_the rcmod_w_product prep_product_of det_weights_the pobj_of_weights prep_weights_of det_rules_the pobj_of_rules partmod_rules_used prep_used_in det_derivation_a pobj_in_derivation prep_e_d det_weight_The nn_weight_rule pobj_d_weight nn_X_w appos_weight_X parataxis_borrow_is prep_is_in pobj_in_turn partmod_turn_defined prep_defined_as nsubj_2Pw_P number_-RRB-_1P parataxis_borrow_2Pw pobj_-RRB-_3Pw number_-RRB-_4 number_-RRB-_exp dobj_2Pw_5 advmod_frequencies_where nsubj_frequencies_P cc_-RRB-_and conj_-RRB-_P cop_frequencies_are det_frequencies_the amod_frequencies_relative rcmod_5_frequencies prep_frequencies_of pobj_of_and cc_borrow_and nsubj_176_Pw cc_-RRB-_and conj_-RRB-_Pw aux_176_are conj_borrow_176 det_weights_the amod_weights_lexical dobj_176_weights ' +o,5050,'However , their decoder is outperformed by phrase-based decoders such as , , and ',Koehn,'advmod_outperformed_However poss_decoder_their nsubjpass_outperformed_decoder auxpass_outperformed_is prep_outperformed_by amod_decoders_phrase-based pobj_by_decoders dep_as_such prep_decoders_as nn__ conj__ cc__and pobj_as_ ' +o,5051,'Many machine learning techniques have been developed to tackle such random process tasks , which include Hidden Markov Models -LRB- HMMs -RRB- , Maximum Entropy Models -LRB- MEs -RRB- , Support Vector Machines -LRB- SVMs -RRB- , etc Among them , SVMs have high memory capacity and show high performance , especially when the target classification requires the consideration of various features ',Ratnaparkhi,'amod_techniques_Many nn_techniques_machine nn_techniques_learning nsubjpass_developed_techniques aux_developed_have auxpass_developed_been aux_tackle_to xcomp_developed_tackle amod_tasks_such amod_tasks_random nn_tasks_process dobj_tackle_tasks nsubj_include_which rcmod_tasks_include nn__Hidden nn__Markov nn__Models abbrev__HMMs dobj_include_ nn__Maximum nn__Entropy nn__Models abbrev__MEs appos__ nn__Support nn__Vector nn__Machines abbrev__SVMs appos__ appos__etc prep_have_Among pobj_Among_them nsubj_have_SVMs rcmod_etc_have amod_capacity_high nn_capacity_memory dobj_have_capacity cc_capacity_and nn_performance_show amod_performance_high conj_capacity_performance advmod_when_especially advmod_requires_when det_classification_the nn_classification_target nsubj_requires_classification dep__requires det_consideration_the dobj_requires_consideration prep_consideration_of amod_features_various pobj_of_features ' +o,5052,'Strube and , and for Coreference Resolution ',Ponzetto,'cc_Strube_and conj_Strube_ cc_Strube_and conj_Strube_for nn__Coreference nn__Resolution pobj_for_ ' +o,5053,'Finally , we would like to investigate the incorporation of unsupervised methods for WSD , such as the heuristically-based methods of and , and the theoretically purer bootstrapping method of ',Yarowsky,'advmod_like_Finally nsubj_like_we aux_like_would aux_investigate_to xcomp_like_investigate det_incorporation_the dobj_investigate_incorporation prep_incorporation_of amod_methods_unsupervised pobj_of_methods prep_methods_for pobj_for_WSD dep_as_such prep_methods_as det_methods_the amod_methods_heuristically-based pobj_as_methods prep_methods_of pobj_of_ cc__and conj__ cc_methods_and det_method_the amod_method_theoretically nn_method_purer amod_method_bootstrapping conj_methods_method prep_method_of ' +p,5054,'Studies on the supervised task have shown that straightforward baselines -LRB- eg models based on source only , target only , or the union of the data -RRB- achieve a relatively high performance level and are surprisingly difficult to beat ','Daume III','nsubj_shown_Studies prep_Studies_on det_task_the amod_task_supervised pobj_on_task aux_shown_have complm_achieve_that amod_baselines_straightforward nsubj_achieve_baselines nn_models_eg dep_baselines_models partmod_models_based prep_based_on pobj_on_source advmod_source_only conj_source_target advmod_target_only cc_source_or det_union_the conj_source_union prep_union_of det_data_the pobj_of_data ccomp_shown_achieve det_level_a advmod_high_relatively amod_level_high nn_level_performance dobj_achieve_level cc_achieve_and cop_difficult_are advmod_difficult_surprisingly conj_achieve_difficult aux_beat_to xcomp_difficult_beat ' +o,5055,'We use a program to label syntactic arguments with the roles they are playing , and the rules for complement\\/adjunct distinction given by to never allow deletion of the complement ',Collins,'nsubj_use_We det_program_a nsubj_label_program aux_label_to xcomp_use_label amod_arguments_syntactic dobj_label_arguments prep_label_with det_roles_the pobj_with_roles nsubj_playing_they aux_playing_are dep_use_playing dobj_playing_ cc__and det_rules_the conj__rules prep_rules_for amod_distinction_complement\\/adjunct pobj_for_distinction prep_playing_given dep_given_by pcomp_by_ aux_allow_to neg_allow_never xcomp__allow dobj_allow_deletion prep_deletion_of det_complement_the pobj_of_complement ' +o,5056,'To quickly -LRB- and approximately -RRB- evaluate this phenomenon , we trained the statistical IBM wordalignment model 4 ,1 using the GIZA + + software for the following language pairs : ChineseEnglish , Italian English , and DutchEnglish , using the IWSLT-2006 corpus for the first two language pairs , and the Europarl corpus for the last one ',Koehn,'aux_evaluate_To advmod_evaluate_quickly dep_-LRB-_and dep_-LRB-_approximately dep_trained_evaluate det_phenomenon_this dobj_evaluate_phenomenon nsubj_trained_we det_model_the amod_model_statistical nn_model_IBM nn_model_wordalignment iobj_trained_model number__4 dobj_trained_ dobj_trained_,1 xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__software dobj_using_ prep__for det_pairs_the amod_pairs_following nn_pairs_language pobj_for_pairs nn_English_ChineseEnglish amod_English_Italian dep__English cc_English_and conj_English_DutchEnglish partmod_English_using det__the nn__IWSLT-2006 nn__corpus dobj_using_ prep__for det_pairs_the amod_pairs_first num_pairs_two nn_pairs_language pobj_for_pairs cc__and det__the nn__Europarl nn__corpus conj__ prep__for det_one_the amod_one_last pobj_for_one ' +n,5057,'Among the applications of collocational analysis for lexical acquisition are: the derivation of syntactic disambiguation cues (Basili et al. 1991, 1993a; Hindle and Rooths 1991,1993; Sekine 1992) (Bogges et al. 1992), sense preference (Yarowski 1992), acquisition of selectional restrictions (Basili et al. 1992b, 1993b; Utsuro et al. 1993), lexical preference in generation (Smadjia 1991), word clustering (Pereira 1993; Hindle 1990; Basili et al. 1993c), etc. In the majority of these papers, even though the (precedent or subsequent) statistical processing reduces the number of accidental associations, very large corpora (10,000,000 words) are necessary to obtain reliable data on a \'large enough\' number of words.',Hindle,'' +o,5058,'However , in yet unpublished work we found that at least for the computation of synonyms and related words neither syntactical analysis nor singular value decomposition lead to significantly better results than the approach described here when applied to the monolingual case , so we did not try to include these methods in our system ',Dunning,'advmod_try_However prep_try_in advmod_work_yet amod_work_unpublished pobj_in_work nsubj_found_we rcmod_work_found complm_lead_that dep_for_at dep_at_least prep_lead_for det_computation_the pobj_for_computation prep_computation_of amod_words_synonyms cc_synonyms_and conj_synonyms_related pobj_of_words preconj_analysis_neither amod_analysis_syntactical nsubj_lead_analysis cc_analysis_nor nn_decomposition_singular nn_decomposition_value conj_analysis_decomposition ccomp_found_lead prep_lead_to advmod_better_significantly amod_results_better pobj_to_results mark_described_than det_approach_the nsubj_described_approach advcl_lead_described advmod_described_here advmod_applied_when advcl_described_applied prep_applied_to det__the amod__monolingual nn__case pobj_to_ advmod_we_so nsubj_try_we aux_try_did neg_try_not aux_include_to xcomp_try_include det_methods_these dobj_include_methods prep_include_in poss_system_our pobj_in_system ' +o,5059,'Similar adaptations of the Matrix-Tree Theorem have been developed independently and simultaneouslyby andMcDonaldand Satta -LRB- 2007 -RRB- ; see Section 5 for more discussion ',Smith,'amod_adaptations_Similar nsubjpass_developed_adaptations prep_adaptations_of det_Theorem_the nn_Theorem_Matrix-Tree pobj_of_Theorem aux_developed_have auxpass_developed_been advmod_developed_independently cc_developed_and advmod_andMcDonaldand_simultaneouslyby advmod_andMcDonaldand_ conj_developed_andMcDonaldand dobj_andMcDonaldand_Satta appos_Satta_2007 rcmod_Satta_see dobj_see_Section num_Section_5 prep_Section_for amod_discussion_more pobj_for_discussion ' +o,5060,' reports 88 \% labeled precision and recall on individual parse constituents on data from the Penn Treebank , roughly consistent with our finding of at least 13 \% error ',Collins,'amod_reports_ nsubj_\%_reports num_\%_88 partmod_\%_labeled dobj_labeled_precision cc_precision_and conj_precision_recall prep_precision_on amod_constituents_individual nn_constituents_parse pobj_on_constituents prep_labeled_on pobj_on_data prep_labeled_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank advmod_consistent_roughly amod_error_consistent prep_consistent_with poss_finding_our pobj_with_finding prep_finding_of quantmod_13_at dep_at_least num_\%_13 pobj_of_\% dobj_\%_error ' +o,5061,'The agreement was statistically significant -LRB- Kappa = 0650 -RRB- 001 for Japanese and Kappa = 0748,0 -RRB- 001 for English -RRB- ',Carletta,'det_agreement_The nsubj_significant_agreement cop_significant_was advmod_significant_statistically nsubj_0650_Kappa dep_0650_= dep_significant_0650 dep_significant_001 prep_001_for pobj_for_Japanese cc_001_and conj_001_Kappa dep_0748,0_= rcmod_Kappa_0748,0 dep_significant_001 prep_001_for nn__English pobj_for_ ' +o,5062,'In the following experiments , the NIST BLEU score is used as the evaluation metric , which is reported as a percentage in the following sections ',Papineni,'prep_used_In det_experiments_the amod_experiments_following pobj_In_experiments det_score_the nn_score_NIST nn_score_BLEU nsubjpass_used_score auxpass_used_is prep_used_as det__the nn__evaluation amod__metric pobj_as_ nsubjpass_reported_which auxpass_reported_is rcmod__reported prep_reported_as det_percentage_a pobj_as_percentage prep_percentage_in det_sections_the amod_sections_following pobj_in_sections ' +o,5063,'The majority of this research was done on extending the tree structure -LRB- finding new synsets or enriching WN with new relationships -RRB- rather than improving the quality of existing concept\\/synset nodes ',Snow,'det_majority_The nsubjpass_done_majority prep_majority_of det_research_this pobj_of_research auxpass_done_was prep_done_on pcomp_on_extending det_structure_the nn_structure_tree dobj_extending_structure parataxis_extending_finding amod_synsets_new dobj_finding_synsets advmod_finding_ cc_finding_or conj_finding_enriching dobj_enriching_WN prep_WN_with amod__new nn__relationships pobj_with_ cc_extending_rather dep_rather_than conj_extending_improving det_quality_the dobj_improving_quality prep_quality_of amod_nodes_existing nn_nodes_concept\\/synset pobj_of_nodes ' +o,5064,' 9417 Li and Roth 9302 9464 Table 2 : Baseline results on three shallow parsing tasks : the NP-Chunking task ; the CoNLL-2000 Chunking task ; and the Li & Roth task , which is the same as CoNLL-2000 but with more training data and a different test section ',Ramshaw,'dobj__9417 nsubj__Li cc_Li_and conj_Li_Roth partmod_Li_ number_9464_9302 num_Table_9464 dobj__Table num_Table_2 amod_results_Baseline dep_Li_results prep_results_on num_tasks_three nn_tasks_shallow nn_tasks_parsing pobj_on_tasks det__the amod__NP-Chunking nn__task dep_Li_ det__the amod__CoNLL-2000 nn__Chunking nn__task dep__ cc_Li_and det_Li_the conj_Li_Li cc_Li_& nn__Roth nn__task conj_Li_ nsubj_section_which cop_section_is det_section_the amod_section_same advmod_CoNLL-2000_as amod_section_CoNLL-2000 cc_CoNLL-2000_but conj_CoNLL-2000_with amod_data_more nn_data_training pobj_with_data cc_data_and det_test_a amod_test_different conj_data_test rcmod_Li_section ' +o,5065,'In computational linguistics , our pattern discovery procedure extends over previous approaches that use surface patterns as indicators of semantic relations between nouns or verbs -LRB- inter alia -RRB- ',Davidov,'prep_extends_In amod_linguistics_computational pobj_In_linguistics poss_procedure_our nn_procedure_pattern nn_procedure_discovery nsubj_extends_procedure prep_extends_over amod_approaches_previous pobj_over_approaches nsubj_use_that rcmod_approaches_use nn_patterns_surface dobj_use_patterns prep_use_as pobj_as_indicators prep_indicators_of amod_relations_semantic pobj_of_relations prep_relations_between pobj_between_nouns cc_nouns_or conj_nouns_verbs num_alia_ nn_alia_inter appos_approaches_alia ' +o,5066,'Our intuition comes from an observation by regarding multiple tokens of words in documents ',Yarowsky,'poss_intuition_Our nsubj_comes_intuition prep_comes_from det_observation_an pobj_from_observation prep_comes_by pobj_by_ prep__regarding amod_tokens_multiple pobj_regarding_tokens prep_tokens_of pobj_of_words prep_words_in pobj_in_documents ' +o,5067,'In such cases , additional information may be coded into the HMM model to achieve higher accuracy ',Cutting,'prep_coded_In amod_cases_such pobj_In_cases amod_information_additional nsubjpass_coded_information aux_coded_may auxpass_coded_be prep_coded_into det_model_the nn_model_HMM pobj_into_model aux_achieve_to purpcl_coded_achieve amod_accuracy_higher dobj_achieve_accuracy ' +o,5068,'22 Table 5 : Comparison with previous best results : -LRB- Top : POS tagging , Bottom : Text Chunking -RRB- POS tagging F = 1 Perceptron 9711 Dep ',Collins,'num_Table_22 num_Table_5 nsubj_POS_Comparison prep_Comparison_with amod_results_previous amod_results_best pobj_with_results dep_-LRB-_Top dep_tagging_POS dep_-LRB-_tagging appos_tagging_Bottom nn_Chunking_Text dep_tagging_Chunking dep_Table_POS xcomp_POS_tagging nsubj_Dep_F dep_Dep_= num_Dep_1 nn_Dep_Perceptron nn_Dep_ num_Dep_9711 ccomp_tagging_Dep ' +p,5069,'Results from show that under these definitions the following guarantee holds : LogLossUpda , k , BestWtk , a C20 BestLossk , a So it can be seen that the update from a to Upda , k , BestWtk , a is guaranteed to decrease LogLoss by at least W k q C0 W C0 k qC16C17 2 From these results , the algorithms in Figures 3 and 4 could be altered to take the revised definitions of W k and W C0 k into account ',Collins,'nsubj_show_Results prep_Results_from pobj_from_ complm_holds_that prep_holds_under det_definitions_these pobj_under_definitions det_guarantee_the amod_guarantee_following nsubj_holds_guarantee ccomp_show_holds nsubjpass_seen_LogLossUpda dep_seen_k nsubjpass_seen_BestWtk det_BestLossk_a nn_BestLossk_C20 nsubjpass_seen_BestLossk dep_So_a advmod_seen_So nsubjpass_seen_it aux_seen_can auxpass_seen_be ccomp_holds_seen complm_guaranteed_that det_update_the nsubjpass_guaranteed_update prep_update_from pobj_from_a prep_a_to nn_BestWtk_Upda dep_BestWtk_k pobj_to_BestWtk dep_BestWtk_a auxpass_guaranteed_is ccomp_seen_guaranteed aux_decrease_to xcomp_guaranteed_decrease dobj_decrease_LogLoss prep_decrease_by dep_k_at dep_at_least nn_k_W nn_k_k nn_k_q nn_k_C0 nn_k_W nn_k_C0 dep_qC16C17_k dep_altered_qC16C17 num_qC16C17_2 prep_altered_From det_results_these pobj_From_results det_algorithms_the nsubjpass_altered_algorithms prep_algorithms_in pobj_in_Figures num_Figures_3 cc_Figures_and conj_Figures_4 aux_altered_could auxpass_altered_be pcomp_by_altered aux_take_to purpcl_altered_take det_definitions_the amod_definitions_revised dobj_take_definitions prep_definitions_of nn_k_W pobj_of_k cc_altered_and nn_C0_W dep_k_C0 conj_altered_k prep_k_into pobj_into_account ' +o,5070,'A totally different approach to improving the accuracy of our parser is to use the idea of selftraining described in ',McClosky,'det_approach_A advmod_different_totally amod_approach_different nsubj_is_approach prep_approach_to pcomp_to_improving det_accuracy_the dobj_improving_accuracy prep_accuracy_of poss_parser_our pobj_of_parser aux_use_to xcomp_is_use det_idea_the dobj_use_idea prep_idea_of pobj_of_selftraining partmod_selftraining_described prep_described_in ' +o,5071,'For instance , the most relaxed IBM Model-1 , which assumes that any source word can be generated by any target word equally regardless of distance , can be improved by demanding a Markov process of alignments as in HMM-based models , or implementing a distribution of number of target words linked to a source word as in IBM fertility-based models ',Brown,'prep_improved_For pobj_For_instance det_Model-1_the amod_Model-1_most amod_Model-1_relaxed nn_Model-1_IBM nsubjpass_improved_Model-1 nsubj_assumes_which rcmod_Model-1_assumes complm_generated_that det_word_any nn_word_source nsubjpass_generated_word aux_generated_can auxpass_generated_be ccomp_assumes_generated prep_generated_by det_word_any nn_word_target pobj_by_word advmod_regardless_equally dep_of_regardless prep_word_of pobj_of_distance aux_improved_can auxpass_improved_be prep_improved_by pcomp_by_demanding det_process_a nn_process_Markov dobj_demanding_process prep_process_of pobj_of_alignments dep_in_as prep_alignments_in amod__HMM-based nn__models pobj_in_ cc_demanding_or conj_demanding_implementing det_distribution_a dobj_implementing_distribution prep_distribution_of pobj_of_number prep_number_of nn_words_target pobj_of_words partmod_words_linked prep_linked_to det_word_a nn_word_source pobj_to_word prep_linked_as dep_as_in nn_models_IBM amod_models_fertility-based pobj_in_models ' +o,5072,'Metrics in the Rouge family allow for skip n-grams ; Kauchak and Barzilay take paraphrasing into account ; metrics such as METEOR and GTM calculate both recall and precision ; METEOR is also similar to SIA in that word class information is used ',Banerjee,'nsubj_allow_Metrics prep_Metrics_in det_family_the nn_family_Rouge pobj_in_family prep_allow_for amod__skip amod__n-grams pobj_for_ nn__Kauchak cc_Kauchak_and conj_Kauchak_Barzilay nsubj_take_ parataxis_allow_take xcomp_take_paraphrasing prep_paraphrasing_into pobj_into_account nsubj_calculate_metrics dep_as_such prep_metrics_as nn__METEOR pobj_as_ cc__and nn__GTM conj__ parataxis_take_calculate preconj_recall_both dobj_calculate_recall cc_recall_and conj_recall_precision nsubj_similar_METEOR cop_similar_is advmod_similar_also parataxis_take_similar prep_similar_to nn__SIA pobj_to_ rel_used_in pobj_in_that nn_information_word nn_information_class nsubjpass_used_information auxpass_used_is rcmod__used ' +p,5073,'The notion of incrementally merging classes of lexical items is intuitively satisfying and is explored in detail in ',Brown,'det_notion_The nsubjpass_satisfying_notion prep_notion_of advmod_merging_incrementally pcomp_of_merging dobj_merging_classes prep_classes_of amod_items_lexical pobj_of_items cop_satisfying_is advmod_satisfying_intuitively cc_satisfying_and auxpass_explored_is conj_satisfying_explored prep_explored_in pobj_in_detail prep_explored_in ' +o,5074,'Lexical cues of differing complexities have been used , including single words and Ngrams -LRB- eg , -RRB- , as well as phrases and lexico-syntactic patterns -LRB- eg , -RRB- ',Turney,'amod_cues_Lexical nsubjpass_used_cues prep_cues_of amod_complexities_differing pobj_of_complexities aux_used_have auxpass_used_been prep_used_including amod_words_single pobj_including_words cc_words_and conj_words_Ngrams appos_words_eg dep_eg_ dep_well_as cc_words_well dep_well_as conj_words_phrases cc_phrases_and amod_patterns_lexico-syntactic conj_phrases_patterns appos_patterns_eg dep_eg_ ' +n,5075,'Due to limited variations in the N-Best list , the nature of ranking , and more importantly , the non-differentiable objective functions used for MT -LRB- such as BLEU -RRB- , one often found only local optimal solutions to , with no clue to walk out of the riddles ',Papineni,'dep_to_Due prep_found_to amod_variations_limited pobj_to_variations prep_variations_in det_list_the amod_list_N-Best pobj_in_list det_nature_the conj_found_nature prep_nature_of pobj_of_ranking cc_found_and advmod_importantly_more advmod_found_importantly det_functions_the amod_functions_non-differentiable amod_functions_objective conj_found_functions partmod_functions_used prep_used_for pobj_for_MT dep_as_such dep_MT_as nn__BLEU pobj_as_ dep_often_one advmod_MT_often amod_solutions_only amod_solutions_local amod_solutions_optimal dobj_found_solutions prep_found_to prep_found_with det_clue_no pobj_with_clue aux_walk_to infmod_clue_walk prt_walk_out prep_walk_of det_riddles_the pobj_of_riddles ' +o,5076,'The simple idea that words in a source chunk are typically aligned to words in a single possible target chunk is used to discard alignments which link words from 2We use IBM-1 to IBM-5 models implemented with GIZA + + ',Brown,'det_idea_The amod_idea_simple nsubjpass_used_idea complm_aligned_that nsubjpass_aligned_words prep_words_in det_chunk_a nn_chunk_source pobj_in_chunk auxpass_aligned_are advmod_aligned_typically dep_idea_aligned prep_aligned_to pobj_to_words prep_words_in det_chunk_a amod_chunk_single amod_chunk_possible nn_chunk_target pobj_in_chunk auxpass_used_is aux_discard_to xcomp_used_discard dobj_discard_alignments nsubj_link_which rcmod_alignments_link dobj_link_words prep_words_from amod_IBM-1_2We nn_IBM-1_use pobj_from_IBM-1 prep_link_to nn_models_IBM-5 pobj_to_models advmod_implemented_ partmod_models_implemented prep_implemented_with pobj_with_GIZA amod_GIZA_+ cc_+_+ conj_+_ ' +o,5077,'There is also work on grouping senses of other inventories using information in the inventory along with information retrieval techniques ',Dolan,'expl_is_There advmod_is_also nsubj_is_work prep_work_on amod_senses_grouping pobj_on_senses prep_senses_of amod_inventories_other pobj_of_inventories partmod_inventories_using dobj_using_information prep_using_in det_inventory_the pobj_in_inventory partmod_inventory_ prt__along prep__with nn_techniques_information nn_techniques_retrieval pobj_with_techniques advmod__ ' +o,5078,'In our experiments , we used a dependency parser only in English -LRB- a version of the Collins parser that has been adapted for building dependencies -RRB- but not in the other language ',Collins,'prep_used_In poss_experiments_our pobj_In_experiments nsubj_used_we det_parser_a amod_parser_dependency dobj_used_parser dep_in_only prep_used_in pobj_in_English det_version_a dep_English_version prep_version_of det__the nn__Collins nn__parser pobj_of_ nsubjpass_adapted_that aux_adapted_has auxpass_adapted_been rcmod__adapted prep_adapted_for nn_dependencies_building pobj_for_dependencies dep_not_but cc_in_not conj_in_in det_language_the amod_language_other pobj_in_language ' +p,5079,'61 Interand Intra-annotator agreement We measured pairwise agreement among annotators usingthekappacoefficient -LRB- K -RRB- whichiswidelyused in computational linguistics for measuring agreement in category judgments ',Carletta,'num_agreement_61 nn_agreement_Interand nn_agreement_Intra-annotator nsubj_measured_We dep_agreement_measured amod_agreement_pairwise dobj_measured_agreement prep_measured_among amod_usingthekappacoefficient_annotators pobj_among_usingthekappacoefficient appos_usingthekappacoefficient_K partmod_usingthekappacoefficient_whichiswidelyused prep_whichiswidelyused_in amod_linguistics_computational pobj_in_linguistics prep_whichiswidelyused_for pcomp_for_measuring dobj_measuring_agreement prep_measuring_in nn_judgments_category pobj_in_judgments ' +o,5080,'Automatic Evaluation Measures A variety of automatic evaluation methods have been recently proposed in the machine translation community ',Papineni,'nn_Measures_Automatic nn_Measures_Evaluation det_variety_A nsubjpass_proposed_variety prep_variety_of amod_methods_automatic nn_methods_evaluation pobj_of_methods aux_proposed_have auxpass_proposed_been advmod_proposed_recently dep_Measures_proposed prep_proposed_in det_community_the nn_community_machine nn_community_translation pobj_in_community ' +o,5081,'According to current tagger comparisons , and according to a comparsion of the results presented here with those in , the Maximum Entropy framework seems to be the only other approach yielding comparable results to the one presented here ',Ratnaparkhi,'prep_seems_According dep_According_to amod_comparisons_current nn_comparisons_tagger pobj_to_comparisons amod_comparisons_ cc__and conj__according dep_according_to det_comparsion_a pobj_to_comparsion prep_comparsion_of det_results_the pobj_of_results partmod_comparisons_presented advmod_presented_here prep_presented_with pobj_with_those prep_those_in pobj_in_ det_framework_the nn_framework_Maximum nn_framework_Entropy nsubj_seems_framework aux_approach_to cop_approach_be det_approach_the advmod_approach_only amod_approach_other xcomp_seems_approach partmod_approach_yielding amod_results_comparable dobj_yielding_results prep_yielding_to det_one_the pobj_to_one partmod_one_presented advmod_presented_here ' +p,5082,'In the supervised setting , a recent paper by shows that a simple feature augmentation method for SVM is able to effectively use both labeled target and source data to provide the best domainadaptation results in a number of NLP tasks ','Daume III','prep_shows_In det_setting_the amod_setting_supervised pobj_In_setting det_paper_a amod_paper_recent nsubj_shows_paper prep_paper_by pobj_by_ complm_able_that det_method_a amod_method_simple nn_method_feature nn_method_augmentation nsubj_able_method prep_method_for pobj_for_SVM cop_able_is ccomp_shows_able aux_use_to advmod_use_effectively xcomp_able_use dobj_use_both partmod_both_labeled nn_data_target cc_target_and conj_target_source dobj_labeled_data aux_provide_to xcomp_labeled_provide det_results_the amod_results_best nn_results_domainadaptation dobj_provide_results prep_results_in det_number_a pobj_in_number prep_number_of nn_tasks_NLP pobj_of_tasks ' +o,5083,'One is to find unknown words from corpora and put them into a dictionary -LRB- eg , -RRB- , and the other is to estimate a model that can identify unknown words correctly -LRB- eg , -RRB- ',Berger,'nsubj_is_One aux_find_to xcomp_is_find amod_words_unknown dobj_find_words prep_find_from pobj_from_corpora cc_find_and conj_find_put dobj_put_them prep_put_into det_dictionary_a pobj_into_dictionary appos_dictionary_eg dep_eg_ cc_is_and det_other_the nsubj_is_other conj_is_is aux_estimate_to xcomp_is_estimate det_model_a dobj_estimate_model nsubj_identify_that aux_identify_can rcmod_model_identify amod_words_unknown dobj_identify_words advmod_identify_correctly appos_model_eg dep_eg_ ' +o,5084,'Early work employed a diverse range of features in a linear classifier -LRB- commonly referred to as feature-based approaches -RRB- , including lexical features , syntactic parse features , dependency features and semantic features ',Jiang,'advmod_work_Early nsubj_employed_work det_range_a amod_range_diverse dobj_employed_range prep_range_of pobj_of_features prep_features_in det_classifier_a amod_classifier_linear pobj_in_classifier advmod_referred_commonly dep_features_referred prep_referred_to advmod_feature-based_as amod_approaches_feature-based pobj_to_approaches prep_employed_including amod_features_lexical pobj_including_features amod_features_syntactic nn_features_parse conj_features_features amod_features_dependency conj_features_features cc_features_and amod_features_semantic conj_features_features ' +o,5085,'Since an existing study incorporates these relations ad hoc , they are apparently crucial in accurate disambiguation ',Collins,'mark_incorporates_Since det_study_an amod_study_existing nsubj_incorporates_study advcl_crucial_incorporates det_relations_these nsubj__relations amod_relations_ad dep_ad_hoc xcomp_incorporates_ nsubj_crucial_they cop_crucial_are advmod_crucial_apparently prep_crucial_in amod_disambiguation_accurate pobj_in_disambiguation ' +o,5086,'Collocation : Collocations were extracted from a seven million word sample of the Longman English Language Corpus using the association ratio and outputted to a lexicon ',Church,'dep_extracted_Collocation nsubjpass_extracted_Collocations auxpass_extracted_were prep_extracted_from det_sample_a number_million_seven num_sample_million nn_sample_word pobj_from_sample prep_sample_of det_Corpus_the nn_Corpus_Longman nn_Corpus_English nn_Corpus_Language pobj_of_Corpus xcomp_extracted_using det__the nn__association nn__ratio dobj_using_ cc_extracted_and conj_extracted_outputted prep_outputted_to det_lexicon_a pobj_to_lexicon ' +o,5087,'The underlying translation model is Model 2 from ',Brown,'det_model_The amod_model_underlying nn_model_translation nsubj_Model_model aux_Model_is dobj_Model_2 prep_Model_from ' +o,5088,'Agglomerative clustering -LRB- eg , -RRB- can produce hierarchical word categories from an unannotated corpus ',Brown,'amod_clustering_Agglomerative nsubj_produce_clustering appos_clustering_eg dep_eg_ aux_produce_can amod_categories_hierarchical nn_categories_word dobj_produce_categories prep_produce_from det_corpus_an amod_corpus_unannotated pobj_from_corpus ' +o,5089,'Much research has been carried out recently in this area ',Brown,'amod_research_Much nsubjpass_carried_research aux_carried_has auxpass_carried_been prt_carried_out advmod_carried_recently prep_carried_in det__this nn__area pobj_in_ ' +o,5090,'A superset of the parallel data was word aligned by GIZA union and EMD ',Fraser,'det_superset_A nsubj_word_superset prep_superset_of det_data_the amod_data_parallel pobj_of_data cop_word_was partmod_word_aligned prep_aligned_by nn__GIZA nn__union pobj_by_ cc__and conj__EMD ' +o,5091,'1 Empty categories however seem different , in that , for the most part , their location and existence is determined , not by observable data , but by explicitly constructed linguistic principles , which 1 Both and Higgins -LRB- 2003 : 100 -RRB- are explicit about this predisposition ',Collins,'num_categories_1 amod_categories_Empty nsubj_different_categories advmod_different_however cop_different_seem rel_determined_in pobj_in_that prep_determined_for det_part_the amod_part_most pobj_for_part poss_location_their nsubjpass_determined_location cc_location_and conj_location_existence auxpass_determined_is rcmod_different_determined dep_by_not prep_determined_by amod_data_observable pobj_by_data cc_by_but conj_by_by advmod_constructed_explicitly amod_principles_constructed amod_principles_linguistic pobj_by_principles dobj_explicit_which nsubj_explicit_1 det__Both dep_1_ cc_1_and conj_1_Higgins dep_Higgins_2003 dep_2003_100 cop_explicit_are ccomp_determined_explicit prep_explicit_about det_predisposition_this pobj_about_predisposition ' +o,5092,'David showed it was accurate in the word sense disambiguation ',Yarowsky,'nn__David nsubj_showed_ nsubj_accurate_it cop_accurate_was ccomp_showed_accurate prep_accurate_in det_disambiguation_the nn_disambiguation_word nn_disambiguation_sense pobj_in_disambiguation ' +o,5093,'Wu adopted chammls that eliminate syntactically unlikely alignments and Wang et al ',Berger,'nn__Wu nsubj_adopted_ dobj_adopted_chammls nsubj_eliminate_that rcmod_chammls_eliminate advmod_unlikely_syntactically amod_alignments_unlikely dobj_eliminate_alignments cc_alignments_and conj_alignments_Wang cc_Wang_et conj_Wang_al ' +o,5094,'Generative and discriminative models have been comparedanddiscussedagreatdeal , including for NLP models ',Klein,'nn_models_Generative cc_Generative_and conj_Generative_discriminative nsubj_comparedanddiscussedagreatdeal_models aux_comparedanddiscussedagreatdeal_have dep_comparedanddiscussedagreatdeal_been tmod_comparedanddiscussedagreatdeal_ prep_comparedanddiscussedagreatdeal_including dep_including_for nn_models_NLP pobj_for_models ' +o,5095,'As comparison , used seed sets consisting of 7 words in their word valence annotation experiments , while Turney used minimal seed sets consisting of only one positive and one negative word -LRB- excellent and poor -RRB- in his experiments on review classification ',Turney,'prep_used_As pobj_As_comparison nsubj_used_ nn_sets_seed dobj_used_sets xcomp_used_consisting prep_consisting_of num_words_7 pobj_of_words prep_words_in poss_word_their pobj_in_word amod_experiments_valence nn_experiments_annotation tmod_used_experiments mark_used_while nn__Turney nsubj_used_ dep_experiments_used amod_sets_minimal nn_sets_seed dobj_used_sets xcomp_used_consisting prep_consisting_of advmod_positive_only num_positive_one pobj_of_positive cc_positive_and num_word_one amod_word_negative conj_positive_word dep_word_excellent cc_excellent_and conj_excellent_poor prep_word_in poss_experiments_his pobj_in_experiments prep_consisting_on nn_classification_review pobj_on_classification ' +o,5096,'An acceptable agreement for most NLP classification tasks lies between 07 and 08 ',Carletta,'det_agreement_An amod_agreement_acceptable nsubj_lies_agreement prep_agreement_for amod_tasks_most nn_tasks_NLP nn_tasks_classification pobj_for_tasks quantmod_07_between dobj_lies_07 cc_07_and conj_07_08 ' +o,5097,'We performed experiments with two statistical classifiers : the decision tree induction system C45 and the Tilburg Memory-Based Learner -LRB- TiMBL -RRB- ',Marcus,'nsubj_performed_We dobj_performed_experiments prep_performed_with num_classifiers_two amod_classifiers_statistical pobj_with_classifiers det__the nn__decision nn__tree nn__induction nn__system nn__C45 nsubj__ cc__and det_Learner_the nn_Learner_Tilburg nn_Learner_Memory-Based conj__Learner appos_Learner_TiMBL parataxis_performed_ ' +o,5098,'We also trained a baseline model with GIZA + + following a regimen of 5 iterations of Model 1 , 5 iterations of HMM , and 5 iterations of Model 4 ',Och,'nsubj_trained_We advmod_trained_also det_model_a amod_model_baseline dobj_trained_model prep_model_with nn__GIZA nn__+ nn__+ pobj_with_ prep__following det_regimen_a pobj_following_regimen prep_regimen_of num_iterations_5 pobj_of_iterations prep_iterations_of pobj_of_Model num_Model_1 num_iterations_5 conj_iterations_iterations prep_iterations_of pobj_of_HMM cc_iterations_and num_iterations_5 conj_iterations_iterations prep_iterations_of pobj_of_Model num_Model_4 ' +o,5099,'Both and Wagner et al ',Okanohara,'det__Both cc__and conj__Wagner cc_Wagner_et conj_Wagner_al ' +o,5100,'The four models we compare are a maximum a posteriori -LRB- MAP -RRB- method and three discriminative training methods , namely the boosting algorithm , the average perceptron and the minimum sample risk method ',Collins,'det_models_The num_models_four nsubj_compare_we rcmod_models_compare cop_maximum_are det_maximum_a ccomp_compare_maximum det__a amod__posteriori abbrev__MAP conj__method cc__and num_methods_three amod_methods_discriminative nn_methods_training conj__methods advmod__namely det__the amod__boosting nn__algorithm conj_methods_ det__the amod__average nn__perceptron conj_methods_ cc_methods_and det_risk_the amod_risk_minimum nn_risk_sample conj_methods_risk nn__method dep_models_ ' +o,5101,'Secondly , while all taggers use lexical information , and , indeed , it is well-known that lexical probabilities are much more revealing than tag sequence probabilities , most taggers make quite limited use of lexical probabilities -LRB- compared with , for example , the bilexical probabilities commonly used in current statistical parsers -RRB- ',Marcus,'advmod_make_Secondly mark_use_while det_taggers_all nsubj_use_taggers dep_make_use amod_information_lexical dobj_use_information cc_use_and advmod_well-known_indeed nsubj_well-known_it cop_well-known_is conj_use_well-known complm_revealing_that amod_probabilities_lexical nsubj_revealing_probabilities cop_revealing_are advmod_revealing_much advmod_revealing_more ccomp_well-known_revealing mark__than nn_probabilities_tag nn_probabilities_sequence nsubj__probabilities ccomp_revealing_ amod_taggers_most nsubj_make_taggers advmod_use_quite amod_use_limited dobj_make_use prep_use_of amod_probabilities_lexical pobj_of_probabilities dep_probabilities_compared prep_compared_with prep_compared_for pobj_for_example det_probabilities_the amod_probabilities_bilexical appos_example_probabilities advmod_used_commonly partmod_probabilities_used prep_used_in amod_parsers_current amod_parsers_statistical pobj_in_parsers ' +o,5102,'METRIC FORMULA Frequency (Guiliano, 1964) x yf Pointwise Mutual Information [PMI] (Church & Hanks, 1990) ( )xy x y2log /P P P True Mutual Information [TMI] (Manning, 1999) ( )xy 2 xy x ylog /P P P P Chi-Squared ( 2 ) (Church and Gale, 1991) { }{ },, 2( ) i X X Y Y i j i j i j j f T-Score (Church & Hanks, 1990) 1 2 2 2 1 2 1 2 x x s s n n + C-Values4 (Frantzi, Anadiou & Mima 2000) 2 is not nested 2 log ( ) log ( ) 1 ( ) ( ) a a b T a f f f b P T where is the candidate string f( ) is its frequency in the corpus T is the set of candidate terms that contain P(T ) is the number of these candidate terms 609 1,700 of the three-word phrases are attested in the Lexile corpus.',Church,'' +o,5103,' avoided the problem by precomputing the oracle translations in advance ',Tillmann,'nsubj_avoided_ det_problem_the dobj_avoided_problem prep_avoided_by pcomp_by_precomputing det_translations_the amod_translations_oracle dobj_precomputing_translations prep_precomputing_in pobj_in_advance ' +o,5104,'Probabilistic generative models like IBM 1-5 (Brown et al., 1993), HMM (Vogel et al., 1996), ITG (Wu, 1997), and LEAF (Fraser and Marcu, 2007) define formulas for P(f | e) or P(e, f), with ok-voon ororok sprok at-voon bichat dat erok sprok izok hihok ghirok totat dat arrat vat hilat ok-drubel ok-voon anok plok sprok at-drubel at-voon pippat rrat dat ok-voon anok drok brok jok at-voon krat pippat sat lat wiwok farok izok stok totat jjat quat cat lalok sprok izok jok stok wat dat krat quat cat lalok farok ororok lalok sprok izok enemok wat jjat bichat wat dat vat eneat lalok brok anok plok nok iat lat pippat rrat nnat wiwok nok izok kantok ok-yurp totat nnat quat oloat at-yurp lalok mok nok yorok ghirok clok wat nnat gat mat bat hilat lalok nok crrrok hihok yorok zanzanok wat nnat arrat mat zanzanat lalok rarok nok izok hihok mok wat nnat forat arrat vat gat Figure 1: Word alignment exercise (Knight, 1997).',Wu,'' +o,5105,'Much of this work has utilized the fundamental concept of semantic orientation , ; however , sentiment analysis still lacks a unified field theory ',Turney,'nsubj_utilized_Much prep_Much_of det_work_this pobj_of_work aux_utilized_has det_concept_the amod_concept_fundamental dobj_utilized_concept prep_concept_of amod_orientation_semantic pobj_of_orientation appos_orientation_ advmod_lacks_however nn_analysis_sentiment nsubj_lacks_analysis advmod_lacks_still parataxis_utilized_lacks det_theory_a amod_theory_unified nn_theory_field dobj_lacks_theory ' +o,5106,'Official DUC scoring utilizes the jackknife procedure and assesses significance using bootstrapping resampling ',Lin,'nn_DUC_Official nsubj_utilizes_DUC partmod_DUC_scoring det_procedure_the amod_procedure_jackknife dobj_utilizes_procedure cc_procedure_and nn_significance_assesses conj_procedure_significance partmod_procedure_using xcomp_using_bootstrapping dobj_bootstrapping_resampling ' +o,5107,'Many approaches for POS tagging have been developed in the past , including rule-based tagging , HMM taggers , maximum-entropy models , cyclic dependency networks , memory-based learning , etc All of these approaches require either a large amount of annotated training data -LRB- for supervised tagging -RRB- or a lexicon listing all possible tags for each word -LRB- for unsupervised tagging -RRB- ',Brown,'amod_approaches_Many nsubjpass_developed_approaches prep_approaches_for dep_tagging_POS pobj_for_tagging aux_developed_have auxpass_developed_been prep_developed_in det_past_the pobj_in_past prep_past_including amod__rule-based nn__tagging pobj_including_ nn__HMM nn__taggers appos__ amod__maximum-entropy nn__models appos__ amod__cyclic amod__dependency nn__networks appos__ amod__memory-based nn__learning appos__ appos__etc nsubj_require_All prep_All_of det_approaches_these pobj_of_approaches dep_developed_require advmod_amount_either det_amount_a amod_amount_large dobj_require_amount prep_amount_of amod_data_annotated nn_data_training pobj_of_data dep_data_for amod_tagging_supervised pobj_for_tagging cc_data_or det_lexicon_a conj_data_lexicon partmod_lexicon_listing det_tags_all amod_tags_possible dobj_listing_tags prep_tags_for det_word_each pobj_for_word dep_lexicon_for amod_tagging_unsupervised pobj_for_tagging ' +o,5108,'The quality of the translation output is mainly evaluated using BLEU , with NIST and METEOR as complementary metrics ',Banerjee,'det_quality_The nsubjpass_evaluated_quality prep_quality_of det_output_the nn_output_translation pobj_of_output auxpass_evaluated_is advmod_evaluated_mainly xcomp_evaluated_using dobj_using_BLEU prep_evaluated_with nn__NIST pobj_with_ cc__and nn__METEOR conj__ prep__as amod_metrics_complementary pobj_as_metrics ' +p,5109,'To speed our computations , we use the cube pruning method of with a fixed beam size ',Huang,'aux_speed_To dep_use_speed poss_computations_our dobj_speed_computations nsubj_use_we det_method_the nn_method_cube nn_method_pruning dobj_use_method prep_method_of pobj_of_ prep_use_with det_size_a amod_size_fixed nn_size_beam pobj_with_size ' +o,5110,'Note that although the source of the data is the same as in Section 5 , as did ',Yarowsky,'complm_same_that dep_same_although det_source_the nsubj_same_source prep_source_of det_data_the pobj_of_data cop_same_is det_same_the ccomp_Note_same advmod_same_as dep_as_in pobj_in_Section num_Section_5 mark_did_as nsubj_did_ advcl_same_did ' +o,5111,'The statistical significance often evaluate whether two words are independant using hypothesis tests such as t-score , the X2 , the log-likelihood and Fishers exact test ',Dunning,'det_significance_The amod_significance_statistical nsubj_evaluate_significance advmod_evaluate_often complm_independant_whether num_words_two nsubj_independant_words cop_independant_are ccomp_evaluate_independant dep_independant_using nn_tests_hypothesis dobj_using_tests dep_as_such prep_tests_as amod__t-score pobj_as_ det_X2_the conj__X2 det__the amod__log-likelihood conj__ cc__and nn__Fishers amod__exact nn__test conj__ ' +o,5112,'C , A , and B are computed for training dataset D as C = summationtext M m = 1 y -LRB- m -RRB- y -LRB- m -RRB- , A = summationtext M m = 1 y -LRB- m -RRB- , and B = summationtext M m = 1 y -LRB- m -RRB- In , y -LRB- m -RRB- was approximated by using the discriminative and logistic functions shown in Eqs ',Jansche,'nn_B_C conj_B_A cc_B_and nsubjpass_computed_B auxpass_computed_are prep_computed_for pcomp_for_training amod_D_dataset dobj_training_D prep_training_as pobj_as_C dep_C_= amod_m_summationtext nn_m_M iobj_=_m number_1_= num_y_1 nn_y_y appos_y_m dobj_=_y appos_y_m det_M_A amod_M_= nn_M_summationtext nsubj_m_M ccomp_computed_m dep_m_= iobj_=_1 dobj_=_y appos_y_m cc_computed_and nsubj_=_B conj_computed_= amod_m_summationtext nn_m_M nsubj_=_m ccomp_=_= num_y_1 dobj_=_y appos_y_m prep_approximated_In pobj_In_ nsubjpass_approximated_y appos_y_m auxpass_approximated_was rcmod_y_approximated prep_approximated_by pcomp_by_using det_functions_the amod_functions_discriminative cc_discriminative_and conj_discriminative_logistic dobj_using_functions partmod_functions_shown prep_shown_in pobj_in_Eqs ' +o,5113,'Various methods of automatically acquiring synonyms have been proposed ',Lin,'amod__Various nn__methods nsubjpass_proposed_ prep__of advmod_acquiring_automatically pcomp_of_acquiring dobj_acquiring_synonyms aux_proposed_have auxpass_proposed_been ' +o,5114,'Then the same system weights are applied to both IncHMM and Joint Decoding - based approaches , and the feature weights of them are trained using the max-BLEU training method proposed by and refined by Moore and Quirk -LRB- 2008 -RRB- ',Och,'advmod_applied_Then det_weights_the amod_weights_same nn_weights_system nsubjpass_applied_weights auxpass_applied_are prep_applied_to preconj_Decoding_both nn_Decoding_IncHMM cc_IncHMM_and conj_IncHMM_Joint pobj_to_Decoding dep_to_based dep_based_approaches cc_applied_and det_weights_the nn_weights_feature nsubjpass_trained_weights prep_weights_of pobj_of_them auxpass_trained_are conj_applied_trained xcomp_trained_using det_method_the amod_method_max-BLEU nn_method_training dobj_using_method dep_method_proposed prep_proposed_by pobj_by_ cc_proposed_and conj_proposed_refined prep_refined_by pobj_by_Moore cc_Moore_and conj_Moore_Quirk dep_refined_2008 ' +o,5115,'Various machine learning strategies have been proposed to address this problem , including semi-supervised learning , domain adaptation , multi-task learning , self-taught learning , etc A commonality among these methods is that they all require the training data and test data to be in the same feature space ',Blitzer,'amod_strategies_Various nn_strategies_machine nn_strategies_learning nsubjpass_proposed_strategies aux_proposed_have auxpass_proposed_been aux_address_to xcomp_proposed_address det_problem_this dobj_address_problem prep_proposed_including amod__semi-supervised nn__learning pobj_including_ nn__domain nn__adaptation appos__ amod__multi-task nn__learning appos__ amod__self-taught nn__learning appos__ appos__etc det_commonality_A nsubj_is_commonality prep_commonality_among det_methods_these pobj_among_methods dep_proposed_is complm_require_that nsubj_require_they advmod_require_all ccomp_is_require det_data_the nn_data_training dobj_require_data cc_data_and nn_data_test conj_data_data aux_be_to xcomp_require_be prep_be_in det_space_the amod_space_same nn_space_feature pobj_in_space ' +o,5116,'One possible conclusion from the POS tagging literature is that accuracy is approaching the limit , and any remaining improvement is within the noise of the Penn Treebank training data ',Ratnaparkhi,'num_conclusion_One amod_conclusion_possible nsubj_is_conclusion prep_conclusion_from det_literature_the dep_literature_POS nn_literature_tagging pobj_from_literature complm_approaching_that nsubj_approaching_accuracy aux_approaching_is ccomp_is_approaching det_limit_the dobj_approaching_limit cc_is_and det_improvement_any amod_improvement_remaining nsubj_is_improvement conj_is_is prep_is_within det_noise_the pobj_within_noise prep_noise_of det_data_the nn_data_Penn nn_data_Treebank nn_data_training pobj_of_data ' +o,5117,'Stress is an attribute of syllables , but syllabification is a non-trivial task in itself ',Bartlett,'nsubj_attribute_Stress cop_attribute_is det_attribute_an prep_attribute_of pobj_of_syllables cc_attribute_but nsubj_task_syllabification cop_task_is det_task_a amod_task_non-trivial conj_attribute_task prep_task_in pobj_in_itself ' +o,5118,'One promising approach extends standard Statistical Machine Translation -LRB- SMT -RRB- techniques to the problems of monolingual paraphrase identification and generation ',Brown,'num_approach_One amod_approach_promising nsubj_extends_approach amod_Translation_standard nn_Translation_Statistical nn_Translation_Machine nn_techniques_Translation abbrev_techniques_SMT nsubj__techniques xcomp_extends_ prep__to det_problems_the pobj_to_problems prep_problems_of amod_identification_monolingual dep_monolingual_paraphrase pobj_of_identification cc_identification_and conj_identification_generation ' +o,5119,'Other commonly used measures include kappa and relative utility , both of which take into account the performance of a summarizer that randomly picks passages from the original document to produce an extract ',Carletta,'amod_measures_Other advmod_used_commonly amod_measures_used nsubj_include_measures nn__kappa dobj_include_ cc__and amod__relative nn__utility conj__ dep_of_both nsubj_take_of pobj_of_which rcmod__take prep_take_into pobj_into_account det_performance_the dobj_take_performance prep_performance_of det_summarizer_a pobj_of_summarizer nsubj_picks_that advmod_picks_randomly rcmod_summarizer_picks dobj_picks_passages prep_picks_from det_document_the amod_document_original pobj_from_document aux_produce_to xcomp_take_produce det_extract_an dobj_produce_extract ' +o,5120,'45 proposed an MI-based measure , which he used to show that nouns could be reliably clustered based on their verb co-occurrences ',Hindle,'num__45 nsubj_proposed_ det_measure_an amod_measure_MI-based dobj_proposed_measure rel_used_which nsubj_used_he rcmod_measure_used aux_show_to xcomp_used_show complm_clustered_that nsubjpass_clustered_nouns aux_clustered_could auxpass_clustered_be advmod_clustered_reliably ccomp_show_clustered prep_clustered_based dep_based_on poss_co-occurrences_their nn_co-occurrences_verb pobj_on_co-occurrences ' +o,5121,'So far , this approach has been taken by a lot of researchers ',Pang,'dep_far_So advmod_taken_far det_approach_this nsubjpass_taken_approach aux_taken_has auxpass_taken_been prep_taken_by det_lot_a pobj_by_lot prep_lot_of pobj_of_researchers ' +o,5122,'Three recent papers in this area are , Hindle -LRB- 1990 -RRB- , and Smadja and McKeown -LRB- 1990 -RRB- ',Church,'num_papers_Three amod_papers_recent nsubj_Hindle_papers prep_papers_in det_area_this pobj_in_area cop_Hindle_are nn_Hindle_ appos_Hindle_1990 cc_Hindle_and conj_Hindle_Smadja cc_Smadja_and conj_Smadja_McKeown appos_Smadja_1990 ' +o,5123,'We utilize the OpenNLP MaxEnt implementation2 of the maximum entropy classification algorithm to train classification models for each lemma and part-of-speech combination in the training corpus ',Berger,'nsubj_utilize_We det_implementation2_the nn_implementation2_OpenNLP nn_implementation2_MaxEnt dobj_utilize_implementation2 prep_implementation2_of det_algorithm_the amod_algorithm_maximum amod_algorithm_entropy nn_algorithm_classification pobj_of_algorithm partmod_algorithm_ aux_train_to xcomp__train nn_models_classification dobj_train_models prep_train_for det_combination_each nn_combination_lemma cc_lemma_and conj_lemma_part-of-speech pobj_for_combination prep_combination_in det_corpus_the nn_corpus_training pobj_in_corpus ' +o,5124,'This generates tens of millions features , so we prune those features that occur fewer than 10 total times , as in ',Smith,'nsubj_generates_This dep_millions_tens quantmod_millions_of num_features_millions dobj_generates_features mark_prune_so nsubj_prune_we advcl_generates_prune det_features_those dobj_prune_features nsubj_occur_that rcmod_features_occur dep_than_fewer quantmod_10_than num_times_10 amod_times_total dobj_occur_times mark_in_as advcl_occur_in ' +o,5125,'Recently some researchers have pointed out the importance of the lexicon and proposed lexicalized models ',Collins,'advmod_researchers_Recently det_researchers_some nsubj_pointed_researchers aux_pointed_have prt_pointed_out det_importance_the dobj_pointed_importance prep_importance_of det_lexicon_the pobj_of_lexicon cc_lexicon_and amod_models_proposed amod_models_lexicalized conj_lexicon_models ' +n,5126,'We also compare our performance against and and find that we manage higher relative improvement than existing work despite starting from a very competitive baseline CRF ',Finkel,'nsubj_compare_We advmod_compare_also poss_performance_our dobj_compare_performance prep_performance_against pobj_against_ cc__and conj__ cc_compare_and conj_compare_find complm_manage_that nsubj_manage_we ccomp_find_manage amod_improvement_higher amod_improvement_relative dobj_manage_improvement prep_improvement_than amod_work_existing pobj_than_work prep_manage_despite pcomp_despite_starting prep_starting_from det_CRF_a advmod_competitive_very amod_CRF_competitive nn_CRF_baseline pobj_from_CRF ' +o,5127,'A reranking parser -LRB- see also -RRB- is a layered model : the base layer is a generative statistical PCFG parser that creates a ranked list of k parses -LRB- say , 50 -RRB- , and the second layer is a reranker that reorders these parses using more detailed features ',Koo,'det_parser_A amod_parser_reranking nsubj_model_parser dep_parser_see advmod__also acomp_see_ cop_model_is det_model_a amod_model_layered det_layer_the nn_layer_base nsubj_parser_layer cop_parser_is det_parser_a amod_parser_generative amod_parser_statistical nn_parser_PCFG parataxis_model_parser nsubj_creates_that rcmod_parser_creates det_list_a amod_list_ranked dobj_creates_list prep_list_of nn_parses_k pobj_of_parses dep_-LRB-_say dep_say_50 cc_model_and det_layer_the amod_layer_second nsubj_reranker_layer cop_reranker_is det_reranker_a conj_model_reranker nsubj_reorders_that rcmod_reranker_reorders det_parses_these nsubj_using_parses dep_reorders_using amod_features_more amod_features_detailed dobj_using_features ' +o,5128,'To achieve step -LRB- 1 -RRB- , we first apply a set of headfinding rules which are similar to those described in ',Collins,'aux_achieve_To dep_apply_achieve dobj_achieve_step appos_step_1 nsubj_apply_we advmod_apply_first det_set_a dobj_apply_set prep_set_of amod_rules_headfinding pobj_of_rules nsubj_similar_which cop_similar_are rcmod_set_similar prep_similar_to pobj_to_those partmod_those_described prep_described_in ' +p,5129,' information-theoretic similarity measure is commonly used in lexicon acquisition tasks and has demonstrated good performance in unsupervised WSD ',Lin,'num_measure_ amod_measure_information-theoretic nn_measure_similarity nsubjpass_used_measure auxpass_used_is advmod_used_commonly prep_used_in nn_tasks_lexicon nn_tasks_acquisition pobj_in_tasks cc_used_and aux_demonstrated_has conj_used_demonstrated amod_performance_good dobj_demonstrated_performance prep_demonstrated_in amod__unsupervised nn__WSD pobj_in_ ' +o,5130,'Obviously , these productions are not in the normal form of an ITG , but with the method described in , they can be normalized ',Wu,'advmod_are_Obviously det_productions_these nsubj_are_productions dep_are_not dep_not_in det_form_the amod_form_normal pobj_in_form prep_form_of det_ITG_an pobj_of_ITG cc_not_but mark_described_with det_method_the nsubj_described_method conj_not_described prep_normalized_in pobj_in_ nsubjpass_normalized_they aux_normalized_can auxpass_normalized_be ccomp_described_normalized ' +p,5131,'1 Introduction State-of-the-art Statistical Machine Translation -LRB- SMT -RRB- systems usually adopt a two-pass search strategy as shown in Figure 1 ',Koehn,'num_Translation_1 nn_Translation_Introduction nn_Translation_State-of-the-art nn_Translation_Statistical nn_Translation_Machine abbrev_Translation_SMT nsubj_adopt_systems advmod_adopt_usually dep_Translation_adopt det_strategy_a amod_strategy_two-pass nn_strategy_search dobj_adopt_strategy dep_as_ advmod_shown_as partmod_strategy_shown prep_shown_in pobj_in_Figure num_Figure_1 ' +o,5132,'We set all weights by optimizing Bleu using minimum error rate training -LRB- MERT -RRB- on a separate development set of 2,000 sentences -LRB- Indonesian or Spanish -RRB- , and we used them in a beam search decoder to translate 2,000 test sentences -LRB- Indonesian or Spanish -RRB- into English ',Och,'nsubj_set_We det_weights_all dobj_set_weights prep_set_by pcomp_by_optimizing nn__Bleu nsubj_using_ dep_optimizing_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training abbrev_training_MERT partmod_training_ prep__on det_set_a amod_set_separate nn_set_development pobj_on_set prep_set_of num_sentences_2,000 pobj_of_sentences appos_training_Indonesian cc_Indonesian_or conj_Indonesian_Spanish cc_set_and nsubj_used_we conj_set_used dobj_used_them prep_used_in det__a nn__beam nn__search nn__decoder pobj_in_ aux_translate_to xcomp_used_translate num_sentences_2,000 nn_sentences_test dobj_translate_sentences appos_sentences_Indonesian cc_Indonesian_or conj_Indonesian_Spanish prep_translate_into pobj_into_English ' +o,5133,'31 Context Extraction We adopted dependency structure as the context of words since it is the most widely used and wellperforming contextual information in the past studies ',Lin,'num_Extraction_31 nn_Extraction_Context nsubj_structure_We dep_structure_adopted amod_structure_dependency dep_Extraction_structure prep_structure_as det_context_the pobj_as_context prep_context_of pobj_of_words mark_used_since nsubj_used_it cop_used_is dep_used_the advmod_used_most advmod_used_widely advcl_structure_used cc_used_and amod_information_wellperforming amod_information_contextual conj_used_information prep_information_in det_studies_the amod_studies_past pobj_in_studies ' +o,5134,' considered the same problem and presented a set of supervised machine learning approaches to it ',Pang,'nsubj_considered_ det_problem_the amod_problem_same dobj_considered_problem cc_considered_and conj_considered_presented det_set_a dobj_presented_set prep_set_of amod_machine_supervised pobj_of_machine xcomp_presented_learning dobj_learning_approaches prep_learning_to pobj_to_it ' +o,5135,'The first one is a hypotheses testing approach while the second one is closer to a model estimating approach ',Brown,'det_one_The amod_one_first nsubj_hypotheses_one cop_hypotheses_is det_hypotheses_a dep__testing nn__approach partmod_hypotheses_ mark_estimating_while det_one_the amod_one_second nsubj_estimating_one aux_estimating_is advmod_estimating_closer dep_closer_to det_model_a pobj_to_model advcl__estimating dobj_estimating_approach ' +o,5136,'Since this trade-off is also affected by the settings of various pruning parameters , we compared decoding time and translation quality , as measured by BLEU score , for the two models on our first test set over a broad range of settings for the decoder pruning parameters ',Papineni,'mark_affected_Since det_trade-off_this nsubjpass_affected_trade-off auxpass_affected_is advmod_affected_also advcl_compared_affected prep_affected_by det_settings_the pobj_by_settings prep_settings_of amod_parameters_various nn_parameters_pruning pobj_of_parameters nsubj_compared_we xcomp_compared_decoding nn_quality_time cc_time_and conj_time_translation dobj_decoding_quality mark_measured_as advcl_compared_measured prep_measured_by amod__BLEU nn__score pobj_by_ prep_measured_for det_models_the num_models_two pobj_for_models prep_models_on poss_test_our amod_test_first pobj_on_test partmod_test_set prep_set_over det_range_a amod_range_broad pobj_over_range prep_range_of pobj_of_settings prep_settings_for det_parameters_the amod_parameters_decoder nn_parameters_pruning pobj_for_parameters ' +o,5137,' showed that it is possible to use only a few of those semantically oriented words -LRB- namely , excellent and poor -RRB- to label other phrases co-occuring with them as positive or negative ',Turney,'nsubj_showed_ complm_possible_that nsubj_possible_it cop_possible_is ccomp_showed_possible aux_use_to xcomp_possible_use quantmod_few_only quantmod_few_a dobj_use_few prep_few_of det_words_those advmod_oriented_semantically amod_words_oriented pobj_of_words dep_excellent_namely dep_words_excellent cc_excellent_and conj_excellent_poor aux_label_to xcomp_use_label amod_phrases_other dobj_label_phrases amod_phrases_co-occuring prep_co-occuring_with pobj_with_them mark_positive_as dep_phrases_positive cc_positive_or conj_positive_negative ' +o,5138,'However , much recent work in machine learning and statistics has turned away from maximum-likelihood in favor of Bayesian methods , and there is increasing interest in Bayesian methods in computational linguistics as well ',Finkel,'advmod_turned_However amod_work_much amod_work_recent nsubj_turned_work prep_work_in nn_learning_machine pobj_in_learning cc_learning_and conj_learning_statistics aux_turned_has prt_turned_away prep_turned_from amod_methods_maximum-likelihood prep_maximum-likelihood_in pobj_in_favor prep_favor_of pobj_of_Bayesian pobj_from_methods cc_turned_and expl_is_there conj_turned_is dep_is_increasing dobj_increasing_interest prep_interest_in amod_methods_Bayesian pobj_in_methods prep_methods_in amod_linguistics_computational pobj_in_linguistics advmod_well_as advmod_increasing_well ' +o,5139,'After the parser produces a semantic feature structure representation of the sentence , predicate mapping rules then match against that representation in order to produce a predicate language representation in the style of Davidsonian event based semantics , as mentioned above ',Hobbs,'mark_produces_After det_parser_the nsubj_produces_parser advcl_match_produces det_representation_a amod_representation_semantic nn_representation_feature nn_representation_structure dobj_produces_representation prep_representation_of det_sentence_the pobj_of_sentence amod_rules_predicate nn_rules_mapping nsubj_match_rules advmod_match_then prep_match_against det_representation_that pobj_against_representation mark_produce_in dep_produce_order aux_produce_to purpcl_match_produce det_representation_a amod_representation_predicate nn_representation_language dobj_produce_representation prep_produce_in det_style_the pobj_in_style prep_style_of nn_event_Davidsonian pobj_of_event prep_produce_based nn__semantics dep_based_ mark_mentioned_as dep__mentioned prep_mentioned_above ' +o,5140,'2 The alignment Algorithm 21 Estimation of translation probabilities The translation probabilities are estimated using a method based on , which is summarized in the following subsection , 211 ',Brown,'dep_Estimation_2 det_Estimation_The amod_Estimation_alignment nn_Estimation_Algorithm num_Estimation_21 nsubjpass_estimated_Estimation prep_Estimation_of nn_probabilities_translation pobj_of_probabilities det_probabilities_The nn_probabilities_translation dep_Estimation_probabilities auxpass_estimated_are xcomp_estimated_using det_method_a dobj_using_method prep_using_based dep_based_on pobj_on_ nsubjpass_summarized_which auxpass_summarized_is rcmod__summarized prep_summarized_in det_subsection_the amod_subsection_following pobj_in_subsection appos_subsection_211 ' +o,5141,'In the Link Grammar framework , strictly local contexts are naturally combined with long-distance information coming from long-range trigrams ',Brown,'prep_combined_In det__the nn__Link nn__Grammar nn__framework pobj_In_ amod_contexts_strictly amod_contexts_local nsubjpass_combined_contexts auxpass_combined_are advmod_combined_naturally prep_combined_with amod_information_long-distance pobj_with_information partmod_information_coming prep_coming_from amod_trigrams_long-range pobj_from_trigrams ' +o,5142,'Minimum-error-rate training are conducted on dev-set to optimize feature weights maximizing the BLEU score up to 4grams , and the obtained feature weights are blindly applied on the test-set ',Och,'amod__Minimum-error-rate nn__training nsubjpass_conducted_ auxpass_conducted_are prep_conducted_on pobj_on_dev-set aux_optimize_to xcomp_conducted_optimize nn_weights_feature dobj_optimize_weights xcomp_optimize_maximizing det_score_the amod_score_BLEU dobj_maximizing_score prt_maximizing_up prep_maximizing_to pobj_to_4grams cc_conducted_and det_weights_the amod_weights_obtained nn_weights_feature nsubjpass_applied_weights auxpass_applied_are advmod_applied_blindly conj_conducted_applied prep_applied_on det_test-set_the pobj_on_test-set ' +o,5143,'Given the training pairs , any sequence predictor can be used , for example a Conditional Random Field -LRB- CRF -RRB- or a structured perceptron ',Collins,'prep_used_Given det_pairs_the nn_pairs_training dep_Given_pairs det_predictor_any nn_predictor_sequence appos_pairs_predictor aux_used_can auxpass_used_be prep_used_for pobj_for_example det_Field_a nn_Field_Conditional nn_Field_Random nsubjpass_used_Field abbrev_Field_CRF dep_Field_ cc__or det_perceptron_a amod_perceptron_structured conj__perceptron ' +o,5144,'For example , in the IBM Models , each word ti independently generates 0 , 1 , or more 2Note that we refer to t as the target sentence , even though in the source-channel model , t is the source sentence which goes through the channel model P -LRB- s t -RRB- to produce the observed sentence s words in the source language ',Brown,'prep_0_For pobj_For_example prep_0_in det__the nn__IBM nn__Models pobj_in_ det_ti_each nn_ti_word nsubj_0_ti advmod_0_independently dep_0_generates conj_0_1 cc_0_or amod_2Note_more conj_0_2Note complm_refer_that nsubj_refer_we dep_2Note_refer aux_t_to xcomp_refer_t prep_t_as det_sentence_the nn_sentence_target pobj_as_sentence advmod_sentence_even dep_sentence_though prep_sentence_in det_model_the amod_model_source-channel pobj_in_model nsubj_sentence_t cop_sentence_is det_sentence_the nn_sentence_source dep_0_sentence nsubj_goes_which rcmod_sentence_goes prep_goes_through det_P_the nn_P_channel nn_P_model pobj_through_P nsubj_t_s parataxis_goes_t aux_produce_to xcomp_goes_produce det_words_the amod_words_observed nn_words_sentence nn_words_s dobj_produce_words prep_produce_in det_language_the nn_language_source pobj_in_language ' +o,5145,'We do not use particular lexicosyntactic patterns , as previous attempts have ',Fleischman,'nsubj_use_We aux_use_do neg_use_not amod_patterns_particular amod_patterns_lexicosyntactic dobj_use_patterns mark_have_as amod_attempts_previous nsubj_have_attempts advcl_use_have ' +o,5146,' , a trigram target language model , an order model , word count , phrase count , average phrase size functions , and whole-sentence IBM Model 1 logprobabilities in both directions ',Och,'det_model_a nn_model_trigram nn_model_target nn_model_language conj__model det_model_an nn_model_order conj__model nn_count_word conj__count nn_count_phrase conj__count amod_functions_average nn_functions_phrase nn_functions_size conj__functions cc__and amod_Model_whole-sentence nn_Model_IBM conj__Model num_logprobabilities_1 dep__logprobabilities prep_logprobabilities_in det_directions_both pobj_in_directions ' +o,5147,'Our approach thus provides an even more extreme version of automatic con rmation generation than that used by where only a small eort is required by the developer ',Chu-Carroll,'poss_approach_Our nsubj_provides_approach advmod_provides_thus det_version_an advmod_extreme_even advmod_extreme_more amod_version_extreme dobj_provides_version prep_version_of amod_generation_automatic nn_generation_con nn_generation_rmation pobj_of_generation prep_provides_than pobj_than_that partmod_that_used prep_used_by pcomp_by_ advmod_required_where advmod_eort_only det_eort_a amod_eort_small nsubjpass_required_eort auxpass_required_is advcl__required prep_required_by det_developer_the pobj_by_developer ' +o,5148,'For example , in machine translation , BLEU score is developed to assess the quality of machine translated sentences ',Papineni,'prep_developed_For pobj_For_example prep_developed_in nn_translation_machine pobj_in_translation amod__BLEU nn__score nsubjpass_developed_ auxpass_developed_is aux_assess_to xcomp_developed_assess det_quality_the dobj_assess_quality prep_quality_of nn_sentences_machine amod_sentences_translated pobj_of_sentences ' +o,5149,'The first one , GIZA-Lex , is obtained by running the GIZA + +2 implementation of the IBM word alignment models on the initial parallel corpus ',Brown,'det_one_The amod_one_first nsubjpass_obtained_one appos_one_GIZA-Lex auxpass_obtained_is prep_obtained_by pcomp_by_running det_implementation_the nn_implementation_GIZA nn_implementation_+ num_implementation_+2 dobj_running_implementation prep_implementation_of det__the nn__IBM nn__word nn__alignment nn__models pobj_of_ prep_running_on det_corpus_the amod_corpus_initial amod_corpus_parallel pobj_on_corpus ' +o,5150,'But such general word lists were shown to perform worse than statistical models built on sufficiently large in-domain training sets of movie reviews ',Pang,'cc_shown_But amod_lists_such amod_lists_general nn_lists_word nsubjpass_shown_lists auxpass_shown_were aux_perform_to xcomp_shown_perform acomp_perform_worse prep_perform_than amod_models_statistical pobj_than_models partmod_models_built prep_built_on advmod_large_sufficiently amod_sets_large amod_sets_in-domain nn_sets_training pobj_on_sets prep_sets_of nn_reviews_movie pobj_of_reviews ' +o,5151,'It assumes that the distance of the positions relative to the diagonal of the -LRB- j , i -RRB- plane is the dominating factor : r -LRB- i _ j I -RRB- p -LRB- ilj , J , I -RRB- = -LRB- 7 -RRB- , Ei , = l r -LRB- i \' j -RRB- As described in , the EM algorithm can be used to estimate the parameters of the model ',Brown,'nsubj_assumes_It complm_=_that det_distance_the nsubj_=_distance prep_distance_of det_positions_the pobj_of_positions advmod_positions_relative dep_relative_to det_diagonal_the pobj_to_diagonal prep_positions_of det_j_the pobj_of_j dep_plane_i appos_j_plane prep_plane_is det_factor_the amod_factor_dominating dep_is_factor dep_plane_r amod_I_i nn_I__ nn_I_j appos_r_I dep_r_p dep_positions_ilj dep_ilj_J dep_ilj_I ccomp_assumes_= dobj_=_7 appos_7_Ei amod_l_= dep_7_l dobj_used_r dep_\'_i dep_r_j mark_described_As advcl_used_described prep_described_in pobj_in_ det_algorithm_the nn_algorithm_EM nsubjpass_used_algorithm aux_used_can auxpass_used_be rcmod_7_used aux_estimate_to purpcl_used_estimate det_parameters_the dobj_estimate_parameters prep_parameters_of det_model_the pobj_of_model ' +o,5152,'1 Introduction Corpus-derived distributional semantic spaces have proved valuable in tackling a variety of tasks , ranging from concept categorization to relation extraction to many others ',Turney,'num_spaces_1 nn_spaces_Introduction nn_spaces_Corpus-derived amod_spaces_distributional nn_spaces_semantic nsubj_proved_spaces aux_proved_have acomp_proved_valuable prep_valuable_in pcomp_in_tackling det_variety_a dobj_tackling_variety prep_variety_of pobj_of_tasks dep_tackling_ranging prep_ranging_from nn_categorization_concept pobj_from_categorization prep_categorization_to nn_extraction_relation pobj_to_extraction prep_ranging_to amod_others_many pobj_to_others ' +o,5153,'One of the advantages of these methods is that a wide variety of features such as dependency trees and sequences of words can easily be incorporated ',Pang,'nsubj_is_One prep_One_of det_advantages_the pobj_of_advantages prep_advantages_of det_methods_these pobj_of_methods complm_incorporated_that det_variety_a amod_variety_wide nsubjpass_incorporated_variety prep_variety_of pobj_of_features dep_as_such prep_features_as amod_trees_dependency pobj_as_trees cc_trees_and conj_trees_sequences prep_trees_of pobj_of_words aux_incorporated_can advmod_incorporated_easily auxpass_incorporated_be ccomp_is_incorporated ' +o,5154,'We then built separate directed word alignments for EnglishX andXEnglish -LRB- X -LCB- Indonesian , Spanish -RCB- -RRB- using IBM model 4 , combined them using the intersect + grow heuristic , and extracted phrase-level translation pairs of maximum length seven using the alignment template approach ',Brown,'nsubj_built_We advmod_built_then amod_alignments_separate amod_alignments_directed nn_alignments_word dobj_built_alignments prep_built_for nn_andXEnglish_EnglishX pobj_for_andXEnglish dep_andXEnglish_X appos_andXEnglish_Indonesian dep_Indonesian_Spanish partmod_andXEnglish_using nn_model_IBM nsubj__model num__4 xcomp_using_ partmod__combined nsubj_using_them dep_combined_using det__the amod__intersect cc_intersect_+ conj_intersect_grow amod__heuristic dobj_using_ cc_built_and conj_built_extracted amod_pairs_phrase-level nn_pairs_translation dobj_extracted_pairs prep_pairs_of nn_length_maximum pobj_of_length num_length_seven xcomp_built_using det__the amod__alignment amod__template nn__approach dobj_using_ ' +o,5155,'Intuitively speaking , the gaps on the target-side will lead to exponential complexity in decoding with integrated language models -LRB- see Section 3 -RRB- , as well as synchronous parsing ',Zhang,'amod_speaking_Intuitively nsubj_lead_speaking det_gaps_the nsubj_lead_gaps prep_gaps_on det_target-side_the pobj_on_target-side aux_lead_will prep_lead_to amod_complexity_exponential pobj_to_complexity prep_complexity_in pcomp_in_decoding prep_decoding_with amod_models_integrated nn_models_language pobj_with_models dep_models_see dobj_see_Section num_Section_3 dep_well_as cc_complexity_well dep_well_as conj_complexity_synchronous amod_synchronous_parsing ' +o,5156,'Now with the availability of large-scale corpus , automatic acquisition of word compositions , especially word collocations from them have been extensively studie ',Smadja,'dep_with_Now prep_studie_with det_availability_the pobj_with_availability prep_availability_of amod_corpus_large-scale pobj_of_corpus amod_acquisition_automatic nsubj_studie_acquisition prep_acquisition_of nn_compositions_word pobj_of_compositions advmod_collocations_especially nn_collocations_word nsubj_studie_collocations prep_collocations_from pobj_from_them aux_studie_have dep_studie_been advmod_studie_extensively ' +o,5157,'Figure 1 gives an example dependency graph for the sentence Mr Tomash will remain as a director emeritus , whichhasbeenextractedfromthe Penn Treebank ',Marcus,'nsubj_gives_Figure num_Figure_1 det_example_an iobj_gives_example amod_graph_dependency dobj_gives_graph prep_graph_for det_sentence_the pobj_for_sentence nn_Tomash_Mr nsubj_remain_Tomash aux_remain_will dep_gives_remain prep_remain_as det_emeritus_a nn_emeritus_director pobj_as_emeritus nn_Treebank_whichhasbeenextractedfromthe nn_Treebank_Penn appos_emeritus_Treebank ' +o,5158,' use pointwise KLdivergence between multiple language models for scoring both phraseness and informativeness of phrases ',Tomokiyo,'amod_use_ nsubj_pointwise_use dobj_pointwise_KLdivergence prep_KLdivergence_between amod_models_multiple nn_models_language pobj_between_models prep_pointwise_for pcomp_for_scoring preconj_phraseness_both dobj_scoring_phraseness cc_phraseness_and conj_phraseness_informativeness prep_phraseness_of pobj_of_phrases ' +o,5159,'We parsed a 125-million word newspaper corpus with Minipar , 1 a descendent of Principar , and extracted dependency relationships from the parsed corpus ',Dunning,'nsubj_parsed_We det_corpus_a amod_corpus_125-million nn_corpus_word nn_corpus_newspaper dobj_parsed_corpus prep_parsed_with pobj_with_Minipar dep_descendent_1 det_descendent_a appos_Minipar_descendent prep_descendent_of nn__Principar pobj_of_ cc_parsed_and conj_parsed_extracted amod_relationships_dependency dobj_extracted_relationships prep_extracted_from det_corpus_the amod_corpus_parsed pobj_from_corpus ' +o,5160,'Discriminative training has been used mainly for translation model combination and with the exception of , has not been used to directly train parameters of a translation model ',Tillmann,'amod_training_Discriminative nsubjpass_used_training aux_used_has auxpass_used_been rcmod_training_used advmod_used_mainly prep_used_for nn__translation nn__model nn__combination pobj_for_ cc_for_and conj_for_with det_exception_the pobj_with_exception prep_exception_of pobj_of_ aux_used_has neg_used_not auxpass_used_been aux_train_to advmod_train_directly xcomp_used_train dobj_train_parameters prep_parameters_of det_model_a nn_model_translation pobj_of_model ' +o,5161,'In WASP , GIZA + + is used to obtain the best alignments from the training examples ',Och,'prep_used_In pobj_In_WASP nn__GIZA nn__+ nn__+ nsubjpass_used_ auxpass_used_is aux_obtain_to xcomp_used_obtain det_alignments_the amod_alignments_best dobj_obtain_alignments prep_alignments_from det_examples_the nn_examples_training pobj_from_examples ' +o,5162,'We ran the baseline semisupervised system for two iterations -LRB- line 2 -RRB- , and in contrast with we found that the best symmetrization heuristic for this system was union , which is most likely due to our use of fully linked alignments which was discussed at the end of Section 3 ',Fraser,'nsubj_ran_We det_system_the amod_system_baseline amod_system_semisupervised dobj_ran_system prep_ran_for num_iterations_two pobj_for_iterations appos_iterations_line num_line_2 cc_ran_and dep_with_in dep_with_contrast prep_found_with pobj_with_ nsubj_found_we conj_ran_found complm_union_that det_heuristic_the amod_heuristic_best nn_heuristic_symmetrization nsubj_union_heuristic prep_heuristic_for det_system_this pobj_for_system cop_union_was ccomp_found_union nsubj_due_which cop_due_is advmod_due_most advmod_due_likely rcmod_union_due prep_due_to poss_use_our pobj_to_use prep_use_of advmod_linked_fully amod_alignments_linked pobj_of_alignments nsubjpass_discussed_which auxpass_discussed_was rcmod_alignments_discussed prep_discussed_at det_end_the pobj_at_end prep_end_of pobj_of_Section num_Section_3 ' +p,5163,'Following we can avoid unnecessary false positives by not querying for the longer n-gram in such cases ',Talbot,'pobj_Following_ nsubj_avoid_we aux_avoid_can rcmod__avoid amod_positives_unnecessary amod_positives_false dobj_avoid_positives prep_avoid_by neg_querying_not pcomp_by_querying prep_querying_for det_n-gram_the amod_n-gram_longer pobj_for_n-gram prep_n-gram_in amod_cases_such pobj_in_cases ' +n,5164,'Although several methods have already been proposed to incorporate non-local features , these present a problem that the types of non-local features are somewhat constrained ',Finkel,'mark_proposed_Although amod_methods_several nsubjpass_proposed_methods aux_proposed_have advmod_proposed_already auxpass_proposed_been advcl_problem_proposed aux_incorporate_to xcomp_proposed_incorporate amod_features_non-local dobj_incorporate_features advmod_incorporate_ nsubj_problem_these dep_problem_present det_problem_a complm_constrained_that det_types_the nsubjpass_constrained_types prep_types_of amod_features_non-local pobj_of_features auxpass_constrained_are advmod_constrained_somewhat ccomp_problem_constrained ' +n,5165,'The morphological processing in PairClass is more sophisticated than in ',Turney,'det_processing_The amod_processing_morphological nsubj_sophisticated_processing prep_processing_in nn__PairClass pobj_in_ cop_sophisticated_is advmod_sophisticated_more prep_sophisticated_than dep_than_in ' +o,5166,'Recent work on the automatic acquisition of multilingual LFG resources from treebanks for Chinese , German and Spanish has shown that given a suitable treebank , it is possible to automatically acquire high quality LFG resources in a very short space of time ',Cahill,'amod_work_Recent nsubj_shown_work prep_work_on det_acquisition_the amod_acquisition_automatic pobj_on_acquisition prep_acquisition_of amod_resources_multilingual nn_resources_LFG pobj_of_resources prep_resources_from pobj_from_treebanks prep_treebanks_for amod__Chinese amod__German cc_German_and conj_German_Spanish pobj_for_ aux_shown_has complm_possible_that prep_possible_given det_treebank_a amod_treebank_suitable dep_given_treebank nsubj_possible_it cop_possible_is ccomp_shown_possible aux_acquire_to advmod_acquire_automatically xcomp_possible_acquire amod_resources_high nn_resources_quality nn_resources_LFG dobj_acquire_resources prep_resources_in det_space_a advmod_short_very amod_space_short pobj_in_space prep_space_of pobj_of_time ' +o,5167,'In Machine Translation , for example , sentences are produced using application-specific decoders , inspired by work on speech recognition , whereas in Summarization , summaries are produced as either extracts or using task-specific strategies ',Brown,'prep_produced_In nn_Translation_Machine pobj_In_Translation prep_produced_for pobj_for_example nsubjpass_produced_sentences auxpass_produced_are ccomp_produced_using amod_decoders_application-specific dobj_using_decoders partmod_decoders_inspired prep_inspired_by pobj_by_work prep_work_on nn__speech nn__recognition pobj_on_ mark_produced_whereas prep_produced_in pobj_in_Summarization nsubjpass_produced_summaries auxpass_produced_are dep_decoders_produced prep_produced_as det_extracts_either pobj_as_extracts cc_using_or conj_using_using amod_strategies_task-specific dobj_using_strategies advmod_using_ ' +o,5168,'Maximum Entropy models implement the intuition that the best model is the one that is consistent with the set of constraints imposed by the evidence but otherwise is as uniform as possible ',Berger,'nn_models_Entropy nsubj_Maximum_models dep_intuition_implement det_intuition_the dep_Maximum_intuition complm_one_that det_model_the amod_model_best nsubj_one_model cop_one_is det_one_the ccomp_intuition_one nsubj_consistent_that cop_consistent_is rcmod_one_consistent prep_consistent_with det_set_the pobj_with_set prep_set_of pobj_of_constraints partmod_constraints_imposed prep_imposed_by det_evidence_the pobj_by_evidence cc_consistent_but advmod_consistent_otherwise cop_uniform_is advmod_uniform_as conj_consistent_uniform prep_uniform_as pobj_as_possible ' +o,5169,'Endemic structural ambiguity , which can lead to such difficulties as trying to cope with the many thousands of possible parses that a grammar can assign to a sentence , can be greatly reduced by adding empirically derived probabilities to grammar rules and by computing statistical measures of lexical association ',Marcus,'amod_ambiguity_Endemic amod_ambiguity_structural nsubjpass_reduced_ambiguity nsubj_lead_which aux_lead_can rcmod_ambiguity_lead prep_lead_to amod_difficulties_such pobj_to_difficulties prep_lead_as pcomp_as_trying aux_cope_to xcomp_trying_cope prep_cope_with det_thousands_the amod_thousands_many pobj_with_thousands prep_thousands_of amod_parses_possible pobj_of_parses complm_assign_that det_grammar_a nsubj_assign_grammar aux_assign_can ccomp_cope_assign prep_assign_to det_sentence_a pobj_to_sentence aux_reduced_can auxpass_reduced_be advmod_reduced_greatly prep_reduced_by pcomp_by_adding advmod_derived_empirically amod_probabilities_derived dobj_adding_probabilities prep_adding_to nn__grammar nn__rules pobj_to_ cc_by_and conj_by_by amod_measures_computing amod_measures_statistical pobj_by_measures prep_measures_of amod__lexical nn__association pobj_of_ ' +o,5170,', 1989 -RRB- , eg , lexicography , information retrieval , text input , etc This paper will touch on its feasibility in topic identification ',Church,'dep_text_1989 rcmod_1989_eg advmod__lexicography amod__ nn__information nn__retrieval dobj_eg_ prep_etc_text nn__input pobj_text_ det_paper_This nsubj_touch_paper aux_touch_will ccomp_etc_touch prep_touch_on poss_feasibility_its pobj_on_feasibility prep_etc_in nn_identification_topic pobj_in_identification ' +p,5171,'It is explored extensively in ',Weeds,'nsubj_explored_It aux_explored_is advmod_explored_extensively prep_explored_in ' +o,5172,'However , it can not handle long-distance reorderings properly and does not exploit discontinuous phrases and linguistically syntactic structure features ',Quirk,'advmod_handle_However nsubj_handle_it aux_handle_can neg_handle_not amod_reorderings_long-distance dobj_handle_reorderings advmod_handle_properly cc_handle_and aux_exploit_does neg_exploit_not conj_handle_exploit amod_phrases_discontinuous dobj_exploit_phrases cc_phrases_and amod_features_linguistically amod_features_syntactic nn_features_structure conj_phrases_features ' +o,5173,'Salience Feature Pronoun Name Nominal TOP 075 017 008 HIGH 055 028 017 MID 039 040 021 LOW 020 045 035 NONE 000 088 012 Table 2 : Posterior distribution of mention type given salience -LRB- taken from -RRB- 33 Modifications to the H&K Model Next , we discuss the potential weaknesses of H&K s model and propose three modifications to it ',Haghighi,'nn_Pronoun_Salience nn_Pronoun_Feature dep_Pronoun_Name nn_Table_Nominal nn_Table_TOP num_Table_075 num_Table_017 num_Table_008 nn_Table_HIGH num_Table_055 num_Table_028 num_Table_017 nn_Table_MID num_Table_039 num_Table_040 num_Table_021 nn_Table_LOW num_Table_020 num_Table_045 num_Table_035 nn_Table_NONE num_Table_000 nn_Table_088 num_Table_012 nsubj_2_Table xcomp_Name_2 nn_distribution_Posterior dep_2_distribution prep_distribution_of nn_type_mention pobj_of_type partmod_type_given dep_given_salience parataxis_Name_taken prep_taken_from pobj_from_ num_Modifications_33 dobj_Name_Modifications prep_Modifications_to det_Next_the nn_Next_H&K nn_Next_Model pobj_to_Next nsubj_discuss_we dep_Name_discuss det_weaknesses_the amod_weaknesses_potential dobj_discuss_weaknesses prep_weaknesses_of nn_model_H&K nn_model_s pobj_of_model cc_discuss_and conj_discuss_propose num_modifications_three dobj_propose_modifications prep_propose_to pobj_to_it ' +o,5174,'Our human word alignments do not distinguish between Sure and Probable links ',Och,'poss_alignments_Our amod_alignments_human nn_alignments_word nsubj_distinguish_alignments aux_distinguish_do neg_distinguish_not prep_distinguish_between amod_links_Sure cc_Sure_and conj_Sure_Probable pobj_between_links ' +o,5175,'521 Generate English Annotated Corpus from Wikipedia Wikipedia provides a variety of data resources for NER and other NLP research ',Richman,'nn_Corpus_521 nn_Corpus_Generate nn_Corpus_English nn_Corpus_Annotated nsubj_provides_Corpus prep_Corpus_from nn_Wikipedia_Wikipedia pobj_from_Wikipedia det_variety_a dobj_provides_variety prep_variety_of nn_resources_data pobj_of_resources prep_resources_for pobj_for_NER cc_NER_and amod_research_other nn_research_NLP conj_NER_research ' +p,5176,'We compare semisupervised LEAF with a previous state of the art semi-supervised system ',Fraser,'nsubj_compare_We amod_LEAF_semisupervised dobj_compare_LEAF prep_compare_with det_state_a amod_state_previous pobj_with_state prep_state_of det_system_the nn_system_art amod_system_semi-supervised pobj_of_system ' +p,5177,'The main application of these techniques to written input has been in the robust , lexical tagging of corpora with part-of-speech labels ',Cutting,'det_application_The amod_application_main nsubj_been_application prep_application_of det_techniques_these pobj_of_techniques prep_techniques_to amod_input_written pobj_to_input aux_been_has prep_been_in det_tagging_the amod_tagging_robust amod_tagging_lexical pobj_in_tagging prep_tagging_of pobj_of_corpora prep_been_with amod_labels_part-of-speech pobj_with_labels ' +o,5178,'Results are reported using lowercase BLEU ',Papineni,'nsubjpass_reported_Results auxpass_reported_are xcomp_reported_using amod_BLEU_lowercase dobj_using_BLEU ' +o,5179,'We have processed the Susanne corpus and Penn treebank to provide tables of word and subtree alignments ',Marcus,'nsubj_processed_We aux_processed_have det__the nn__Susanne nn__corpus dobj_processed_ cc__and nn__Penn nn__treebank conj__ aux_provide_to xcomp_processed_provide dobj_provide_tables prep_tables_of nn_alignments_word cc_word_and conj_word_subtree pobj_of_alignments ' +o,5180,'Parameters were tuned with MERT algorithm on the NIST evaluation set of 2003 -LRB- MT03 -RRB- for both the baseline systems and the system combination model ',Och,'nsubjpass_tuned_Parameters auxpass_tuned_were prep_tuned_with amod__MERT nn__algorithm pobj_with_ prep__on det_evaluation_the amod_evaluation_NIST pobj_on_evaluation partmod_evaluation_set prep_set_of num_MT03_2003 pobj_of_MT03 prep_MT03_for preconj_systems_both det_systems_the nn_systems_baseline pobj_for_systems cc_systems_and det_model_the nn_model_system nn_model_combination conj_systems_model ' +o,5181,'Any way to enforce linguistic constraints will result in a reduced need for data , and ultimately in more complete models , given the same amount of data ',Koehn,'det_way_Any nsubj_result_way aux_enforce_to infmod_way_enforce amod_constraints_linguistic dobj_enforce_constraints aux_result_will prep_result_in det_need_a amod_need_reduced pobj_in_need prep_need_for pobj_for_data cc_result_and conj_result_ultimately prep_ultimately_in advmod_complete_more amod_models_complete pobj_in_models partmod_models_given det_amount_the amod_amount_same dobj_given_amount prep_amount_of pobj_of_data ' +o,5182,'1 Introduction Phrase-based modeling method is a simple , but powerful mechanism to machine translation since it can model local reorderings and translations of multiword expressions well ',Och,'num__1 nn__Introduction amod__Phrase-based nn__modeling nn__method nsubj_mechanism_ cop_mechanism_is det_mechanism_a amod_mechanism_simple cc_simple_but conj_simple_powerful prep_mechanism_to nn_translation_machine pobj_to_translation mark_model_since nsubj_model_it aux_model_can advcl_mechanism_model amod_reorderings_local dobj_model_reorderings cc_reorderings_and conj_reorderings_translations prep_reorderings_of amod_expressions_multiword pobj_of_expressions advmod_model_well ' +o,5183,'In the area of statistical machine translation -LRB- SMT -RRB- , recently a combination of the BLEU evaluation metric and the bootstrap method for statistical significance testing has become popular ',Och,'prep_become_In det_area_the pobj_In_area prep_area_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_SMT advmod_combination_recently det_combination_a nsubj_become_combination prep_combination_of det__the amod__BLEU nn__evaluation amod__metric pobj_of_ cc_combination_and det_method_the nn_method_bootstrap conj_combination_method prep_method_for amod__statistical nn__significance nn__testing pobj_for_ aux_become_has dep_become_popular ' +o,5184,'It can be proven that the probability distribution p satisfying the above assumption is the one with the highest entropy, is unique and has the following expone ntial form (Berger et al. 1996): (1) = = k j cajf jcZcap 1 ),( )( 1)|( a where Z(c) is a normalization factor, fj(a,c) are the values of k features of the pair (a,c) and correspond to the linguistic cues of c that are relevant to predict the outcome a. Features are extracted from the training data and define the constraints that the probabilistic model p must satisfy.',Berger,'' +o,5185,'It generates a vector of 5 numeric values for each phrase pair: phrase translation probability: ( f|e) = count( f, e) count(e),(e| f) = count( f, e) count( f) 2http://www.phramer.org/ Java-based open-source phrase based SMT system 3http://www.isi.edu/licensed-sw/carmel/ 4http://www.speech.sri.com/projects/srilm/ 5http://www.iccs.inf.ed.ac.uk/pkoehn/training.tgz 150 lexical weighting (Koehn et al. , 2003): lex( f|e,a) = nproductdisplay i=1 1 |{j|(i, j) a}| summationdisplay (i,j)a w(fi|ej) lex(e|f,a) = mproductdisplay j=1 1 |{i|(i, j) a}| summationdisplay (i,j)a w(ej|fi) phrase penalty: ( f|e) = e; log(( f|e)) = 1 2.2 Decoding We used the Pharaoh decoder for both the Minimum Error Rate Training (Och, 2003) and test dataset decoding.',Koehn,'' +o,5186,'This algorithm and its many variants are widely used in the computational linguistics community ',Collins,'det_algorithm_This nsubjpass_used_algorithm cc_algorithm_and poss_variants_its amod_variants_many conj_algorithm_variants auxpass_used_are advmod_used_widely prep_used_in det_community_the amod_community_computational nn_community_linguistics pobj_in_community ' +o,5187,'32 73 Unknown Words and Parts of Speech When the parser encounters an unknown word , the first-best tag delivered by tagger is used ',Ratnaparkhi,'num_Words_32 num_Words_73 nn_Words_Unknown nsubjpass_used_Words cc_Words_and conj_Words_Parts prep_Words_of pobj_of_Speech advmod_encounters_When det_parser_the nsubj_encounters_parser dep_Speech_encounters det_word_an amod_word_unknown dobj_encounters_word det_tag_the amod_tag_first-best appos_word_tag partmod_tag_delivered prep_delivered_by amod_tagger_ pobj_by_tagger auxpass_used_is ' +o,5188,'Recently , sentiment classification has become popular because of its wide applications ',Pang,'advmod_become_Recently nn_classification_sentiment nsubj_become_classification aux_become_has dep_become_popular mark_of_because advcl_become_of poss_applications_its amod_applications_wide pobj_of_applications ' +o,5189,'This information can be annotated reliably -LRB- a1a3a2a5a4a7a6a9a8 a10a12a11a14a13a16a15 and a1a17a2a5a4a19a18a20a8 a10a12a11a14a13a16a21 -RRB- 4 4Following , we use the a22 statistic to estimate reliability of annotation ',Carletta,'det_information_This nsubjpass_annotated_information aux_annotated_can auxpass_annotated_be advmod_annotated_reliably amod_a10a12a11a14a13a16a15_a1a3a2a5a4a7a6a9a8 dep_annotated_a10a12a11a14a13a16a15 cc_a10a12a11a14a13a16a15_and nn_a10a12a11a14a13a16a21_a1a17a2a5a4a19a18a20a8 conj_a10a12a11a14a13a16a15_a10a12a11a14a13a16a21 num_4Following_4 measure__4Following ccomp_use_ nsubj_use_we dep_annotated_use det_statistic_the amod_statistic_a22 dobj_use_statistic aux_estimate_to infmod_statistic_estimate dobj_estimate_reliability prep_reliability_of pobj_of_annotation ' +o,5190,'a0 subsequence S1 S2 a0 subsequence S1 S2 a0 subsequence S1 S2 Becoming 1 1 Becoming-is a1 a2 a1 a2 astronaut-DREAM 0 a1 a2 DREAM 1 1 Becoming-my a1a4a3a5a1a4a3 astronaut-ambition 0 a1 a2 SPACEMAN 1 1 SPACEMAN-DREAM a1a4a3a5a1 a2 astronaut-is 0 1 a 1 0 SPACEMAN-ambition 0 a1 a2 astronaut-my 0 a1 ambition 0 1 SPACEMAN-dream a1 a3 0 cosmonaut-DREAM a1 a3 0 1 an 0 1 SPACEMAN-great a1 a2 0 cosmonaut-dream a1 a3 0 astronaut 0 1 SPACEMAN-is 1 1 cosmonaut-great a1 a2 0 cosmonaut 1 0 SPACEMAN-my a1a6a1 cosmonaut-is 1 0 dream 1 0 a-DREAM a1 a7 0 cosmonaut-my a1 0great 1 0 a-SPACEMAN 1 0 great-DREAM 1 0 is 1 1 2 a-cosmonaut 1 0 2 great-dream 1 0 my 1 1 a-dream a1 a7 0 is-DREAM a1 a2 a1 Becoming-DREAM a1a4a8a5a1 a7 a-great a1 a3 0 is-ambition 0 a1 Becoming-SPACEMAN a1a6a1 a-is a1 0 is-dream a1 a2 0 Becoming-a 1 0 a-my a1 a2 0 is-great a1 0 Becoming-ambition 0 a1 a7 an-DREAM 0 a1 a3 is-my 1 1 2 Becoming-an 0 1 an-SPACEMAN 0 1 my-DREAM a1 1 Becoming-astronaut 0 a1 an-ambition 0 a1 a3 my-ambition 0 1 Becoming-cosmonaut a1 0 an-astronaut 0 1 my-dream a1 0 Becoming-dream a1a4a8 0 an-is 0 a1 my-great 1 0 Becoming-great a1 a7 0 an-my 0 a1 a2 2002; Lin and Hovy, 2003; Lin, 2004a; Lin, 2004b; Soricut and Brill, 2004).',Lin,'' +o,5191,'As expected , as we double the size of the data , the BLEU score increases ',Papineni,'prep_score_As pcomp_As_expected mark_double_as nsubj_double_we dep_expected_double det_size_the dobj_double_size prep_size_of det_data_the pobj_of_data det_BLEU_the nsubj_score_BLEU amod_increases_ dobj_score_increases ' +o,5192,'In Section 3 we review s method for recovering English NLDs in treebank-based LFG approximations ',Cahill,'prep_review_In pobj_In_Section num_Section_3 nsubj_review_we nsubj_method_ dep_method_s ccomp_review_method prep_method_for pcomp_for_recovering nn_NLDs_English dobj_recovering_NLDs prep_NLDs_in amod_approximations_treebank-based nn_approximations_LFG pobj_in_approximations ' +o,5193,'The use of structured prediction to SMT is also investigated by ',Tillmann,'det_use_The nsubjpass_investigated_use prep_use_of amod_prediction_structured pobj_of_prediction prep_prediction_to pobj_to_SMT auxpass_investigated_is advmod_investigated_also prep_investigated_by ' +o,5194,'There has also been previous work on determining whether a given text is factual or expresses opinion ; again this work uses a binary distinction , and supervised rather than unsupervised approaches ',Pang,'expl_been_There aux_been_has advmod_been_also amod_work_previous dobj_been_work prep_work_on pcomp_on_determining complm_factual_whether det_text_a amod_text_given nsubj_factual_text cop_factual_is ccomp_determining_factual cc_been_or conj_been_expresses nn__opinion dobj_expresses_ advmod_uses_again det_work_this nsubj_uses_work parataxis_been_uses det_distinction_a amod_distinction_binary dobj_uses_distinction cc_distinction_and amod_approaches_supervised cc_supervised_rather dep_rather_than conj_supervised_unsupervised conj_distinction_approaches ' +o,5195,'Finally, methods in the literature more focused on a specific disambiguation task include statistical methods for the attachment of hyponyms under the most likely hypernym in the WordNet taxonomy (Snow et al., 2006), structural approaches based on semantic clusters and distance metrics (Pennacchiotti and Pantel, 2006), supervised machine learning methods for the disambiguation of meronymy relations (Girju et al., 2003), etc. 6 Conclusions In this paper we presented a novel approach to disambiguate the glosses of computational lexicons and machine-readable dictionaries, with the aim of alleviating the knowledge acquisition bottleneck.',Snow,'' +o,5196,'Then we compute the same ratio of machine translation sentence to source sentence , and take the output of p-norm function as a feature : -RRB- __ \\/ __ -LRB- -RRB- -LRB- s csrcoflengthtoflenght Ptf norm = -LRB- 7 -RRB- Features based on parse score The usual practice to model the wellformedness of a sentence is to employ the n-gram language model or compute the syntactic structure similarity ',Liu,'nsubj_compute_we rcmod_Then_compute det_ratio_the amod_ratio_same dobj_compute_ratio prep_ratio_of nn_sentence_machine nn_sentence_translation pobj_of_sentence prep_compute_to nn_sentence_source pobj_to_sentence cc_compute_and conj_compute_take det_output_the dobj_take_output prep_output_of amod_function_p-norm pobj_of_function prep_output_as det_feature_a pobj_as_feature dep_Features___ punct_Features_\\/ dep_Ptf___ dep_Ptf_s dep_Ptf_csrcoflengthtoflenght amod_Features_Ptf nn_Features_norm nn_Features_= appos_Features_7 dep_Then_Features partmod_Features_based prep_based_on amod_score_parse pobj_on_score det_practice_The amod_practice_usual nsubj_is_practice aux_model_to infmod_practice_model det_wellformedness_the dobj_model_wellformedness prep_wellformedness_of det_sentence_a pobj_of_sentence rcmod_score_is aux_employ_to xcomp_is_employ det_model_the amod_model_n-gram nn_model_language dobj_employ_model cc_employ_or conj_employ_compute det_similarity_the amod_similarity_syntactic nn_similarity_structure dobj_compute_similarity ' +o,5197,' and Jing -LRB- 2000 -RRB- propose a cut-and-paste strategy as a computational process of automatic abstracting and a sentence reduction strategy to produce concise sentences ',Jing,'nsubj_propose_ cc__and conj__Jing appos__2000 det_strategy_a amod_strategy_cut-and-paste dobj_propose_strategy prep_propose_as det_process_a amod_process_computational pobj_as_process prep_process_of amod_abstracting_automatic pobj_of_abstracting cc_abstracting_and det_strategy_a nn_strategy_sentence nn_strategy_reduction conj_abstracting_strategy aux_produce_to xcomp_propose_produce amod_sentences_concise dobj_produce_sentences ' +o,5198,'In this paper , we employed the Chinese word segmentation tool that achieved about 093-096 recall\\/precision rates in the SIGHAN-3 word segmentation task ',Levow,'prep_employed_In det_paper_this pobj_In_paper nsubj_employed_we det__the amod__Chinese nn__word nn__segmentation nn__tool dobj_employed_ nsubj_achieved_that rcmod__achieved quantmod_093-096_about num_rates_093-096 amod_rates_recall\\/precision dobj_achieved_rates prep_rates_in det_task_the nn_task_SIGHAN-3 nn_task_word nn_task_segmentation pobj_in_task ' +n,5199,'In addition , the performance of the adapted model for Joint S&T obviously surpass that of , which achieves an F1 of 9341 \% for Joint S&T , although with more complicated models and features ',Jiang,'prep_surpass_In pobj_In_addition det_performance_the nsubj_surpass_performance prep_performance_of det_model_the amod_model_adapted pobj_of_model prep_model_for nn_S&T_Joint pobj_for_S&T advmod_surpass_obviously dobj_surpass_that prep_that_of pobj_of_ nsubj_achieves_which rcmod__achieves det_F1_an dobj_achieves_F1 prep_F1_of num_\%_9341 pobj_of_\% prep_\%_for nn_S&T_Joint pobj_for_S&T mark_with_although dep_\%_with advmod_complicated_more amod_models_complicated pobj_with_models cc_surpass_and conj_surpass_features ' +p,5200,'2 Head Lexicalization As previously shown -LRB- Charniak , , Carroll and Rooth -LRB- 1998 -RRB- , etc -RRB- , ContextFree Grammars -LRB- CFGs -RRB- can be transformed to lexicalized CFGs , provided that a head-marking scheme for rules is given ',Collins,'num_Lexicalization_2 nn_Lexicalization_Head nsubj_provided_Lexicalization mark_shown_As advmod_shown_previously advcl_transformed_shown nn__Charniak dep_shown_ nn__ conj__ conj__Carroll cc__and conj__Rooth appos_Rooth_1998 appos_Rooth_etc nn_Grammars_ContextFree nsubjpass_transformed_Grammars abbrev_Grammars_CFGs aux_transformed_can auxpass_transformed_be rcmod_Lexicalization_transformed prep_transformed_to amod_CFGs_lexicalized pobj_to_CFGs complm_given_that det_scheme_a amod_scheme_head-marking nsubjpass_given_scheme prep_scheme_for pobj_for_rules auxpass_given_is ccomp_provided_given ' +o,5201,'52 Bleu : Automatic Evaluation BLEU is a system for automatic evaluation of machine translation ',Papineni,'num_Bleu_52 dep_system_Bleu nn__Automatic nn__Evaluation nn__BLEU nsubj_system_ cop_system_is det_system_a prep_system_for amod_evaluation_automatic pobj_for_evaluation prep_evaluation_of nn_translation_machine pobj_of_translation ' +o,5202,'The part of the 1Release 2 of this data set can be obtained t ` rmn the Linguistic Data Consortium with Catalogue number LDC94T4B -LRB- http://wwwldcupennedu/ldc/nofranmhtml -RRB- 2There are 48 labels defined in , however , three of ttmm do not appear in the corpus ',Marcus,'det_part_The nsubjpass_t_part prep_part_of det_2_the amod_2_1Release pobj_of_2 prep_2_of det_set_this nn_set_data pobj_of_set aux_t_can auxpass_t_be dep_t_obtained nn_2There_rmn det_Consortium_the nn_Consortium_Linguistic nn_Consortium_Data nn_2There_Consortium prep_Consortium_with amod_LDC94T4B_Catalogue nn_LDC94T4B_number pobj_with_LDC94T4B appos_LDC94T4B_http://wwwldcupennedu/ldc/nofranmhtml nsubjpass_defined_2There auxpass_defined_are num_labels_48 dep_defined_labels dep_t_defined prep_defined_in pobj_in_ advmod_appear_however nsubj_appear_three prep_three_of pobj_of_ttmm aux_appear_do neg_appear_not dep_defined_appear prep_appear_in det_corpus_the pobj_in_corpus ' +o,5203,' , we used the MXPOST tagger trained on training data to provide part-of-speech tags for the development and the test set , and we used 10way jackknifing to generate tags for the training set ',Ratnaparkhi,'advmod_used_ nsubj_used_we det_tagger_the nn_tagger_MXPOST nn_tagger_ dobj_used_tagger partmod_tagger_trained prep_trained_on nn_data_training pobj_on_data aux_provide_to xcomp_trained_provide amod_tags_part-of-speech dobj_provide_tags prep_provide_for det_development_the pobj_for_development cc_used_and det_test_the nsubj_set_test conj_used_set cc_used_and nsubj_used_we conj_used_used dobj_used_10way partmod_10way_jackknifing aux_generate_to xcomp_jackknifing_generate dobj_generate_tags prep_tags_for det_set_the nn_set_training pobj_for_set ' +o,5204,'4.1 Baseline Our baseline system is a fairly typical phrasebased machine translation system (Finch and Sumita, 2008a) built within the framework of a feature-based exponential model containing the following features: Table 1: Language Resources Corpus Train Dev Eval NC Spanish sentences 74K 2,001 2,007 words 2,048K 49,116 56,081 vocab 61K 9,047 8,638 length 27.6 24.5 27.9 OOV (\%) 5.2 / 2.9 1.4 / 0.9 English sentences 74K 2,001 2,007 words 1,795K 46,524 49,693 vocab 47K 8,110 7,541 length 24.2 23.2 24.8 OOV (\%) 5.2 / 2.9 1.2 / 0.9 perplexity 349 / 381 348 / 458 EP Spanish sentences 1,404K 1,861 2,000 words 41,003K 50,216 61,293 vocab 170K 7,422 8,251 length 29.2 27.0 30.6 OOV (\%) 2.4 / 0.1 2.4 / 0.2 English sentences 1,404K 1,861 2,000 words 39,354K 48,663 59,145 vocab 121K 5,869 6,428 length 28.0 26.1 29.6 OOV (\%) 1.8 / 0.1 1.9 / 0.1 perplexity 210 / 72 305 / 125 Table 2: Testset 2009 Corpus Test NC Spanish sentences 3,027 words 80,591 vocab 12,616 length 26.6 Source-target phrase translation probability Inverse phrase translation probability Source-target lexical weighting probability Inverse lexical weighting probability Phrase penalty Language model probability Lexical reordering probability Simple distance-based distortion model Word penalty For the training of the statistical models, standard word alignment (GIZA++ (Och and Ney, 2003)) and language modeling (SRILM (Stolcke, 2002)) tools were used.',Och,'' +p,5205,'Other well-known metrics are WER , NIST , GTM , ROUGE , METEOR , and TER , just to name a few ',Banerjee,'amod_metrics_Other amod_metrics_well-known nsubj__metrics cop__are nn__WER nn__NIST conj__ nn__GTM conj__ nn__ROUGE conj__ nn__METEOR conj__ cc__and nn__TER conj__ advmod_name_just aux_name_to xcomp__name det_few_a dobj_name_few ' +o,5206,'The classifier uses mutual information -LRB- MI -RRB- scores rather than the raw frequences of the occurring patterns ',Church,'det_classifier_The nsubj_uses_classifier amod_scores_mutual nn_scores_information abbrev_scores_MI dobj_uses_scores cc_scores_rather dep_rather_than det_frequences_the amod_frequences_raw conj_scores_frequences prep_frequences_of det_patterns_the amod_patterns_occurring pobj_of_patterns ' +o,5207,'31 Golden-standard-based criteria In the domain of machine translation systems , an increasingly accepted way to measure the quality of a system is to compare the outputs it produces with a set of reference translations , considered as an approximation of a golden standard ',Papineni,'num_criteria_31 amod_criteria_Golden-standard-based dep_In_criteria prep_is_In det_domain_the pobj_In_domain prep_domain_of nn_systems_machine nn_systems_translation pobj_of_systems det_way_an advmod_way_increasingly amod_way_accepted nsubj_is_way aux_measure_to infmod_way_measure det_quality_the dobj_measure_quality prep_quality_of det_system_a pobj_of_system aux_compare_to xcomp_is_compare det_outputs_the dobj_compare_outputs nsubj_produces_it dep_is_produces prep_produces_with det_set_a pobj_with_set prep_set_of nn_translations_reference pobj_of_translations partmod_set_considered prep_considered_as det_approximation_an pobj_as_approximation prep_approximation_of det_standard_a amod_standard_golden pobj_of_standard ' +o,5208,'Also , PMI-IR is useful for calculating semantic orientation and rating reviews ',Turney,'advmod_useful_Also nsubj_useful_PMI-IR cop_useful_is prep_useful_for pcomp_for_calculating amod_orientation_semantic dobj_calculating_orientation cc_orientation_and nn_reviews_rating conj_orientation_reviews ' +o,5209,'Using this heuristic , BABAR identifies existential definite NPs in the training corpus using our previous learning algorithm and resolves all occurrences of the same existential NP with each another ',Bean,'dep_identifies_Using det_heuristic_this dobj_Using_heuristic nsubj_identifies_BABAR amod_NPs_existential amod_NPs_definite dobj_identifies_NPs prep_identifies_in det_corpus_the nn_corpus_training pobj_in_corpus partmod_corpus_using poss__our amod__previous nn__learning nn__algorithm dobj_using_ cc_identifies_and conj_identifies_resolves det_occurrences_all dobj_resolves_occurrences prep_occurrences_of det_NP_the amod_NP_same amod_NP_existential pobj_of_NP prep_resolves_with dep_another_each pobj_with_another ' +o,5210,'6 Related Work A description of the IBM models for statistical machine translation can be found in ',Brown,'num_Work_6 amod_Work_Related det_description_A nsubjpass_found_description prep_description_of det_models_the nn_models_IBM pobj_of_models prep_models_for amod_translation_statistical nn_translation_machine pobj_for_translation aux_found_can auxpass_found_be dep_Work_found prep_found_in ' +o,5211,'This is similar to work by several other groups which aims to induce semantic classes through syntactic co-occurrence analysis , although in our case the contexts are limited to selected patterns , relevant to the scenario ',Dagan,'nsubj_similar_This cop_similar_is aux_work_to xcomp_similar_work prep_work_by amod_groups_several amod_groups_other pobj_by_groups nsubj_aims_which rcmod_groups_aims aux_induce_to xcomp_aims_induce amod_classes_semantic dobj_induce_classes prep_induce_through amod__syntactic amod__co-occurrence nn__analysis pobj_through_ mark_limited_although prep_limited_in poss_case_our pobj_in_case det_contexts_the nsubj_limited_contexts cop_limited_are advcl_similar_limited prep_limited_to amod_patterns_selected pobj_to_patterns acomp_limited_relevant prep_relevant_to det_scenario_the pobj_to_scenario ' +o,5212,' proposed a structured model based on CRFs for jointly classifying the sentiment of text at varying levels of granularity ',McDonald,'nsubj_proposed_ det_model_a amod_model_structured dobj_proposed_model prep_proposed_based dep_based_on pobj_on_CRFs prep_proposed_for advmod_classifying_jointly pcomp_for_classifying det_sentiment_the dobj_classifying_sentiment prep_sentiment_of pobj_of_text prep_proposed_at amod_levels_varying pobj_at_levels prep_levels_of pobj_of_granularity ' +n,5213,'Some are the result of inconsistency in labeling in the training data , which usually reflects a lack of linguistic clarity or determination of the correct part of speech in context ',Ratnaparkhi,'nsubj_result_Some cop_result_are det_result_the prep_result_of pobj_of_inconsistency prep_inconsistency_in pcomp_in_labeling prep_labeling_in det__the nn__training nn__data pobj_in_ nsubj_reflects_which advmod_reflects_usually rcmod__reflects det_lack_a dobj_reflects_lack prep_lack_of amod_clarity_linguistic pobj_of_clarity cc_clarity_or conj_clarity_determination prep_lack_of det_part_the amod_part_correct pobj_of_part prep_part_of pobj_of_speech prep_speech_in pobj_in_context ' +o,5214,'164 and Itai , 1990 ; Dagan et al , 1995 ; Kennedy and Boguraev , 1996a ; Kennedy and Boguraev , 1996b -RRB- ',Berger,'cc_164_and conj_164_Itai appos_164_1990 dep_164_Dagan cc_Dagan_et conj_Dagan_al appos_Dagan_1995 dep_164_Kennedy cc_Kennedy_and conj_Kennedy_Boguraev appos_Kennedy_1996a dep_Kennedy_Kennedy cc_Kennedy_and conj_Kennedy_Boguraev appos_Kennedy_1996b ' +o,5215,'We then scored each query pair -LRB- q1 , q2 -RRB- in this subset using the log-likelihood ratio between q1 and q2 , which measures the mutual dependence within the context of web search queries ',Dunning,'nsubj_scored_We advmod_scored_then det__each advmod_pair_query amod__pair dep_pair_q1 dep_q1_q2 prep_pair_in det_subset_this pobj_in_subset partmod_subset_using det__the amod__log-likelihood nn__ratio dobj_using_ prep__between pobj_between_q1 cc_q1_and conj_q1_q2 nsubj_measures_which rcmod_q1_measures det_dependence_the amod_dependence_mutual dobj_measures_dependence prep_measures_within det_context_the pobj_within_context prep_context_of nn_queries_web nn_queries_search pobj_of_queries dobj_scored_ ' +o,5216,'Using the log-linear form to model p -LRB- e f -RRB- gives us the flexibility to introduce overlapping features that can represent global context while decoding -LRB- searching the space of candidate translations -RRB- and rescoring -LRB- ranking a set of candidate translations before performing the argmax operation -RRB- , albeit at the cost of the traditional source-channel generative model of translation proposed in ',Brown,'det_form_the amod_form_log-linear dobj_Using_form prep_Using_to nn_p_model pobj_to_p dep_f_e dep_-LRB-_f dep_Using_gives iobj_gives_us det_flexibility_the dobj_gives_flexibility aux_introduce_to infmod_flexibility_introduce amod_features_overlapping dobj_introduce_features nsubj_represent_that aux_represent_can rcmod_features_represent amod_context_global dobj_represent_context prep_represent_while pcomp_while_decoding parataxis_decoding_searching det_space_the dobj_searching_space prep_space_of nn_translations_candidate pobj_of_translations cc_decoding_and conj_decoding_rescoring parataxis_decoding_ranking det_set_a dobj_ranking_set prep_set_of nn_translations_candidate pobj_of_translations prep_ranking_before pcomp_before_performing det_operation_the nn_operation_argmax dobj_performing_operation advmod_represent_albeit dep_albeit_at det_cost_the pobj_at_cost prep_cost_of det_generative_the amod_generative_traditional amod_generative_source-channel pobj_of_generative nsubj_Using_model prep_model_of pobj_of_translation partmod_translation_proposed prep_proposed_in ' +o,5217,'Research in the first category aims to identify specific types of nonanaphoric phrases , with some identifying pleonastic it -LRB- using heuristics -LRB- eg , Paice and Husk , Lappin and Leass , Kennedy and Boguraev -RRB- , supervised approaches -LRB- eg , Evans , Muller , Versley et al ',Berger,'nsubj_aims_Research prep_Research_in det_category_the amod_category_first pobj_in_category ccomp_supervised_aims aux_identify_to xcomp_aims_identify amod_types_specific dobj_identify_types prep_types_of amod_phrases_nonanaphoric pobj_of_phrases prep_identify_with pobj_with_some partmod_some_identifying dep_identifying_pleonastic dep_pleonastic_it dep_some_using nn__heuristics nn__eg conj_eg_Paice cc_eg_and conj_eg_Husk nn__ conj__Lappin cc__and conj__Leass nn__ conj__Kennedy cc__and conj__Boguraev dobj_using_ nsubj_supervised_approaches dep_approaches_eg nn__Evans dep_eg_ nn__Muller appos_approaches_ conj__Versley cc__et conj__al ' +o,5218,'Relative frequency ratio -LRB- RFR -RRB- of terms between two different corpora can also be used to discover domain-oriented multi-word terms that are characteristic of a corpus when compared with another ',Dunning,'amod_ratio_Relative nn_ratio_frequency nsubjpass_used_ratio abbrev_ratio_RFR prep_ratio_of pobj_of_terms prep_terms_between num_corpora_two amod_corpora_different pobj_between_corpora aux_used_can advmod_used_also auxpass_used_be aux_discover_to purpcl_used_discover amod_terms_domain-oriented amod_terms_multi-word dobj_discover_terms nsubj_characteristic_that cop_characteristic_are rcmod_terms_characteristic prep_characteristic_of det_corpus_a pobj_of_corpus advmod_compared_when advcl_characteristic_compared prep_compared_with det__another pobj_with_ ' +o,5219,'Semantic features are used for classifying entities into semantic types such as name of person , organization , or place , while syntactic features characterize the kinds of dependency 5It is worth noting that the present approach can be recast into one based on constraint relaxation ',Tromble,'amod_features_Semantic nsubjpass_used_features auxpass_used_are prep_used_for pcomp_for_classifying dobj_classifying_entities prep_classifying_into amod_types_semantic pobj_into_types dep_as_such prep_types_as pobj_as_name prep_name_of nn_organization_person pobj_of_organization cc_organization_or conj_organization_place mark_characterize_while amod_features_syntactic nsubj_characterize_features dep_organization_characterize det_kinds_the nsubj_worth_kinds prep_kinds_of amod_5It_dependency pobj_of_5It cop_worth_is ccomp_characterize_worth xcomp_worth_noting complm_recast_that det_approach_the amod_approach_present nsubjpass_recast_approach aux_recast_can auxpass_recast_be ccomp_noting_recast prep_recast_into pobj_into_one prep_recast_based dep_based_on nn_relaxation_constraint pobj_on_relaxation ' +o,5220,'Other factors that distinguish us from previous work are the use of all phrases proposed by a phrase-based system , and the use of a dependency language model that also incorporates constituent information -LRB- although see for related approaches -RRB- ',Shen,'amod_factors_Other complm_use_that csubj_use_distinguish dobj_distinguish_us prep_distinguish_from amod_work_previous pobj_from_work cop_use_are det_use_the dep_factors_use prep_use_of det_phrases_all pobj_of_phrases partmod_phrases_proposed prep_proposed_by det_system_a amod_system_phrase-based pobj_by_system cc_factors_and det_use_the conj_factors_use prep_use_of det_model_a amod_model_dependency nn_model_language pobj_of_model nsubj_incorporates_that advmod_incorporates_also rcmod_use_incorporates amod_information_constituent dobj_incorporates_information mark_see_although dep_factors_see dobj_see_ prep__for amod_approaches_related pobj_for_approaches ' +o,5221,'Discriminative training with hidden variables has been handled in this probabilistic framework , but we choose Equation 3 for efficiency ',Koo,'amod_training_Discriminative nsubjpass_handled_training prep_training_with amod_variables_hidden pobj_with_variables aux_handled_has auxpass_handled_been prep_handled_in det__this amod__probabilistic nn__framework pobj_in_ cc_handled_but nsubj_choose_we conj_handled_choose dobj_choose_Equation num_Equation_3 prep_Equation_for pobj_for_efficiency ' +o,5222,'Dependency representation has been used for language modeling , textual entailment and machine translation , to name a few tasks ',Shen,'amod_representation_Dependency nsubjpass_used_representation aux_used_has auxpass_used_been prep_used_for nn_modeling_language pobj_for_modeling amod_entailment_textual conj_modeling_entailment cc_modeling_and nn__machine nn__translation conj_modeling_ aux_name_to xcomp_used_name det_tasks_a amod_tasks_few dobj_name_tasks ' +o,5223,'Many adaptation methods operate by simple augmentations of the target feature space , as we have donehere ','Daume III','amod_methods_Many nn_methods_adaptation nsubj_operate_methods prep_operate_by amod_augmentations_simple pobj_by_augmentations prep_augmentations_of det_space_the nn_space_target nn_space_feature pobj_of_space mark_have_as nsubj_have_we advcl_operate_have amod__donehere dobj_have_ ' +o,5224,'17 The justification for this is that there is an estimated 3 \% error rate in the hand-assigned POS tags in the treebank , and we didnt want this noise to contribute to dependency errors ',Ratnaparkhi,'det_justification_The nsubj_is_justification prep_justification_for pobj_for_this dep_17_is complm_is_that expl_is_there ccomp_is_is det_rate_an dep_3_estimated dep_\%_3 dep_rate_\% nn_rate_error nsubj_is_rate prep_rate_in det_tags_the amod_tags_hand-assigned dep_tags_POS pobj_in_tags prep_tags_in det__the nn__treebank pobj_in_ cc_is_and nsubj_didnt_we conj_is_didnt xcomp_didnt_want det_noise_this dobj_want_noise aux_contribute_to infmod_noise_contribute prep_contribute_to amod_errors_dependency pobj_to_errors ' +o,5225,'We develop this intuition into a technique called synchronous binarization which binarizes a synchronous production or treetranduction rule on both source and target sides simultaneously ',Zhang,'nsubj_develop_We det_intuition_this dobj_develop_intuition prep_develop_into det_technique_a pobj_into_technique partmod_technique_called amod__synchronous nn__binarization dobj_called_ nsubj_binarizes_which rcmod__binarizes det_production_a amod_production_synchronous dobj_binarizes_production cc_production_or nn_rule_treetranduction conj_production_rule prep_production_on preconj_sides_both nn_sides_source cc_source_and conj_source_target pobj_on_sides advmod_binarizes_simultaneously ' +o,5226,'To deal with this question , we use ATIS p-o-s trees as found in the Penn Treebank ',Marcus,'aux_deal_To dep_use_deal prep_deal_with det_question_this pobj_with_question nsubj_use_we nn_trees_ATIS amod_trees_p-o-s nsubj_found_trees advmod_trees_as ccomp_use_found prep_found_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +p,5227,'Studies reveal that statistical alignment models outperform the simple Dice coefficient ',Och,'nsubj_reveal_Studies complm_outperform_that amod_models_statistical nn_models_alignment nsubj_outperform_models ccomp_reveal_outperform det_coefficient_the amod_coefficient_simple nn_coefficient_Dice dobj_outperform_coefficient ' +p,5228,'One possible approach is to employ state-of-the-art techniques for coreference and zeroanaphora resolution in preprocessing cooccurrence samples ',Iida,'num_approach_One amod_approach_possible nsubj_is_approach aux_employ_to xcomp_is_employ amod_techniques_state-of-the-art dobj_employ_techniques prep_techniques_for nn__coreference cc_coreference_and conj_coreference_zeroanaphora nn__resolution pobj_for_ prep_employ_in pcomp_in_preprocessing amod_samples_cooccurrence dobj_preprocessing_samples ' +o,5229,'In this paper we present MapReduce implementations of training algorithms for two kinds of models commonly used in statistical MT today : a phrasebased translation model and word alignment models based on pairwise lexical translation trained using expectation maximization ',Koehn,'prep_present_In det_paper_this pobj_In_paper nsubj_present_we amod_implementations_MapReduce dobj_present_implementations prep_implementations_of nn_algorithms_training pobj_of_algorithms prep_present_for num_kinds_two pobj_for_kinds prep_kinds_of pobj_of_models advmod_used_commonly partmod_models_used prep_used_in amod_MT_statistical pobj_in_MT tmod_present_today det__a amod__phrasebased nn__translation nn__model nsubj_based_ cc__and nn_models_word nn_models_alignment conj__models parataxis_present_based prep_based_on amod_translation_pairwise amod_translation_lexical pobj_on_translation partmod_translation_trained xcomp_trained_using nn__expectation nn__maximization dobj_using_ ' +o,5230,'Then the two models and a search module are used to decode the best translation ',Koehn,'dep_used_the num_models_two nsubjpass_used_models cc_models_and det_module_a nn_module_search conj_models_module auxpass_used_are dep_Then_used aux_decode_to xcomp_used_decode det_translation_the amod_translation_best dobj_decode_translation ' +o,5231,'Graph-based algorithms for classification into subjective\\/objective or positive\\/negative language units have been mostly used at the sentence and document level , instead of aiming at dictionary annotation as we do ',Thomas,'amod_algorithms_Graph-based nsubjpass_used_algorithms prep_algorithms_for pobj_for_classification prep_classification_into amod_units_subjective\\/objective cc_subjective\\/objective_or conj_subjective\\/objective_positive\\/negative nn_units_language pobj_into_units aux_used_have auxpass_used_been advmod_used_mostly prep_used_at det_sentence_the pobj_at_sentence cc_sentence_and nn__document nn__level conj_sentence_ dep_of_instead prep_used_of pcomp_of_aiming prep_aiming_at nn_annotation_dictionary pobj_at_annotation mark_do_as nsubj_do_we advcl_aiming_do ' +o,5232,'We used the heuristic combination described in and extracted phrasal translation pairs from this combined alignment as described in ',Och,'nsubj_used_We det_combination_the amod_combination_heuristic dobj_used_combination dep_combination_described prep_described_in pobj_in_ cc_described_and conj_described_extracted amod_pairs_phrasal nn_pairs_translation dobj_extracted_pairs prep_extracted_from det_alignment_this amod_alignment_combined pobj_from_alignment advmod_described_as amod_alignment_described prep_described_in pobj_in_ ' +o,5233,'Most work on corpora of naturally occurring language 244 Michael R Brent From Grammar to Lexicon either uses no a priori grammatical knowledge , or else it relies on a large and complex grammar ',Hindle,'nsubj_work_Most prep_work_on pobj_on_corpora prep_corpora_of advmod_occurring_naturally amod_language_occurring pobj_of_language num_R_244 nn_R_Michael nsubj_Brent_R dep_work_Brent prep_Brent_From pobj_From_Grammar prep_Brent_to pobj_to_Lexicon dep_Brent_either conj_Brent_uses det__no dep_priori_a advmod_grammatical_priori amod__grammatical nn__knowledge dobj_uses_ cc_Brent_or advmod_it_else nsubj_relies_it conj_Brent_relies prep_relies_on det_grammar_a amod_grammar_large cc_large_and conj_large_complex pobj_on_grammar ' +o,5234,' -RRB- -RRB- , and others identifying non-anaphoric definite descriptions -LRB- using rule-based techniques -LRB- eg , Vieira and Poesio -RRB- and unsupervised techniques -LRB- eg , -RRB- -RRB- ',Bean,'nn_others_and dobj__others partmod_others_identifying amod_descriptions_non-anaphoric amod_descriptions_definite dobj_identifying_descriptions dep_others_using amod_techniques_rule-based dobj_using_techniques dep_techniques_eg conj_eg_Vieira cc_eg_and nn__Poesio conj_eg_ cc_techniques_and amod_techniques_unsupervised conj_techniques_techniques appos_techniques_eg dep_eg_ ' +o,5235,'In the following section , we follow the notation in ',Collins,'prep_follow_In det_section_the amod_section_following pobj_In_section nsubj_follow_we det_notation_the dobj_follow_notation prt_follow_in ' +o,5236,'Several approaches have been proposed in the context of word sense disambiguation , named entity -LRB- NE -RRB- classification , patternacquisitionforIE , or dimensionality reduction for text categorization -LRB- TC -RRB- ',Yarowsky,'amod_approaches_Several nsubjpass_proposed_approaches aux_proposed_have auxpass_proposed_been ccomp_named_proposed prep_proposed_in det_context_the pobj_in_context prep_context_of nn__word nn__sense nn__disambiguation pobj_of_ dobj_named_entity abbrev_entity_NE nn__classification nsubj_named_ amod__patternacquisitionforIE conj__ cc__or amod_reduction_dimensionality conj__reduction prep_reduction_for nn_categorization_text pobj_for_categorization abbrev_categorization_TC advmod_categorization_ ' +o,5237,'3 Margin Perceptron Algorithm for Sequence Labeling Weextendedaperceptronwithamargin to sequence labeling in this study , as extended the perceptron algorithm to sequence labeling ',Collins,'num_Margin_3 nn_Algorithm_Perceptron nsubj__Algorithm prep_Algorithm_for nn_Weextendedaperceptronwithamargin_Sequence nn_Weextendedaperceptronwithamargin_Labeling pobj_for_Weextendedaperceptronwithamargin dep_Margin_ prep__to nn_labeling_sequence pobj_to_labeling prep__in det_study_this pobj_in_study mark_extended_as nsubj_extended_ advcl__extended det_algorithm_the nn_algorithm_perceptron dobj_extended_algorithm prep_extended_to nn_labeling_sequence pobj_to_labeling ' +o,5238,'Our work in sentence reformulation is different from cut-and-paste summarization in many ways ',Jing,'poss_work_Our nsubj_different_work prep_work_in nn_reformulation_sentence pobj_in_reformulation cop_different_is prep_different_from amod__cut-and-paste nn__summarization pobj_from_ prep__in amod_ways_many pobj_in_ways ' +o,5239,'Statistical approaches , which depend on a set of unknown parameters that are learned from training data , try to describe the relationship between a bilingual sentence pair ',Berger,'nn_approaches_Statistical nsubj_try_approaches nsubj_depend_which rcmod_approaches_depend prep_depend_on det_set_a pobj_on_set prep_set_of amod_parameters_unknown pobj_of_parameters nsubjpass_learned_that auxpass_learned_are rcmod_set_learned prep_learned_from nn_data_training pobj_from_data aux_describe_to xcomp_try_describe det_relationship_the dobj_describe_relationship prep_relationship_between det__a amod__bilingual nn__sentence nn__pair pobj_between_ ' +o,5240,'To compute the degree of interaction between two proteins D4 BD and D4 BE , we use the information-theoretic measure of pointwise mutual information , which is computed based on the following quantities : 1 ',Church,'aux_compute_To dep_use_compute det_degree_the dobj_compute_degree prep_degree_of pobj_of_interaction prep_interaction_between num_BD_two nn_BD_proteins nn_BD_D4 pobj_between_BD cc_BD_and nn_BE_D4 conj_BD_BE nsubj_use_we det_measure_the amod_measure_information-theoretic dobj_use_measure prep_measure_of amod__pointwise amod__mutual nn__information pobj_of_ nsubj_computed_which cop_computed_is rcmod__computed prep_computed_based dep_based_on det_quantities_the amod_quantities_following pobj_on_quantities attr_computed_1 ' +o,5241,'Most clustering schemes use the average entropy reduction to decide when two words fall into the same cluster ',Brown,'amod__Most amod__clustering nn__schemes nsubj_use_ det_reduction_the amod_reduction_average nn_reduction_entropy nsubj_decide_reduction aux_decide_to xcomp_use_decide advmod_fall_when num_words_two nsubj_fall_words advcl_decide_fall prep_fall_into det_cluster_the amod_cluster_same pobj_into_cluster ' +p,5242,'The evaluation results also confirm the argument of , who suggested G2 as a more robust alternative to X2 ',Dunning,'det_results_The nn_results_evaluation nsubj_confirm_results advmod_confirm_also dep_suggested_confirm det_argument_the dobj_confirm_argument prep_argument_of pobj_of_ nsubj_suggested_who dobj_suggested_G2 prep_suggested_as det_alternative_a advmod_robust_more amod_alternative_robust pobj_as_alternative prep_alternative_to pobj_to_X2 ' +o,5243,'If we assign a probability a13a15a14a17a16 a10a12a11a5a19a18a2 a3a5a21a20 to each pair of strings a16 a10 a11a5a12a22 a2a4a3a5 a20, then according to Bayes decision rule, we have to choose the English string that maximizes the product of the English language model a13a23a14a24a16 a10 a11a5 a20 and the string translation model a13a15a14a17a16a25a2 a3a5a26a18a10a27a11a5a28a20 . Many existing systems for statistical machine translation (Wang and Waibel, 1997; Nieen et al. , 1998; Och and Weber, 1998) make use of a special way of structuring the string translation model like proposed by (Brown et al. , 1993): The correspondence between the words in the source and the target string is described by alignments which assign one target word position to each source word position.',Brown,'' +o,5244,'The Spanish corpus was parsed using the MST dependency parser trained using dependency trees generated from the the English Penn Treebank and Spanish CoNLL-X data ',Marcus,'det_corpus_The amod_corpus_Spanish nsubjpass_parsed_corpus auxpass_parsed_was xcomp_parsed_using det__the nn__MST nn__dependency nn__parser dobj_using_ partmod__trained xcomp_trained_using amod_trees_dependency dobj_using_trees partmod_trees_generated prep_generated_from dep__the det__the nn__English nn__Penn nn__Treebank pobj_from_ cc__and amod__Spanish amod__CoNLL-X nn__data conj__ ' +o,5245,'It is today common practice to use phrases as translation units instead of the original word-based approach ',Koehn,'nsubj_practice_It cop_practice_is advmod_practice_today amod_practice_common aux_use_to infmod_practice_use dobj_use_phrases prep_use_as nn__translation nn__units pobj_as_ dep_of_instead prep__of det_approach_the amod_approach_original amod_approach_word-based pobj_of_approach ' +o,5246,'This setting is reminiscent of the problem of optimizing feature weights for reranking of candidate machine translation outputs , and we employ an optimization technique similar to that used by for machine translation ',Och,'det_setting_This nsubj_reminiscent_setting cop_reminiscent_is prep_reminiscent_of det_problem_the pobj_of_problem prep_problem_of pcomp_of_optimizing nn_weights_feature dobj_optimizing_weights prep_optimizing_for pobj_for_reranking prep_reranking_of nn_outputs_candidate nn_outputs_machine nn_outputs_translation pobj_of_outputs cc_reminiscent_and nsubj_employ_we conj_reminiscent_employ det_technique_an nn_technique_optimization nsubj_similar_technique xcomp_employ_similar prep_similar_to pobj_to_that partmod_that_used prep_used_by pobj_by_ prep__for nn_translation_machine pobj_for_translation ' +o,5247,'For more detail , explanations and experiments see ',Titov,'prep_see_For amod_detail_more pobj_For_detail nsubj_see_explanations cc_explanations_and conj_explanations_experiments ' +n,5248,'Therefore , sublanguage techniques such as Sager and do not work ',Smadja,'advmod_work_Therefore amod_techniques_sublanguage nsubj_work_techniques dep_as_such prep_techniques_as nn__Sager pobj_as_ cc__and conj__ aux_work_do neg_work_not ' +o,5249,'51 Agreement between translators In an attempt to quantify the agreement between the two groups of translators , we computed the Kappa coefficient for annotation tasks , as defined by ',Carletta,'num_Agreement_51 prep_Agreement_between pobj_between_translators dep_Agreement_In det_attempt_an pobj_In_attempt aux_quantify_to infmod_attempt_quantify det_agreement_the dobj_quantify_agreement prep_agreement_between det_groups_the num_groups_two pobj_between_groups prep_groups_of pobj_of_translators nsubj_computed_we parataxis_quantify_computed det_coefficient_the nn_coefficient_Kappa dobj_computed_coefficient prep_computed_for nn_tasks_annotation pobj_for_tasks mark_defined_as advcl_quantify_defined prep_defined_by ' +o,5250,'Using the ME principle , we can combine information from a variety of sources into the same language model ',Berger,'dep_combine_Using det_principle_the nn_principle_ME dobj_Using_principle nsubj_combine_we aux_combine_can dobj_combine_information prep_combine_from det_variety_a pobj_from_variety prep_variety_of pobj_of_sources prep_combine_into det_model_the amod_model_same nn_model_language pobj_into_model ' +o,5251,'CRF (baseline)] 97.18 97.21 Table 7: POS tagging results of the previous top systems for PTB III data evaluated by label accuracy system test additional resources JESS-CM (CRF/HMM) 95.15 1G-word unlabeled data 94.67 15M-word unlabeled data (Ando and Zhang, 2005) 94.39 15M-word unlabeled data (Suzuki et al., 2007) 94.36 17M-word unlabeled data (Zhang et al., 2002) 94.17 full parser output (Kudo and Matsumoto, 2001) 93.91 [supervised CRF (baseline)] 93.88 Table 8: Syntactic chunking results of the previous top systems for CoNLL00 shared task data (F=1 score) 30-31 Aug. 1996 and 6-7 Dec. 1996 Reuters news articles, respectively.',Suzuki,'' +o,5252,'Note that generative hybrids are the norm in SMT , where translation scores are provided by a discriminative combination of generative models ',Och,'complm_norm_that amod_hybrids_generative nsubj_norm_hybrids cop_norm_are det_norm_the ccomp_Note_norm prep_norm_in pobj_in_SMT advmod_provided_where nn_scores_translation nsubjpass_provided_scores auxpass_provided_are rcmod_SMT_provided prep_provided_by det_combination_a amod_combination_discriminative pobj_by_combination prep_combination_of amod_models_generative pobj_of_models ' +o,5253,'For instance , on unsupervised part-ofspeech tagging , EM requires over 100 iterations to reach its peak performance on the Wall-Street Journal ',Johnson,'prep_requires_For pobj_For_instance prep_requires_on amod_tagging_unsupervised amod_tagging_part-ofspeech pobj_on_tagging nsubj_requires_EM prep_requires_over num_iterations_100 pobj_over_iterations aux_reach_to xcomp_requires_reach poss_performance_its amod_performance_peak dobj_reach_performance prep_reach_on det_Journal_the nn_Journal_Wall-Street pobj_on_Journal ' +o,5254,'When evaluated against the state-of-the-art, phrase-based decoder Pharaoh (Koehn, 2004), using the same experimental conditions translation table trained on the FBIS corpus (7.2M Chinese words and 9.2M English words of parallel text), trigram language model trained on 155M words of English newswire, interpolation weights a65 (Equation 2) trained using discriminative training (Och, 2003) (on the 2002 NIST MT evaluation set), probabilistic beam a90 set to 0.01, histogram beam a58 set to 10 and BLEU (Papineni et al. , 2002) as our metric, the WIDL-NGLM-Aa86 a129 algorithm produces translations that have a BLEU score of 0.2570, while Pharaoh translations have a BLEU score of 0.2635.',Koehn,'' +o,5255,'These transtbr rules are pairs of corresponding rooted substructures , where a substructure is a connected set of arcs and nodes ',Brown,'det_rules_These nn_rules_transtbr nsubj_pairs_rules cop_pairs_are prep_pairs_of amod_substructures_corresponding amod_substructures_rooted pobj_of_substructures advmod_set_where det__a amod__substructure nsubj_set_ cop_set_is det_set_a amod_set_connected rcmod_substructures_set prep_set_of pobj_of_arcs cc_arcs_and conj_arcs_nodes ' +o,5256,'Therefore , introduced skip-bigram statistics for the evaluation of machine translation ',Lin,'advmod_introduced_Therefore nsubj_introduced_ amod_statistics_skip-bigram dobj_introduced_statistics prep_introduced_for det_evaluation_the pobj_for_evaluation prep_evaluation_of nn_translation_machine pobj_of_translation ' +o,5257,' and manual methods ',Lin,'cc__and amod_methods_manual conj__methods ' +o,5258,'And indeed , the agreement figures went up from K = 063 to K = 068 -LRB- ignoring doubts -RRB- when we did so , ie , within the ` tentative \' margins of agreement according to -LRB- 068 -LRB- _ x -LRB- 08 -RRB- ',Carletta,'cc_went_And advmod_went_indeed det_figures_the nn_figures_agreement nsubj_went_figures prt_went_up prep_went_from pobj_from_K dep_063_= rcmod_K_063 prep_went_to pobj_to_K dep_068_= rcmod_K_068 parataxis_went_ignoring dobj_ignoring_doubts advmod_did_when nsubj_did_we advcl_went_did advmod_did_so ccomp_did_ie prep_ie_within det_margins_the amod_margins_tentative pobj_within_margins prep_margins_of pobj_of_agreement prep_ie_according dep_according_to nn_068_ pobj_to_068 nn_x__ dep_068_x appos_068_08 ' +o,5259,'The reason may be that shorter dependencies are often modifier of nouns such as determiners or adjectives or pronouns modifying their direct neighbors , while longer dependencies typically represent modifiers of the root or the main verb in a sentenc ',McDonald,'det_reason_The nsubj_be_reason aux_be_may complm_modifier_that amod_dependencies_shorter nsubj_modifier_dependencies cop_modifier_are advmod_modifier_often ccomp_be_modifier prep_modifier_of pobj_of_nouns dep_as_such prep_nouns_as pobj_as_determiners cc_determiners_or conj_determiners_adjectives cc_nouns_or conj_nouns_pronouns partmod_pronouns_modifying poss_neighbors_their amod_neighbors_direct dobj_modifying_neighbors mark_represent_while amod_dependencies_longer nsubj_represent_dependencies advmod_represent_typically advcl_modifying_represent dobj_represent_modifiers prep_modifiers_of det_root_the pobj_of_root cc_modifiers_or det_verb_the amod_verb_main conj_modifiers_verb prep_verb_in det_sentenc_a pobj_in_sentenc ' +n,5260,'2Mutual information , though potentially of interest as a measure of collocational status , was not tested due to its well-known property of overemphasising the significance of rare events ',Church,'amod_information_2Mutual nsubjpass_tested_information dep_potentially_though amod_information_potentially prep_potentially_of pobj_of_interest prep_interest_as det_measure_a pobj_as_measure prep_measure_of amod_status_collocational pobj_of_status auxpass_tested_was neg_tested_not dep_to_due prep_tested_to poss_property_its amod_property_well-known pobj_to_property prep_property_of pcomp_of_overemphasising det_significance_the dobj_overemphasising_significance prep_significance_of amod_events_rare pobj_of_events ' +o,5261,'For the MER training , we modify Koehns MER trainer to train our system ',Och,'prep_modify_For det__the nn__MER nn__training pobj_For_ nsubj_modify_we nn__Koehns nn__MER nn__trainer nsubj_train_ aux_train_to xcomp_modify_train poss_system_our dobj_train_system ' +o,5262,'The parameters of the refined productions Ax By Cz , where Ax is a subcategory of A , By of B , and Cz of C , can then be estimated in various ways ; past work has included both generative and discriminative approaches ',Matsuzaki,'det_parameters_The prep_parameters_of det_Ax_the amod_Ax_refined nn_Ax_productions pobj_of_Ax prep_estimated_By pobj_By_Cz advmod_subcategory_where nsubj_subcategory_Ax cop_subcategory_is det_subcategory_a dep_estimated_subcategory prep_subcategory_of pobj_of_A prep_estimated_By dep_By_of pobj_of_B cc_B_and conj_B_Cz prep_Cz_of pobj_of_C aux_estimated_can advmod_estimated_then auxpass_estimated_be dep_parameters_estimated prep_estimated_in amod_ways_various pobj_in_ways amod_work_past nsubj_included_work aux_included_has parataxis_estimated_included preconj__both amod__generative dobj_included_ cc__and amod__discriminative nn__approaches conj__ ' +o,5263,'In the work of on extracting collocations , preference was given to constructions whose constituents appear in a fixed order , a similar -LRB- and more generally implemented -RRB- version of our assumption here that asymmetric constructions are more idiomatic than symmetric ones ',Smadja,'prep_given_In det_work_the pobj_In_work prep_work_of pcomp_of_ prep__on pcomp_on_extracting dobj_extracting_collocations nsubjpass_given_preference auxpass_given_was prep_given_to pobj_to_constructions poss_constituents_whose nsubj_appear_constituents rcmod_constructions_appear prep_appear_in det_order_a amod_order_fixed pobj_in_order det_version_a amod_version_similar dep_more_and dep_version_more advmod_implemented_generally partmod_more_implemented appos_order_version prep_version_of poss_assumption_our pobj_of_assumption advmod_given_here complm_idiomatic_that amod_constructions_asymmetric nsubj_idiomatic_constructions cop_idiomatic_are advmod_idiomatic_more ccomp_given_idiomatic prep_idiomatic_than amod_ones_symmetric pobj_than_ones ' +o,5264,'4 Structural Correspondence Learning SCL -LRB- Structural Correspondence Learning -RRB- is a recently proposed domain adaptation technique which uses unlabeled data from both source and target domain to learn correspondences between features from different domains ',Blitzer,'dep_Structural_4 amod__Structural nn__Correspondence nn__Learning nn__SCL nn_Learning_Structural nn_Learning_Correspondence appos__Learning nsubj_technique_ cop_technique_is det_technique_a advmod_proposed_recently amod_technique_proposed nn_technique_domain nn_technique_adaptation nsubj_uses_which rcmod_technique_uses amod_data_unlabeled dobj_uses_data prep_uses_from preconj_domain_both nn_domain_source cc_source_and conj_source_target pobj_from_domain aux_learn_to xcomp_uses_learn dobj_learn_correspondences prep_correspondences_between pobj_between_features prep_learn_from amod_domains_different pobj_from_domains ' +o,5265,'al 2006 -RRB- , we are interested in applying alternative metrics such a Meteor ',Banerjee,'dep_we_2006 dobj_al_we cop_interested_are rcmod_we_interested prep_interested_in pcomp_in_applying amod_metrics_alternative dobj_applying_metrics prep_we_such det_Meteor_a dep_such_Meteor ' +o,5266,'52 Assigning complex ambiguity tags In the tagging literature -LRB- eg , -RRB- an ambiguity class is often composed of the set of every possible tag for a word ',Cutting,'num_tags_52 nn_tags_Assigning amod_tags_complex nn_tags_ambiguity dep_In_tags prep_composed_In det_literature_the amod_literature_tagging pobj_In_literature appos_literature_eg dep_eg_ det_class_an amod_class_ambiguity nsubjpass_composed_class auxpass_composed_is advmod_composed_often prep_composed_of det_set_the pobj_of_set prep_set_of det_tag_every amod_tag_possible pobj_of_tag prep_tag_for det_word_a pobj_for_word ' +o,5267,'Finally , we plan to apply the model to other paraphrasing tasks including fully abstractive document summarisation ','Daume III','advmod_plan_Finally nsubj_plan_we aux_apply_to xcomp_plan_apply det_model_the dobj_apply_model prep_apply_to amod_tasks_other amod_tasks_paraphrasing pobj_to_tasks prep_tasks_including advmod_abstractive_fully amod_summarisation_abstractive nn_summarisation_document pobj_including_summarisation ' +o,5268,'The other 5 have been suggested for Dutch by ',Dunning,'det_5_The amod_5_other nsubjpass_suggested_5 aux_suggested_have auxpass_suggested_been prep_suggested_for pobj_for_Dutch prep_suggested_by pobj_by_ ' +o,5269,'We performed a comparison between the existing CFG filtering techniques for LTAG and HPSG , using strongly equivalent grammars obtained by converting LTAGs extracted from the Penn Treebank into HPSG-style ',Marcus,'nsubj_performed_We det_comparison_a dobj_performed_comparison prep_comparison_between det_techniques_the amod_techniques_existing nn_techniques_CFG amod_techniques_filtering pobj_between_techniques prep_techniques_for nn__LTAG pobj_for_ cc__and nn__HPSG conj__ xcomp_performed_using advmod_equivalent_strongly amod_grammars_equivalent dobj_using_grammars partmod_grammars_obtained prep_obtained_by pcomp_by_converting dobj_converting_LTAGs partmod_LTAGs_extracted prep_extracted_from det__the nn__Penn nn__Treebank pobj_from_ prep_extracted_into pobj_into_HPSG-style ' +p,5270,'We decided to use the class of maximum entropy models , which are probabilistically sound , can make use of possibly many overlapping features , and can be trained efficiently ',Berger,'nsubjpass_decided_We aux_use_to xcomp_decided_use det_class_the dobj_use_class prep_class_of amod_models_maximum nn_models_entropy pobj_of_models nsubj_sound_which cop_sound_are advmod_sound_probabilistically rcmod_models_sound aux_make_can conj_decided_make dobj_make_use prep_use_of advmod_features_possibly amod_features_many nn_features_overlapping pobj_of_features cc_decided_and aux_trained_can auxpass_trained_be conj_decided_trained advmod_trained_efficiently ' +o,5271,'For the MER training , Koehns MER trainer is modified for our system ',Och,'prep_modified_For det__the nn__MER nn__training pobj_For_ nn__Koehns nn__MER nn__trainer nsubjpass_modified_ auxpass_modified_is prep_modified_for poss_system_our pobj_for_system ' +o,5272,'Therefore , estimating a natural language model based on the maximum entropy -LRB- ME -RRB- method has been highlighted recently ',Berger,'advmod_based_Therefore dep_based_estimating det_language_a amod_language_natural dobj_estimating_language nsubj_based_model prep_based_on det_entropy_the amod_entropy_maximum pobj_on_entropy abbrev_entropy_ME nn__method nsubjpass_highlighted_ aux_highlighted_has auxpass_highlighted_been dep_based_highlighted advmod_highlighted_recently ' +o,5273,'953 2 Bilexicalization of Inversion Transduction Grammar The Inversion Transduction Grammar of models word alignment between a translation pair of sentences by assuming a binary synchronous tree on top of both sides ',Wu,'number_2_953 num_Bilexicalization_2 dep_Grammar_Bilexicalization prep_Bilexicalization_of pobj_of_Inversion nn_Grammar_Transduction dep_Grammar_Grammar det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction dep_alignment_Grammar prep_Grammar_of amod_models_ pobj_of_models nn_alignment_word prep_alignment_between det_pair_a nn_pair_translation pobj_between_pair prep_pair_of pobj_of_sentences prep_sentences_by pcomp_by_assuming det_tree_a advmod_synchronous_binary amod_tree_synchronous dobj_assuming_tree prep_assuming_on pobj_on_top prep_top_of det_sides_both pobj_of_sides ' +o,5274,'Even before the 2006 shared task , the parsers of and Charniak -LRB- 2000 -RRB- , originally developed for English , had been adapted for dependency parsing of Czech , and the parsing methodology proposed by Kudo and Matsumoto and Yamada and Matsumoto had been evaluated on both Japanese and English ',Collins,'advmod_shared_Even dep_shared_before dep_2006_the nsubj_shared_2006 advcl_adapted_shared dobj_shared_task det_parsers_the nsubjpass_adapted_parsers prep_parsers_of pobj_of_ cc__and conj__Charniak appos_parsers_2000 advmod_developed_originally partmod_parsers_developed prep_developed_for pobj_for_English aux_adapted_had auxpass_adapted_been prep_adapted_for amod_parsing_dependency pobj_for_parsing prep_parsing_of pobj_of_Czech cc_adapted_and det_methodology_the amod_methodology_parsing nsubjpass_evaluated_methodology partmod_methodology_proposed prep_proposed_by nn__Kudo cc_Kudo_and conj_Kudo_Matsumoto pobj_by_ cc__and nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto conj__ aux_evaluated_had auxpass_evaluated_been conj_adapted_evaluated prep_evaluated_on preconj_Japanese_both pobj_on_Japanese cc_Japanese_and conj_Japanese_English ' +o,5275,'A few studies addressed this defect by selecting the appropriate translation rules for an input span based on its context in the input sentence ',He,'det__A amod__few nn__studies nsubj_addressed_ det_defect_this dobj_addressed_defect prep_addressed_by pcomp_by_selecting det_rules_the amod_rules_appropriate nn_rules_translation dobj_selecting_rules prep_rules_for det_span_an nn_span_input pobj_for_span prep_selecting_based dep_based_on poss_context_its pobj_on_context prep_context_in det_sentence_the nn_sentence_input pobj_in_sentence ' +o,5276,'We first determine lexical heads of nonterminal nodes by using Bikels implementation of Collins head detection algorithm9 ',Collins,'nsubj_determine_We advmod_determine_first amod_heads_lexical dobj_determine_heads prep_heads_of amod_nodes_nonterminal pobj_of_nodes prep_determine_by pcomp_by_using nn_implementation_Bikels dobj_using_implementation prep_implementation_of nn_algorithm9_Collins nn_algorithm9_head nn_algorithm9_detection pobj_of_algorithm9 ' +p,5277,'Whereas dependency based semantic spaces have been shown to surpass other word space models for a number of problems , for the task of categorisation simple pattern based spaces have been shown to perform equally good if not better ',Lin,'mark_shown_Whereas advmod_based_dependency amod_spaces_based amod_spaces_semantic nsubjpass_shown_spaces aux_shown_have auxpass_shown_been aux_surpass_to xcomp_shown_surpass amod_models_other nn_models_word nn_models_space dobj_surpass_models prep_surpass_for det_number_a pobj_for_number prep_number_of nn__problems pobj_of_ prep_shown_for det_task_the pobj_for_task prep_task_of amod_pattern_categorisation amod_pattern_simple pobj_of_pattern partmod_pattern_based nsubjpass_shown_spaces aux_shown_have auxpass_shown_been ccomp_based_shown aux_perform_to xcomp_shown_perform advmod_good_equally amod__good dep_good_if neg_better_not dep_good_better dobj_perform_ ' +o,5278,'For instance , word alignment models are often trained using the GIZA + + toolkit ; error minimizing training criteria such as the Minimum Error Rate Training are employed in order to learn feature function weights for log-linear models ; and translation candidates are produced using phrase-based decoders in combination with n-gram language models ',Koehn,'prep_trained_For pobj_For_instance nn_models_word nn_models_alignment nsubjpass_trained_models auxpass_trained_are advmod_trained_often xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__toolkit dobj_using_ nsubjpass_employed_error partmod_error_minimizing nn_criteria_training dobj_minimizing_criteria dep_as_such prep_criteria_as det__the nn__Minimum nn__Error nn__Rate nn__Training pobj_as_ auxpass_employed_are parataxis_trained_employed prep_employed_in pobj_in_order aux_learn_to xcomp_employed_learn nn_weights_feature nn_weights_function dobj_learn_weights prep_learn_for amod_models_log-linear pobj_for_models cc_trained_and nn_candidates_translation nsubjpass_produced_candidates auxpass_produced_are conj_trained_produced xcomp_produced_using amod_decoders_phrase-based dobj_using_decoders partmod_decoders_ prep__in pobj_in_combination prep__with amod_models_n-gram nn_models_language pobj_with_models advmod__ ' +o,5279,'But the lack of corpora has led to a situation where much of the current work on parsing is performed on a single domain using training data from that domain the Wall Street Journal -LRB- WSJ -RRB- section of the Penn Treebank ',Marcus,'cc_led_But det_lack_the nsubj_led_lack prep_lack_of pobj_of_corpora aux_led_has prep_led_to det_situation_a pobj_to_situation advmod_performed_where nsubjpass_performed_much prep_much_of det_work_the amod_work_current pobj_of_work prep_work_on pobj_on_parsing auxpass_performed_is rcmod_situation_performed prep_performed_on det_domain_a amod_domain_single pobj_on_domain partmod_domain_using nn_data_training dobj_using_data prep_using_from det_domain_that pobj_from_domain det_section_the nn_Journal_Wall nn_Journal_Street nn_section_Journal abbrev_Journal_WSJ nsubj_Treebank_section prep_section_of det_Penn_the pobj_of_Penn dep_led_Treebank ' +o,5280,'In , as well as other similar works , only left-toright search was employed ',Collins,'prep_employed_In pobj_In_ dep_well_as cc__well dep_well_as amod_works_other amod_works_similar conj__works partmod_works_ advmod_search_only amod_search_left-toright nsubjpass_employed_search auxpass_employed_was ' +o,5281,'This operation does not change the collection of phrases or rules extracted from a hypothesized alignment , see , for instance , ',Koehn,'det_operation_This nsubj_change_operation aux_change_does neg_change_not det_collection_the dobj_change_collection prep_collection_of pobj_of_phrases cc_phrases_or conj_phrases_rules partmod_phrases_extracted prep_extracted_from det_alignment_a amod_alignment_hypothesized pobj_from_alignment dep_change_see prep_change_for pobj_for_instance ' +o,5282,'We tested the techniques described above with the previous Bakeoffs data5 ',Levow,'nsubj_tested_We det_techniques_the nsubj_described_techniques ccomp_tested_described advmod_described_above prep_described_with det_data5_the amod_data5_previous nn_data5_Bakeoffs pobj_with_data5 ' +o,5283,'The modify features involve the dependency parse tree for the sentence , obtained by first parsing the sentence and then converting the tree into its dependency representation ',Collins,'det_features_The amod_features_modify nsubj_involve_features det_tree_the amod_tree_dependency nn_tree_parse dobj_involve_tree prep_involve_for det_sentence_the pobj_for_sentence partmod_sentence_obtained prep_obtained_by advmod_parsing_first pcomp_by_parsing det__the nn__sentence dobj_parsing_ cc_parsing_and advmod_parsing_then conj_parsing_converting det_tree_the dobj_converting_tree prep_converting_into poss__its amod__dependency nn__representation pobj_into_ ' +o,5284,'In a factored translation model other factors than surface form can be used , such as lemma or part-of-speech ',Koehn,'det_model_a amod_model_factored nn_model_translation pobj_In_model amod_factors_other nsubjpass_used_factors prep_factors_than nn_form_surface pobj_than_form aux_used_can auxpass_used_be rcmod_model_used dep_as_such prep_model_as pobj_as_lemma cc_lemma_or conj_lemma_part-of-speech ' +o,5285,'Therefore , P -LRB- g l e -RRB- is the sum of the probabilities of generating g from e over all possible alignments A , in which the position i in the target sentence g is aligned to the position ai in the source sentence e : P -LRB- gle -RRB- = I l m e ~ , ~ \' IT t -LRB- g # le = jla -LRB- a ~ Ij , l , m -RRB- = al = 0 amm0j = l m ! e 1 \' I ~ t -LRB- g # l e , -RRB- a -LRB- ilj , t , m -RRB- -LRB- 3 -RRB- j = l i = 0 also described how to use the EM algorithm to estimate the parameters a -LRB- i I j , l , m -RRB- and $ -LRB- g I e -RRB- in the aforementioned model ',Brown,'advmod_sum_Therefore nsubj_sum_P nn_e_g nn_e_l appos_P_e cop_sum_is det_sum_the prep_sum_of det_probabilities_the pobj_of_probabilities prep_probabilities_of amod_g_generating pobj_of_g prep_g_from dep_from_e prep_g_over det_A_all amod_A_possible nn_A_alignments pobj_over_A rel_m_in pobj_in_which det_position_the nsubj_m_position nsubjpass_aligned_i prep_i_in det_g_the nn_g_target nn_g_sentence pobj_in_g auxpass_aligned_is rcmod_position_aligned prep_aligned_to det_ai_the nn_ai_position pobj_to_ai prep_ai_in det_sentence_the nn_sentence_source pobj_in_sentence dep_P_e ccomp_aligned_P nsubj_I_gle dep_I_= ccomp_P_I tmod_aligned_l rcmod_A_m dep_m_e dep_e_~ dep_e_~ amod_t_IT nsubj_m_t partmod_t_g dep_=_# number_=_le num_jla_= dobj_g_jla det_Ij_a nn_Ij_~ appos_jla_Ij dep_Ij_l dep_Ij_m xcomp_g_= nsubj_amm0j_al dep_amm0j_= num_amm0j_0 ccomp_=_amm0j dep_m_= amod_m_l rcmod_~_m dep_I_e poss_I_1 dep_~_I partmod_I_~ dobj_~_t dep_t_g dep_e_# number_e_l dep_g_e amod_t_a nn_m_ilj dep_m_t dep_a_m dep_described_3 nsubj_described_j dep_j_= iobj_=_l nn_l_i dep_l_= num__0 dobj_=_ advmod__also dep_sum_described advmod_use_how aux_use_to ccomp_described_use det_algorithm_the nn_algorithm_EM nsubj_estimate_algorithm aux_estimate_to xcomp_use_estimate det_a_the amod_a_parameters dobj_estimate_a num_j_i nn_j_I appos_a_j dep_j_l dep_j_m cc_a_and conj_a_$ nn_e_g nn_e_I dep_aforementioned_e advmod_e_in dep_in_the dep_model_aforementioned num_$_model ' +o,5286,'12 As such , we resort to an approximation : Voted Perceptron training ',Collins,'prep_to_As pobj_As_such nsubj_to_we aux_to_resort dep_12_to det_training_an nn_training_approximation nn_training_Voted nn_training_Perceptron dobj_to_training ' +o,5287,'(see Brown et al. , 1993 for a detailed mathematical description of the model and the formula for computing the probability of an alignment and target string given a source string).',Brown,'' +o,5288,'Our evaluation metric is case-insensitive BLEU-4 , as defined by NIST , that is , using the shortest -LRB- as opposed to closest -RRB- reference sentence length for the brevity penalty ',Papineni,'poss_metric_Our nn_metric_evaluation nsubj__metric cop__is amod__case-insensitive nn__BLEU-4 advmod_defined_as partmod__defined prep_defined_by pobj_by_NIST nsubj_is_that rcmod_NIST_is partmod__using det_length_the nn_length_shortest advmod_opposed_as dep_shortest_opposed prep_opposed_to pobj_to_closest nn_length_reference nn_length_sentence dobj_using_length prep_length_for det_penalty_the amod_penalty_brevity pobj_for_penalty ' +p,5289,'In addition , parsing re-ranking has also been shown to be another effective technique to improve parsing performance ',Collins,'prep_shown_In pobj_In_addition amod__parsing amod__re-ranking nsubjpass_shown_ aux_shown_has advmod_shown_also auxpass_shown_been aux_technique_to cop_technique_be det_technique_another amod_technique_effective xcomp_shown_technique aux_improve_to infmod_technique_improve amod_performance_parsing dobj_improve_performance ' +o,5290,'211 Pointwise Mutual Information This measure for word similarity was first used in this context by ',Church,'nn_Information_211 nn_Information_Pointwise nn_Information_Mutual det_measure_This nsubjpass_used_measure prep_measure_for nn_similarity_word pobj_for_similarity auxpass_used_was advmod_used_first dep_Information_used prep_used_in det_context_this pobj_in_context prep_used_by ' +o,5291,'2 Related Work The issue of MWE processing has attracted much attention from the Natural Language Processing -LRB- NLP -RRB- community , including Smadja , 1993 ; Dagan and Church , 1994 ; Daille , 1995 ; 1995 ; McEnery et al , 1997 ; Wu , 1997 ; Michiels and Dufour , 1998 ; Maynard and Ananiadou , 2000 ; Merkel and Andersson , 2000 ; Piao and McEnery , 2001 ; Sag et al , 2001 ; Tanaka and Baldwin , 2003 ; Dias , 2003 ; Baldwin et al , 2003 ; Nivre and Nilsson , 2004 Pereira et al , ',Dunning,'num_Work_2 amod_Work_Related det_issue_The nsubj_attracted_issue prep_issue_of nn_processing_MWE pobj_of_processing aux_attracted_has dep_Work_attracted amod_attention_much dobj_attracted_attention prep_attracted_from det_community_the nn_Processing_Natural nn_Processing_Language nn_community_Processing abbrev_Processing_NLP pobj_from_community prep_community_including pobj_including_Smadja appos_Smadja_1993 dep_Smadja_Dagan cc_Dagan_and conj_Dagan_Church appos_Dagan_1994 dep_Smadja_Daille appos_Daille_1995 dep_Smadja_1995 dep_1995_McEnery cc_McEnery_et conj_McEnery_al appos_McEnery_1997 dep_Smadja_Wu appos_Wu_1997 dep_Smadja_Michiels cc_Michiels_and conj_Michiels_Dufour appos_Michiels_1998 dep_Smadja_Maynard cc_Maynard_and conj_Maynard_Ananiadou appos_Maynard_2000 dep_Smadja_Merkel cc_Merkel_and conj_Merkel_Andersson appos_Merkel_2000 dep_Smadja_Piao cc_Piao_and conj_Piao_McEnery appos_Piao_2001 dep_Smadja_Sag cc_Sag_et conj_Sag_al appos_Sag_2001 dep_Smadja_Tanaka cc_Tanaka_and conj_Tanaka_Baldwin appos_Tanaka_2003 dep_Smadja_Dias appos_Dias_2003 dep_Smadja_Baldwin cc_Baldwin_et conj_Baldwin_al appos_Baldwin_2003 dep_Smadja_Nivre cc_Nivre_and conj_Nivre_Nilsson nn_Pereira_2004 appos_Smadja_Pereira cc_Pereira_et conj_Pereira_al ' +o,5292,'405 PRF 1 proposed 383 437 408 multinomial mixture 360 374 367 Newman 318 353 334 cosine 603 114 192 - skew divergence 730 155 255 Lins similarity 691 096 169 CBC 981 060 114 Table 3 : Precision , recall , and F-measure ',Lin,'num_PRF_405 num_PRF_1 dep_PRF_proposed number_437_383 num_mixture_437 amod_mixture_408 amod_mixture_multinomial dobj_proposed_mixture number_374_360 num_mixture_374 num_603_367 nn_603_Newman nn_603_ nn_603_318 num_603_353 num_603_334 amod_603_cosine dep_Table_603 number_192_114 dep_603_192 nn_730_skew nn_730_divergence nn_730_ dep_192_730 number_255_155 num_Lins_255 dep_730_Lins dep_Table_similarity num_similarity_ num_Table_691 number_169_096 num_Table_169 nn_Table_CBC nn_Table_ num_Table_981 number_114_060 num_Table_114 dep_mixture_Table dep_Table_3 dep_PRF_Precision conj_Precision_recall cc_Precision_and conj_Precision_F-measure ' +o,5293,'The model parameters are trained using minimum error-rate training ',Och,'det_parameters_The nn_parameters_model nsubjpass_trained_parameters auxpass_trained_are xcomp_trained_using amod_training_minimum amod_training_error-rate dobj_using_training ' +o,5294,'Forced decoding arises in online discriminative training , where model updates are made toward the most likely derivation of a gold translation ',Liang,'amod_decoding_Forced nsubj_arises_decoding prep_arises_in amod_training_online nn_training_discriminative pobj_in_training advmod_made_where nn_updates_model nsubjpass_made_updates auxpass_made_are rcmod_training_made prep_made_toward det_derivation_the advmod_likely_most amod_derivation_likely pobj_toward_derivation prep_derivation_of det_translation_a amod_translation_gold pobj_of_translation ' +o,5295,'We used treebank grammars induced directly from the local trees of the entire WSJ section of the Penn Treebank -LRB- release 3 -RRB- ',Marcus,'nsubj_used_We nn_grammars_treebank nsubj_induced_grammars ccomp_used_induced advmod_induced_directly prep_induced_from det_trees_the amod_trees_local pobj_from_trees prep_trees_of det_section_the amod_section_entire nn_section_WSJ pobj_of_section prep_section_of det__the nn__Penn nn__Treebank pobj_of_ dep_induced_release num_release_3 ' +o,5296,' finds significant bigrams using an estimate of z-score -LRB- deviation from an expected mean -RRB- ',Smadja,'nsubj_finds_ amod_bigrams_significant dobj_finds_bigrams partmod_bigrams_using det_estimate_an dobj_using_estimate prep_estimate_of pobj_of_z-score dep_z-score_deviation prep_deviation_from det_mean_an amod_mean_expected pobj_from_mean ' +o,5297,'The results so far mainly come from studies where a parser originally developed for English , such as the Collins parser , is applied to a new language , which often leads to a signicant decrease in the measured accuracy ',Collins,'det_results_The nsubjpass_applied_results advmod_far_so advmod_come_far advmod_come_mainly partmod_results_come prep_come_from pobj_from_studies advmod_developed_where det_parser_a nsubj_developed_parser advmod_developed_originally advcl_come_developed prep_developed_for pobj_for_English dep_as_such prep_results_as det__the nn__Collins nn__parser pobj_as_ auxpass_applied_is prep_applied_to det_language_a amod_language_new pobj_to_language nsubj_leads_which advmod_leads_often rcmod_language_leads prep_leads_to det_decrease_a amod_decrease_signicant pobj_to_decrease prep_decrease_in det_accuracy_the amod_accuracy_measured pobj_in_accuracy ' +o,5298,'Tag test data using the POS-tagger described in ',Ratnaparkhi,'nn_data_test dobj_Tag_data partmod_data_using det_POS-tagger_the dobj_using_POS-tagger partmod_POS-tagger_described prep_described_in ' +n,5299,'While in traditional word-based statistical models the atomic unit that translation operates on is the word , phrase-based methods acknowledge the significant role played in language by multiword expressions , thus incorporating in a statistical framework the insight behind Example-Based Machine Translation ',Brown,'mark_word_While prep_word_in amod_models_traditional amod_models_word-based amod_models_statistical pobj_in_models partmod_models_ det_unit_the amod_unit_atomic nsubj_word_unit complm_operates_that nsubj_operates_translation dep_unit_operates prt_operates_on cop_word_is det_word_the advcl_acknowledge_word amod_methods_phrase-based nsubj_acknowledge_methods det_role_the amod_role_significant nsubj_played_role ccomp_acknowledge_played prep_played_in pobj_in_language prep_played_by amod_expressions_multiword pobj_by_expressions advmod_incorporating_thus xcomp_played_incorporating prep_incorporating_in det_framework_a amod_framework_statistical pobj_in_framework det_insight_the dobj_incorporating_insight prep_insight_behind amod__Example-Based nn__Machine nn__Translation pobj_behind_ ' +o,5300,'The third estimates the equivalence based on word alignment composed using templates or translation probabilities derived from a set of parallel text ',Barzilay,'det_third_The nsubj_estimates_third det_equivalence_the dobj_estimates_equivalence prep_estimates_based dep_based_on nn_alignment_word pobj_on_alignment partmod_alignment_composed xcomp_composed_using dobj_using_templates cc_templates_or nn_probabilities_translation conj_templates_probabilities partmod_templates_derived prep_derived_from det_set_a pobj_from_set prep_set_of amod_text_parallel pobj_of_text ' +o,5301,'Classes can be induced directly from the corpus using distributional clustering or taken from a manually crafted taxonomy ',Brown,'nsubjpass_induced_Classes aux_induced_can auxpass_induced_be advmod_induced_directly prep_induced_from det_corpus_the pobj_from_corpus partmod_corpus_using amod__distributional nn__clustering dobj_using_ cc_induced_or conj_induced_taken prep_taken_from det_manually_a pobj_from_manually partmod_manually_crafted advmod__taxonomy dep_crafted_ ' +o,5302,'Table 2 shows the dependency accuracy , root accuracy and complete match scores for our best parser -LRB- Model 2 with label set B -RRB- in comparison with -LRB- Model 3 -RRB- , Charniak -LRB- 2000 -RRB- , and Yamada and Matsumoto 5 It is clear that , with respect to unlabeled accuracy , our parser does not quite reach state-of-the-art performance , even if we limit the competition to deterministic methods such as that of Yamada and Matsumoto ',Collins,'nsubj_shows_Table num_Table_2 det_accuracy_the amod_accuracy_dependency dobj_shows_accuracy nn_accuracy_root conj_accuracy_accuracy cc_accuracy_and amod_scores_complete nn_scores_match conj_accuracy_scores prep_shows_for poss_parser_our amod_parser_best pobj_for_parser dep_parser_Model num_Model_2 prep_Model_with nn_B_label nn_B_set pobj_with_B prep_parser_in pobj_in_comparison prep_comparison_with pobj_with_ appos_comparison_Model num_Model_3 appos_comparison_Charniak appos_Charniak_2000 cc_parser_and nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto conj_parser_ num__5 nsubj_clear_It cop_clear_is rcmod__clear complm_reach_that prep_reach_with pobj_with_respect prep_respect_to amod_accuracy_unlabeled pobj_to_accuracy poss_parser_our nsubj_reach_parser aux_reach_does neg_reach_not advmod_reach_quite ccomp_clear_reach amod_performance_state-of-the-art dobj_reach_performance advmod_limit_even dep_limit_if nsubj_limit_we dep_shows_limit det_competition_the dobj_limit_competition prep_limit_to amod_methods_deterministic pobj_to_methods dep_as_such prep_methods_as pobj_as_that prep_that_of nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto pobj_of_ ' +o,5303,'Within NLP , applications include sentiment-analysis problems and content selection for text generation ',Pang,'prep_include_Within pobj_Within_NLP nsubj_include_applications amod_problems_sentiment-analysis nsubj_selection_problems amod_selection_ cc__and conj__content xcomp_include_selection prep_selection_for nn__text nn__generation pobj_for_ ' +o,5304,'Some researchers apply shallow or partial parsers to acquiring specific patterns from texts ',Hindle,'det_researchers_Some nsubj_apply_researchers nn__shallow cc_shallow_or conj_shallow_partial nn__parsers dobj_apply_ prep_apply_to pcomp_to_acquiring amod_patterns_specific dobj_acquiring_patterns prep_acquiring_from pobj_from_texts ' +o,5305,'The task of classifying several different uses of definite descriptions is somewhat analogous to that for bare nouns ',Bean,'det_task_The nsubj_analogous_task prep_task_of pcomp_of_classifying amod_uses_several amod_uses_different dobj_classifying_uses prep_uses_of amod__definite nn__descriptions pobj_of_ cop_analogous_is advmod_analogous_somewhat prep_analogous_to pobj_to_that prep_that_for amod_nouns_bare pobj_for_nouns ' +o,5306,'This results in two forbidden alignment structures , shown in Figure 1 , called inside-out transpositions in ',Wu,'det_results_This nsubj_called_results prep_results_in num_structures_two amod_structures_forbidden nn_structures_alignment pobj_in_structures partmod_results_shown prep_shown_in pobj_in_Figure num_Figure_1 amod_transpositions_inside-out dobj_called_transpositions prep_called_in ' +n,5307,'Note that the minimum error rate training uses only the target sentence with the maximum posterior probability whereas , here , the whole probability distribution is taken into account ',Och,'complm_sentence_that det__the amod__minimum nn__error nn__rate nn__training nsubj_sentence_ dep_sentence_uses advmod_sentence_only det_sentence_the nn_sentence_target ccomp_Note_sentence prep_sentence_with det_probability_the amod_probability_maximum nn_probability_posterior pobj_with_probability mark_taken_whereas advmod_taken_here det_distribution_the amod_distribution_whole nn_distribution_probability nsubjpass_taken_distribution auxpass_taken_is advcl_sentence_taken prep_taken_into pobj_into_account ' +o,5308,'In the model is optimized to produce a block orientation and the target sentence is used only for computing a sentence level BLEU ',Tillmann,'prep_optimized_In pobj_In_ det_model_the nsubjpass_optimized_model auxpass_optimized_is aux_produce_to xcomp_optimized_produce det_orientation_a nn_orientation_block nsubjpass_used_orientation cc_orientation_and det_sentence_the nn_sentence_target conj_orientation_sentence auxpass_used_is ccomp_produce_used dep_for_only prep_used_for pcomp_for_computing det_BLEU_a nn_BLEU_sentence nn_BLEU_level dobj_computing_BLEU ' +o,5309,'For extracting simple noun phrases we first used Ramshaw and Marcuss base NP chunker ',Ramshaw,'prep_used_For pcomp_For_extracting amod_phrases_simple nn_phrases_noun dobj_extracting_phrases nsubj_used_we advmod_used_first nsubj_base_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcuss dep_used_base nn_chunker_NP dobj_base_chunker ' +o,5310,'re-ranking 1 uses the score of the rst model as a feature in addition to the non-local features as in ',Collins,'dep_uses_re-ranking tmod_re-ranking_1 det_score_the nsubj_uses_score prep_score_of det_model_the nn_model_rst pobj_of_model prep_model_as det_feature_a pobj_as_feature prep_feature_in dep_in_addition dep_in_to det_features_the amod_features_non-local pobj_in_features prep_uses_as dep_as_in ' +o,5311,'The optimal bilingual parsing tree for a given sentence-pair can be computed using dynamic programming -LRB- DP -RRB- algorith ',Wu,'det_tree_The amod_tree_optimal amod_tree_bilingual nn_tree_parsing nsubjpass_computed_tree prep_tree_for det_sentence-pair_a amod_sentence-pair_given pobj_for_sentence-pair aux_computed_can auxpass_computed_be purpcl_computed_using amod_programming_dynamic nsubj_algorith_programming abbrev_programming_DP xcomp_using_algorith ' +o,5312,'The hierarchical phrase translation pairs are extracted in a standard way : First , the bilingual data are word alignment annotated by running GIZA + + in two directions ',Och,'det_pairs_The amod_pairs_hierarchical nn_pairs_phrase nn_pairs_translation nsubjpass_extracted_pairs auxpass_extracted_are prep_extracted_in det__a amod__standard nn__way pobj_in_ advmod_alignment_First det_data_the amod_data_bilingual nsubj_alignment_data cop_alignment_are nn_alignment_word parataxis_extracted_alignment partmod_alignment_annotated prep_annotated_by pcomp_by_running nn__GIZA nn__+ nn__+ dobj_running_ prep__in num_directions_two pobj_in_directions ' +o,5313,'We were already using a generative statistical model for part-of-speech tagging , and more recently , had begun using a generative statistical model for name finding ',Marcus,'nsubj_were_We advmod_using_already dep_were_using det_model_a amod_model_generative amod_model_statistical dobj_using_model prep_model_for amod__part-of-speech amod__tagging pobj_for_ cc_using_and advmod_recently_more advmod_begun_recently aux_begun_had conj_using_begun xcomp_begun_using det_model_a amod_model_generative amod_model_statistical dobj_using_model prep_using_for nn__name nn__finding pobj_for_ ' +o,5314,'Motivated by our goal of representing syntax , we used part-of-speech -LRB- POS -RRB- tags as labeled by a maximum entropy tagger ',Ratnaparkhi,'ccomp_used_Motivated prep_Motivated_by poss_goal_our pobj_by_goal prep_goal_of pcomp_of_representing dobj_representing_syntax nsubj_used_we amod_tags_part-of-speech appos_tags_POS dobj_used_tags mark_labeled_as advcl_used_labeled prep_labeled_by det_tagger_a amod_tagger_maximum nn_tagger_entropy pobj_by_tagger ' +o,5315,' , but its performance was worse than our centroid baseline ',Lin,'cc_worse_but poss_performance_its nsubj_worse_performance cop_worse_was ccomp__worse prep_worse_than poss_baseline_our amod_baseline_centroid pobj_than_baseline ' +o,5316,'12 Evaluation In this paper we report results using the BLEU metric , however as the evaluation criterion in GALE is HTER , we also report in TER ',Papineni,'num_Evaluation_12 dep_report_Evaluation dep_Evaluation_In det_paper_this pobj_In_paper nsubj_report_we parataxis_report_report nsubj_using_results dep_report_using det__the amod__BLEU amod__metric dobj_using_ advmod__however dep__as det_criterion_the nn_criterion_evaluation nsubj__criterion prep_criterion_in pobj_in_GALE cop__is nn__HTER ccomp_report_ nsubj_report_we advmod_report_also prep_report_in nn__TER pobj_in_ ' +o,5317,'24 METEOR Given a pair of strings to compare -LRB- a system translation and a reference translation -RRB- , METEOR first creates a word alignment between the two strings ',Banerjee,'num_METEOR_24 prep_creates_Given det_pair_a dep_Given_pair prep_pair_of pobj_of_strings aux_compare_to infmod_strings_compare det_translation_a nn_translation_system dep_pair_translation cc_translation_and det_translation_a nn_translation_reference conj_translation_translation nn__METEOR nsubj_creates_ advmod_creates_first dep_METEOR_creates det_alignment_a nn_alignment_word dobj_creates_alignment prep_alignment_between det_strings_the num_strings_two pobj_between_strings ' +o,5318,'For this present work , we use Dunnings log-likelihood ratio statistics defined as follows : sim = aloga + blogb + clogc + dlogd -LRB- a + b -RRB- log -LRB- a + b -RRB- -LRB- a + c -RRB- log -LRB- a + c -RRB- -LRB- b + d -RRB- log -LRB- b + d -RRB- -LRB- c + d -RRB- log -LRB- c + d -RRB- + -LRB- a + b + c + d -RRB- log -LRB- a + b + c + d -RRB- For each bilingual pattern EiJj , we compute its similarity score and qualify it as a bilingual sequence-to-sequence correspondence if no equally strong or stronger association for monolingual constituent is found ',Dunning,'prep_use_For det_work_this amod_work_present pobj_For_work nsubj_use_we nn_statistics_Dunnings amod_statistics_log-likelihood nn_statistics_ratio nsubj_defined_statistics advmod_defined_ ccomp_use_defined mark_follows_as advcl_defined_follows nsubj_=_sim ccomp_follows_= nn_dlogd_aloga cc_blogb_+ amod_dlogd_blogb cc_blogb_+ conj_blogb_clogc nn_dlogd_+ dobj_=_dlogd det_+_a dep_dlogd_+ dep_+_b prep_dlogd_log det_+_a dep_log_+ dep_+_b det_+_a dep_log_+ dep_+_c dep_log_log det_c_a amod_c_+ dep_log_c amod_log_b cc_b_+ conj_b_d nsubj_log_log dep_+_b dep_d_+ dep_-LRB-_d dep_+_c dep_d_+ dep_-LRB-_d rcmod_log_log dep_+_c dep_d_+ dep_-LRB-_d ccomp_log_+ det_d_a cc_b_+ amod_d_b cc_b_+ conj_b_c amod_d_+ dep_+_d dobj_+_log det_d_a cc_b_+ amod_d_b cc_b_+ conj_b_c amod_d_+ appos_log_d prep_compute_For det_EiJj_each amod_EiJj_bilingual nn_EiJj_pattern pobj_For_EiJj nsubj_compute_we dep_use_compute poss_score_its nn_score_similarity dobj_compute_score cc_compute_and dep_it_qualify conj_compute_it prep_it_as det_correspondence_a amod_correspondence_bilingual amod_correspondence_sequence-to-sequence pobj_as_correspondence mark_found_if det_association_no advmod_strong_equally amod_association_strong cc_strong_or conj_strong_stronger nsubjpass_found_association prep_association_for amod_constituent_monolingual pobj_for_constituent auxpass_found_is advcl_it_found ' +p,5319,'However , attempts to retrofit syntactic information into the phrase-based paradigm have not met with enormous success 1 , and purely phrase-based machine translation systems continue to outperform these syntax\\/phrase-based hybrids ',Koehn,'advmod_met_However nsubj_met_attempts aux_retrofit_to infmod_attempts_retrofit amod_information_syntactic dobj_retrofit_information prep_retrofit_into det_paradigm_the amod_paradigm_phrase-based pobj_into_paradigm aux_met_have neg_met_not prep_met_with amod_success_enormous pobj_with_success number_1_ amod_systems_1 cc_1_and advmod_phrase-based_purely conj_1_phrase-based nn_systems_machine nn_systems_translation nsubj_continue_systems dep_met_continue aux_outperform_to xcomp_continue_outperform det_hybrids_these amod_hybrids_syntax\\/phrase-based dobj_outperform_hybrids ' +o,5320,'Mathematical details are fully described in ',Brown,'amod_details_Mathematical nsubjpass_described_details auxpass_described_are advmod_described_fully dep_described_in ' +o,5321,'Following the broad shift in the field from finite state transducers to grammar transducers , recent approaches to phrase-based alignment have used synchronous grammar formalisms permitting polynomial time inference ',Wu,'prep_transducers_Following det_shift_the amod_shift_broad pobj_Following_shift prep_shift_in det_field_the pobj_in_field prep_field_from amod_state_finite pobj_from_state prep_transducers_to nn_transducers_grammar pobj_to_transducers nsubj_transducers_ amod_approaches_recent appos__approaches prep_approaches_to amod_alignment_phrase-based pobj_to_alignment aux_used_have rcmod_alignment_used amod_formalisms_synchronous nn_formalisms_grammar dobj_used_formalisms xcomp_used_permitting amod_inference_polynomial nn_inference_time dobj_permitting_inference ' +o,5322,'In the domain adaptation track , participants were provided with English training data from the Wall Street Journal portion of the Penn Treebank converted to dependencies to train parsers to be evaluated on material in the biological -LRB- development set -RRB- and chemical -LRB- test set -RRB- domains , and optionally on text from the CHILDES database ',Marcus,'prep_provided_In det_track_the nn_track_domain nn_track_adaptation pobj_In_track nsubjpass_provided_participants auxpass_provided_were prep_provided_with amod_data_English nn_data_training pobj_with_data prep_provided_from det_portion_the nn_portion_Wall nn_portion_Street nn_portion_Journal pobj_from_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ partmod__converted prep_converted_to pobj_to_dependencies partmod_dependencies_ aux_train_to xcomp__train nsubjpass_evaluated_parsers aux_evaluated_to auxpass_evaluated_be xcomp_train_evaluated prep_evaluated_on amod__material prep_material_in det_biological_the pobj_in_biological nn_set_development appos_biological_set cc_biological_and conj_biological_chemical nn_set_test appos_biological_set nn__domains pobj_on_ cc_on_and conj_on_optionally conj_on_on pobj_on_text prep_evaluated_from det__the nn__CHILDES nn__database pobj_from_ ' +o,5323,'31 Definition The following set-up , adapted from , was used for all three discriminative training methods : 266 Training data is a set of input-output pairs ',Collins,'num_Definition_31 det_set-up_The amod_set-up_following nsubjpass_used_set-up partmod_set-up_adapted prep_adapted_from pobj_from_ auxpass_used_was dep_Definition_used prep_used_for det_methods_all num_methods_three amod_methods_discriminative nn_methods_training pobj_for_methods num_data_266 nn_data_Training nsubj_set_data cop_set_is det_set_a parataxis_used_set prep_set_of amod_pairs_input-output pobj_of_pairs ' +o,5324,'The parameters , j , were trained using minimum error rate training to maximize the BLEU score on a 150 sentence development set ',Papineni,'det_parameters_The nsubjpass_trained_parameters appos_parameters_j auxpass_trained_were xcomp_trained_using amod__minimum nn__error nn__rate nn__training dobj_using_ aux_maximize_to xcomp_using_maximize det__the amod__BLEU nn__score dobj_maximize_ prep_maximize_on det_set_a num_set_150 nn_set_sentence nn_set_development pobj_on_set ' +p,5325,'In their seminal paper on SMT , Brownand his colleagues highlighted the problems weface aswe go from IBM Models 1-2 to 3-5 3 : Asweprogress from Model1toModel5 , evaluating the expectations that gives us counts becomes increasingly difficult ',Brown,'prep_highlighted_In poss_paper_their amod_paper_seminal pobj_In_paper prep_paper_on nn_Brownand_SMT pobj_on_Brownand poss_colleagues_his nsubj_highlighted_colleagues det_aswe_the nn_aswe_problems nn_aswe_weface nsubj_go_aswe ccomp_highlighted_go prep_go_from nn_Models_IBM pobj_from_Models number_3-5_1-2 dep_3-5_to num__3-5 dobj_go_ dep__3 dep__Asweprogress prep_Asweprogress_from pobj_from_Model1toModel5 xcomp_go_evaluating det_expectations_the dobj_evaluating_expectations nsubj_gives_that rcmod_expectations_gives iobj_gives_us dobj_gives_counts cop_difficult_becomes advmod_difficult_increasingly rcmod_counts_difficult ' +o,5326,'Typically , a phrase-based SMT system includes a feature that scores phrase pairs using lexical weights which are computed for two directions : source to target and target to source ',Koehn,'advmod_includes_Typically det_system_a amod_system_phrase-based nn_system_SMT nsubj_includes_system det_feature_a dobj_includes_feature complm__that nn_pairs_scores nn_pairs_phrase nsubj__pairs partmod_pairs_using amod_weights_lexical dobj_using_weights ccomp_includes_ nsubjpass_computed_which auxpass_computed_are dep_source_computed prep_computed_for num_directions_two pobj_for_directions dobj__source aux_target_to infmod_source_target cc_target_and conj_target_target prep_target_to pobj_to_source ' +o,5327,'As was demonstrated in , even a minimal set of local explicit features achieves results which are non-significantly different from a carefully chosen set of explicit features , given the language independent definition of locality described in section 2 ',Titov,'mark_demonstrated_As auxpass_demonstrated_was advcl_achieves_demonstrated prep_demonstrated_in pobj_in_ advmod_set_even det_set_a amod_set_minimal nsubj_achieves_set prep_set_of amod_features_local amod_features_explicit pobj_of_features dobj_achieves_results nsubj_different_which cop_different_are advmod_different_non-significantly rcmod_results_different prep_different_from det_set_a advmod_chosen_carefully amod_set_chosen pobj_from_set prep_set_of amod_features_explicit pobj_of_features partmod_features_given det_language_the iobj_given_language amod_definition_independent dobj_given_definition prep_definition_of pobj_of_locality partmod_locality_described prep_described_in pobj_in_section num_section_2 ' +o,5328,'The Penn Treebank results reported here for the Markov model approach are at least equivalent to those reported for the Maximum Entropy approach in ',Ratnaparkhi,'det_results_The nn_results_Penn nn_results_Treebank nsubj_reported_results advmod_reported_here mark_equivalent_for det_approach_the nn_approach_Markov nn_approach_model nsubj_equivalent_approach cop_equivalent_are advmod_equivalent_at dep_at_least advcl_reported_equivalent prep_equivalent_to pobj_to_those partmod_those_reported prep_reported_for det_approach_the nn_approach_Maximum nn_approach_Entropy pobj_for_approach prep_reported_in ' +o,5329,'Although a large number of studies have been made on learning paraphrases , for example , there are only a few studies which address the connotational difference of paraphrases ',Barzilay,'mark_made_Although det_number_a amod_number_large nsubjpass_made_number prep_number_of pobj_of_studies aux_made_have auxpass_made_been advcl_are_made prep_made_on pcomp_on_learning dobj_learning_paraphrases prep_are_for nn__example pobj_for_ expl_are_there quantmod_few_only quantmod_few_a num_studies_few nsubj_are_studies nsubj_address_which rcmod_studies_address det_difference_the amod_difference_connotational dobj_address_difference prep_difference_of pobj_of_paraphrases ' +o,5330,'3 Bi-Stream HMMs for Transliteration Standard IBM translation models can be used to obtain letter-to-letter translations ',Brown,'num_HMMs_3 amod_HMMs_Bi-Stream nsubjpass_used_HMMs prep_HMMs_for nn__Transliteration nn__Standard nn__IBM nn__translation nn__models pobj_for_ aux_used_can auxpass_used_be aux_obtain_to purpcl_used_obtain amod_translations_letter-to-letter dobj_obtain_translations ' +o,5331,'It turns out that while problems of coverage and ambiguity prevent straightforward lookup , injection of gazetteer matches as features in machine-learning based approaches is critical for good performance ',Kazama,'nsubj_turns_It prt_turns_out complm_matches_that mark_prevent_while nsubj_prevent_problems prep_problems_of pobj_of_coverage cc_coverage_and conj_coverage_ambiguity advcl_matches_prevent amod_lookup_straightforward dobj_prevent_lookup nsubj_matches_injection prep_injection_of pobj_of_gazetteer ccomp_turns_matches mark_critical_as nsubj_critical_features prep_features_in pobj_in_machine-learning partmod_machine-learning_based dobj_based_approaches cop_critical_is advcl_matches_critical prep_critical_for amod_performance_good pobj_for_performance ' +p,5332,'1 Introduction Phrase-based modeling method is a simple , but powerful mechanism to machine translation since it can model local reorderings and translations of multiword expressions well ',Koehn,'num__1 nn__Introduction amod__Phrase-based nn__modeling nn__method nsubj_mechanism_ cop_mechanism_is det_mechanism_a amod_mechanism_simple cc_simple_but conj_simple_powerful prep_mechanism_to nn_translation_machine pobj_to_translation mark_model_since nsubj_model_it aux_model_can advcl_mechanism_model amod_reorderings_local dobj_model_reorderings cc_reorderings_and conj_reorderings_translations prep_reorderings_of amod_expressions_multiword pobj_of_expressions advmod_model_well ' +o,5333,'3 The Framework 31 The Algorithm Our transductive learning algorithm , Algorithm 1 , is inspired by the Yarowsky algorithm ',Yarowsky,'det_algorithm_The nn_algorithm_Framework num_algorithm_31 nn_algorithm_The nn_algorithm_Algorithm nn_algorithm_Our nn_algorithm_transductive nn_algorithm_learning nsubjpass_inspired_algorithm appos_algorithm_Algorithm num_Algorithm_1 auxpass_inspired_is dep_3_inspired prep_inspired_by det_algorithm_the nn_algorithm_Yarowsky pobj_by_algorithm ' +o,5334,'42 Binarization Schemes Besides the baseline and iterative cost reduction binarization methods , we also perform right-heavy and random synchronous binarizations for comparison ',Zhang,'num_Schemes_42 nn_Schemes_Binarization prep_perform_Besides det__the amod__baseline pobj_Besides_ cc__and amod_methods_iterative nn_methods_cost nn_methods_reduction nn_methods_binarization conj__methods nsubj_perform_we advmod_perform_also dep_Schemes_perform amod_binarizations_right-heavy cc_right-heavy_and conj_right-heavy_random amod_binarizations_synchronous dobj_perform_binarizations prep_binarizations_for pobj_for_comparison ' +o,5335,'Independently , in AI an effort arose to encode large amounts of commonsense knowledge ',Hobbs,'dep_arose_Independently dep_Independently_in pobj_in_AI det_effort_an nsubj_arose_effort aux_encode_to xcomp_arose_encode amod_amounts_large dobj_encode_amounts prep_amounts_of amod_knowledge_commonsense pobj_of_knowledge ' +p,5336,'And again , we see this insight informing statistical machine translation systems , for instance , in the phrase-based approaches of Och and et al ',Koehn,'cc_see_And advmod_see_again nsubj_see_we det_insight_this nsubj_informing_insight dep_see_informing amod_systems_statistical nn_systems_machine nn_systems_translation dobj_informing_systems prep_systems_for pobj_for_instance prep_informing_in det_approaches_the amod_approaches_phrase-based pobj_in_approaches prep_approaches_of nn__Och pobj_of_ cc__and conj__ cc__et conj__al ' +o,5337,'1 The Baseline Maximum Entropy Model We started with a maximum entropy based tagger that uses features very similar to the ones proposed in ',Ratnaparkhi,'det_Model_The nn_Model_Baseline nn_Model_Maximum nn_Model_Entropy nsubj_features_Model nsubj_started_We rcmod_Model_started prep_started_with det_entropy_a amod_entropy_maximum pobj_with_entropy partmod_entropy_based dobj_based_tagger nsubj_uses_that rcmod_tagger_uses dep_1_features advmod_similar_very acomp_features_similar prep_similar_to det_ones_the pobj_to_ones partmod_ones_proposed prep_proposed_in ' +o,5338,'We collect substring rationales for a sentiment classification task and use them to obtain significant accuracy improvements for each annotator ',Pang,'nsubj_collect_We amod_rationales_substring dobj_collect_rationales prep_rationales_for det__a nn__sentiment nn__classification nn__task pobj_for_ cc_collect_and conj_collect_use nsubj_obtain_them aux_obtain_to xcomp_use_obtain amod_improvements_significant nn_improvements_accuracy dobj_obtain_improvements prep_obtain_for det_annotator_each pobj_for_annotator ' +o,5339,'Chunking For NP chunking , used data extracted from section 15-18 of the WSJ as a fixed train set and section 20 as a fixed test set , the same data as ',Ramshaw,'dep_Chunking_For nn_data_NP nn_data_chunking dep_data_ amod_data_used pobj_For_data partmod_data_extracted prep_extracted_from pobj_from_section num_section_15-18 prep_section_of det_WSJ_the pobj_of_WSJ prep_extracted_as det_set_a amod_set_fixed nn_set_train pobj_as_set cc_set_and conj_set_section num_section_20 prep_section_as det_set_a amod_set_fixed nn_set_test pobj_as_set det_data_the amod_data_same dep_as_data dep_as_as ' +o,5340,'Firstly , we run GIZA + + on the training corpus in both directions and then apply the ogrow-diag-finalprefinement rule to obtain many-to-many word alignments ',Och,'advmod_run_Firstly nsubj_run_we dobj_run_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_run_on det_corpus_the nn_corpus_training pobj_on_corpus prep_corpus_in det_directions_both pobj_in_directions cc_run_and advmod_run_then dep_run_apply det__the amod__ogrow-diag-finalprefinement nn__rule dobj_apply_ aux_obtain_to xcomp_apply_obtain amod_alignments_many-to-many nn_alignments_word dobj_obtain_alignments ' +o,5341,'In the multilingual parsing track , participants train dependency parsers using treebanks provided for ten languages : Arabic , Basque , Catalan , Chinese , Czech , English , Greek , Hungarian , Italian , and Turkish ',Marcus,'prep_train_In det_track_the amod_track_multilingual nn_track_parsing pobj_In_track nsubj_train_participants amod_parsers_dependency dobj_train_parsers partmod_parsers_using dobj_using_treebanks prep_using_provided dep_provided_for amod_languages_ten pobj_for_languages nn__Arabic dep_parsers_ amod__Basque conj__ nn__Catalan conj__ amod__Chinese conj__ amod__Czech conj__ nn__English conj__ amod__Greek conj__ amod__Hungarian conj__ amod__Italian conj__ cc__and amod__Turkish conj__ ' +o,5342,' , 1996 -RRB- , a single inconsistency in a test set tree will very likely yield a zero percent parse accuracy for the particular test set sentence ',Ratnaparkhi,'nsubj_yield_ dep_inconsistency_1996 det_inconsistency_a amod_inconsistency_single appos__inconsistency prep_inconsistency_in det_test_a pobj_in_test partmod_test_set dobj_set_tree aux_yield_will advmod_likely_very advmod_yield_likely det_accuracy_a num_accuracy_zero nn_accuracy_percent nn_accuracy_parse dobj_yield_accuracy prep_yield_for det_sentence_the amod_sentence_particular nn_sentence_test nn_sentence_set pobj_for_sentence ' +p,5343,' reports interesting results of this kind based on literal collocations , where he parses the corpus into predicate-argument structures and applies a mutual information measure to weigh the association between the predicate and each of its arguments ',Hindle,'advmod_reports_ amod_results_interesting dobj_reports_results prep_results_of det_kind_this pobj_of_kind partmod_kind_based prep_based_on amod_collocations_literal pobj_on_collocations advmod_parses_where nsubj_parses_he rcmod_collocations_parses det__the nn__corpus dobj_parses_ prep_parses_into amod_structures_predicate-argument pobj_into_structures cc_parses_and conj_parses_applies det__a amod__mutual nn__information nn__measure nsubj_reports_ aux_weigh_to infmod__weigh det_association_the dobj_weigh_association prep_association_between det_predicate_the pobj_between_predicate cc_association_and conj_association_each prep_each_of poss_arguments_its pobj_of_arguments ' +o,5344,'32 -LRB- m , n -RRB- - cousin Classification The classifier for learning coordinate terms relies on the notion of distributional similarity , ie , the idea that two words with similar meanings will be used in similar contexts ',Hindle,'appos_32_m dep_m_n nsubj_relies_cousin partmod_cousin_Classification det_classifier_The dobj_Classification_classifier prep_Classification_for pcomp_for_learning amod_terms_coordinate dobj_learning_terms dep_32_relies prep_relies_on det_notion_the pobj_on_notion prep_notion_of amod_similarity_distributional pobj_of_similarity appos_similarity_ie det_idea_the dep_32_idea complm_used_that num_words_two nsubjpass_used_words prep_words_with amod_meanings_similar pobj_with_meanings aux_used_will auxpass_used_be dep_idea_used prep_used_in amod_contexts_similar pobj_in_contexts ' +o,5345,'5-gram word language models in English are trained on a variety of monolingual corpora ',Brants,'amod_models_5-gram nn_models_word nn_models_language nsubjpass_trained_models prep_models_in pobj_in_English auxpass_trained_are prep_trained_on det_variety_a pobj_on_variety prep_variety_of amod_corpora_monolingual pobj_of_corpora ' +o,5346,'2 Related Work There has been a large and diverse body of research in opinion mining , with most research at the text , sentence or word level ',Pang,'num_Work_2 amod_Work_Related expl_been_There aux_been_has dep_Work_been det_body_a amod_body_large cc_large_and conj_large_diverse dobj_been_body prep_body_of pobj_of_research prep_research_in nn_mining_opinion pobj_in_mining prep_been_with amod_research_most pobj_with_research prep_been_at det__the nn__text pobj_at_ nn__sentence conj__ cc__or nn_level_word nn_level_ conj__level ' +o,5347,'This was overcome by a probabilistic model which provides probabilities of discriminating a correct parse tree among candidates of parse trees in a log-linear model or maximum entropy model with many features for parse trees ',Berger,'nsubjpass_overcome_This auxpass_overcome_was prep_overcome_by det_model_a amod_model_probabilistic pobj_by_model nsubj_provides_which rcmod_model_provides dobj_provides_probabilities prep_probabilities_of pcomp_of_discriminating det_tree_a amod_tree_correct nn_tree_parse dobj_discriminating_tree prep_discriminating_among pobj_among_candidates prep_candidates_of amod_trees_parse pobj_of_trees prep_trees_in det_model_a amod_model_log-linear pobj_in_model cc_model_or amod__maximum amod__entropy nn__model conj_model_ prep_discriminating_with amod_features_many pobj_with_features prep_features_for amod__parse nn__trees pobj_for_ ' +o,5348,'2.2 Automatic evaluation metric Since the official evaluation criterion for WMT09 is human sentence ranking, we chose to minimize a linear combination of two common evaluation metrics, BLEU and TER (Papineni et al., 2002; Snover et al., 2006), during system development and tuning: TERBLEU 2 Although we are not aware of any work demonstrating that this combination of metrics correlates better than either individually in sentence ranking, Yaser Al-Onaizan (personal communication) reports that it correlates well with the human evaluation metric HTER.',Papineni,'' +o,5349,'Examples of such contexts are verb-object relations and noun-modifier relations , which were traditionally used in word similarity tasks from non-parallel corpora ',Brown,'nsubj_relations_Examples prep_Examples_of amod_contexts_such pobj_of_contexts cop_relations_are amod_relations_verb-object cc_relations_and nn_relations_noun-modifier conj_relations_relations nsubjpass_used_which auxpass_used_were advmod_used_traditionally rcmod_relations_used prep_used_in nn_tasks_word nn_tasks_similarity pobj_in_tasks prep_used_from amod__non-parallel nn__corpora pobj_from_ ' +o,5350,'The refined grammar is estimated using a variant of the forward-backward algorithm ',Matsuzaki,'det_grammar_The amod_grammar_refined nsubjpass_estimated_grammar auxpass_estimated_is xcomp_estimated_using det_variant_a dobj_using_variant prep_variant_of det_algorithm_the amod_algorithm_forward-backward pobj_of_algorithm ' +o,5351,'(McDonald and Satta, 2007; Smith and Smith, 2007).',Smith,'' +n,5352,'A word order correlation bias , as well as the phrase structure biases in Models 4 and 5 , would be less beneficial with noisier training bitexts or for language pairs with less similar word order ',Brown,'det_bias_A nn_bias_word nn_bias_order nn_bias_correlation nsubj_beneficial_bias dep_well_as cc_bias_well dep_well_as det_biases_the nn_biases_phrase nn_biases_structure conj_bias_biases prep_biases_in num_Models_ pobj_in_Models num_Models_4 cc_4_and conj_4_5 aux_beneficial_would cop_beneficial_be advmod_beneficial_less prep_beneficial_with nn_bitexts_noisier nn_bitexts_training pobj_with_bitexts cc_with_or conj_with_for nn_pairs_language pobj_for_pairs prep_pairs_with advmod_similar_less amod_order_similar nn_order_word pobj_with_order ' +o,5353,' and discuss different ways of generalizing the tree-level crosslinguistic correspondence relation , so it is not confined to single tree nodes , thereby avoiding a continuity assumption ',Galley,'nsubj_ways_ cc__and conj__ dep_ways_discuss amod_ways_different prep_ways_of pcomp_of_generalizing det_relation_the amod_relation_tree-level amod_relation_crosslinguistic nn_relation_correspondence dobj_generalizing_relation mark_confined_so nsubjpass_confined_it auxpass_confined_is neg_confined_not advcl_ways_confined prep_confined_to amod_nodes_single nn_nodes_tree pobj_to_nodes advmod_avoiding_thereby xcomp_confined_avoiding det_assumption_a nn_assumption_continuity dobj_avoiding_assumption ' +p,5354,'31 A History-Based Model The history-based -LRB- HB -RRB- approach which incorporates more context information has worked well in parsing ',Collins,'num_Model_31 nn_Model_A nn_Model_History-Based det_approach_The amod_approach_history-based abbrev_approach_HB nsubj_worked_approach nsubj_incorporates_which rcmod_approach_incorporates amod_information_more nn_information_context dobj_incorporates_information aux_worked_has dep_Model_worked dep_in_well prep_worked_in pcomp_in_parsing ' +o,5355,'1 Introduction : Defining SCMs The work presented here was done in the context of phrase-based MT ',Och,'num_Introduction_1 nn_SCMs_Defining nsubjpass_done_SCMs det_work_The nsubj_presented_work rcmod_SCMs_presented advmod_done_here auxpass_done_was dep_Introduction_done prep_done_in det_context_the pobj_in_context prep_context_of amod_MT_phrase-based pobj_of_MT ' +o,5356,'This model is very similar to ',Smith,'det_model_This nsubj_similar_model cop_similar_is advmod_similar_very prep_similar_to ' +o,5357,'We also report the result of our translation quality in terms of both BLEU and TER against four human reference translations ',Papineni,'nsubj_report_We advmod_report_also det_result_the dobj_report_result prep_result_of poss_quality_our nn_quality_translation pobj_of_quality prep_report_in pobj_in_terms prep_terms_of preconj__both nn__BLEU pobj_of_ cc__and nn__TER conj__ prep_report_against num_translations_four amod_translations_human nn_translations_reference pobj_against_translations ' +o,5358,'One could use the estimated co-occurrences from a small sample to compute the test statistics , most commonly Pearsons chi-squared test , the likelihood ratio test , Fishers exact test , cosine similarity , or resemblance -LRB- Jaccard coefficient -RRB- ',Dunning,'nsubj_use_One aux_use_could ccomp_Pearsons_use det_co-occurrences_the amod_co-occurrences_estimated dobj_use_co-occurrences prep_use_from det_sample_a amod_sample_small pobj_from_sample aux_compute_to infmod_sample_compute det_statistics_the nn_statistics_test dobj_compute_statistics advmod_commonly_most advmod_Pearsons_commonly amod_test_chi-squared nsubj_Pearsons_test det_test_the nn_test_likelihood nn_test_ratio conj_test_test nn_test_Fishers amod_test_exact conj_test_test amod_similarity_cosine conj_test_similarity cc_test_or conj_test_resemblance nn_coefficient_Jaccard appos_resemblance_coefficient ' +o,5359,'Similar observations have been made in the context of tagging problems using maximum-entropy models ',Klein,'amod_observations_Similar nsubjpass_made_observations aux_made_have auxpass_made_been prep_made_in det_context_the pobj_in_context prep_context_of amod_problems_tagging pobj_of_problems partmod_problems_using amod_models_maximum-entropy dobj_using_models ' +o,5360,'Since most phrases appear only a few times in training data , a phrase pair translation is also evaluated by lexical weights or term weighting as additional features to avoid overestimation ',Koehn,'mark_appear_Since amod_phrases_most nsubj_appear_phrases advcl_evaluated_appear quantmod_few_only quantmod_few_a num_times_few dep_in_times prep_appear_in amod_data_training pobj_in_data det_translation_a nn_translation_phrase nn_translation_pair nsubjpass_evaluated_translation auxpass_evaluated_is advmod_evaluated_also prep_evaluated_by amod__lexical nn__weights pobj_by_ cc__or nn__term nn__weighting conj__ prep_evaluated_as amod_features_additional pobj_as_features aux_avoid_to xcomp_evaluated_avoid dobj_avoid_overestimation ' +o,5361,'2 Our statistical engine 21 The statistical models In this study , we built an SMT engine designed to translate from French to English , following the noisy-channel paradigm flrst described by ',Brown,'dep_21_2 poss_engine_Our amod_engine_statistical nsubj_21_engine det_models_The amod_models_statistical dobj_21_models prep_built_In det_study_this pobj_In_study nsubj_built_we dep_21_built det_engine_an nn_engine_SMT dobj_built_engine partmod_engine_designed aux_translate_to xcomp_designed_translate prep_translate_from pobj_from_French prep_translate_to pobj_to_English prep_designed_following det_flrst_the amod_flrst_noisy-channel nn_flrst_paradigm pobj_following_flrst partmod_flrst_described prep_described_by ' +o,5362,'A la , and Kudo and Matsumato -LRB- 2000 -RRB- , we use the IOB tagging style for modeling and classification ',Ramshaw,'det__A nn__la cc__and conj__Kudo cc_Kudo_and conj_Kudo_Matsumato appos_Kudo_2000 nsubj_use_we dep__use det_style_the nn_style_IOB amod_style_tagging dobj_use_style prep_style_for pobj_for_modeling cc_modeling_and conj_modeling_classification ' +o,5363,'CIT -RRB- ',Matsuzaki,'nn_-RRB-_CIT ' +o,5364,'5 External Knowledge Sources 51 Lexical Dependencies Features derived from n-grams of words and tags in the immediate vicinity of the word being tagged have underpinned the world of POS tagging for many years , and have proven to be useful features in WSD ',Ratnaparkhi,'num_Sources_5 nn_Sources_External nn_Sources_Knowledge num_Features_51 nn_Features_Lexical nn_Features_Dependencies nsubj_underpinned_Features partmod_Features_derived prep_derived_from pobj_from_n-grams prep_n-grams_of pobj_of_words cc_words_and conj_words_tags prep_derived_in det_vicinity_the amod_vicinity_immediate pobj_in_vicinity prep_vicinity_of det_word_the pobj_of_word auxpass_tagged_being partmod_word_tagged aux_underpinned_have dep_Sources_underpinned det_world_the dobj_underpinned_world prep_world_of pobj_of_POS dep_Sources_tagging prep_tagging_for amod__many nn__years pobj_for_ cc_tagging_and aux_proven_have conj_tagging_proven aux_features_to cop_features_be amod_features_useful xcomp_proven_features prep_features_in nn__WSD pobj_in_ ' +o,5365,'NULL -RRB- Compared with the B-Chunk and I-Chunk used in , structural relations 99 and 90 correspond to B-Chunk which represents the first word of the chunk , and structural relations 00 and 09 correspond to I-Chunk which represnts each other in the chunk while 90 also means the beginning of the sentence and 09 means the end of the sentence ',Ramshaw,'dobj_NULL_Compared prep_NULL_with det_B-Chunk_the pobj_with_B-Chunk cc_B-Chunk_and conj_B-Chunk_I-Chunk partmod_B-Chunk_used prep_used_in amod_relations_ amod_relations_structural pobj_in_relations nsubj_NULL_99 cc_99_and conj_99_90 advmod_99_correspond dep_correspond_to pobj_to_B-Chunk nsubj_represents_which rcmod_99_represents det_word_the amod_word_first dobj_represents_word prep_word_of det_chunk_the pobj_of_chunk cc_word_and amod_relations_structural conj_word_relations num_correspond_00 cc_00_and conj_00_09 dep_relations_correspond prep_represents_to pobj_to_I-Chunk nsubj_represnts_which rcmod_I-Chunk_represnts det_other_each dobj_represnts_other prep_represnts_in det_chunk_the pobj_in_chunk mark_means_while nsubj_means_90 advmod_means_also advcl_represnts_means det_beginning_the nsubj_means_beginning prep_beginning_of det_sentence_the pobj_of_sentence cc_sentence_and conj_sentence_09 ccomp_means_means det_end_the dobj_means_end prep_end_of det_sentence_the pobj_of_sentence ' +o,5366,'We trained three Arabic-English syntax-based statistical MT systems using max-B training : one on a newswire development set , one on a weblog development set , and one on a combined development set containing documents from both genres ',Galley,'nsubj_trained_We num_systems_three amod_systems_Arabic-English amod_systems_syntax-based amod_systems_statistical nn_systems_MT dobj_trained_systems partmod_systems_ xcomp__using amod__max-B nn__training dobj_using_ dep_systems_one prep_one_on det_set_a nn_set_newswire nn_set_development pobj_on_set conj_systems_one prep_one_on det_set_a amod_set_weblog nn_set_development pobj_on_set cc_systems_and conj_systems_one prep_one_on det_development_a amod_development_combined pobj_on_development partmod_development_set xcomp_set_containing dobj_containing_documents prep_trained_from det_genres_both pobj_from_genres ' +o,5367,'The reliability for the two annotation tasks -LRB- - statistics -RRB- was of 094 and 090 respectively ',Carletta,'det_reliability_The nsubj_was_reliability prep_reliability_for det_tasks_the num_tasks_two nn_tasks_annotation pobj_for_tasks measure__statistics dep_reliability_ prep_was_of pobj_of_094 cc_094_and conj_094_090 dep_of_respectively ' +o,5368,'Different models have been presented in the literature , see for instance ',Brown,'amod_models_Different nsubjpass_presented_models aux_presented_have auxpass_presented_been prep_presented_in det_literature_the pobj_in_literature ccomp_presented_see prep_see_for pobj_for_instance ' +o,5369,'Thus , our generative model is a quasi-synchronous grammar , exactly as in 3 When training on target sentences w , therefore , we tune the model parameters to maximize notsummationtextt p -LRB- t , w -RRB- as in ordinary EM , but rather 3Our task here is new ; they used it for alignment ',Smith,'advmod_grammar_Thus poss_model_our amod_model_generative nsubj_grammar_model cop_grammar_is det_grammar_a amod_grammar_quasi-synchronous dep_in_exactly dep_in_as prep_grammar_in pobj_in_ num__3 advmod_w_When nsubj_w_training prep_training_on nn_sentences_target pobj_on_sentences dep__w advmod_w_therefore nsubj_tune_we dep_w_tune det_parameters_the nn_parameters_model nsubj_maximize_parameters aux_maximize_to xcomp_tune_maximize amod_p_notsummationtextt dobj_maximize_p appos_p_t dep_t_w prep_maximize_as dep_as_in amod_EM_ordinary pobj_in_EM cc_tune_but advmod_new_rather amod_task_3Our nsubj_new_task advmod_new_here cop_new_is conj_tune_new nsubj_used_they parataxis_grammar_used dobj_used_it prep_used_for pobj_for_alignment ' +o,5370,'We then train IBM models using the GIZA + + package ',Brown,'nsubj_train_We advmod_train_then nn_models_IBM nsubj__models ccomp_train_ xcomp__using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__package dobj_using_ ' +o,5371,' shows that parsing a binary SCFG is in O -LRB- w 6 -RRB- while parsing SCFG is NP-hard in general ',Wu,'advmod_shows_ dep_is_that csubj_is_parsing det_SCFG_a amod_SCFG_binary dobj_parsing_SCFG dep_shows_is prep_is_in pobj_in_O number_6_w dep_O_6 mark_NP-hard_while amod_SCFG_parsing nsubj_NP-hard_SCFG cop_NP-hard_is advcl_is_NP-hard prep_NP-hard_in amod__general pobj_in_ ' +o,5372,'Beyond WordNet , a wide range of resources has been developed and utilized , including extensions to WordNet and resources based on automatic distributional similarity methods ',Snow,'prep_developed_Beyond nn__WordNet pobj_Beyond_ det_range_a amod_range_wide nsubjpass_developed_range prep_range_of pobj_of_resources aux_developed_has auxpass_developed_been cc_developed_and conj_developed_utilized conj_developed_including pobj_including_extensions aux_WordNet_to infmod_extensions_WordNet dobj_WordNet_ cc__and conj__resources prep_WordNet_based dep_based_on amod__automatic amod__distributional nn__similarity nn__methods pobj_on_ ' +p,5373,'In previous work , I described a Maximum Entropy\\/Minimum Divergence -LRB- MEMD -RRB- model for p -LRB- w -LRB- hi , s -RRB- which incorporates a trigram language model and a translation component which is an analog of the well-known IBM translation model 1 ',Brown,'prep_described_In amod__previous nn__work pobj_In_ nsubj_described_I det__a nn_Divergence_Maximum nn_Divergence_Entropy\\/Minimum nn__Divergence abbrev_Divergence_MEMD nn__model dobj_described_ prep__for pobj_for_p dep_p_w dep_p_hi dep_hi_s nsubj_incorporates_which rcmod__incorporates det_model_a nn_model_trigram nn_model_language dobj_incorporates_model cc_model_and det_component_a nn_component_translation conj_model_component nsubj_analog_which cop_analog_is det_analog_an rcmod_component_analog prep_analog_of det_model_the amod_model_well-known nn_model_IBM nn_model_translation pobj_of_model num_model_1 ' +o,5374,'The model was trained on sections 221 from the English Penn Treebank ',Marcus,'det_model_The nsubjpass_trained_model auxpass_trained_was prep_trained_on number_221_sections pobj_on_221 prep_trained_from det_Treebank_the nn_Treebank_English nn_Treebank_Penn pobj_from_Treebank ' +o,5375,'The modifications are made to deal with the efficiency issue due to the fact that there is a very large number of features and training samples in our task , compared to only 8 features used in ',Och,'det_modifications_The nsubjpass_made_modifications auxpass_made_are aux_deal_to xcomp_made_deal prep_deal_with det_issue_the nn_issue_efficiency pobj_with_issue amod_issue_due prep_due_to det_fact_the pobj_to_fact complm_is_that expl_is_there ccomp_deal_is det_number_a advmod_large_very amod_number_large nsubj_is_number prep_number_of pobj_of_features cc_features_and nn_samples_training conj_features_samples prep_features_in poss_task_our pobj_in_task prep_made_compared dep_compared_to quantmod_8_only num_features_8 pobj_to_features partmod_features_used prep_used_in ' +o,5376,'Let W1 , W2 be the vocabulary sizes of the two languages , and N = -LCB- A1 , , AN -RCB- be the set of nonterminals with indices 1 , , N also showed that ITGs can be equivalently be defined in two other ways ',Wu,'dep_W1_Let dep_showed_W1 nsubj_sizes_W2 cop_sizes_be det_sizes_the nn_sizes_vocabulary parataxis_showed_sizes prep_sizes_of det_languages_the num_languages_two pobj_of_languages cc_sizes_and nsubj_=_N conj_sizes_= nsubj_set_A1 nsubj_set_AN cop_set_be det_set_the ccomp_=_set prep_set_of pobj_of_nonterminals prep_set_with amod_1_indices pobj_with_1 nn__N nsubj_showed_ advmod_showed_also complm_equivalently_that nsubj_equivalently_ITGs aux_equivalently_can cop_equivalently_be ccomp_showed_equivalently auxpass_defined_be ccomp_equivalently_defined prep_defined_in num_ways_two amod_ways_other pobj_in_ways ' +o,5377,'It is shown that -LRB- 2,2 -RRB- - BRCGs induce inside-out alignments and cross-serial discontinuous translation units -LRB- CDTUs -RRB- ; both phenomena can be shown to occur frequently in many hand-aligned parallel corpora ',Wu,'nsubjpass_shown_It auxpass_shown_is dobj_shown_that appos_that_2,2 nsubj_induce_BRCGs rcmod_that_induce amod__inside-out amod__alignments dobj_induce_ cc__and amod_units_cross-serial amod_units_discontinuous nn_units_translation conj__units abbrev__CDTUs det_phenomena_both nsubjpass_shown_phenomena aux_shown_can auxpass_shown_be parataxis_shown_shown aux_occur_to purpcl_shown_occur advmod_occur_frequently prep_occur_in amod_corpora_many amod_corpora_hand-aligned amod_corpora_parallel pobj_in_corpora ' +o,5378,'ps -LRB- arc -RRB- is increased by 1110 1 \\/ -LRB- k +1 -RRB- if the hypothesis ranking k in the system s contains the arc ',Rosti,'nsubjpass_increased_ps appos_ps_arc auxpass_increased_is prep_increased_by number_1_1110 num_\\/_1 pobj_by_\\/ number_+1_k dep_\\/_+1 mark_contains_if det_k_the nn_k_hypothesis amod_k_ranking nsubj_contains_k prep_k_in det_s_the nn_s_system pobj_in_s advcl_increased_contains det_arc_the dobj_contains_arc ' +o,5379,'1 Introduction Word alignment is a critical component in training statistical machine translation systems and has received a significant amount of research , for example , , including work leveraging syntactic parse trees , eg , ',Fraser,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubj_component_alignment cop_component_is det_component_a amod_component_critical prep_component_in amod_systems_training amod_systems_statistical nn_systems_machine nn_systems_translation pobj_in_systems cc_component_and aux_received_has conj_component_received det_amount_a amod_amount_significant dobj_received_amount prep_amount_of pobj_of_research prep_received_for pobj_for_example conj_component_ conj_component_including pobj_including_work partmod_work_leveraging amod_trees_syntactic nn_trees_parse nsubj__trees appos_trees_eg xcomp_leveraging_ ' +o,5380,'21 Word Sequence Classification Similar to English text chunking , the word sequence classification model aims to classify each word via encoding its context features ',Ramshaw,'num_Classification_21 nn_Classification_Word nn_Classification_Sequence nsubj_aims_Classification amod_Classification_Similar prep_Similar_to amod__English nn__text nn__chunking pobj_to_ det_model_the nn_model_word nn_model_sequence nn_model_classification appos__model aux_classify_to xcomp_aims_classify det_word_each dobj_classify_word prep_word_via pcomp_via_encoding poss_features_its nn_features_context dobj_encoding_features ' +o,5381,'Finally , following Haghighi and Klein and we can instead insist that at most one HMM state can be mapped to any part-of-speech tag ',Johnson,'advmod_insist_Finally dep_insist_following nsubj__Haghighi cc_Haghighi_and conj_Haghighi_Klein xcomp_following_ cc__and conj__ nsubj_insist_we aux_insist_can advmod_insist_instead dobj_insist_that dep_most_at advmod_insist_most num_state_one nn_state_HMM nsubjpass_mapped_state aux_mapped_can auxpass_mapped_be dep_insist_mapped prep_mapped_to det_tag_any amod_tag_part-of-speech pobj_to_tag ' +o,5382,'WLCS -LRB- w , d -RRB- = summationtextmi = 0 f -LRB- ki -RRB- We then compute the following quantities , where is word length , and f1 is the inverse of f P -LRB- w , d -RRB- = f1 -LRB- WLCS -LRB- w , d -RRB- f -LRB- w -RRB- -RRB- R -LRB- w , d -RRB- = f1 -LRB- WLCS -LRB- w , d -RRB- f -LRB- d -RRB- -RRB- F -LRB- w , d -RRB- = -LRB- 1 +2 -RRB- R -LRB- w , d -RRB- P -LRB- w , d -RRB- R -LRB- w , d -RRB- +2 P -LRB- w , d -RRB- In effect , P -LRB- w , d -RRB- examines how close the longest common substring is to w and R -LRB- w , d -RRB- how close it is to d Following , we use = 8 , assigninggreaterimportancetoR -LRB- w , d -RRB- ',Lin,'dep_quantities_w appos_w_d dep_quantities_= dobj_=_summationtextmi dep_summationtextmi_= iobj_=_0 dobj_=_f appos_f_ki nsubj_quantities_We advmod_quantities_then dep_quantities_compute det_quantities_the amod_quantities_following dep_WLCS_quantities advmod_length_where cop_length_is nn_length_word advcl_quantities_length cc_quantities_and nsubj_inverse_f1 cop_inverse_is det_inverse_the conj_quantities_inverse prep_inverse_of pobj_of_f nsubj_=_P dep_P_w appos_w_d rcmod_f_= dobj_=_f1 dep_f1_WLCS dep_WLCS_w appos_w_d dep_WLCS_f appos_f_w partmod_f1_R nsubj_=_w appos_w_d ccomp_R_= amod_F_f1 dep_F_WLCS dep_WLCS_w appos_w_d dep_WLCS_f dep_WLCS_d dobj_=_F dep_F_w appos_w_d dep_F_= number_+2_1 dep_f1_+2 nsubj_P_R dep_R_w appos_w_d rcmod_f_P dep_P_w appos_w_d dobj_P_R dep_R_w appos_w_d num_P_+2 dep_examines_P dep_P_w appos_w_d prep_examines_In pobj_In_effect nsubj_examines_P dep_P_w appos_w_d rcmod_R_examines advmod_is_how advmod_how_close det_substring_the amod_substring_longest amod_substring_common nsubj_is_substring ccomp_examines_is prep_is_to pobj_to_w cc_w_and conj_w_R dep_w_w appos_w_d advmod_is_how advmod_how_close nsubj_is_it dep_w_is prep_is_to pobj_to_d prep_d_Following pobj_Following_ nsubj_use_we dep_examines_use dep_8_= ccomp_use_8 nsubj_WLCS_assigninggreaterimportancetoR dep_assigninggreaterimportancetoR_w appos_w_d ' +o,5383,'Our corpora were automatically aligned with Giza + + in both directions between source and target and symmetrised using the intersection heuristic ',Koehn,'poss_corpora_Our nsubjpass_aligned_corpora auxpass_aligned_were advmod_aligned_automatically prep_aligned_with pobj_with_Giza amod_Giza_+ cc_+_+ conj_+_ prep_aligned_in det_directions_both pobj_in_directions prep_directions_between pobj_between_source cc_source_and conj_source_target cc_aligned_and conj_aligned_symmetrised xcomp_symmetrised_using det_intersection_the nsubj_heuristic_intersection xcomp_using_heuristic ' +o,5384,'The initial state contains terminal items , whose labels are the POS tags given by the tagger of ',Ratnaparkhi,'det_state_The amod_state_initial nsubj_contains_state amod_items_terminal dobj_contains_items poss_labels_whose nsubj_tags_labels cop_tags_are det_tags_the dep_tags_POS rcmod_items_tags prep_tags_given dep_given_by det_tagger_the pobj_by_tagger prep_tagger_of ' +o,5385,'Previous approaches include supervised learning , , vectorial similarity computed between an initial abstract and sentences in the given document , intradocument similarities , or graph algorithms , , ',Mihalcea,'amod_approaches_Previous nsubj_include_approaches amod__supervised nn__learning dobj_include_ conj__ amod_similarity_vectorial conj__similarity partmod_similarity_computed prep_computed_between det_abstract_an amod_abstract_initial pobj_between_abstract cc_abstract_and conj_abstract_sentences prep_computed_in det_document_the amod_document_given pobj_in_document amod_similarities_intradocument conj__similarities partmod_similarities_ cc__or nn_algorithms_graph conj__algorithms amod_algorithms_ dep__ dep__ ' +o,5386,'We then describe the two main paradigms for learning and inference , in this years shared task as well as in last years , which we call transition-based parsers -LRB- section 52 -RRB- and graph-based parsers -LRB- section 53 -RRB- , adopting the terminology of 5 Finally , we give an overview of the domain adaptation methods that were used -LRB- section 54 -RRB- ',McDonald,'nsubj_describe_We advmod_describe_then det_paradigms_the num_paradigms_two amod_paradigms_main dobj_describe_paradigms prep_paradigms_for pobj_for_learning cc_learning_and conj_learning_inference prep_describe_in det_years_this pobj_in_years partmod_years_shared dobj_shared_task dep_well_as cc_in_well dep_well_as conj_in_in amod_years_last pobj_in_years dobj_call_which nsubj_call_we rcmod_years_call amod_parsers_transition-based dobj_call_parsers appos_parsers_section num_section_52 cc_parsers_and amod_parsers_graph-based conj_parsers_parsers appos_parsers_section num_section_53 partmod_parsers_adopting det_terminology_the dobj_adopting_terminology prep_terminology_of pobj_of_ num__5 advmod_overview_Finally nsubj_overview_we dep_overview_give det_overview_an dep_describe_overview prep_overview_of det_domain_the pobj_of_domain nn_methods_adaptation dobj_overview_methods complm_used_that auxpass_used_were ccomp_overview_used dep_used_section num_section_54 ' +o,5387,'This is similar to Model 3 of , but without null-generated elements or re-ordering ',Brown,'nsubj_similar_This cop_similar_is prep_similar_to pobj_to_Model num_Model_3 prep_Model_of pobj_of_ cc_similar_but conj_similar_without amod_elements_null-generated pobj_without_elements cc_elements_or conj_elements_re-ordering ' +o,5388,' , , , , , -RRB- , and to pick those ingredients which are known to be con ~ i -RRB- utationally ` tractable \' in some sense ',Marcus,'dep__ dobj__ dep__ dep__ dep__ cc__and aux_pick_to conj__pick det_ingredients_those dobj_pick_ingredients nsubjpass_known_which auxpass_known_are rcmod_ingredients_known aux_~_to cop_~_be amod_~_con xcomp_known_~ dep_sense_i nn_tractable_utationally nsubj_sense_tractable prep_tractable_in pobj_in_some dep__sense ' +o,5389,'From this data , we use the the GHKM minimal-rule extraction algorithm of to yield rules like : NP-C -LRB- x0 : NPB PP -LRB- IN -LRB- of x1 : NPB -RRB- -RRB- $ x1 de x0 Though this rule can be used in either direction , here we use it right-to-left -LRB- Chinese to English -RRB- ',Galley,'prep_use_From det_data_this pobj_From_data nsubj_use_we dep_algorithm_the det_algorithm_the nn_algorithm_GHKM amod_algorithm_minimal-rule nn_algorithm_extraction nsubj_yield_algorithm prep_algorithm_of pobj_of_ aux_yield_to xcomp_use_yield dobj_yield_rules prep_rules_like nn_x0_NP-C pobj_like_x0 nn_PP_NPB dep_x0_PP dep_rules_IN dep_NPB_of pobj_of_x1 dep_-LRB-_NPB dep_rules_$ num_$_x1 prep_$_de pobj_de_x0 mark_used_Though det_rule_this nsubjpass_used_rule aux_used_can auxpass_used_be advcl_use_used prep_used_in det_direction_either pobj_in_direction advmod_use_here nsubj_use_we dep_use_use nsubj_right-to-left_it xcomp_use_right-to-left dep_right-to-left_Chinese prep_Chinese_to pobj_to_English ' +o,5390,'In the years since the appearance of the first papers on using statistical models for bilingual lexicon compilation and machine translation , large amount of human effort and time has been invested in collecting parallel corpora of translated texts ',Brown,'prep_invested_In det_years_the pobj_In_years prep_years_since det_appearance_the pobj_since_appearance prep_appearance_of det_papers_the amod_papers_first pobj_of_papers prep_papers_on pcomp_on_using amod_models_statistical dobj_using_models prep_invested_for amod_compilation_bilingual nn_compilation_lexicon pobj_for_compilation cc_compilation_and nn__machine nn__translation conj_compilation_ amod_amount_large nsubjpass_invested_amount prep_amount_of amod_effort_human pobj_of_effort cc_effort_and conj_effort_time aux_invested_has auxpass_invested_been prep_invested_in pcomp_in_collecting amod_corpora_parallel dobj_collecting_corpora prep_corpora_of amod_texts_translated pobj_of_texts ' +o,5391,'Much previous work on unsupervised grammar induction has used gold-standard partof-speech tags ',Smith,'amod_work_Much amod_work_previous nsubj_used_work prep_work_on amod_induction_unsupervised nn_induction_grammar pobj_on_induction aux_used_has amod_tags_gold-standard amod_tags_partof-speech dobj_used_tags ' +o,5392,'The translation model used in (Koehn et al. , 2003) is the product of translation probability a34a35a4 a29 a0 a33 a6 a29 a2 a33 a8 and distortion probability a36a37a4a39a38 a33a41a40a43a42a44a33a46a45 a32 a8, a3a5a4a35a29 a0 a30 a32 a6 a29 a2 a30 a32 a8 a10 a30 a47 a33a49a48 a32 a34a35a4 a29 a0a22a33 a6 a29 a2 a33a50a8 a36a51a4a39a38 a33 a40a52a42 a33a53a45 a32 a8 (1) where a38 a33 denotes the start position of the source phrase translated into the a54 -th target phrase, and a42 a33a53a45 a32 denotes the end position of the source phrase translated into the a4a53a54 a40a56a55 a8 -th target phrase.',Koehn,'' +o,5393,'All corpora are formatted in the IOB sequence representation ',Ramshaw,'det_corpora_All nsubjpass_formatted_corpora auxpass_formatted_are prep_formatted_in det_representation_the nn_representation_IOB nn_representation_sequence pobj_in_representation ' +o,5394,'For a full discussion of previous work , please see , or see for work relating to synonym resolution ',Yates,'prep_please_For det_discussion_a amod_discussion_full pobj_For_discussion prep_discussion_of amod_work_previous pobj_of_work dep_please_see dobj_see_ cc_see_or conj_see_see dobj_see_ prep_see_for pobj_for_work partmod_work_relating aux_synonym_to xcomp_relating_synonym dobj_synonym_resolution ' +o,5395,'37 3 Semi-supervised Domain Adaptation 31 Structural Correspondence Learning Structural Correspondence Learning exploits unlabeled data from both source and target domain to find correspondences among features from different domains ',Blitzer,'number_3_37 num_data_3 amod_data_Semi-supervised nn_data_Domain nn_data_Adaptation num_data_31 amod_data_Structural nn_data_Correspondence nn_data_Learning nn_data_Structural nn_data_Correspondence nn_data_Learning num_exploits_ measure_unlabeled_exploits amod_data_unlabeled prep_data_from preconj_domain_both nn_domain_source cc_source_and conj_source_target pobj_from_domain aux_find_to infmod_data_find dobj_find_correspondences prep_find_among pobj_among_features prep_features_from amod_domains_different pobj_from_domains ' +o,5396,'The syntactic and part-of-speech informations were obtained from the part of the corpus processed in the Penn Treebank project ',Marcus,'det_informations_The amod_informations_syntactic cc_syntactic_and conj_syntactic_part-of-speech nsubjpass_obtained_informations auxpass_obtained_were prep_obtained_from det_part_the pobj_from_part prep_part_of det_corpus_the pobj_of_corpus partmod_corpus_processed prep_processed_in det_project_the nn_project_Penn nn_project_Treebank pobj_in_project ' +o,5397,'The recall problem is usually addressed by increasing the amount of text data for extraction -LRB- taking larger collections -RRB- or by developing more surface patterns ',Fleischman,'det_problem_The nn_problem_recall nsubjpass_addressed_problem auxpass_addressed_is advmod_addressed_usually prep_addressed_by pcomp_by_increasing det_amount_the dobj_increasing_amount prep_amount_of nn_data_text pobj_of_data prep_increasing_for pobj_for_extraction dep_for_taking amod_collections_larger dobj_taking_collections advmod_taking_ cc_for_or conj_for_by pcomp_by_developing amod_patterns_more nn_patterns_surface dobj_developing_patterns advmod_developing_ ' +n,5398,'Although various approaches to SMT system combination have been explored , including enhanced combination model structure , better word alignment between translations and improved confusion network construction , most previous work simply used the ensemble of SMT systems based on different models and paradigms at hand and did not tackle the issue of how to obtain the ensemble in a principled way ',Rosti,'mark_explored_Although amod_approaches_various nsubj_explored_approaches prep_approaches_to nn_combination_SMT nn_combination_system pobj_to_combination aux_explored_have dep_explored_been advcl_used_explored prep_explored_including amod__enhanced nn__combination nn__model nn__structure pobj_including_ amod_alignment_better nn_alignment_word appos__alignment prep_alignment_between nn__translations pobj_between_ cc__and amod__improved nn__confusion nn__network nn__construction conj__ advmod_work_most amod_work_previous nsubj_used_work advmod_used_simply det_ensemble_the dobj_used_ensemble prep_ensemble_of nn_systems_SMT pobj_of_systems partmod_systems_based prep_based_on amod_models_different pobj_on_models cc_models_and conj_models_paradigms prep_based_at pobj_at_hand cc_used_and aux_tackle_did neg_tackle_not conj_used_tackle det_issue_the dobj_tackle_issue prep_issue_of advmod_obtain_how aux_obtain_to pcomp_of_obtain det_ensemble_the dobj_obtain_ensemble prep_ensemble_in det_way_a amod_way_principled pobj_in_way ' +p,5399,'a list of pilot terms ranked from the most representative of the corpus to the least thanks to the Loglikelihood coefficient introduced by ',Dunning,'det_list_a nsubj_ranked_list prep_list_of nn_terms_pilot pobj_of_terms prep_ranked_from det_representative_the amod_representative_most pobj_from_representative prep_representative_of det_corpus_the pobj_of_corpus prep_ranked_to det_least_the pobj_to_least dep_to_thanks advmod_introduced_to det_coefficient_the nn_coefficient_Loglikelihood pobj_to_coefficient partmod_least_introduced prep_introduced_by ' +o,5400,'The sentences were processed using Collins parser to generate parse-trees automatically ',Collins,'det_sentences_The nsubjpass_processed_sentences auxpass_processed_were xcomp_processed_using nn__Collins nn__parser nsubj_generate_ aux_generate_to xcomp_using_generate dobj_generate_parse-trees advmod_generate_automatically ' +o,5401,' on CTB 50 and on CTB 40 since they reported the best performances on joint word segmentation and POS tagging using the training materials only derived from the corpora ',Zhang,'prep__on pobj_on_CTB num_CTB_50 cc__and conj__ prep__on pobj_on_CTB num_CTB_40 mark_reported_since nsubj_reported_they advcl__reported det_performances_the amod_performances_best dobj_reported_performances prep_performances_on amod_segmentation_joint nn_segmentation_word pobj_on_segmentation cc_segmentation_and dep_tagging_POS conj_segmentation_tagging partmod_segmentation_using det_materials_the nn_materials_training dobj_using_materials advmod_derived_only partmod_materials_derived prep_derived_from det_corpora_the pobj_from_corpora ' +o,5402,'Our method is based on the Extended String Subsequence Kernel -LRB- ESK -RRB- which is a kind of convolution kernel ',Lin,'poss_method_Our nsubjpass_based_method auxpass_based_is prep_based_on det_String_the nn_String_Extended pobj_on_String nn__Subsequence nn__Kernel abbrev__ESK tmod_on_ nsubj_kind_which cop_kind_is det_kind_a dep__kind prep_kind_of nn__convolution nn__kernel pobj_of_ ' +o,5403,' -RRB- ',Banerjee,'nn_-RRB-_ ' +o,5404,'Related work includes , Zens and Ney -LRB- 2003 -RRB- and Wellington et al ',Wu,'amod_work_Related nsubj_includes_work dobj_includes_ conj__Zens cc__and conj__Ney appos__2003 cc__and conj__Wellington cc_Wellington_et conj_Wellington_al ' +o,5405,'Lastly , collocations are domain-dependent and language-dependent ',Smadja,'advmod__Lastly nsubj__collocations cop__are amod__domain-dependent cc__and conj__language-dependent ' +o,5406,'To further emphasize the importance of morphology in MT to Czech , we compare the standard BLEU of a baseline phrasebased translation with BLEU which disregards word forms -LRB- lemmatized MT output is compared to lemmatized reference translation -RRB- ',Papineni,'aux_emphasize_To advmod_emphasize_further dep__emphasize det_importance_the dobj_emphasize_importance prep_importance_of pobj_of_morphology prep_emphasize_in pobj_in_MT prep_emphasize_to pobj_to_Czech nsubj__we dep__compare det__the amod__standard amod__BLEU prep__of det_translation_a amod_translation_baseline amod_translation_phrasebased pobj_of_translation prep_translation_with pobj_with_BLEU nsubj_disregards_which rcmod__disregards nn_forms_word dobj_disregards_forms amod_output_lemmatized nn_output_MT nsubj_is_output parataxis__is prep_is_compared dep_compared_to amod_translation_lemmatized nn_translation_reference pobj_to_translation ' +o,5407,'Word features are introduced primarily to help with unknown words , as in ',Marcus,'amod_features_Word nsubjpass_introduced_features auxpass_introduced_are advmod_introduced_primarily aux_help_to xcomp_introduced_help prep_help_with amod_words_unknown pobj_with_words dep_with_as dep_with_in pobj_in_ ' +p,5408,'This paper is heavily indebted to prior work on unsupervised learning of position categories such as et al 1992 , Schtze 1997 , Higgins 2002 , and others cited there ',Brown,'det_paper_This nsubj_indebted_paper cop_indebted_is advmod_indebted_heavily dep_cited_indebted prep_indebted_to amod_work_prior pobj_to_work prep_work_on amod_learning_unsupervised pobj_on_learning prep_learning_of nn_categories_position pobj_of_categories dep_as_such prep_categories_as nn_al_ nn_al_et pobj_as_al num_al_1992 conj_al_Schtze num_Schtze_1997 conj_al_Higgins num_Higgins_2002 cc_al_and conj_al_others nsubj_cited_there ' +o,5409,'As a basis mapping function we used a generalisation of the one used by Grefenstette and ',Lin,'prep_used_As det_function_a nn_function_basis nn_function_mapping pobj_As_function nsubj_used_we det_generalisation_a dobj_used_generalisation prep_generalisation_of det_one_the pobj_of_one partmod_one_used prep_used_by nn_and_Grefenstette nn_and_ pobj_by_and ' +n,5410,'1 Introduction The field of machine translation has seen many advances in recent years , most notably the shift from word-based to phrasebased models which use token n-grams as translation units ',Brown,'num_Introduction_1 det_field_The nsubj_seen_field prep_field_of nn_translation_machine pobj_of_translation aux_seen_has dep_Introduction_seen amod_advances_many dobj_seen_advances prep_seen_in amod_years_recent pobj_in_years advmod_notably_most advmod_years_notably det_shift_the dep_years_shift prep_shift_from amod__word-based pobj_from_ prep_seen_to amod_models_phrasebased pobj_to_models nsubj_use_which rcmod_models_use amod_n-grams_token dobj_use_n-grams prep_use_as nn__translation nn__units pobj_as_ ' +o,5411,'Given a set of features and a training corpus , the MaxEnt estimation process produces a model in which every feature fi has a weight i We can compute the conditional probability as : p -LRB- o h -RRB- = 1Z -LRB- h -RRB- productdisplay i ifi -LRB- h , o -RRB- -LRB- 1 -RRB- Z -LRB- h -RRB- = summationdisplay o productdisplay i ifi -LRB- h , o -RRB- -LRB- 2 -RRB- The conditional probability of the outcome is the product of the weights of all active features , normalized over the products of all the features ',Berger,'prep_produces_Given det_set_a dep_Given_set prep_set_of pobj_of_features cc_features_and det_corpus_a nn_corpus_training conj_features_corpus det_process_the nn_process_MaxEnt nn_process_estimation nsubj_produces_process det_model_a dobj_produces_model rel_has_in pobj_in_which det_fi_every nn_fi_feature nsubj_has_fi rcmod_model_has det_i_a nn_i_weight dobj_has_i nsubj_compute_We aux_compute_can rcmod_i_compute det_probability_the amod_probability_conditional dobj_compute_probability prep_compute_as pobj_as_ nsubj_=_p nn_h_o appos_p_h rcmod_model_= amod_ifi_1Z appos_ifi_h advmod_i_productdisplay amod_ifi_i dobj_=_ifi appos_ifi_h dep_h_o dep_product_1 nn_h_Z nsubj_product_h dep_h_= amod_o_summationdisplay iobj_=_o advmod_i_productdisplay amod_ifi_i dobj_=_ifi appos_ifi_h dep_h_o dep_probability_2 det_probability_The amod_probability_conditional dep_h_probability prep_probability_of det_outcome_the pobj_of_outcome cop_product_is det_product_the dep_produces_product prep_product_of det_weights_the pobj_of_weights prep_weights_of det_features_all amod_features_active pobj_of_features partmod_features_normalized prep_normalized_over det_products_the pobj_over_products prep_products_of predet_features_all det_features_the pobj_of_features ' +o,5412,'As discussed in , undirected graphical models do not seem to be suitable for history-based parsing models ',Titov,'mark_discussed_As advcl_seem_discussed prep_discussed_in pobj_in_ amod_models_undirected amod_models_graphical nsubj_seem_models aux_seem_do neg_seem_not aux_suitable_to cop_suitable_be xcomp_seem_suitable prep_suitable_for amod_models_history-based amod_models_parsing pobj_for_models ' +o,5413,'One is to use a stochastic gradient descent -LRB- SGD -RRB- or Perceptron like online learning algorithm to optimize the weights of these features directly for MT ',Liang,'nsubj_is_One aux_use_to xcomp_is_use det_descent_a amod_descent_stochastic nn_descent_gradient dobj_use_descent abbrev_descent_SGD cc_descent_or conj_descent_Perceptron prep_Perceptron_like amod_algorithm_online nn_algorithm_learning pobj_like_algorithm aux_optimize_to xcomp_use_optimize det_weights_the dobj_optimize_weights prep_weights_of det_features_these pobj_of_features dep_for_directly prep_optimize_for pobj_for_MT ' +o,5414,'For instance , one might be interested in frequencies of co-occurences of a word with other words and phrases -LRB- collocations -RRB- , or one might be interested in inducing wordclasses from the text by collecting frequencies of the left and right context words for a word in focus ',Smadja,'prep_interested_For pobj_For_instance nsubj_interested_one aux_interested_might cop_interested_be prep_interested_in pobj_in_frequencies prep_frequencies_of pobj_of_co-occurences prep_co-occurences_of det_word_a pobj_of_word prep_interested_with amod_words_other pobj_with_words cc_interested_and nsubj__phrases appos_phrases_collocations conj_interested_ cc_interested_or nsubj_interested_one aux_interested_might cop_interested_be conj_interested_interested prep_interested_in pcomp_in_inducing dobj_inducing_wordclasses prep_inducing_from det_text_the pobj_from_text prep_inducing_by pcomp_by_collecting dobj_collecting_frequencies prep_frequencies_of det_words_the amod_words_left cc_left_and conj_left_right nn_words_context pobj_of_words prep_words_for det_word_a pobj_for_word prep_word_in nn__focus pobj_in_ ' +o,5415,'Similar to Goldwater and Griffiths and , Toutanova and also use Bayesian inference for POS tagging ',Johnson,'dep_use_Similar prep_Similar_to nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths pobj_to_ cc__and nn__ conj__ conj__Toutanova cc__and nn__ conj__ advmod_use_also nn_inference_Bayesian nsubj_use_inference prep_use_for dep_tagging_POS pobj_for_tagging ' +o,5416,'Starting from the parallel training corpus , provided with direct and inverted alignments , the socalled union alignment is computed ',Och,'prep_computed_Starting dep_Starting_from det_corpus_the amod_corpus_parallel nn_corpus_training pobj_from_corpus prep_computed_provided dep_provided_with amod_alignments_direct cc_direct_and conj_direct_inverted pobj_with_alignments det__the amod__socalled nn__union nn__alignment nsubjpass_computed_ auxpass_computed_is ' +o,5417,'While theoretically sound , this approach is computationally challenging both in practice and in theory , may suffer from reference reachability problems , and in the end may lead to inferior translation quality ',Koehn,'mark_sound_While advmod_sound_theoretically advcl_challenging_sound det_approach_this nsubj_challenging_approach cop_challenging_is advmod_challenging_computationally preconj_in_both prep_challenging_in nn__practice pobj_in_ cc_in_and conj_in_in nn__theory pobj_in_ aux_suffer_may ccomp_challenging_suffer prep_suffer_from pobj_from_reference amod_problems_reachability nsubj__problems dep_challenging_ cc__and prep_lead_in det_end_the pobj_in_end aux_lead_may conj__lead prep_lead_to amod_quality_inferior nn_quality_translation pobj_to_quality ' +o,5418,'Proceedings of the 40th Annual Meeting of the Association for In a key step for locating important sentences , NeATS computes the likelihood ratio to identify key concepts in unigrams , bigrams , and trigrams1 , using the ontopic document collection as the relevant set and the off-topic document collection as the irrelevant set ',Dunning,'prep_Proceedings_of det_Meeting_the amod_Meeting_40th amod_Meeting_Annual pobj_of_Meeting prep_Meeting_of det_Association_the pobj_of_Association mark_computes_for prep_computes_In det_step_a amod_step_key pobj_In_step prep_step_for pcomp_for_locating amod_sentences_important dobj_locating_sentences nsubj_computes_NeATS dep_Proceedings_computes det_ratio_the nn_ratio_likelihood nsubj__ratio ccomp_computes_ aux_identify_to xcomp__identify amod_concepts_key dobj_identify_concepts prep_identify_in nn_trigrams1_unigrams conj_trigrams1_bigrams cc_trigrams1_and pobj_in_trigrams1 xcomp__using det_collection_the amod_collection_ontopic nn_collection_document dobj_using_collection prep_using_as det_set_the amod_set_relevant pobj_as_set cc_set_and det_collection_the amod_collection_off-topic nn_collection_document conj_set_collection prep_using_as det_set_the amod_set_irrelevant pobj_as_set ' +o,5419,'5 Results We present results that compare our system against the baseline Pharaoh implementation and MER training scripts provided for this workshop ',Koehn,'num_Results_5 nsubj_present_We dep_Results_present dobj_present_results nsubj_compare_that rcmod_results_compare poss_system_our dobj_compare_system prep_system_against det__the amod__baseline nn__Pharaoh nn__implementation pobj_against_ cc__and nn_scripts_MER nn_scripts_training conj__scripts prep_compare_provided dep_provided_for det_workshop_this pobj_for_workshop ' +o,5420,'Statistical approaches , which depend on a set of unknown parameters that are learned from training data , try to describe the relationship between a bilingual sentence pair ',Brown,'nn_approaches_Statistical nsubj_try_approaches nsubj_depend_which rcmod_approaches_depend prep_depend_on det_set_a pobj_on_set prep_set_of amod_parameters_unknown pobj_of_parameters nsubjpass_learned_that auxpass_learned_are rcmod_set_learned prep_learned_from nn_data_training pobj_from_data aux_describe_to xcomp_try_describe det_relationship_the dobj_describe_relationship prep_relationship_between det_pair_a amod_pair_bilingual nn_pair_sentence pobj_between_pair ' +o,5421,'42 Impact of Paraphrases on Machine Translation Evaluation The standard way to analyze the performance of an evaluation metric in machine translation is to compute the Pearson correlation between the automatic metric and human scores ',Papineni,'num_Impact_42 nsubj_is_Impact prep_Impact_of pobj_of_Paraphrases prep_Paraphrases_on nn_Evaluation_Machine nn_Evaluation_Translation pobj_on_Evaluation det_way_The amod_way_standard dep_Evaluation_way aux_analyze_to infmod_Paraphrases_analyze det_performance_the dobj_analyze_performance prep_performance_of det_metric_an nn_metric_evaluation pobj_of_metric prep_analyze_in nn_translation_machine pobj_in_translation aux_compute_to xcomp_is_compute det_correlation_the nn_correlation_Pearson dobj_compute_correlation prep_correlation_between dep_automatic_the pobj_between_automatic dep_automatic_metric cc_automatic_and amod_scores_human conj_automatic_scores ' +o,5422,'In fact , many studies that try to exploit Wikipedia as a knowledge source have recently emerged ',Ponzetto,'prep_emerged_In pobj_In_fact amod_studies_many nsubj_emerged_studies nsubj_try_that rcmod_studies_try aux_exploit_to xcomp_try_exploit dobj_exploit_Wikipedia prep_Wikipedia_as det_source_a nn_source_knowledge pobj_as_source aux_emerged_have advmod_emerged_recently ' +o,5423,'These texts were not seen at the training phase which means that neither the 6Since Brill \'s tagger was trained on the Penn tag-set we provided an additional mapping ',Marcus,'det_texts_These nsubjpass_seen_texts auxpass_seen_were neg_seen_not prep_seen_at det_phase_the nn_phase_training pobj_at_phase nsubj_means_which rcmod_phase_means complm_trained_that dep_6Since_neither det_6Since_the nsubjpass_trained_6Since poss_tagger_Brill possessive_Brill_\'s dep_6Since_tagger auxpass_trained_was ccomp_means_trained prep_trained_on det__the nn__Penn amod__tag-set pobj_on_ nsubj_provided_we dep_seen_provided det_mapping_an amod_mapping_additional dobj_provided_mapping ' +o,5424,'also and Nivre , 2007 -RRB- ',McDonald,'advmod__also cc__and conj__Nivre appos__2007 ' +o,5425,'4 Corpus Annotation For our corpus , we selected 1,000 sentences containing at least one comma from the Penn Treebank WSJ section 00 , and manually annotated them with comma information3 ',Marcus,'num_Annotation_4 nn_Annotation_Corpus prep_selected_For poss_corpus_our pobj_For_corpus nsubj_selected_we dep_Annotation_selected num_sentences_1,000 dobj_selected_sentences dep_Annotation_containing quantmod_one_at dep_at_least num_comma_one dobj_containing_comma prep_containing_from det_section_the nn_section_Penn nn_section_Treebank nn_section_ nn_section_WSJ pobj_from_section num_section_00 cc_containing_and advmod_containing_manually conj_containing_annotated dobj_annotated_them prep_annotated_with nn_information3_comma pobj_with_information3 ' +o,5426,'But we did not use any LM estimate to achieve early stopping as suggested by ',Huang,'cc_use_But nsubj_use_we aux_use_did neg_use_not det_estimate_any nn_estimate_LM nsubj_achieve_estimate aux_achieve_to xcomp_use_achieve amod_stopping_early dobj_achieve_stopping mark_suggested_as advcl_achieve_suggested prep_suggested_by ' +p,5427,'Recent work by shows a practically ef cient approach that binarizes linguistically SCFG rules when possible ',Zhang,'amod_work_Recent nsubj_shows_work prep_work_by pobj_by_ det_approach_a advmod_approach_practically amod_approach_ef nn_approach_cient dobj_shows_approach nsubj_binarizes_that rcmod_approach_binarizes amod_rules_linguistically nn_rules_SCFG dobj_binarizes_rules dep_possible_when dep_binarizes_possible ' +o,5428,'More details on these standard criteria can be found for instance in ',Och,'amod_details_More nsubjpass_found_details prep_details_on det_criteria_these amod_criteria_standard pobj_on_criteria aux_found_can auxpass_found_be prep_found_for pobj_for_instance prep_found_in ' +o,5429,'3However , the binary-branching SCFGs used by and Alshawi et al ',Wu,'det_SCFGs_the amod_SCFGs_binary-branching appos_3However_SCFGs partmod_SCFGs_used prep_used_by pobj_by_ cc__and conj__Alshawi cc_Alshawi_et conj_Alshawi_al ' +o,5430,'Recently there have been some studies addressing domain adaptation from different perspectives ',Blitzer,'advmod_been_Recently expl_been_there aux_been_have det_studies_some nsubj_addressing_studies dep_been_addressing nn_adaptation_domain dobj_addressing_adaptation prep_addressing_from amod_perspectives_different pobj_from_perspectives ' +o,5431,'Works on word similarity and word sense disambiguation are generally based on statistical methods designed for large or even very large corpora ',Hindle,'nsubjpass_based_Works prep_Works_on nn_similarity_word pobj_on_similarity cc_similarity_and nn_disambiguation_word nn_disambiguation_sense conj_similarity_disambiguation auxpass_based_are advmod_based_generally prep_based_on amod_methods_statistical pobj_on_methods partmod_methods_designed prep_designed_for amod_corpora_large cc_large_or advmod_large_even advmod_large_very conj_large_large pobj_for_corpora ' +p,5432,'Among the grammar formalisms successfully put into use in syntaxbased SMT are synchronous context-free grammars -LRB- SCFG -RRB- and synchronous treesubstitutiongrammars -LRB- STSG -RRB- ',Wu,'prep_synchronous_Among det_formalisms_the nn_formalisms_grammar pobj_Among_formalisms advmod_put_successfully partmod_formalisms_put prep_put_into pobj_into_use prep_use_in amod_SMT_syntaxbased pobj_in_SMT cop_synchronous_are amod_grammars_context-free nsubj_synchronous_grammars abbrev_grammars_SCFG amod_treesubstitutiongrammars_ cc__and conj__synchronous dep_grammars_treesubstitutiongrammars abbrev_treesubstitutiongrammars_STSG partmod_treesubstitutiongrammars_ ' +o,5433,'In this paper we use the phrase-based system of as our underlying model ',Koehn,'prep_use_In det_paper_this pobj_In_paper nsubj_use_we det_system_the amod_system_phrase-based dobj_use_system prep_system_of pobj_of_ prep_use_as poss_model_our amod_model_underlying pobj_as_model ' +o,5434,'This sequential property is well suited to HMMs , in which the jumps from the current aligned position can only be forward ',Berger,'det_property_This amod_property_sequential nsubjpass_suited_property auxpass_suited_is advmod_suited_well prep_suited_to nn__HMMs pobj_to_ rel_be_in pobj_in_which det_jumps_the nsubj_be_jumps prep_jumps_from det_position_the amod_position_current amod_position_aligned pobj_from_position aux_be_can advmod_be_only rcmod__be advmod_be_forward ' +n,5435,'Although ITA rates and system performance both significantly improve with coarse-grained senses , the question about what level of granularity is needed remains ',Navigli,'mark_improve_Although nn_rates_ITA nsubj_improve_rates cc_rates_and nn_performance_system conj_rates_performance dep_performance_both advmod_improve_significantly advcl_remains_improve prep_improve_with amod__coarse-grained nn__senses pobj_with_ det_question_the nsubj_remains_question rel_needed_about pobj_about_what nsubjpass_needed_level prep_level_of pobj_of_granularity auxpass_needed_is rcmod_question_needed ' +o,5436,'Thus the alignment set is denoted as -RCB- & -RRB- ,1 -LRB- -RRB- , -LCB- -LRB- ialiaiA ii = We adapt the bilingual word alignment model , IBM Model 3 , to monolingual word alignment ',Brown,'det_alignment_the dep_Thus_alignment partmod_alignment_set auxpass_denoted_is ccomp_set_denoted prep_set_as dep_to_& rcmod_&_,1 nn_=_ialiaiA nn_=_ii nsubj_alignment_= dep_adapt_We prep_alignment_adapt det_word_the amod_word_bilingual pobj_adapt_word advcl_,1_alignment dobj_alignment_model nn_Model_IBM dobj_alignment_Model dep__3 amod_Model_ prep_set_to amod_alignment_monolingual nn_alignment_word pobj_to_alignment ' +o,5437,'In particular , we use a feature augmentation technique recently introduced by , and active learning to perform domain adaptation of WSD systems ','Daume III','prep_use_In pobj_In_particular nsubj_use_we det_technique_a nn_technique_feature nn_technique_augmentation nsubj_introduced_technique advmod_introduced_recently xcomp_use_introduced prep_introduced_by pobj_by_ cc_introduced_and conj_introduced_active dep_active_learning nsubj_perform_ aux_perform_to xcomp_learning_perform nn_adaptation_domain dobj_perform_adaptation prep_adaptation_of nn_systems_WSD pobj_of_systems ' +o,5438,'In this paper , we give an overview of NLPWin , a multi-application natural language analysis and generation system under development at Microsoft Research , incorporating analysis systems for 7 languages -LRB- Chinese , English , French , German , Japanese , Korean and Spanish -RRB- ',Marcus,'prep_give_In det_paper_this pobj_In_paper nsubj_give_we det_overview_an dobj_give_overview prep_overview_of pobj_of_NLPWin det_analysis_a amod_analysis_multi-application amod_analysis_natural nn_analysis_language conj_NLPWin_analysis cc_NLPWin_and nn_system_generation conj_NLPWin_system prep_give_under pobj_under_development prep_development_at nn__Microsoft nn__Research pobj_at_ xcomp_give_incorporating nn_systems_analysis dobj_incorporating_systems prep_incorporating_for num_languages_7 pobj_for_languages dep_-LRB-_Chinese dep_French_English amod_Chinese_French conj_French_German conj_French_Japanese conj_French_Korean cc_French_and conj_French_Spanish ' +o,5439,'2 Background 21 Previous Work 211 Research on Phrase-Based SMT The original work on statistical machine translation was carried out by researchers at IBM ',Brown,'num_Work_2 nn_Work_Background num_Work_21 amod_Work_Previous nsubj_211_Work dobj_211_Research prep_Research_on amod_SMT_Phrase-Based pobj_on_SMT det_work_The amod_work_original nsubjpass_carried_work prep_work_on amod_translation_statistical nn_translation_machine pobj_on_translation auxpass_carried_was dep_211_carried prt_carried_out prep_carried_by pobj_by_researchers prep_researchers_at pobj_at_IBM ' +o,5440,'However , while discriminative models promise much , they have not been shown to deliver significant gains 1We class approaches using minimum error rate training frequency count based as these systems re-scale a handful of generative features estimated from frequency counts and do not support large sets of non-independent features ',Och,'advmod_shown_However mark_promise_while amod_models_discriminative nsubj_promise_models dep_shown_promise advmod_promise_much nsubjpass_shown_they aux_shown_have neg_shown_not auxpass_shown_been aux_deliver_to xcomp_shown_deliver amod_gains_significant dobj_deliver_gains amod_approaches_1We nn_approaches_class nsubj_estimated_approaches partmod_approaches_using amod_count_minimum nn_count_error nn_count_rate nn_count_training nn_count_ nn_count_frequency dobj_using_count partmod_count_based mark_re-scale_as det_systems_these nsubj_re-scale_systems advcl_based_re-scale det_handful_a dobj_re-scale_handful prep_handful_of amod_features_generative pobj_of_features dep_shown_estimated prep_estimated_from nn_counts_frequency pobj_from_counts cc_estimated_and aux_support_do neg_support_not conj_estimated_support amod_sets_large dobj_support_sets prep_sets_of amod_features_non-independent pobj_of_features ' +o,5441,'Recently , many phrase reordering methods have been proposed , ranging from simple distancebased distortion model , flat reordering model , lexicalized reordering model , to hierarchical phrase-based model and classifier-based reordering model with linear features ',Wu,'advmod_proposed_Recently amod_methods_many nn_methods_phrase nn_methods_reordering nsubjpass_proposed_methods aux_proposed_have auxpass_proposed_been xcomp_proposed_ranging prep_ranging_from amod__simple amod__distancebased nn__distortion nn__model nn__ amod__flat nn__reordering nn__model pobj_from_ amod__lexicalized nn__reordering nn__model conj__ aux_hierarchical_to conj__hierarchical amod__phrase-based nn__model dobj_hierarchical_ cc__and amod_model_classifier-based nn_model_reordering conj__model prep_model_with amod__linear nn__features pobj_with_ ' +o,5442,'Examples include Wus ITG and Chiangs hierarchical models ',Wu,'nsubj_include_Examples nn_ITG_Wus nn_ITG_ dobj_include_ITG cc_ITG_and nn__Chiangs amod__hierarchical nn__models conj_ITG_ ' +p,5443,'Hidden Markov models are simple and effective , but unlike discriminative models , such as Maximum Entropy models and Conditional Random Fields , they have more difficulty utilizing a rich set of conditionally dependent features ',Ratnaparkhi,'nn_models_Hidden nn_models_Markov nsubj_simple_models cop_simple_are cc_simple_and conj_simple_effective cc_simple_but conj_simple_unlike amod_models_discriminative pobj_unlike_models dep_as_such dep_simple_as nn__Maximum nn__Entropy nn__models pobj_as_ cc__and nn__Conditional nn__Random nn__Fields conj__ nsubj_have_they dep_simple_have amod_difficulty_more dobj_have_difficulty partmod_difficulty_utilizing det_set_a amod_set_rich dobj_utilizing_set prep_set_of advmod_dependent_conditionally amod_features_dependent pobj_of_features ' +o,5444,'This is possible because of the availability of statistical parsers , which can be trained on human-annotated treebanks for multiple languages ; -LRB- 2 -RRB- The binding theory is used as a guideline and syntactic structures are encoded as features in a maximum entropy coreference system ; -LRB- 3 -RRB- The syntactic features are evaluated on three languages : Arabic , Chinese and English -LRB- one goal is to see if features motivated by the English language can help coreference resolution in other languages -RRB- ',Marcus,'nsubj_possible_This cop_possible_is dep_of_because prep_possible_of det_availability_the pobj_of_availability prep_availability_of amod_parsers_statistical pobj_of_parsers nsubjpass_trained_which aux_trained_can auxpass_trained_be rcmod_parsers_trained prep_trained_on amod_treebanks_human-annotated pobj_on_treebanks partmod_treebanks_ prep__for amod_languages_multiple pobj_for_languages dep_used_2 det_theory_The nn_theory_binding nsubjpass_used_theory auxpass_used_is parataxis_possible_used prep_used_as det_guideline_a pobj_as_guideline cc_possible_and amod_structures_syntactic nsubjpass_encoded_structures auxpass_encoded_are conj_possible_encoded prep_encoded_as pobj_as_features prep_features_in det_system_a amod_system_maximum advmod_coreference_entropy amod_system_coreference pobj_in_system dep_evaluated_3 det_features_The amod_features_syntactic nsubjpass_evaluated_features auxpass_evaluated_are parataxis_encoded_evaluated prep_evaluated_on num_languages_three pobj_on_languages dep_languages_Arabic conj_Arabic_Chinese cc_Arabic_and conj_Arabic_English num_goal_one nsubj_is_goal dep_Arabic_is aux_see_to xcomp_is_see mark_help_if nsubj_help_features partmod_features_motivated prep_motivated_by det_language_the amod_language_English pobj_by_language aux_help_can advcl_see_help amod_resolution_coreference dobj_help_resolution prep_help_in amod_languages_other pobj_in_languages ' +o,5445,'Much like kappa statistics proposed by , existing employments of majority class baselines assume an equal set of identical potential mark-ups , ie attributes and their values , for all markables ',Carletta,'dep_Much_like nn_statistics_kappa pobj_like_statistics amod_statistics_proposed prep_statistics_by pobj_by_ amod_employments_existing appos_statistics_employments prep_employments_of nn_baselines_majority nn_baselines_class pobj_of_baselines rcmod_baselines_assume det_set_an amod_set_equal dobj_assume_set prep_set_of amod_mark-ups_identical amod_mark-ups_potential pobj_of_mark-ups nn_attributes_ie conj_mark-ups_attributes cc_mark-ups_and poss_values_their conj_mark-ups_values prep_employments_for det_markables_all pobj_for_markables ' +o,5446,'A very common case of this in the CoNLL dataset is that of documents containing references to both The China Daily , a newspaper , and China , the country ',Finkel,'det_case_A advmod_common_very amod_case_common nsubj_that_case prep_case_of pobj_of_this prep_this_in det_dataset_the nn_dataset_CoNLL pobj_in_dataset cop_that_is prep_that_of pobj_of_documents partmod_documents_containing dobj_containing_references prep_containing_to preconj_Daily_both det_Daily_The nn_Daily_China pobj_to_Daily det_newspaper_a conj_Daily_newspaper cc_Daily_and conj_Daily_China det_country_the appos_China_country ' +o,5447,'Many of the previous studies of Bio-NER tasks have been based on machine learning techniques including Hidden Markov Models -LRB- HMMs -RRB- , the dictionary HMM model and Maximum Entropy Markov Models -LRB- MEMMs -RRB- ',Finkel,'nsubjpass_based_Many prep_Many_of det_studies_the amod_studies_previous pobj_of_studies prep_studies_of amod_tasks_Bio-NER pobj_of_tasks aux_based_have auxpass_based_been prep_based_on pobj_on_machine xcomp_based_learning dobj_learning_techniques prep_techniques_including nn__Hidden nn__Markov nn__Models abbrev__HMMs pobj_including_ det__the nn__dictionary nn__HMM nn__model conj__ cc__and nn_Models_Maximum nn_Models_Entropy nn_Models_Markov conj__Models abbrev_Models_MEMMs ' +o,5448,'Finally , in section 4 we add additional features to the maxent model , and chain these models into a conditional markov model -LRB- CMM -RRB- , as used for tagging or earlier NER work ',Ratnaparkhi,'advmod_add_Finally dep_section_in dep_add_section num_section_4 nsubj_add_we amod_features_additional dobj_add_features prep_add_to det_model_the amod_model_maxent pobj_to_model cc_add_and conj_add_chain det_models_these dobj_chain_models prep_chain_into det_model_a amod_model_conditional amod_model_markov pobj_into_model abbrev_model_CMM mark_used_as dep_model_used prep_used_for amod__tagging pobj_for_ cc__or amod__earlier nn__NER nn__work conj__ ' +o,5449,'Previous approaches , eg , and , have all used the Brown algorithm for clustering ',Brown,'amod_approaches_Previous nsubj_used_approaches appos_approaches_eg conj_eg_ cc_eg_and conj_eg_ aux_used_have advmod_used_all det_algorithm_the nn_algorithm_Brown dobj_used_algorithm prep_used_for pcomp_for_clustering ' +o,5450,'We used the Berkeley Parser4 to learn such grammars from Sections 2-21 of the Penn Treebank ',Marcus,'nsubj_used_We det_Parser4_the nn_Parser4_Berkeley dobj_used_Parser4 aux_learn_to xcomp_used_learn amod_grammars_such dobj_learn_grammars prep_learn_from pobj_from_Sections num_Sections_2-21 prep_Sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,5451,'Along similar lines , combine a generative model of word alignment with a log-linear discriminative model trained on a small set of hand aligned sentences ',Fraser,'prep_combine_Along amod_lines_similar pobj_Along_lines nsubj_combine_ det_model_a amod_model_generative dobj_combine_model prep_model_of nn_alignment_word pobj_of_alignment prep_combine_with det_model_a amod_model_log-linear nn_model_discriminative pobj_with_model partmod_model_trained prep_trained_on det_set_a amod_set_small pobj_on_set prep_set_of pobj_of_hand partmod_hand_aligned dobj_aligned_sentences ' +p,5452,'8Interestingly , in work on the automated classification of nouns , also noted problems with \'em pty \' words that depend on their complements for meaning ',Hindle,'nsubj_noted_8Interestingly prep_noted_in pobj_in_work prep_work_on det_classification_the amod_classification_automated pobj_on_classification prep_classification_of pobj_of_nouns nsubj_noted_ advmod_noted_also dobj_noted_problems prep_problems_with amod_pty_\'em poss_words_pty pobj_with_words nsubj_depend_that rcmod_problems_depend prep_depend_on poss_complements_their pobj_on_complements prep_depend_for pcomp_for_meaning ' +o,5453,'2 Related Work Previous work on polarity disambiguation has used contextual clues and reversal words ',Kanayama,'num_work_2 nn_work_Related nn_work_Work amod_work_Previous nsubj_used_work prep_work_on amod_disambiguation_polarity pobj_on_disambiguation aux_used_has amod_clues_contextual dobj_used_clues cc_clues_and nn_words_reversal conj_clues_words ' +n,5454,'three models in are susceptible to the O -LRB- n 3 -RRB- method -LRB- cf ',Collins,'num_models_three nsubj_susceptible_models prep_models_in pobj_in_ cop_susceptible_are prep_susceptible_to det_O_the pobj_to_O appos_O_n num_n_3 nn_cf_method dep_O_cf ' +o,5455,'In the context of headline generation , simple statistical models are used for aligning documents and headlines , based on IBM Model 1 ',Brown,'prep_used_In det_context_the pobj_In_context prep_context_of nn_generation_headline pobj_of_generation amod_models_simple amod_models_statistical nsubjpass_used_models auxpass_used_are prep_used_for pcomp_for_aligning dobj_aligning_documents cc_documents_and conj_documents_headlines advmod_aligning_ prep_aligning_based dep_based_on nn_Model_IBM pobj_on_Model num_Model_1 ' +o,5456,' defined two local search operations for their 1-to-N alignment models 3 , 4 and 5 ',Brown,'nsubj_defined_ num_operations_two amod_operations_local nn_operations_search dobj_defined_operations prep_defined_for poss_models_their amod_models_1-to-N nn_models_alignment pobj_for_models tmod_defined_3 num_3_4 dep_3_and dep_3_5 ' +o,5457,'In order to create the necessary SMT language and translation models, they used: Giza++ (Och & Ney, 2003);2 the CMU-Cambridge statistical toolkit;3 the ISI ReWrite Decoder.4 Translation was performed from EnglishFrench and FrenchEnglish, and the resulting translations were evaluated using a range of automatic metrics: BLEU (Papineni et al. , 2002), Precision and Recall 2http://www.isi.edu/och/Giza++.html 3http://mi.eng.cam.ac.uk/prc14/toolkit.html 4http://www.isi.edu/licensed-sw/rewrite-decoder/ 185 (Turian et al. , 2003), and Wordand Sentence Error Rates.',Papineni,'' +o,5458,'The sentences included in the gold standard were chosen at random from the BNC , subject to the condition that they contain a verb which does not occur in the training sections of the WSJ section of the PTB ',Marcus,'det_sentences_The nsubjpass_chosen_sentences partmod_sentences_included prep_included_in det_standard_the amod_standard_gold pobj_in_standard auxpass_chosen_were prep_chosen_at amod_sections_random prep_random_from det_BNC_the pobj_from_BNC amod_BNC_subject prep_subject_to det_condition_the pobj_to_condition complm_contain_that nsubj_contain_they dep_condition_contain det_verb_a dobj_contain_verb nsubj_occur_which aux_occur_does neg_occur_not rcmod_verb_occur prep_occur_in det_training_the pobj_in_training pobj_at_sections prep_sections_of det_section_the nn_section_WSJ pobj_of_section prep_section_of det_PTB_the pobj_of_PTB ' +o,5459,'Only recently have robust knowledge-based methods for some of these tasks begun to appear , and their performance is still not very good , as seen above in our discussion of using WordNet as a semantic network ; 33 as for checking the plausibility of a hypothesis on the basis of causal knowledge about the world , we now have a much better theoretical grasp of how such inferences could be made , but we are still quite a long way from a general inference engine ',Marcus,'advmod_recently_Only advmod_have_recently amod_methods_robust amod_methods_knowledge-based dobj_have_methods prep_methods_for pobj_for_some prep_some_of det_tasks_these pobj_of_tasks partmod_tasks_begun aux_appear_to xcomp_begun_appear cc_good_and poss_performance_their nsubj_good_performance cop_good_is advmod_good_still neg_good_not advmod_good_very parataxis_appear_good mark_seen_as advcl_appear_seen dep_in_above prep_seen_in poss_discussion_our pobj_in_discussion prep_discussion_of pcomp_of_using dobj_using_WordNet prep_WordNet_as det_network_a amod_network_semantic pobj_as_network dep_WordNet_33 mark_have_as prep_have_for pcomp_for_checking det_plausibility_the dobj_checking_plausibility prep_plausibility_of det_hypothesis_a pobj_of_hypothesis prep_hypothesis_on det_basis_the pobj_on_basis prep_basis_of amod_knowledge_causal pobj_of_knowledge prep_checking_about det_world_the pobj_about_world nsubj_have_we advmod_have_now dep_have_have det_grasp_a advmod_better_much amod_grasp_better amod_grasp_theoretical dobj_have_grasp prep_grasp_of advmod_made_how amod_inferences_such nsubjpass_made_inferences aux_made_could auxpass_made_be pcomp_of_made dep_made_ cc_have_but nsubj_way_we cop_way_are advmod_way_still advmod_way_quite det_way_a amod_way_long conj_have_way prep_way_from det_engine_a amod_engine_general nn_engine_inference pobj_from_engine ' +o,5460,'The principle of maximum entropy states that when one searches among probability distributions that model the observed data -LRB- evidence -RRB- , the preferred one is the one that maximizes the entropy -LRB- a measure of the uncertainty of the model -RRB- ',Berger,'det_principle_The prep_principle_of amod_states_maximum nn_states_entropy pobj_of_states complm_one_that advmod_searches_when nsubj_searches_one dep_one_searches prep_searches_among nn_distributions_probability pobj_among_distributions nsubj_model_that rcmod_distributions_model det_data_the amod_data_observed dobj_model_data appos_distributions_evidence det_one_the amod_one_preferred nsubj_one_one cop_one_is det_one_the dep_principle_one nsubj_maximizes_that rcmod_one_maximizes det_entropy_the dobj_maximizes_entropy det_measure_a dep_entropy_measure prep_measure_of det_uncertainty_the pobj_of_uncertainty prep_uncertainty_of det_model_the pobj_of_model ' +o,5461,'Usually the IBM Model 1 , developed in the statistical machine translation field , is used to construct translation models for retrieval purposes in practice ',Brown,'det_Model_the nn_Model_IBM dep_Usually_Model nsubjpass_used_1 partmod_1_developed prep_developed_in det__the amod__statistical nn__machine nn__translation nn__field pobj_in_ auxpass_used_is rcmod_Model_used aux_construct_to xcomp_used_construct nn_models_translation dobj_construct_models prep_construct_for nn_purposes_retrieval pobj_for_purposes prep_purposes_in pobj_in_practice ' +o,5462,'et al , 1994 ; Brill and Resnik , 1994 ; and Brooks , 1995 ; Merlo et al , 1997 -RRB- ',Collins,'nn_al_et appos_al_1994 dep_al_Brill cc_Brill_and conj_Brill_Resnik appos_Brill_1994 dep_al_ cc__and conj__Brooks appos__1995 dep__Merlo cc_Merlo_et conj_Merlo_al appos_Merlo_1997 ' +o,5463,'52 Adding lexical information Gildea shows that removing the lexical dependencies in Model 1 of -LRB- that is , not conditioning on w h when generating w s -RRB- decreases labeled precision and recall by only 05 \% ',Collins,'nsubj_labeled_52 partmod_52_Adding amod_information_lexical dobj_Adding_information nn__Gildea nsubj_shows_ rcmod_information_shows complm_decreases_that csubj_decreases_removing det_dependencies_the amod_dependencies_lexical dobj_removing_dependencies prep_dependencies_in pobj_in_Model num_Model_1 prep_dependencies_of pobj_of_ nsubj_is_that dep__is neg_is_not dep_is_conditioning prep_conditioning_on nn_h_w pobj_on_h advmod_s_when amod_w_generating nsubj_s_w advcl_conditioning_s ccomp_shows_decreases dobj_labeled_precision cc_precision_and conj_precision_recall prep_labeled_by quantmod_05_only num_\%_05 pobj_by_\% ' +o,5464,'The translation quality on the TransType2 task in terms of WER , PER , BLEU score , and NIST score is given in Table 4 ',Papineni,'det_quality_The nn_quality_translation nsubjpass_given_quality prep_quality_on det_task_the amod_task_TransType2 pobj_on_task prep_task_in pobj_in_terms prep_terms_of pobj_of_WER conj_WER_PER amod__BLEU nn__score conj_WER_ cc_WER_and amod__NIST nn__score conj_WER_ auxpass_given_is prep_given_in pobj_in_Table num_Table_4 ' +o,5465,'81 The Averaged Perceptron Algorithm with Separating Plane The averaged perceptron algorithm has previously been applied to various NLP tasks for discriminative reranking ',Collins,'det_Algorithm_The nn_Algorithm_Averaged nn_Algorithm_Perceptron nsubj_averaged_Algorithm prep_Algorithm_with nn_The_Separating nn_The_Plane pobj_with_The dep_81_averaged nn__perceptron nn__algorithm nsubjpass_applied_ aux_applied_has advmod_applied_previously auxpass_applied_been ccomp_averaged_applied prep_applied_to amod_tasks_various nn_tasks_NLP pobj_to_tasks partmod_tasks_ prep__for amod_reranking_discriminative pobj_for_reranking ' +o,5466,'Other researchers have also reported similar problems of excessive resource demands with the ` collect all neighbors \' model ',Brown,'amod_researchers_Other nsubj_reported_researchers aux_reported_have advmod_reported_also amod_problems_similar dobj_reported_problems prep_problems_of amod_demands_excessive nn_demands_resource pobj_of_demands prep_reported_with det_model_the dep_model_collect det_neighbors_all dobj_collect_neighbors pobj_with_model amod_model_ ' +o,5467,'Many corpus based statistical methods have been proposed to solve this problem , including supervised learning algorithms , weakly supervised learning algorithms , unsupervised learning algorithms -LRB- or word sense discrimination -RRB- , and knowledge based algorithms ',Yarowsky,'amod_corpus_Many nsubjpass_proposed_corpus partmod_corpus_based amod_methods_statistical dobj_based_methods aux_proposed_have auxpass_proposed_been aux_solve_to xcomp_proposed_solve det_problem_this dobj_solve_problem prep_proposed_including amod_algorithms_supervised nn_algorithms_learning pobj_including_algorithms amod_algorithms_ advmod__weakly dep__supervised num_algorithms_learning measure__algorithms amod_algorithms_ conj_algorithms_unsupervised xcomp_unsupervised_learning amod__algorithms dep_word_or dep__word nn_discrimination_sense dep_word_discrimination dobj_learning_ cc_algorithms_and conj_algorithms_knowledge partmod_knowledge_based amod__algorithms dobj_based_ ' +n,5468,'Even with the current incomplete set of semantic templates , the hypertagger brings realizer performance roughly up to state-of-the-art levels , as our overall test set BLEU score slightly exceeds that of , though at a coverage of 96 \% insteadof98 \% ',Cahill,'dep_with_Even advcl_brings_with det_set_the amod_set_current amod_set_incomplete pobj_with_set prep_set_of amod_templates_semantic pobj_of_templates det_hypertagger_the nsubj_brings_hypertagger nn_performance_realizer iobj_brings_performance quantmod_up_roughly dobj_brings_up prep_up_to amod_levels_state-of-the-art pobj_to_levels mark_set_as poss_test_our amod_test_overall nsubj_set_test advcl_brings_set amod__BLEU nn__score nsubj_exceeds_ advmod_exceeds_slightly ccomp_set_exceeds dobj_exceeds_that prep_that_of pobj_of_ dep_at_though prep_exceeds_at det_coverage_a pobj_at_coverage prep_coverage_of num_\%_96 pobj_of_\% num_\%_insteadof98 dep_\%_\% ' +o,5469,'Two disjoint corpora are used in steps 2 and 5 , both consisting of complete articles taken from the Wall Street Journal Treebank Corpus ',Marcus,'num_corpora_Two nn_corpora_disjoint nsubjpass_used_corpora auxpass_used_are prep_used_in pobj_in_steps num_steps_2 cc_steps_and conj_steps_5 nsubj_consisting_both rcmod_steps_consisting prep_consisting_of amod_articles_complete pobj_of_articles partmod_articles_taken prep_taken_from det_Corpus_the nn_Corpus_Wall nn_Corpus_Street nn_Corpus_Journal nn_Corpus_Treebank pobj_from_Corpus ' +o,5470,'But there is also extensive research focused on including linguistic knowledge in metrics among others ',Liu,'cc_is_But expl_is_there advmod_is_also amod_research_extensive nsubj_is_research partmod_research_focused prt_focused_on prep_focused_including amod_knowledge_linguistic pobj_including_knowledge prep_knowledge_in pobj_in_metrics partmod_metrics_ prep__among pobj_among_others ' +o,5471,' tested the claim on about 37,000 examples and found that when a polysemous word appeared more than once in a discourse , they took on the majority sense for the discourse 998 \% of the time on average ',Yarowsky,'nsubj_tested_ det_claim_the dobj_tested_claim prep_tested_on advmod_examples_about num_examples_37,000 pobj_on_examples cc_tested_and conj_tested_found complm_took_that advmod_than_when det_word_a nn_word_polysemous nsubj_than_word cop_than_appeared quantmod_than_more dep_took_than quantmod_than_once prep_than_in det_discourse_a pobj_in_discourse nsubj_took_they ccomp_found_took prep_took_on det_sense_the nn_sense_majority pobj_on_sense prep_took_for det_\%_the nn_\%_discourse num_\%_998 pobj_for_\% prep_\%_of det_time_the pobj_of_time prep_time_on pobj_on_average ' +o,5472,'One popular approach is to use a log-linear parsing model and maximize the conditional likelihood function ',Collins,'num_approach_One amod_approach_popular nsubj_is_approach aux_use_to xcomp_is_use det_model_a amod_model_log-linear amod_model_parsing dobj_use_model cc_use_and conj_use_maximize det__the amod__conditional nn__likelihood nn__function dobj_maximize_ ' +p,5473,'Several studies have reported alignment or translation performance for syntactically augmented translation models and these results have been promising ',Wu,'amod_studies_Several nsubj_reported_studies aux_reported_have nn_performance_alignment cc_alignment_or conj_alignment_translation dobj_reported_performance prep_reported_for advmod_augmented_syntactically amod_models_augmented nn_models_translation pobj_for_models nsubj_promising_ cc__and det_results_these conj__results aux_promising_have aux_promising_been dep_reported_promising ' +o,5474,'This is based on the idea from that rare words in the training set are similar to unknown words in the test set , and can be used to learn how to tag the unknown words that will be encountered during testing ',Ratnaparkhi,'nsubjpass_based_This auxpass_based_is prep_based_on det_idea_the pobj_on_idea prep_based_from pcomp_from_ complm_similar_that amod_words_rare nsubjpass_similar_words prep_words_in det_set_the nn_set_training pobj_in_set cop_similar_are ccomp__similar prep_similar_to amod_words_unknown pobj_to_words prep_words_in det_set_the nn_set_test pobj_in_set cc_similar_and aux_used_can auxpass_used_be conj_similar_used aux_learn_to purpcl_used_learn advmod_tag_how aux_tag_to ccomp_learn_tag det_words_the amod_words_unknown dobj_tag_words nsubjpass_encountered_that aux_encountered_will auxpass_encountered_be rcmod_words_encountered prep_encountered_during pobj_during_testing ' +o,5475,'Then , to solve p E C in equation -LRB- 8 -RRB- is equivalent to solve h that maximize the loglikelihood : = -LRB- x -RRB- log zj , -LRB- z -RRB- + x i -LRB- 10 -RRB- h = argmax kV -LRB- h -RRB- Such h can be solved by one of the numerical algorithm called the Improved Iteratire Scaling Algorithm ',Berger,'advmod_equivalent_Then aux_solve_to parataxis_equivalent_solve nn_C_p nn_C_E dobj_solve_C prep_C_in pobj_in_equation appos_equation_8 cop_equivalent_is aux_solve_to xcomp_equivalent_solve dobj_solve_h nsubj_maximize_that rcmod_h_maximize det_loglikelihood_the dobj_maximize_loglikelihood dep_maximize_= dep_called_x nsubj_called_log dep_log_zj dep_-LRB-_z amod_h_+ amod_h_x nn_h_i appos_h_10 nsubj_=_h rcmod_log_= nn_kV_argmax dobj_=_kV appos_kV_h amod_h_Such nsubjpass_solved_h aux_solved_can auxpass_solved_be rcmod_kV_solved prep_solved_by pobj_by_one prep_one_of det_algorithm_the amod_algorithm_numerical pobj_of_algorithm dep_equivalent_called det_Algorithm_the nn_Algorithm_Improved nn_Algorithm_Iteratire nn_Algorithm_Scaling dobj_called_Algorithm ' +o,5476,'There are many techniques for transliteration and back-transliteration , and they vary along a number of dimensions : phoneme substitution vs character substitution heuristic vs generative vs discriminative models manual vs automatic knowledge acquisition We explore the third dimension , where we see several techniques in use : Manually-constructed transliteration models , eg , ',Hermjakob,'expl_are_There amod_techniques_many nsubj_are_techniques prep_techniques_for pobj_for_transliteration cc_transliteration_and conj_transliteration_back-transliteration cc_are_and nsubj_vary_they conj_are_vary prt_vary_along det_number_a dobj_vary_number prep_number_of pobj_of_dimensions amod_substitution_phoneme dep_number_substitution cc_heuristic_vs nn_heuristic_character nn_heuristic_substitution nsubj_manual_heuristic cc_heuristic_vs nn_models_generative cc_generative_vs conj_generative_discriminative conj_heuristic_models rcmod_substitution_manual cc_substitution_vs amod_acquisition_automatic nn_acquisition_knowledge conj_substitution_acquisition nsubj_explore_We rcmod_acquisition_explore det_dimension_the amod_dimension_third dobj_explore_dimension advmod_see_where nsubj_see_we rcmod_dimension_see amod_techniques_several dobj_see_techniques prep_techniques_in pobj_in_use amod_models_Manually-constructed nn_models_transliteration dep_acquisition_models appos_models_eg ' +o,5477,'The annotation scheme is modeled to a certain extent on that of the Penn Treebank , with crucial differences ',Marcus,'det__The nn__annotation nn__scheme nsubjpass_modeled_ auxpass_modeled_is prep_modeled_to det_extent_a amod_extent_certain pobj_to_extent prep_modeled_on pobj_on_that prep_that_of det__the nn__Penn nn__Treebank pobj_of_ prep_modeled_with amod_differences_crucial pobj_with_differences ' +o,5478,'The mutual information of a cooccurrence pair , which measures the degree of association between the two words , is defined as : P -LRB- xly -RRB- I -LRB- x , y -RRB- - log 2 P -LRB- x , y -RRB- _ log 2 -LRB- 1 -RRB- P -LRB- x -RRB- P -LRB- y -RRB- P -LRB- x -RRB- = log 2 P -LRB- y -LRB- x -RRB- P -LRB- Y -RRB- where P -LRB- x -RRB- and P -LRB- y -RRB- are the probabilities of the events x and y -LRB- occurrences of words , in our case -RRB- and P -LRB- x , y -RRB- is the probability of the joint event -LRB- a cooccurrence pair -RRB- ',Church,'det_information_The amod_information_mutual nsubjpass_defined_information prep_information_of det_pair_a amod_pair_cooccurrence pobj_of_pair nsubj_measures_which rcmod_information_measures det_degree_the dobj_measures_degree prep_degree_of pobj_of_association prep_association_between det_words_the num_words_two pobj_between_words advmod_measures_ auxpass_defined_is mark_log_as nsubj_log_ dep__P appos_P_xly dep_P_I dep_I_x appos_x_y advcl_defined_log num_P_2 nsubj___P dep_P_x appos_x_y ccomp_log__ number_2_log num_P_2 appos_P_1 nn_P_P appos_P_x nsubj_P_P appos_P_y ccomp___P dep_log_x nsubj_log_= dep_P_log num_P_2 dobj_log_P dep_P_y dep_P_x dep_y_P appos_P_Y advmod_probabilities_where nsubj_probabilities_P dep_P_x cc_P_and conj_P_P appos_P_y cop_probabilities_are det_probabilities_the rcmod_P_probabilities prep_probabilities_of det_events_the pobj_of_events dep_events_x cc_events_and conj_events_y dep_y_occurrences prep_occurrences_of pobj_of_words prep_y_in poss_case_our pobj_in_case cc_log_and nsubj_probability_P dep_P_x appos_x_y cop_probability_is det_probability_the conj_log_probability prep_probability_of det_event_the amod_event_joint pobj_of_event det_pair_a nn_pair_cooccurrence appos_event_pair ' +o,5479,'Following , we use syntactic dependencies between words to model their semantic properties ',Lin,'prep_use_Following pobj_Following_ nsubj_use_we amod_dependencies_syntactic nsubj_model_dependencies prep_dependencies_between pobj_between_words aux_model_to xcomp_use_model poss_properties_their amod_properties_semantic dobj_model_properties ' +o,5480,'32 ITG Constraints In this section , we describe the ITG constraints ',Wu,'num_Constraints_32 nn_Constraints_ITG prep_describe_In det_section_this pobj_In_section nsubj_describe_we dep_Constraints_describe det_constraints_the nn_constraints_ITG dobj_describe_constraints ' +o,5481,'We therefore ran the dependency model on a test corpus tagged with the POS-tagger of , which is trained on the original Penn Treebank -LRB- see HWDep -LRB- + tagger -RRB- in Table 3 -RRB- ',Ratnaparkhi,'nsubj_ran_We advmod_ran_therefore det_model_the amod_model_dependency dobj_ran_model prep_model_on det_corpus_a nn_corpus_test pobj_on_corpus partmod_model_tagged prep_tagged_with det_POS-tagger_the pobj_with_POS-tagger prep_POS-tagger_of pobj_of_ nsubjpass_trained_which auxpass_trained_is rcmod_POS-tagger_trained prep_trained_on det_Treebank_the amod_Treebank_original nn_Treebank_Penn pobj_on_Treebank dep_model_see dobj_see_HWDep amod_tagger_+ appos_HWDep_tagger prep_HWDep_in pobj_in_Table num_Table_3 ' +o,5482,'Several authors have used mutual information and similar statistics as an objective function for word clustering , for automatic determination of phonemic baseforms , and for language modeling for speech recognition ',Brown,'amod_authors_Several nsubj_used_authors aux_used_have amod_information_mutual dobj_used_information cc_information_and amod_statistics_similar conj_information_statistics prep_information_as det_function_an amod_function_objective pobj_as_function prep_used_for nn__word nn__clustering pobj_for_ prep_used_for amod_determination_automatic pobj_for_determination prep_determination_of amod__phonemic nn__baseforms pobj_of_ cc_for_and conj_for_for nn_modeling_language pobj_for_modeling prep_modeling_for nn__speech nn__recognition pobj_for_ ' +o,5483,'We use a standard data set consisting of sections 15-19 of the WSJ corpus as training and section 20 as testing ',Ramshaw,'nsubj_use_We det_data_a amod_data_standard nsubj_set_data ccomp_use_set nn_consisting_ dobj_set_consisting prep_consisting_of pobj_of_sections num_sections_15-19 prep_sections_of det_corpus_the nn_corpus_WSJ pobj_of_corpus prep_set_as pobj_as_training cc_training_and conj_training_section num_section_20 prep_set_as pobj_as_testing ' +o,5484,'In one set of experiments , we generated lexicons for PEOPLE and ORGANIZATIONS using 2500 Wall Street Journal articles from the Penn Treebank ',Marcus,'prep_generated_In num_set_one pobj_In_set prep_set_of pobj_of_experiments nsubj_generated_we dobj_generated_lexicons prep_generated_for pobj_for_PEOPLE cc_PEOPLE_and conj_PEOPLE_ORGANIZATIONS xcomp_generated_using num_articles_2500 nn_articles_Wall nn_articles_Street nn_articles_Journal dobj_using_articles prep_using_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,5485,'A comparison of the two approaches can be found in Koehn , ',Och,'det_comparison_A nsubjpass_found_comparison prep_comparison_of det_approaches_the num_approaches_two pobj_of_approaches aux_found_can auxpass_found_be prep_found_in pobj_in_Koehn ' +o,5486,'a65 The rest of the factors denote distorsion probabilities -LRB- d -RRB- , which capture the probability that words change their position when translated from one language into another ; the probability of some French words being generated from an invisible English NULL element -LRB- pa6 -RRB- , etc See or for a detailed discussion of this translation model and a description of its parameters ',Brown,'det_rest_The dobj_a65_rest prep_rest_of det_factors_the pobj_of_factors amod_probabilities_denote nn_probabilities_distorsion nsubj_a65_probabilities dep_probabilities_d nsubj_capture_which dep_d_capture det_probability_the dobj_capture_probability complm_change_that nsubj_change_words ccomp_capture_change poss_position_their dobj_change_position advmod_translated_when advcl_change_translated prep_translated_from num_language_one pobj_from_language prep_translated_into pobj_into_another det_probability_the dep_d_probability prep_probability_of det_words_some amod_words_French pobj_of_words auxpass_generated_being partmod_words_generated prep_generated_from det_element_an amod_element_invisible nn_element_English nn_element_NULL pobj_from_element appos_element_pa6 nn__etc nn__See conj_d_ cc__or conj__ prep__for det_discussion_a amod_discussion_detailed pobj_for_discussion prep_discussion_of det_model_this nn_model_translation pobj_of_model cc_d_and det_description_a conj_d_description prep_description_of poss_parameters_its pobj_of_parameters ' +o,5487,'We use ROUGE to assess summary quality using common n-gram counts and longest common subsequence -LRB- LCS -RRB- measures ',Lin,'nsubj_use_We nn__ROUGE nsubj_assess_ aux_assess_to xcomp_use_assess nn_quality_summary dobj_assess_quality partmod_quality_using amod_counts_common amod_counts_n-gram dobj_using_counts cc_counts_and conj_counts_longest amod_measures_common nn_measures_subsequence abbrev_measures_LCS dep_longest_measures ' +o,5488,'Dynamic programming is applied to bilingual sentence alignment in most of previous works ',Brown,'nn_programming_Dynamic nsubjpass_applied_programming auxpass_applied_is prep_applied_to amod_alignment_bilingual nn_alignment_sentence pobj_to_alignment prep_alignment_in pobj_in_most prep_most_of amod_works_previous pobj_of_works ' +p,5489,'22 Perceptron-based training To tune the parameters w of the model , we use the averaged perceptron algorithm because of its efficiency and past success on various NLP tasks ',Collins,'num_training_22 amod_training_Perceptron-based aux_tune_To dep_use_tune det_w_the nn_w_parameters dobj_tune_w prep_w_of det_model_the pobj_of_model nsubj_use_we dep_training_use det__the amod__averaged nn__perceptron nn__algorithm dobj_use_ dep_of_because prep__of poss_efficiency_its pobj_of_efficiency cc__and amod_success_past conj__success prep_success_on amod_tasks_various nn_tasks_NLP pobj_on_tasks ' +o,5490,'To evaluate the performance of a parser , NP chunks can usefully be evaluated by a gold standard ; many systems use the Penn Treebank for this type of evaluation ',Ramshaw,'aux_evaluate_To dep_evaluated_evaluate det_performance_the dobj_evaluate_performance prep_performance_of det_parser_a pobj_of_parser nn_chunks_NP nsubjpass_evaluated_chunks aux_evaluated_can advmod_evaluated_usefully auxpass_evaluated_be prep_evaluated_by det_standard_a amod_standard_gold pobj_by_standard amod_systems_many nsubj_use_systems advmod_use_ parataxis_evaluated_use det_Treebank_the nn_Treebank_Penn dobj_use_Treebank prep_Treebank_for det_type_this pobj_for_type prep_type_of pobj_of_evaluation ' +o,5491,'In particular , it shows systematically better F-Measure and Accuracy measures over all other metrics showing an improvement of -LRB- 1 -RRB- at least 286 \% in terms of F-Measure and 396 \% in terms of Accuracy and -LRB- 2 -RRB- at most 661 \% in terms of FMeasure and 674 \% in terms of Accuracy compared to the second best metric which is also systematically the word N-gram overlap similarity measure used by ',Barzilay,'prep_shows_In pobj_In_particular nsubj_shows_it advmod_shows_systematically amod_F-Measure_better dobj_shows_F-Measure cc_F-Measure_and nn_measures_Accuracy conj_F-Measure_measures prep_F-Measure_over det_metrics_all amod_metrics_other pobj_over_metrics partmod_metrics_showing det_improvement_an dobj_showing_improvement prep_improvement_of dep_of_1 advmod_\%_at dep_at_least num_\%_286 pobj_of_\% prep_\%_in pobj_in_terms prep_terms_of pobj_of_F-Measure cc_terms_and num_\%_396 conj_terms_\% prep_\%_in pobj_in_terms prep_terms_of pobj_of_Accuracy cc_\%_and appos_\%_2 quantmod_661_at dep_at_most num_\%_661 conj_\%_\% prep_\%_in pobj_in_terms prep_terms_of pobj_of_FMeasure cc_\%_and num_\%_674 conj_\%_\% prep_\%_in pobj_in_terms prep_terms_of pobj_of_Accuracy prep_showing_compared dep_compared_to det_metric_the amod_metric_second dep_second_best pobj_to_metric nsubj_is_which rcmod_metric_is advmod_is_also advmod_is_systematically det_N-gram_the nn_N-gram_word nsubj_overlap_N-gram ccomp_is_overlap nn_measure_similarity dobj_overlap_measure partmod_measure_used prep_used_by ' +o,5492,'Their experiments were performed using a decoder based on IBM Model 4 using the translation techniques developed at IBM ',Brown,'poss_experiments_Their nsubjpass_performed_experiments auxpass_performed_were xcomp_performed_using det_decoder_a dobj_using_decoder prep_using_based dep_based_on nn_Model_IBM pobj_on_Model num_Model_4 partmod_Model_using det_techniques_the nn_techniques_translation dobj_using_techniques partmod_techniques_developed prep_developed_at pobj_at_IBM ' +o,5493,'A superset of the parallel data was word aligned by GIZA union and EMD ',Och,'det_superset_A nsubj_word_superset prep_superset_of det_data_the amod_data_parallel pobj_of_data cop_word_was partmod_word_aligned prep_aligned_by nn__GIZA nn__union pobj_by_ cc__and nn__EMD conj__ ' +o,5494,'To group the letters into classes , we employ a hierarchical clustering algorithm ',Brown,'prep_employ_To pobj_To_group det_letters_the dep_group_letters dep_To_into pobj_into_classes nsubj_employ_we det_algorithm_a amod_algorithm_hierarchical nn_algorithm_clustering dobj_employ_algorithm ' +o,5495,'Previous research in automatic acquisition focuses primarily on the use of statistical techniques , such as bilingual alignment or extraction of syntactic constructions from online dictionaries and corpora ',Church,'amod_research_Previous nsubj_focuses_research prep_research_in amod_acquisition_automatic pobj_in_acquisition advmod_focuses_primarily prep_focuses_on det_use_the pobj_on_use prep_use_of amod_techniques_statistical pobj_of_techniques dep_as_such prep_use_as amod__bilingual nn__alignment pobj_as_ cc__or conj__extraction prep_extraction_of amod_constructions_syntactic pobj_of_constructions prep_constructions_from amod_dictionaries_online pobj_from_dictionaries cc_dictionaries_and nn__corpora conj_dictionaries_ ' +o,5496,'These sentences were parsed with the Collins parser ',Collins,'det_sentences_These nsubjpass_parsed_sentences auxpass_parsed_were prep_parsed_with det_parser_the nn_parser_Collins pobj_with_parser ' +p,5497,'For French\\/English translation we use a state of the art phrase-based MT system similar to ',Koehn,'prep_use_For nn_translation_French\\/English pobj_For_translation nsubj_use_we det_state_a nsubj_similar_state prep_state_of det_system_the nn_system_art amod_system_phrase-based nn_system_MT pobj_of_system xcomp_use_similar prep_similar_to ' +o,5498,'The precision of the extracted information can be improved significantly by using machine learning methods to filter out noise ',Fleischman,'det_precision_The nsubjpass_improved_precision prep_precision_of det_information_the amod_information_extracted pobj_of_information aux_improved_can auxpass_improved_be advmod_improved_significantly prep_improved_by pcomp_by_using dobj_using_machine partmod_machine_learning dobj_learning_methods aux_filter_to xcomp_learning_filter prt_filter_out dobj_filter_noise ' +o,5499,'9 report that , for translation reranking , such local updates -LRB- towards the oracle -RRB- outperform bold updates -LRB- towards the gold standard -RRB- ',Liang,'num__9 nsubj_report_ complm_outperform_that prep_outperform_for pobj_for_translation partmod_translation_reranking amod_updates_such amod_updates_local nsubj_outperform_updates dep_updates_towards det_oracle_the pobj_towards_oracle ccomp_report_outperform amod_updates_bold dobj_outperform_updates prep_updates_towards det_standard_the amod_standard_gold pobj_towards_standard ' +p,5500,'Maximum entropy taggers have been shown to be highly competitive on a number of tagging tasks , such as partof-speech tagging , and namedentity recognition -LRB- Borthwick et ',Ratnaparkhi,'amod_taggers_Maximum nn_taggers_entropy nsubjpass_shown_taggers aux_shown_have auxpass_shown_been aux_competitive_to cop_competitive_be advmod_competitive_highly xcomp_shown_competitive prep_competitive_on det_number_a pobj_on_number prep_number_of amod_tasks_tagging pobj_of_tasks dep_as_such prep_number_as amod__partof-speech amod__tagging pobj_as_ cc__and advmod__namedentity conj__recognition dep_recognition_Borthwick dep_Borthwick_et ' +o,5501,'To achieve robust training , Daume III and Marcu employed the averaged perceptron and ALMA ',Collins,'aux_achieve_To amod_training_robust dobj_achieve_training nn_III_Daume conj_training_III cc_training_and nn__Marcu conj_training_ partmod__employed det__the amod__averaged nn__perceptron dobj_employed_ cc__and nn__ALMA conj__ ' +o,5502,'However , our representation of the model conceptually separates some of the hyperparameters which are not separated in , and we found that setting these hyperparameters with different values from one another was critical for improving performance ','Daume III','advmod_separates_However poss_representation_our nsubj_separates_representation prep_representation_of det_conceptually_the nn_conceptually_model pobj_of_conceptually dobj_separates_some prep_some_of det_hyperparameters_the pobj_of_hyperparameters nsubjpass_separated_which auxpass_separated_are neg_separated_not rcmod_hyperparameters_separated prep_separated_in pobj_in_ cc_separates_and nsubj_found_we conj_separates_found complm_critical_that csubj_critical_setting det_hyperparameters_these dobj_setting_hyperparameters prep_setting_with amod_values_different pobj_with_values prep_setting_from pobj_from_one dep_one_another cop_critical_was ccomp_found_critical prep_critical_for pcomp_for_improving dobj_improving_performance ' +o,5503,'6 Related works After the work of , many machine learning techniques have been applied to the basic chunking task , such as Support Vector Machines , Hidden Markov Model , Memory Based Learning , Conditional Random Fields , and so on ',Ramshaw,'num_works_6 amod_works_Related prep_applied_After det_work_the pobj_After_work prep_work_of pobj_of_ amod_techniques_many nn_techniques_machine nn_techniques_learning nsubjpass_applied_techniques aux_applied_have auxpass_applied_been dep_works_applied prep_applied_to det_task_the amod_task_basic amod_task_chunking pobj_to_task dep_as_such dep__as pobj_as_Support nn__Vector nn__Machines dep_task_ nn__Hidden nn__Markov nn__Model appos__ dep_Based_Memory dep__Based nn__Learning nn__ appos__Conditional nn__Random nn__Fields dep_Based_ cc_Based_and conj_Based_so conj_Based_on ' +o,5504,'Different news articles reporting on the same event are commonly used as monolingual comparable corpora , from which both paraphrase patterns and phrasal paraphrases can be derived ',Barzilay,'amod_articles_Different nn_articles_news nsubjpass_used_articles partmod_articles_reporting prep_reporting_on det_event_the amod_event_same pobj_on_event auxpass_used_are advmod_used_commonly prep_used_as amod_corpora_monolingual amod_corpora_comparable pobj_as_corpora rel_derived_from pobj_from_which det_patterns_both amod_patterns_paraphrase nsubjpass_derived_patterns cc_patterns_and amod_paraphrases_phrasal conj_patterns_paraphrases aux_derived_can auxpass_derived_be rcmod_corpora_derived ' +o,5505,'We ran GIZA + + on the training corpus in both directions with IBM model 4 , and then applied the refinement rule described in to obtain a many-to-many word alignment for each sentence pair ',Koehn,'nsubj_ran_We nn__GIZA nn__+ nn__+ iobj_ran_ prep__on det_corpus_the nn_corpus_training pobj_on_corpus prep_corpus_in det_directions_both pobj_in_directions prep_directions_with nn_model_IBM pobj_with_model dobj_ran_4 cc_ran_and advmod_applied_then conj_ran_applied det_rule_the amod_rule_refinement dobj_applied_rule partmod_rule_described prep_described_in pobj_in_ aux_obtain_to xcomp_described_obtain det_alignment_a amod_alignment_many-to-many nn_alignment_word dobj_obtain_alignment prep_obtain_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,5506,' , -RRB- , source extraction -LRB- eg Bethard et al ',Kim,'amod__ nn_extraction_source appos__extraction nn_al_eg nn_al_Bethard nn_al_et dep__al ' +o,5507,'7For details about the Bleu evaluation metric , see ',Papineni,'amod_details_7For nsubj_see_details prep_details_about det_metric_the nn_metric_Bleu nn_metric_evaluation pobj_about_metric ' +p,5508,'Lexicalized PCFGs use the structural features on the lexical head of phrasal node in a tree , and get significant improvements for parsing ',Collins,'nn_PCFGs_Lexicalized nsubj_use_PCFGs det_features_the amod_features_structural dobj_use_features prep_use_on det_head_the amod_head_lexical pobj_on_head prep_head_of amod_node_phrasal pobj_of_node prep_node_in det_tree_a pobj_in_tree cc_use_and conj_use_get amod_improvements_significant dobj_get_improvements prep_get_for pcomp_for_parsing ' +o,5509,' 8802 -LRB- +082 -RRB- + unlabeled data -LRB- 17M 27M words -RRB- 8841 -LRB- +039 -RRB- + supplied gazetters 8890 -LRB- +049 -RRB- + add dev ',Suzuki,'num_data_8802 amod_data_+082 cc_+082_+ conj_+082_unlabeled dobj__data number_27M_17M num_words_27M appos_data_words num_dev_8841 amod_dev_+039 cc_+039_+ advmod_gazetters_supplied conj_+039_gazetters dep_gazetters_8890 dep_gazetters_+049 cc_gazetters_+ conj_gazetters_add nsubj__dev ' +o,5510,'Interestingly , the interannotator agreement on SWITCHBOARD -LRB- a0a2a1 a3a5a4a7a6a9a8a9a6 -RRB- is higher than on the lecture corpus and higher than the a0 - score reported by for the ICSI meeting data used by Murray et al ',Galley,'advmod_higher_Interestingly det_agreement_the nn_agreement_interannotator nsubj_higher_agreement prep_agreement_on pobj_on_SWITCHBOARD amod_a3a5a4a7a6a9a8a9a6_a0a2a1 appos_agreement_a3a5a4a7a6a9a8a9a6 cop_higher_is prep_higher_than dep_than_on det__the nn__lecture nn__corpus pobj_on_ cc_higher_and conj_higher_higher prep_higher_than det_a0_the pobj_than_a0 nsubj_reported_score parataxis_higher_reported prep_reported_by pobj_by_ prep_reported_for det_data_the nn_data_ICSI nn_data_meeting pobj_for_data partmod_data_used prep_used_by pobj_by_Murray cc_Murray_et conj_Murray_al ' +o,5511,'In the results we describe here , we use mutual information as the metric for neighborhood pruning , pruning which occurs as the network is being generated ',Church,'prep_use_In det_results_the pobj_In_results nsubj_describe_we rcmod_results_describe advmod_describe_here nsubj_use_we amod__mutual nn__information dobj_use_ prep_use_as det_pruning_the amod_pruning_metric prep_metric_for pobj_for_neighborhood pobj_as_pruning appos_pruning_pruning nsubj_occurs_which rcmod_pruning_occurs mark_generated_as det_network_the nsubjpass_generated_network aux_generated_is auxpass_generated_being advcl_occurs_generated ' +o,5512,'Our work builds upon Turneys work on semantic orientation and synonym learning , in which he used a PMI-IR algorithm to measure the similarity of words and phrases based on Web queries ',Turney,'poss_work_Our nsubj_builds_work prep_builds_upon nn_work_Turneys pobj_upon_work prep_builds_on amod__semantic nn__orientation pobj_on_ cc__and nn__synonym nn__learning conj__ rel_used_in pobj_in_which nsubj_used_he rcmod__used det_algorithm_a nn_algorithm_PMI-IR dobj_used_algorithm aux_measure_to xcomp_used_measure det_similarity_the dobj_measure_similarity prep_similarity_of pobj_of_words cc_words_and conj_words_phrases prep_measure_based dep_based_on nn_queries_Web pobj_on_queries ' +o,5513,'2 We illustrate the rule extraction with an example from the tree-to-tree translation model based on tree sequence alignment without losing of generality to most syntactic tree based models ',Zhang,'nsubj_illustrate_We rcmod_2_illustrate det_extraction_the nn_extraction_rule dobj_illustrate_extraction prep_extraction_with det_example_an pobj_with_example prep_illustrate_from det_model_the amod_model_tree-to-tree nn_model_translation pobj_from_model partmod_model_based prep_based_on nn__tree nn__sequence nn__alignment pobj_on_ prep_based_without pcomp_without_losing prep_losing_of pobj_of_generality dep_tree_to advmod_syntactic_most amod_tree_syntactic dep_2_tree partmod_tree_based dobj_based_models ' +o,5514,'MET was carried out using a development set , and the BLEU score evaluated on two test sets ',Och,'nn__MET nsubjpass_carried_ auxpass_carried_was prep_carried_out pcomp_out_using det_development_a dobj_using_development partmod_development_set cc_development_and det_score_the amod_score_BLEU conj_development_score partmod_score_evaluated prep_evaluated_on num_sets_two nn_sets_test pobj_on_sets ' +p,5515,'The most widely used single-word-based statistical alignment models -LRB- SAMs -RRB- have been proposed in ',Brown,'det_models_The advmod_used_most advmod_used_widely amod_models_used amod_models_single-word-based amod_models_statistical nn_models_alignment nsubjpass_proposed_models abbrev_models_SAMs aux_proposed_have auxpass_proposed_been prep_proposed_in ' +o,5516,'However morphosyntactic features alone can not verify the terminological status of the units extracted since they can also select non terms ',Smadja,'advmod_verify_However amod_features_morphosyntactic nsubj_verify_features advmod_verify_alone aux_verify_can neg_verify_not det_status_the amod_status_terminological dobj_verify_status prep_status_of det_units_the pobj_of_units partmod_units_extracted mark_select_since nsubj_select_they aux_select_can advmod_select_also advcl_extracted_select nn_terms_non dobj_select_terms ' +o,5517,'There has thus been a trend recently towards robust wide-coverage semantic construction -LRB- eg , -RRB- ',Zettlemoyer,'expl_been_There aux_been_has advmod_been_thus det_trend_a dobj_been_trend advmod_been_recently prep_been_towards amod_construction_robust amod_construction_wide-coverage amod_construction_semantic pobj_towards_construction appos_construction_eg dep_eg_ ' +o,5518,'This model is trained on approximately 5 million sentence pairs of Hansard -LRB- Canadian parliamentary -RRB- and UN proceedings which have been aligned on a sentence-by-sentence basis by the methods of , and then further aligned on a word-by-word basis by methods similar to ',Brown,'det_model_This nsubjpass_trained_model auxpass_trained_is prep_trained_on quantmod_million_approximately number_million_5 num_pairs_million nn_pairs_sentence pobj_on_pairs prep_pairs_of pobj_of_Hansard nn_parliamentary_Canadian appos_Hansard_parliamentary cc_pairs_and nn_proceedings_UN conj_pairs_proceedings nsubjpass_aligned_which aux_aligned_have auxpass_aligned_been rcmod_proceedings_aligned prep_aligned_on det_basis_a amod_basis_sentence-by-sentence pobj_on_basis prep_aligned_by det_methods_the pobj_by_methods prep_methods_of pobj_of_ cc_aligned_and advmod_aligned_then advmod_aligned_further conj_aligned_aligned prep_aligned_on det_basis_a amod_basis_word-by-word pobj_on_basis prep_aligned_by pobj_by_methods amod_methods_similar prep_similar_to ' +p,5519,'Decision lists have already been successfully applied to lexical ambiguity resolution by where they perfromed well ',Yarowsky,'nn_lists_Decision nsubjpass_applied_lists aux_applied_have advmod_applied_already auxpass_applied_been advmod_applied_successfully aux_lexical_to xcomp_applied_lexical amod_resolution_ambiguity dobj_lexical_resolution prep_lexical_by pobj_by_ advmod_perfromed_where nsubj_perfromed_they advcl_lexical_perfromed advmod_perfromed_well ' +p,5520,'Some of the more popular and more accurate of these approaches to data-driven parsing have been based on generative models that are closely related to probabilistic contextfree grammars ',Collins,'nsubjpass_based_Some prep_Some_of det__the advmod_popular_more amod__popular cc_popular_and advmod_accurate_more conj_popular_accurate prep_accurate_of det_approaches_these pobj_of_approaches prep_approaches_to amod_parsing_data-driven pobj_to_parsing pobj_of_ aux_based_have auxpass_based_been prep_based_on amod_models_generative pobj_on_models nsubjpass_related_that auxpass_related_are advmod_related_closely rcmod_models_related prep_related_to amod_grammars_probabilistic nn_grammars_contextfree pobj_to_grammars ' +o,5521,'In contrast , globally optimized clustering decisions were reported in and , where all clustering possibilities are considered by searching on a Bell tree representation or by using the Learning as Search Optimization -LRB- LaSO -RRB- framework respectively , but the first search is partial and driven by heuristics and the second one only looks back in text ','Daume III','prep_reported_In pobj_In_contrast advmod_reported_globally amod_decisions_optimized nn_decisions_clustering nsubjpass_reported_decisions auxpass_reported_were prep_reported_in pobj_in_ cc__and conj__ advmod_considered_where det_possibilities_all amod_possibilities_clustering nsubjpass_considered_possibilities auxpass_considered_are advcl_reported_considered prep_considered_by pcomp_by_searching prep_searching_on det_representation_a nn_representation_Bell nn_representation_tree pobj_on_representation cc_by_or conj_by_by pcomp_by_using det_framework_the amod_framework_Learning prep_Learning_as pobj_as_Search nn_framework_Optimization appos_framework_LaSO dobj_using_framework partmod_framework_ advmod__respectively cc_reported_but det_search_the amod_search_first nsubjpass_partial_search auxpass_partial_is conj_reported_partial cc_partial_and conj_partial_driven prep_driven_by pobj_by_heuristics cc_reported_and det_one_the amod_one_second nsubj_looks_one advmod_looks_only conj_reported_looks prt_looks_back prep_looks_in pobj_in_text ' +o,5522,' compare taggers trained and tested on the Wall Street Journal and the Lancaster-Oslo-Bergen corpora and find that the results for the WSJ perform significantly worse ',Marcus,'advmod_compare_ dobj_compare_taggers partmod_taggers_trained cc_trained_and conj_trained_tested prep_trained_on det__the nn__Wall nn__Street nn__Journal pobj_on_ cc__and det_corpora_the amod_corpora_Lancaster-Oslo-Bergen amod_corpora_ conj__corpora cc_compare_and conj_compare_find complm_perform_that det_results_the nsubj_perform_results prep_results_for det_WSJ_the pobj_for_WSJ ccomp_find_perform advmod_worse_significantly acomp_perform_worse ' +o,5523,'The training samples are respectively used to create the models PT ^ G , PCHUNK , PBUILD , and PCMECK , all of which have the form : k p -LRB- a , b -RRB- = II _ ij -LRB- o , b ~ j -LRB- 1 -RRB- j -- 1 where a is some action , b is some context , ~ \' is a nor4 Model Categories Description Templates Used TAG See CHUNK chunkandpostag -LRB- n -RRB- \\* BUILD CHECK chunkandpostag -LRB- m , n -RRB- \\* cons -LRB- n -RRB- cons -LRB- re , n -RRB- \\* cons -LRB- m , n , p -RRB- T punctuation checkcons -LRB- n -RRB- \\* checkcons -LRB- m , n -RRB- \\* production surround -LRB- n -RRB- \\* The word , POS tag , and chunk tag of nth leaf ',Ratnaparkhi,'det_training_The nsubjpass_used_samples auxpass_used_are advmod_used_respectively ccomp_training_used aux_create_to xcomp_used_create det_models_the dobj_create_models nn_G_PT nn_G_^ nsubj_=_G nn_PBUILD_PCHUNK appos_G_PBUILD cc_PBUILD_and conj_PBUILD_PCMECK dep_of_all nsubj_have_of pobj_of_which rcmod_PBUILD_have det_form_the dobj_have_form nn_p_k dep_PBUILD_p dep_p_a appos_a_b rcmod_models_= nn___II dobj_=__ partmod___ij dobj_ij_o dep_j_b amod_j_~ dep_training_j appos_j_1 partmod_j_j dep_context_1 advmod_action_where nsubj_action_a cop_action_is det_action_some dep_context_action nsubj_context_b cop_context_is det_context_some dep_j_context nsubj_nor4_~ cop_nor4_is det_nor4_a dep_context_nor4 nn_Templates_Model nn_Templates_Categories nn_Templates_Description nsubj_Used_Templates rcmod_nor4_Used nn_CHUNK_TAG nn_CHUNK_See nn_CHUNK_ iobj_Used_CHUNK dobj_Used_chunkandpostag appos_chunkandpostag_n dep_chunkandpostag_\\* nn_chunkandpostag_BUILD nn_chunkandpostag_CHECK dobj_Used_chunkandpostag appos_chunkandpostag_m dep_m_n dep_chunkandpostag_\\* dobj_Used_cons appos_cons_n dobj_Used_cons appos_cons_re dep_re_n dep_cons_\\* dobj_Used_cons nn_p_m dep_p_n appos_cons_p nn_checkcons_T nn_checkcons_punctuation dobj_Used_checkcons appos_checkcons_n dep_checkcons_\\* dobj_Used_checkcons appos_checkcons_m dep_m_n dep_checkcons_\\* nn_surround_production dobj_Used_surround appos_surround_n dep_word_\\* det_word_The dep_surround_word dep_tag_POS dobj_Used_tag cc_tag_and nn_tag_chunk conj_tag_tag prep_tag_of nn_leaf_nth pobj_of_leaf ' +o,5524,'We also record for each token its derivational root , using the CELEX database ',Brown,'nsubj_record_We advmod_record_also prep_record_for det_token_each pobj_for_token poss_derivational_its nsubj_root_derivational dep_record_root xcomp_root_using det_database_the nn_database_CELEX nn_database_ dobj_using_database ' +o,5525,'1 Church and Hanks thus emphasize the importance of human judgment used in conjunction with these tools ',Church,'nsubj__Church cc_Church_and conj_Church_Hanks dep_1_ advmod__thus dep__emphasize det_importance_the dobj_emphasize_importance prep_importance_of amod_judgment_human pobj_of_judgment partmod_judgment_used prep_used_in pobj_in_conjunction prep_used_with det_tools_these pobj_with_tools ' +o,5526,' , in which we translate a source-language sentence f into the target-language sentence e that maximizes a linear combination of features and weights :1 e , a = argmax e , a score -LRB- e , a , f -RRB- -LRB- 1 -RRB- = argmax e , a Msummationdisplay m = 1 mhm -LRB- e , a , f -RRB- -LRB- 2 -RRB- where a represents the segmentation of e and f into phrases and a correspondence between phrases , and each hm is a R-valued feature with learned weight m The translation is typically found using beam search ',Koehn,'advmod_=_ dep_translate_in pobj_in_which nsubj_translate_we dep_=_translate det_sentence_a amod_sentence_source-language dobj_translate_sentence nsubj_e_f prep_f_into det_sentence_the amod_sentence_target-language pobj_into_sentence dep_sentence_e dep_maximizes_that dep_sentence_maximizes det_combination_a amod_combination_linear dobj_maximizes_combination prep_combination_of pobj_of_features cc_features_and conj_features_weights num_e_:1 nsubj_maximizes_e det_e_a amod_e_= nn_e_argmax appos_e_e det_score_a appos_e_score dep_score_e dep_e_a appos_a_f appos_score_1 dep_e_= nn_e_argmax rcmod_sentence_e det_m_a nn_m_Msummationdisplay nsubj_=_m num_mhm_1 dobj_=_mhm dep_mhm_e dep_e_a appos_a_f appos_mhm_2 advmod_represents_where nsubj_represents_a rcmod_mhm_represents det_segmentation_the dobj_represents_segmentation prep_segmentation_of dep_of_e cc_segmentation_and conj_segmentation_f prep_f_into pobj_into_phrases cc_f_and det_correspondence_a conj_f_correspondence prep_correspondence_between pobj_between_phrases cc_=_and det_hm_each nsubj_feature_hm cop_feature_is det_feature_a amod_feature_R-valued conj_=_feature prep_feature_with amod_m_learned nn_m_weight pobj_with_m det_translation_The nsubjpass_found_translation auxpass_found_is advmod_found_typically rcmod_m_found xcomp_found_using nn_search_beam dobj_using_search ' +o,5527,'Drawing on analysis of the Yarowsky algorithm , we perform bootstrapping by entropy regularization : we maximize a linear combination of conditional likelihood on labeled data and confidence -LRB- negative Renyi entropy -RRB- on unlabeled data ',Abney,'dep_perform_Drawing prep_Drawing_on amod_analysis_ pobj_on_analysis prep_analysis_of det_algorithm_the nn_algorithm_Yarowsky pobj_of_algorithm nsubj_perform_we xcomp_perform_bootstrapping prep_bootstrapping_by amod_regularization_entropy pobj_by_regularization nsubj_maximize_we ccomp_perform_maximize det_combination_a amod_combination_linear dobj_maximize_combination prep_combination_of amod_likelihood_conditional pobj_of_likelihood prep_likelihood_on amod_data_labeled pobj_on_data cc_likelihood_and conj_likelihood_confidence amod_entropy_negative amod_entropy_Renyi appos_confidence_entropy prep_confidence_on amod_data_unlabeled pobj_on_data ' +o,5528,'3 GM Representation of IBM MT Models In this section we present a GM representation for IBM model 3 in fig ',Brown,'num_Representation_3 nn_Representation_GM prep_Representation_of nn_Models_IBM nn_Models_MT pobj_of_Models dep_Representation_In det_section_this pobj_In_section nsubj_present_we rcmod_section_present det_representation_a nn_representation_GM dobj_present_representation prep_present_for nn__IBM nn__model nn__3 pobj_for_ prep__in pobj_in_fig ' +o,5529,'Previous work has demonstrated that this scoring function is able to provide high discrimination power for a variety of applications ',Brown,'amod_work_Previous nsubj_demonstrated_work aux_demonstrated_has complm_able_that det_function_this amod_function_scoring nsubj_able_function cop_able_is ccomp_demonstrated_able aux_provide_to xcomp_able_provide amod_power_high nn_power_discrimination dobj_provide_power prep_provide_for det_variety_a pobj_for_variety prep_variety_of nn__applications pobj_of_ ' +p,5530,'Indeed , recent work has shown that benefits can be made by first separating facts from opinions in a document -LRB- eg , Yu and Hatzivassiloglou -RRB- and classifying the polarity based solely on the subjective portions of the document -LRB- eg , -RRB- ',Pang,'advmod_shown_Indeed amod_work_recent nsubj_shown_work aux_shown_has complm_made_that nsubjpass_made_benefits aux_made_can auxpass_made_be ccomp_shown_made prep_made_by advmod_separating_first pcomp_by_separating dobj_separating_facts prep_separating_from pobj_from_opinions prep_opinions_in det_document_a pobj_in_document dep_document_eg conj_eg_Yu cc_eg_and nn__Hatzivassiloglou conj_eg_ cc_separating_and conj_separating_classifying det_polarity_the dobj_classifying_polarity prep_classifying_based dep_on_solely dep_based_on det_portions_the amod_portions_subjective pobj_on_portions prep_portions_of det_document_the pobj_of_document appos_document_eg dep_eg_ ' +o,5531,'22 Learning Algorithm For learning coreference decisions , we used a Maximum Entropy model ',Berger,'num_Algorithm_22 nn_Algorithm_Learning prep_used_For pcomp_For_learning amod_decisions_coreference dobj_learning_decisions nsubj_used_we dep_Algorithm_used det_model_a amod_model_Maximum nn_model_Entropy nn_model_ dobj_used_model ' +o,5532,'Since we need knowledge-poor Daille , 1996 -RRB- induction , we can not use human-suggested filtering Chi-squared -LRB- G24 -RRB- 2 Z-Score Students t-Score n-gram list in accordance to each probabilistic algorithm ',Dunning,'mark_need_Since nsubj_need_we advcl_use_need amod_induction_knowledge-poor dep_knowledge-poor_Daille appos_Daille_1996 dobj_need_induction nsubj_use_we aux_use_can neg_use_not amod_Chi-squared_human-suggested amod_Chi-squared_filtering dobj_use_Chi-squared appos_Chi-squared_G24 number__2 num_Students_ amod_Students_Z-Score nn_Students_ nsubj__Students advmod__t-Score rcmod_Chi-squared_ amod_list_n-gram dobj__list prep__in pobj_in_accordance prep__to det_algorithm_each amod_algorithm_probabilistic pobj_to_algorithm ' +o,5533,'In the first , a separate language model is trained on each column of the database and these models are then used to segment and label a given text sequence ',Canisius,'prep_trained_In det_first_the pobj_In_first det_model_a amod_model_separate nn_model_language nsubjpass_trained_model auxpass_trained_is dep_used_trained prep_trained_on det_column_each pobj_on_column prep_column_of det_database_the pobj_of_database cc_database_and det_models_these conj_database_models auxpass_used_are advmod_used_then prep_used_to pobj_to_segment cc_segment_and conj_segment_label det_sequence_a amod_sequence_given nn_sequence_text nsubjpass_used_sequence ' +o,5534,'The toolkit also implements suffix-array grammar extraction and minimum error rate training ',Och,'det_toolkit_The nsubj_implements_toolkit advmod_implements_also amod__suffix-array nn__grammar nn__extraction dobj_implements_ cc__and nn_training_minimum nn_training_error nn_training_rate conj__training ' +o,5535,'We use the neural network approximation to perform inference in our model ',Titov,'nsubj_use_We det_approximation_the amod_approximation_neural nn_approximation_network nsubj__approximation ccomp_use_ aux_perform_to xcomp__perform dobj_perform_inference prep_perform_in poss_model_our pobj_in_model ' +p,5536,'1 Introduction Chinese Word Segmentation -LRB- CWS -RRB- has been witnessed a prominent progress in the last three Bakeoffs , , ',Levow,'num_Segmentation_1 nn_Segmentation_Introduction amod_Segmentation_Chinese nn_Segmentation_Word nsubjpass_witnessed_Segmentation abbrev_Segmentation_CWS aux_witnessed_has auxpass_witnessed_been det_progress_a amod_progress_prominent dobj_witnessed_progress prep_progress_in det__the amod__last num__three nn__Bakeoffs pobj_in_ appos__ ' +o,5537,'We compared our system Lynx against a freely available phrase-based decoder Pharaoh ',Koehn,'dep_We_compared poss_Lynx_our nn_Lynx_system dep_compared_Lynx prep_Lynx_against det_Pharaoh_a advmod_available_freely amod_Pharaoh_available amod_Pharaoh_phrase-based nn_Pharaoh_decoder pobj_against_Pharaoh ' +o,5538,'Recently several latent variable models for constituent parsing have been proposed ',Koo,'amod_latent_several dep_Recently_latent amod_models_variable nsubjpass_proposed_models prep_models_for amod_parsing_constituent pobj_for_parsing aux_proposed_have auxpass_proposed_been rcmod_latent_proposed ' +o,5539,' presented randomized language model based on perfect hashing combined with entropy pruning to achieve further memory reductions ',Talbot,'nsubj_presented_ amod_model_randomized nn_model_language dobj_presented_model prep_presented_based dep_based_on amod_hashing_perfect pobj_on_hashing partmod_hashing_combined prep_combined_with amod_pruning_entropy pobj_with_pruning aux_achieve_to xcomp_combined_achieve amod_reductions_further nn_reductions_memory dobj_achieve_reductions ' +o,5540,'In particular , since we treat each individual speech within a debate as a single document , we are considering a version of document-level sentiment-polarity classification , namely , automatically distinguishing between positive and negative documents ',Turney,'prep_considering_In pobj_In_particular mark_treat_since nsubj_treat_we dep_considering_treat det_speech_each amod_speech_individual dobj_treat_speech prep_treat_within det_debate_a pobj_within_debate prep_debate_as det_document_a amod_document_single pobj_as_document nsubj_considering_we aux_considering_are det_version_a dobj_considering_version prep_version_of amod_classification_document-level amod_classification_sentiment-polarity pobj_of_classification advmod_considering_namely advmod_considering_automatically dep_considering_distinguishing prep_distinguishing_between amod_documents_positive cc_positive_and conj_positive_negative pobj_between_documents ' +o,5541,'Most current statistical models treat the aligned sentences in the corpus as sequences of tokens that are meant to be words ; the goal of the alignment process is to find links between source and target words ',Brown,'amod_models_Most amod_models_current amod_models_statistical nsubj__models xcomp__treat det_sentences_the amod_sentences_aligned dobj_treat_sentences prep_sentences_in det_corpus_the pobj_in_corpus prep_treat_as pobj_as_sequences prep_sequences_of pobj_of_tokens nsubjpass_meant_that auxpass_meant_are rcmod_tokens_meant aux_words_to cop_words_be xcomp_meant_words det_goal_the nsubj_is_goal prep_goal_of det_process_the amod_process_alignment pobj_of_process parataxis__is aux_find_to xcomp_is_find dobj_find_links prep_links_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,5542,'In addition , IC is stable even for relatively low frequency words , which can be contrasted with Fano \'s mutual information formula recently used by to compute word cooccurrence patterns in a 44 million word corpus of Associated Press news stories ',Church,'prep_stable_In pobj_In_addition nsubj_stable_IC cop_stable_is advmod_stable_even prep_stable_for advmod_low_relatively amod_words_low nn_words_frequency pobj_for_words nsubjpass_contrasted_which aux_contrasted_can auxpass_contrasted_be rcmod_words_contrasted prep_contrasted_with poss_formula_Fano possessive_Fano_\'s amod_formula_mutual nn_formula_information pobj_with_formula advmod_used_recently partmod_formula_used prep_used_by pobj_by_ aux_compute_to xcomp_used_compute nn_patterns_word nn_patterns_cooccurrence dobj_compute_patterns prep_compute_in det_corpus_a number_million_44 num_corpus_million nn_corpus_word pobj_in_corpus prep_corpus_of nn_stories_Associated nn_stories_Press nn_stories_news pobj_of_stories ' +o,5543,'A similar view underlies the class-based methods cited in Section 243 ',Brown,'det_view_A amod_view_similar nsubj_underlies_view det_methods_the amod_methods_class-based nsubj_cited_methods ccomp_underlies_cited prep_cited_in nn_243_Section pobj_in_243 ' +o,5544,'We have used a state-of-the-art Chinese handwriting recognizer developed by ATC , CCL , ITRI , Taiwan as the basis of our experiments ',Brown,'nsubj_used_We aux_used_have det__a amod__state-of-the-art amod__Chinese nn__handwriting nn__recognizer dobj_used_ partmod__developed prep_developed_by nn_Taiwan_ATC appos_Taiwan_CCL appos_Taiwan_ITRI pobj_by_Taiwan prep_Taiwan_as det_basis_the pobj_as_basis prep_basis_of poss_experiments_our pobj_of_experiments ' +o,5545,'We also use minimum error-rate training to tune our feature weights ',Och,'nsubj_use_We advmod_use_also amod__minimum amod__error-rate nn__training dobj_use_ aux_tune_to xcomp_use_tune poss_weights_our nn_weights_feature dobj_tune_weights ' +o,5546,'Thenthewordalignment is refined by performing grow-diag-final method ',Koehn,'nsubjpass_refined_Thenthewordalignment auxpass_refined_is prep_refined_by pcomp_by_performing amod_method_grow-diag-final dobj_performing_method ' +o,5547,' Hindi is a verb final , flexible word order language and therefore , has frequent occurrences of non-projectivity in its dependency structures ',Kuhlmann,'nn_Hindi_ nsubj_language_Hindi cop_language_is det_language_a amod_language_verb amod_language_final amod_language_flexible nn_language_word nn_language_order cc_language_and conj_language_therefore dep_therefore_has amod_occurrences_frequent dobj_has_occurrences prep_occurrences_of pobj_of_non-projectivity prep_non-projectivity_in poss_structures_its amod_structures_dependency pobj_in_structures ' +o,5548,'173 The standard features for genre classification models include words , part-of-speech -LRB- POS -RRB- tags , and punctuation , but constituent-based syntactic categories have also been explored ',Marcus,'det_features_The amod_features_standard nsubj_include_features prep_features_for amod_models_genre nn_models_classification pobj_for_models dep_173_include dobj_include_words amod_tags_part-of-speech appos_tags_POS conj_words_tags cc_words_and nn__punctuation conj_words_ cc_include_but amod_categories_constituent-based amod_categories_syntactic nsubj__categories aux__have advmod__also dep__been advmod__explored conj_include_ ' +o,5549,'In the following sections , we present the best performing set of feature templates as determined on the development data set using only the supervised training setting ; our feature templates have thus not been influenced nor extended by the unsupervised data13 11The full list of tags , as used by , also makes the underlying Viterbi algorithm unbearably slow ',Shen,'prep_present_In det_sections_the amod_sections_following pobj_In_sections nsubj_present_we det_set_the dep_performing_best amod_set_performing dobj_present_set prep_set_of nn_templates_feature pobj_of_templates advmod_determined_as amod_templates_determined prep_present_on det_data_the nn_data_development pobj_on_data partmod_data_set xcomp_set_using advmod_setting_only det_setting_the amod_setting_supervised nn_setting_training dobj_using_setting poss_templates_our nn_templates_feature nsubjpass_influenced_templates aux_influenced_have advmod_not_thus neg_influenced_not auxpass_influenced_been parataxis_present_influenced cc_influenced_nor conj_influenced_extended prep_extended_by det_list_the amod_list_unsupervised amod_list_data13 amod_list_11The amod_list_full pobj_by_list prep_list_of pobj_of_tags mark_makes_as csubj_makes_used prep_used_by pobj_by_ advmod_used_also advcl_extended_makes det_algorithm_the amod_algorithm_underlying nn_algorithm_Viterbi nsubj_slow_algorithm advmod_slow_unbearably xcomp_makes_slow ' +o,5550,'In this sense , instead of measuring only the categorial agreement between annotators with the kappa statistic or the performance of a system in terms of precision\\/recall , we could take into account the hierarchical organization of the categories or concepts by making use of measures considering the hierarchical distance between two concepts such as proposed by or ',Carletta,'prep_take_In det_sense_this pobj_In_sense dep_of_instead prep_take_of pcomp_of_measuring advmod_agreement_only det_agreement_the amod_agreement_categorial dobj_measuring_agreement prep_agreement_between pobj_between_annotators prep_annotators_with det__the nn__kappa nn__statistic pobj_with_ cc_annotators_or det_performance_the conj_annotators_performance prep_performance_of det_system_a pobj_of_system prep_system_in pobj_in_terms prep_terms_of pobj_of_precision\\/recall nsubj_take_we aux_take_could prep_take_into pobj_into_account det_organization_the amod_organization_hierarchical dobj_take_organization prep_organization_of det_categories_the pobj_of_categories cc_categories_or conj_categories_concepts prep_categories_by pcomp_by_making dobj_making_use prep_use_of pobj_of_measures partmod_measures_considering det_distance_the amod_distance_hierarchical dobj_considering_distance prep_distance_between num_concepts_two pobj_between_concepts dep_as_such prep_concepts_as pobj_as_proposed prep_proposed_by pobj_by_ cc__or conj__ ' +o,5551,'Such a similarity is calculated by using the WordNet : : Similarity tool , and , concretely , the Wu-Palmer measure , as defined in Equation1 ',Pedersen,'predet_similarity_Such det_similarity_a nsubj_calculated_similarity cop_calculated_is prep_calculated_by pcomp_by_using det_WordNet_the dobj_using_WordNet nn__Similarity nn__tool dep_WordNet_ cc_by_and conj_by_concretely det_measure_the amod_measure_Wu-Palmer dep_in_measure advmod_defined_as amod_measure_defined conj_by_in nn__Equation1 pobj_in_ ' +o,5552,'Furthermore , I plan to apply my parsers in other domains -LRB- eg , biomedical data -RRB- besides treebank data , to investigate the effectiveness and generality of my approaches ',Blitzer,'advmod_plan_Furthermore nsubj_plan_I aux_apply_to xcomp_plan_apply poss_parsers_my dobj_apply_parsers prep_apply_in amod_domains_other pobj_in_domains dep_domains_eg amod_data_biomedical appos_eg_data partmod_domains_ prep__besides nn_data_treebank pobj_besides_data aux_investigate_to xcomp__investigate det_effectiveness_the dobj_investigate_effectiveness cc_effectiveness_and conj_effectiveness_generality prep_effectiveness_of poss_approaches_my pobj_of_approaches ' +o,5553,' studied open domain relation extraction , for which they manually identified several common relation patterns ',Banko,'advmod_studied_ amod_extraction_open nn_extraction_domain nn_extraction_relation nsubj_studied_extraction rel_identified_for pobj_for_which nsubj_identified_they advmod_identified_manually rcmod_extraction_identified amod_patterns_several amod_patterns_common nn_patterns_relation dobj_identified_patterns ' +o,5554,'Once we obtain the augmented phrase table , we should run the minimum-error-rate training with the augmented phrase table such that the model parameters are properly adjusted ',Och,'mark_obtain_Once nsubj_obtain_we advcl__obtain det_table_the amod_table_augmented nn_table_phrase dobj_obtain_table nsubj__we aux__should dep__run det__the amod__minimum-error-rate nn__training prep__with det_table_the amod_table_augmented nn_table_phrase pobj_with_table prep__such complm_adjusted_that det_parameters_the nn_parameters_model nsubjpass_adjusted_parameters auxpass_adjusted_are advmod_adjusted_properly ccomp__adjusted ' +o,5555,'As such , we quantify success based on ROUGE scores ',Lin,'prep_quantify_As pobj_As_such nsubj_quantify_we dobj_quantify_success prep_quantify_based dep_based_on nn_scores_ROUGE nn_scores_ pobj_on_scores ' +p,5556,'Of particular interest are lexicalized parsing models such as the ones developed by and Carroll and Rooth -LRB- 1998 -RRB- ',Collins,'prep_lexicalized_Of amod_interest_particular pobj_Of_interest auxpass_lexicalized_are amod_models_parsing nsubjpass_lexicalized_models dep_as_such prep_models_as det_ones_the pobj_as_ones partmod_ones_developed prep_developed_by pobj_by_ cc__and conj__Carroll cc_Carroll_and conj_Carroll_Rooth appos_Carroll_1998 ' +o,5557,'(Ng and Low 2004, Toutanova et al, 2003, Brants 2000, Ratnaparkhi 1996, Samuelsson 1993).',Marcus,'' +o,5558,'In each iteration of local search , we look in the neighborhood of the current best alignment for a better alignment ',Brown,'prep_look_In det_iteration_each pobj_In_iteration prep_iteration_of amod_search_local pobj_of_search nsubj_look_we prep_look_in det_neighborhood_the pobj_in_neighborhood prep_neighborhood_of det_alignment_the amod_alignment_current amod_alignment_best pobj_of_alignment prep_alignment_for det_alignment_a amod_alignment_better pobj_for_alignment ' +o,5559,'To address this drawback , we proposed a new method3 to compute a more reliable and smoothed score in the undefined case , based on the IBM model 1 ',Brown,'aux_address_To dep_proposed_address det_drawback_this dobj_address_drawback nsubj_proposed_we det_method3_a amod_method3_new dobj_proposed_method3 aux_compute_to xcomp_proposed_compute det_score_a advmod_reliable_more amod_score_reliable cc_reliable_and conj_reliable_smoothed dobj_compute_score prep_compute_in det_case_the amod_case_undefined pobj_in_case prep_compute_based dep_based_on det_model_the nn_model_IBM pobj_on_model num_model_1 ' +o,5560,'One is the longest common subsequence -LRB- LCS -RRB- based approach ',Lin,'nsubj_subsequence_One cop_subsequence_is det_subsequence_the amod_subsequence_longest amod_subsequence_common abbrev_subsequence_LCS partmod_subsequence_based dobj_based_approach ' +p,5561,'Recently , graph-based methods have proved useful for a number of NLP and IR tasks such as document re-ranking in ad hoc IR and analyzing sentiments in text ',Pang,'advmod_proved_Recently amod_methods_graph-based nsubj_proved_methods aux_proved_have acomp_proved_useful prep_useful_for det_number_a pobj_for_number prep_number_of nn_tasks_NLP cc_NLP_and conj_NLP_IR pobj_of_tasks dep_as_such prep_tasks_as nn_re-ranking_document pobj_as_re-ranking prep_re-ranking_in amod__ad dep_ad_hoc nn__IR pobj_in_ cc_useful_and conj_useful_analyzing dobj_analyzing_sentiments prep_sentiments_in pobj_in_text ' +o,5562,'5 Experimental Evaluation To perform empirical evaluations of the proposed methods , we considered the task of parsing the Penn Treebank Wall Street Journal corpus ',Marcus,'num_Evaluation_5 nn_Evaluation_Experimental aux_perform_To dep_considered_perform amod_evaluations_empirical dobj_perform_evaluations prep_evaluations_of det_methods_the amod_methods_proposed pobj_of_methods nsubj_considered_we dep_Evaluation_considered det_task_the dobj_considered_task prep_task_of pcomp_of_parsing det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_Wall nn_corpus_Street nn_corpus_Journal dobj_parsing_corpus ' +o,5563,'In Section 3 , we will present a Perceptron like algorithm to obtain the parameters ',Collins,'prep_present_In pobj_In_Section num_Section_3 nsubj_present_we aux_present_will det_Perceptron_a dobj_present_Perceptron prep_Perceptron_like nn__algorithm pobj_like_ aux_obtain_to xcomp_present_obtain det_parameters_the dobj_obtain_parameters ' +o,5564,'F-Struct Feats Grammar Rules -LCB- PRED = PRO , NUM = SG PER = 3 , GEN = FEM -RCB- PRP-nom -LRB- = -RRB- she -LCB- PRED = PRO , NUM = SG PER = 3 , GEN = FEM -RCB- PRP-acc -LRB- = -RRB- her Table 5 : Lexical item rules with case markings 4 A History-Based Generation Model The automatic generation grammar transform presented in provides a solution to coarse-grained and -LRB- in fact -RRB- inappropriate independence assumptions in the basic generation model ',Cahill,'nn_PRP-nom_Feats nn_PRP-nom_Grammar nn_PRP-nom_Rules nn_PRO_PRED nn_PRO_= dep_FEM_PRO nn_PER_NUM nn_PER_= nn_PER_SG appos_PRO_PER dep_3_= rcmod_PER_3 nsubj_FEM_GEN dep_FEM_= dep_PRP-nom_FEM dobj_F-Struct_PRP-nom appos_PRP-nom_= dep_F-Struct_she parataxis_F-Struct_PRED dep_PRO_= ccomp_PRED_PRO advmod_=_NUM dep_PRED_= nn_PER_SG dobj_=_PER dep_3_= dep_PER_3 nsubj_FEM_GEN dep_FEM_= ccomp_PRED_FEM amod_=_PRP-acc dep_Model_= poss_5_her number_5_Table dep_=_5 nn_rules_Lexical nn_rules_item dep_5_rules prep_rules_with nn_markings_case pobj_with_markings dep_Model_4 det_Model_A amod_Model_History-Based nn_Model_Generation dep_provides_Model det_grammar_The amod_grammar_automatic nn_grammar_generation dep_Model_grammar csubj_provides_transform dep_transform_presented prep_presented_in pobj_in_ dep_F-Struct_provides det_solution_a dobj_provides_solution prep_provides_to amod_assumptions_coarse-grained cc_coarse-grained_and dep_coarse-grained_in pobj_in_fact conj_coarse-grained_inappropriate nn_assumptions_independence pobj_to_assumptions prep_assumptions_in det_model_the amod_model_basic nn_model_generation pobj_in_model ' +o,5565,'Moreover , it was -LRB- without imposing determinism -RRB- the inference technique employed in ',Brown,'advmod_technique_Moreover nsubj_technique_it cop_technique_was dep_technique_without pcomp_without_imposing dobj_imposing_determinism det_technique_the amod_technique_inference partmod_technique_employed prep_employed_in pobj_in_ ' +o,5566,'Oxford, UK: Oxford University Press. L A Black C McMeel M McTear N Black R Harper M Lemon Implementing autonomy in a diabetes management system 2005 J Telemed Telecare 11 -care applications, Examples include scheduling appointments over the phone (Zajicek et al. 2004, Wolters et al., submitted), interactive reminder systems (Pollack, 2005), symptom management systems (Black et al. 2005) or environmental control systems (Clarke et al. 2005).',Lin,'' +o,5567,'We dealt with this by either limiting the translation probability from the null word at the hypothetical 0-position over a threshold during the EM training , or setting SHo -LRB- j -RRB- to a small probability 7r instead of 0 for the initial null hypothesis H0 ',Brown,'nsubj_dealt_We prep_dealt_with pobj_with_this prep_dealt_by preconj_limiting_either pcomp_by_limiting det_probability_the nn_probability_translation dobj_limiting_probability prep_limiting_from det__the amod__null nn__word pobj_from_ prep_limiting_at det__the amod__hypothetical amod__0-position pobj_at_ prep__over det_threshold_a pobj_over_threshold prep_limiting_during det_training_the nn_training_EM pobj_during_training cc_limiting_or conj_limiting_setting dobj_setting_SHo appos_SHo_j prep_SHo_to det_7r_a amod_7r_small nn_7r_probability pobj_to_7r dep_of_instead prep_7r_of pobj_of_0 prep_setting_for det_H0_the amod_H0_initial amod_H0_null nn_H0_hypothesis pobj_for_H0 ' +o,5568,'Such metrics have been introduced in other fields , including PARADISE for spoken dialogue systems , BLEU for machine translation ,1 and ROUGE for summarisation ',Lin,'amod_metrics_Such nsubjpass_introduced_metrics aux_introduced_have auxpass_introduced_been prep_introduced_in amod_fields_other pobj_in_fields prep_introduced_including nn__PARADISE pobj_including_ prep__for amod_systems_spoken nn_systems_dialogue pobj_for_systems nn__BLEU conj__ prep__for nn_translation_machine pobj_for_translation num_translation_,1 cc__and nn__ROUGE conj__ prep__for pobj_for_summarisation ' +n,5569,'An alternative method makes decisions at the end but has a high computational requirement ',Wu,'det__An amod__alternative nn__method nsubj_makes_ dobj_makes_decisions prep_makes_at det_end_the pobj_at_end cc_makes_but conj_makes_has det_requirement_a amod_requirement_high amod_requirement_computational dobj_has_requirement ' +o,5570,'Our features were based on those in ',Finkel,'poss_features_Our nsubjpass_based_features auxpass_based_were prep_based_on pobj_on_those prep_based_in ' +p,5571,'For English , after a relatively big jump achieved by , we have seen two significant improvements : and pushed the results by a significant amount each time1 1In our final comparison , we have also included the results of , because it has surpassed as well and we have used this tagger in the data preparation phase ',Shen,'dep_seen_For pobj_For_English mark_achieved_after det_jump_a advmod_big_relatively amod_jump_big nsubj_achieved_jump dep_seen_achieved prep_achieved_by pobj_by_ nsubj_seen_we aux_seen_have num_improvements_two amod_improvements_significant dobj_seen_improvements nsubj_pushed_ cc__and conj__ parataxis_seen_pushed det_results_the dobj_pushed_results prep_results_by det_amount_a amod_amount_significant pobj_by_amount det_1In_each amod_1In_time1 tmod_pushed_1In poss_comparison_our amod_comparison_final dep_1In_comparison nsubj_included_we aux_included_have advmod_included_also conj_seen_included det_results_the dobj_included_results prep_results_of pobj_of_ mark_surpassed_because nsubj_surpassed_it aux_surpassed_has advcl_included_surpassed xcomp_surpassed_ advmod_well_as advmod__well cc_seen_and nsubj_used_we aux_used_have conj_seen_used det_tagger_this dobj_used_tagger prep_tagger_in det_phase_the nn_phase_data nn_phase_preparation pobj_in_phase ' +o,5572,'In prior research , ILP was used as a postprocessing step to remove redundancy and make other global decisions about parameters ',Clarke,'prep_used_In amod_research_prior pobj_In_research nsubjpass_used_ILP auxpass_used_was prep_used_as det_step_a amod_step_postprocessing pobj_as_step aux_remove_to xcomp_used_remove dobj_remove_redundancy cc_remove_and conj_remove_make amod_decisions_other amod_decisions_global dobj_make_decisions prep_make_about pobj_about_parameters ' +p,5573,'The algorithm was one of the first bootstrapping algorithms to become widely known in computational linguistics ',Yarowsky,'det_algorithm_The amod_algorithm_ nsubj_one_algorithm cop_one_was prep_one_of det_algorithms_the amod_algorithms_first amod_algorithms_bootstrapping pobj_of_algorithms aux_known_to auxpass_known_become advmod_known_widely infmod_one_known prep_known_in amod_linguistics_computational pobj_in_linguistics ' +o,5574,'Previous work used all possible pre xes and suf xes ranging in length from 1 to k characters , with k = 4 , and k = 10 ',Ratnaparkhi,'amod_work_Previous nsubj_used_work det_xes_all amod_xes_possible amod_xes_pre nsubj_=_xes cc_xes_and nn_xes_suf conj_xes_xes partmod_xes_ranging prep_ranging_in pobj_in_length prep_ranging_from number_k_1 dep_k_to num_characters_k pobj_from_characters prep_ranging_with pobj_with_k ccomp_used_= number__4 dobj_=_ cc__and conj__k dep__= num__10 rcmod_k_ ' +o,5575,'Texts are represented by dependency parse trees -LRB- using the Minipar parser -RRB- and templates by parse sub-trees ',Lin,'nsubjpass_represented_Texts auxpass_represented_are prep_represented_by amod_trees_dependency nn_trees_parse pobj_by_trees parataxis_represented_using det__the nn__Minipar nn__parser dobj_using_ cc_represented_and conj_represented_templates prep_templates_by amod_sub-trees_parse pobj_by_sub-trees ' +o,5576,' employsthez-scoreinconjunction with several heuristics -LRB- eg , the systematic occurrenceof two lexical items at the same distanceintext -RRB- andextractspredicativecollocations , 1Eg , -LRB- Frantziet al ,2000 ; Pearce ,2001 ; Goldmanet al , 2001 ; ZaiuInkpenandHirst ,2002 ; Dias ,2003 ; Seretanetal ',Smadja,'dobj__employsthez-scoreinconjunction prep__with amod_heuristics_several pobj_with_heuristics dep__eg det_occurrenceof_the amod_occurrenceof_systematic appos_eg_occurrenceof num_items_two amod_items_lexical dep_eg_items dep_eg_at det_distanceintext_the amod_distanceintext_same pobj_at_distanceintext nsubj__andextractspredicativecollocations appos_andextractspredicativecollocations_1Eg nn_al_Frantziet appos_1Eg_al num_al_,2000 dep_1Eg_Pearce dep_Pearce_,2001 nn_al_Goldmanet dep_1Eg_al appos_al_2001 dep_1Eg_ZaiuInkpenandHirst dep_ZaiuInkpenandHirst_,2002 dep_1Eg_Dias num_Dias_,2003 dep_Dias_Seretanetal ' +o,5577,'We evaluated performance by measuring WER -LRB- word error rate -RRB- , PER -LRB- position-independent word error rate -RRB- , BLEU and TER -LRB- translation error rate -RRB- using multiple references ',Papineni,'nsubj_evaluated_We dobj_evaluated_performance prep_evaluated_by pcomp_by_measuring dobj_measuring_WER nn_rate_word nn_rate_error dep_WER_rate appos_WER_PER amod_rate_position-independent nn_rate_word nn_rate_error appos_PER_rate nn__BLEU appos_WER_ cc__and conj__TER nn_rate_translation nn_rate_error dep__rate partmod_WER_ xcomp__using amod_references_multiple dobj_using_references ' +o,5578,'(Zollmann et al., 2008).',Zollmann,'' +o,5579,'Whereas most of the work on English has been based on constituency-based representations , partly inuenced by the availability of data resources such as the Penn Treebank , it has been argued that free constituent order languages can be analyzed more adequately using dependency-based representations , which is also the kind of annotation found , for example , in the Prague Dependency Treebank of Czech ',Marcus,'mark_based_Whereas nsubjpass_based_most prep_most_of det_work_the pobj_of_work prep_work_on pobj_on_English aux_based_has auxpass_based_been advcl_argued_based prep_based_on amod_representations_constituency-based pobj_on_representations advmod_inuenced_partly partmod_representations_inuenced prep_inuenced_by det_availability_the pobj_by_availability prep_availability_of nn_resources_data pobj_of_resources dep_as_such prep_resources_as det__the nn__Penn nn__Treebank pobj_as_ nsubjpass_argued_it aux_argued_has auxpass_argued_been complm_analyzed_that amod_languages_free amod_languages_constituent nn_languages_order nsubjpass_analyzed_languages aux_analyzed_can auxpass_analyzed_be ccomp_argued_analyzed dobj_analyzed_more advmod_using_adequately xcomp_analyzed_using amod_representations_dependency-based dobj_using_representations nsubj_kind_which cop_kind_is advmod_kind_also det_kind_the rcmod_representations_kind prep_kind_of pobj_of_annotation partmod_annotation_found prep_found_for pobj_for_example prep_found_in det_Treebank_the nn_Treebank_Prague nn_Treebank_Dependency pobj_in_Treebank prep_Treebank_of amod__Czech pobj_of_ ' +o,5580,'In our VB experiments we set i = j = 01 , i -LCB- 1 , , T -RCB- , j -LCB- 1 , , V -RCB- , which yielded the best performance on most reported metrics in ',Johnson,'prep_set_In poss_experiments_our nn_experiments_VB pobj_In_experiments nsubj_set_we nsubj_=_i ccomp_set_= dobj_=_j dep_j_= dep_=_01 appos_01_i dep_01_1 appos_=_T appos_=_j dep_=_1 appos_j_V nsubj_yielded_which rcmod_j_yielded det_performance_the amod_performance_best dobj_yielded_performance prep_performance_on advmod_reported_most amod_metrics_reported pobj_on_metrics prep_=_in ' +o,5581,'Note that , since the FrameNet data does not include deep syntactic tree annotation , we processed the FrameNet data with Collins parser , consequently , the experiments on FrameNet relate to automatic syntactic parse trees ',Collins,'complm_processed_that mark_include_since det_data_the nn_data_FrameNet nsubj_include_data aux_include_does neg_include_not advcl_processed_include advmod_syntactic_deep amod_annotation_syntactic nn_annotation_tree dobj_include_annotation nsubj_processed_we ccomp_Note_processed det_data_the nn_data_FrameNet dobj_processed_data prep_processed_with nn__Collins nn__parser pobj_with_ advmod_processed_consequently det_experiments_the nsubj_relate_experiments prep_experiments_on pobj_on_FrameNet dep_processed_relate prep_relate_to amod_trees_automatic amod_trees_syntactic nn_trees_parse pobj_to_trees ' +o,5582,'While transfer learning was proposed more than a decade ago , its application in natural language processing is still a relatively new territory , and its application in relation extraction is still unexplored ','Daume III','mark_proposed_While nn_learning_transfer nsubjpass_proposed_learning auxpass_proposed_was advcl__proposed dep_than_more quantmod_a_than num_decade_a measure_ago_decade advmod_proposed_ago dep_ago_ poss_application_its nsubj__application prep_application_in amod_processing_natural nn_processing_language pobj_in_processing cop__is advmod__still det__a advmod_new_relatively amod__new nn__territory cc__and poss_application_its nsubjpass_unexplored_application prep_application_in nn_extraction_relation pobj_in_extraction auxpass_unexplored_is advmod_unexplored_still conj__unexplored ' +p,5583,'Another widely used discriminative method is the perceptron algorithm , which achieves comparable performance to CRFs with much faster training , so we base this work on the perceptron ',Collins,'det_method_Another advmod_used_widely amod_method_used nn_method_discriminative nsubj__method cop__is det__the nn__perceptron nn__algorithm nsubj_achieves_which rcmod__achieves amod_performance_comparable dobj_achieves_performance prep_achieves_to pobj_to_CRFs prep_CRFs_with advmod_faster_much amod_training_faster pobj_with_training mark_base_so nsubj_base_we advcl_achieves_base det_work_this dobj_base_work prep_work_on det_perceptron_the pobj_on_perceptron ' +o,5584,'NER proves to be a knowledgeintensive task , and it was reassuring to observe that System Resources Used F1 + LBJ-NER Wikipedia , Nonlocal Features , Word-class Model 9080 Semi-supervised on 1Gword unlabeled data 8992 Semi-supervised on 27Mword unlabeled data 8931 Wikipedia 8802 Non-local Features 8724 Non-local Features 8717 + Non-local Features 8686 Table 7 : Results for CoNLL03 data reported in the literature ',Finkel,'nsubj_proves_NER aux_task_to cop_task_be det_task_a amod_task_knowledgeintensive xcomp_proves_task cc_proves_and nsubj_reassuring_it cop_reassuring_was conj_proves_reassuring aux_observe_to xcomp_reassuring_observe complm_reported_that nn_Resources_System nsubj_reported_Resources partmod_Resources_Used nn_Wikipedia_F1 nn_Wikipedia_+ nn_Wikipedia_LBJ-NER dobj_Used_Wikipedia nn_Features_Nonlocal appos_Wikipedia_Features nn_Model_Word-class appos_Wikipedia_Model num__9080 measure_Semi-supervised_ amod_Model_Semi-supervised prep_Semi-supervised_on amod_data_1Gword amod_data_unlabeled pobj_on_data num__8992 measure_Semi-supervised_ dep_on_Semi-supervised advmod_Model_on amod_data_27Mword amod_data_unlabeled pobj_on_data number__8931 num_Features_ nn_Features_Wikipedia nn_Features_8802 nn_Features_ nn_Features_Non-local nn_Features_Features nn_Features_8724 nn_Features_ nn_Features_Non-local nn_Features_Features nn_Features_8717 nn_Features_+ nn_Features_ nn_Features_Non-local dep_Model_Features num_Table_8686 dep_Features_Table num_Table_7 dep_Wikipedia_Results prep_Results_for nn_data_CoNLL03 pobj_for_data ccomp_observe_reported prep_reported_in det_literature_the pobj_in_literature ' +o,5585,'Second , instead of disambiguating phrase senses as in , we model word selection independently of the phrases used in the MT models ',Carpuat,'advmod_model_Second dep_of_instead prep_model_of pcomp_of_disambiguating nn_senses_phrase dobj_disambiguating_senses prep_disambiguating_as dep_as_in pobj_in_ nsubj_model_we nn_selection_word dobj_model_selection advmod_model_independently prep_model_of det_phrases_the pobj_of_phrases partmod_phrases_used prep_used_in det_models_the nn_models_MT pobj_in_models ' +o,5586,' , and et al ',Rosti,'cc__and conj__ nn_al_et dobj__al ' +o,5587,'Early examples of this work include ; more recent models include ',Wu,'advmod_examples_Early nsubj_include_examples prep_examples_of det_work_this pobj_of_work acomp_include_ advmod_models_more amod_models_recent nsubj_include_models parataxis_include_include acomp_include_ ' +p,5588,'For symmetrization , we found that Och and Neys refined technique described in produced the best AER for this data set under all experimental conditions ',Och,'prep_found_For pobj_For_symmetrization nsubj_found_we complm_produced_that nn_technique_Och cc_Och_and conj_Och_Neys nn_technique_refined nsubj_produced_technique partmod_technique_described prep_described_in pobj_in_ ccomp_found_produced det_AER_the amod_AER_best dobj_produced_AER prep_AER_for det_data_this pobj_for_data partmod_data_set prep_set_under det_conditions_all amod_conditions_experimental pobj_under_conditions ' +o,5589,'Instead of computing all intersections , only computes critical intersections where highest-score translations will change ',Och,'dep_of_Instead advcl_computes_of pcomp_of_computing det_intersections_all dobj_computing_intersections nsubj_computes_ advmod_computes_only amod_intersections_critical dobj_computes_intersections advmod_change_where amod_translations_highest-score nsubj_change_translations aux_change_will advcl_computes_change ' +o,5590,' ; -RRB- ',Koehn,'dep__ ' +o,5591,' Model 3 integrates the detection and resolution of WH-traces in relative clauses into a lexicalized PCFG ',Collins,'nn_3_ nn_3_Model nsubj_integrates_3 det_detection_the dobj_integrates_detection cc_detection_and conj_detection_resolution prep_detection_of pobj_of_WH-traces prep_WH-traces_in amod_clauses_relative pobj_in_clauses prep_integrates_into det_PCFG_a amod_PCFG_lexicalized pobj_into_PCFG ' +o,5592,'Given training data consisting of parallel sentences: }1),,{( )()( Sief ii =, our Model-1 training for t(f|e) is as follows: = = S s ss e efefceft 1 )()(1 ),;|()|( Where 1 e is a normalization factor such that 0.1)|( = j j eft ),;|( )()( ss efefc denotes the expected number of times that word e connects to word f. == = = l i i m j jl k k ss eeff eft eft efefc 11 1 )()( ),(),( )|( )|( ),;|( With the conditional probability t(f|e), the probability for an alignment of foreign string F given English string E is in (1): = = + = m j n i ijm eft l EFP 1 0 )|( )1( 1 )|( (1) The probability of alignment F given E: )|( EFP is shown to achieve the global maximum under this EM framework as stated in (Brown et al. ,1993).',Brown,'' +p,5593,'Ramshaw and Marcus successflflly applied Eric Brill \'s transformation-based learning method to the chunking problem ',Ramshaw,'nn__Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nsubj_applied_ advmod_applied_successflflly nn_Brill_Eric poss_method_Brill possessive_Brill_\'s amod_method_transformation-based nn_method_learning dobj_applied_method prep_applied_to det_problem_the amod_problem_chunking pobj_to_problem ' +o,5594,'As such , discourse markers play an important role in the parsing of natural language discourse , and their correspondence with discourse relations can be exploited for the unsupervised learning of discourse relations ',Marcu,'prep_play_As pobj_As_such nn_markers_discourse nsubj_play_markers det_role_an amod_role_important dobj_play_role prep_role_in det_parsing_the pobj_in_parsing prep_parsing_of amod__natural nn__language nn__discourse pobj_of_ cc_play_and poss_correspondence_their nsubjpass_exploited_correspondence prep_correspondence_with nn_relations_discourse pobj_with_relations aux_exploited_can auxpass_exploited_be conj_play_exploited prep_exploited_for det_learning_the amod_learning_unsupervised pobj_for_learning prep_learning_of nn_relations_discourse pobj_of_relations ' +o,5595,'Table 2 shows the results for English projective dependency trees extracted from the Penn Treebank using the rules of Yamada and Matsumoto -LRB- 2003 -RRB- ',Marcus,'nsubj_shows_Table num_Table_2 det_results_the dobj_shows_results prep_shows_for amod_trees_English amod_trees_projective amod_trees_dependency pobj_for_trees partmod_trees_extracted prep_extracted_from det__the nn__Penn nn__Treebank pobj_from_ xcomp_extracted_using det_rules_the dobj_using_rules prep_rules_of pobj_of_Yamada cc_Yamada_and conj_Yamada_Matsumoto appos_rules_2003 ' +o,5596,'Mean number of instances of paraphrase phenomena per sentence ',Barzilay,'nn_number_Mean prep_number_of pobj_of_instances prep_instances_of amod_phenomena_paraphrase pobj_of_phenomena prep_number_per pobj_per_sentence ' +n,5597,'Several teams had approaches that relied (to varying degrees) on an IBM model of statistical machine translation (Brown et al. , 1993), with different improvements brought by different teams, consisting of new submodels, improvements in the HMM model, model combination for optimal alignment, etc. Se-veral teams used symmetrization metrics, as introduced in (Och and Ney, 2003) (union, intersection, refined), most of the times applied on the alignments produced for the two directions sourcetarget and targetsource, but also as a way to combine different word alignment systems.',Brown,'' +o,5598,'1 Introduction Bilingual word alignment is first introduced as an intermediate result in statistical machine translation -LRB- SMT -RRB- ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Bilingual nn_alignment_word nsubjpass_introduced_alignment auxpass_introduced_is advmod_introduced_first prep_introduced_as det_result_an amod_result_intermediate pobj_as_result prep_result_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +p,5599,'METEOR uses the Porter stemmer and synonymmatching via WordNet to calculate recall and precision more accurately ',Banerjee,'nsubj_uses_METEOR det_stemmer_the nn_stemmer_Porter nsubj_calculate_stemmer cc_stemmer_and conj_stemmer_synonymmatching prep_stemmer_via pobj_via_WordNet aux_calculate_to xcomp_uses_calculate dobj_calculate_recall cc_recall_and conj_recall_precision advmod_accurately_more advmod_calculate_accurately ' +o,5600,'We then replaced fi with its associated z-score k $ , e k $ , e is the strength of code frequency f at Lt , and represents the standard deviation above the average of frequency fave , t Referring to Smadja \'s definition , the standard deviation at at Lt and strength kf , t of the code frequencies are defined as shown in formulas 1 and 2 ',Smadja,'nsubj_replaced_We advmod_replaced_then dobj_replaced_fi prep_replaced_with poss_k_its amod_k_associated amod_k_z-score pobj_with_k nsubjpass_defined_$ amod_k_e dep_at_k measure_at_$ amod_is_e dep_t_is det_strength_the advmod_is_strength dep_strength_of nn_frequency_code pobj_of_frequency dep_frequency_f dep_of_at pobj_at_Lt cc_Lt_and amod_deviation_represents det_deviation_the amod_deviation_standard conj_Lt_deviation prep_deviation_above det_average_the pobj_above_average prep_average_of nn_fave_frequency pobj_of_fave dep_Referring_t num_$_Referring prep_$_to poss__Smadja possessive_Smadja_\'s nn__definition pobj_to_ det_deviation_the amod_deviation_standard appos__deviation advmod_k_at dep_Lt_at num_$_Lt cc_$_and nn_kf_strength conj_$_kf appos_kf_t prep_t_of det_frequencies_the nn_frequencies_code pobj_of_frequencies auxpass_defined_are dep_replaced_defined mark_shown_as advcl_defined_shown prep_shown_in pobj_in_formulas num_formulas_1 cc_formulas_and conj_formulas_2 ' +p,5601,'Recent work has demonstrated that randomized encodings can be used to represent n-gram counts for LMs with signficant space-savings , circumventing information-theoretic constraints on lossless data structures by allowing errors with some small probability ',Talbot,'amod__Recent nn__work nsubj_demonstrated_ aux_demonstrated_has complm_used_that amod_encodings_randomized nsubjpass_used_encodings aux_used_can auxpass_used_be ccomp_demonstrated_used aux_represent_to purpcl_used_represent amod_counts_n-gram dobj_represent_counts prep_counts_for pobj_for_LMs prep_represent_with amod_space-savings_signficant pobj_with_space-savings prep_represent_circumventing amod_constraints_information-theoretic pobj_circumventing_constraints prep_constraints_on amod_structures_lossless nn_structures_data pobj_on_structures prep_represent_by pcomp_by_allowing dobj_allowing_errors prep_allowing_with det_probability_some amod_probability_small pobj_with_probability ' +o,5602,'It has been shown that the methods can be ported to other languages and treebanks , including Cast3LB ',Cahill,'nsubjpass_shown_It aux_shown_has auxpass_shown_been complm_ported_that det_methods_the nsubjpass_ported_methods aux_ported_can auxpass_ported_be ccomp_shown_ported prep_ported_to amod_languages_other pobj_to_languages cc_languages_and nn__treebanks conj_languages_ prep_ported_including nn__Cast3LB pobj_including_ ' +p,5603,'Also , slightly restating the advantages of phrase-pairs identified in , these blocks are effective at capturing context including the encoding of non-compositional phrase pairs , and capturing local reordering , but they lack variables -LRB- eg embedding between ne pas in French -RRB- , have sparsity problems , and lack a strategy for global reordering ',Quirk,'advmod_have_Also advmod_effective_slightly dep_effective_restating det_advantages_the dobj_restating_advantages prep_advantages_of pobj_of_phrase-pairs partmod_phrase-pairs_identified prep_identified_in pobj_in_ det_blocks_these nsubj_effective_blocks cop_effective_are dep_have_effective prep_effective_at pcomp_at_capturing dobj_capturing_context prep_context_including det_encoding_the pobj_including_encoding prep_encoding_of amod_pairs_non-compositional nn_pairs_phrase pobj_of_pairs cc_capturing_and conj_capturing_capturing amod_reordering_local dobj_capturing_reordering cc_effective_but nsubj_lack_they conj_effective_lack dobj_lack_variables dep_variables_eg amod_eg_embedding prep_eg_between nn_pas_ne pobj_between_pas prep_pas_in pobj_in_French amod_problems_sparsity dobj_have_problems cc_have_and conj_have_lack det_strategy_a dobj_lack_strategy prep_strategy_for amod_reordering_global pobj_for_reordering ' +o,5604,'Others try to accommodate both syntactic and lexical differences between the candidate translation and the reference , like CDER , which employs a version of edit distance for word substitution and reordering ; METEOR , which uses stemming and WordNet synonymy ; and a linear regression model developed by , which makes use of stemming , WordNet synonymy , verb class synonymy , matching noun phrase heads , and proper name matching ',Banerjee,'nsubj_try_Others aux_accommodate_to xcomp_try_accommodate preconj_syntactic_both amod_differences_syntactic cc_syntactic_and conj_syntactic_lexical dobj_accommodate_differences prep_accommodate_between det_translation_the nn_translation_candidate pobj_between_translation cc_translation_and det_reference_the conj_translation_reference prep_accommodate_like nn__CDER pobj_like_ nsubj_employs_which rcmod__employs det_version_a dobj_employs_version prep_version_of amod_distance_edit pobj_of_distance prep_distance_for nn_substitution_word pobj_for_substitution cc_substitution_and conj_substitution_reordering nn__METEOR dep_version_ nsubj_uses_which rcmod__uses xcomp_uses_stemming cc_stemming_and conj_stemming_WordNet dobj_stemming_synonymy cc_version_and det_model_a amod_model_linear nn_model_regression conj_version_model partmod_model_developed prep_developed_by pobj_by_ nsubj_makes_which rcmod__makes dobj_makes_use prep_use_of pcomp_of_stemming amod_synonymy_WordNet amod_synonymy_synonymy amod_synonymy_verb nn_synonymy_class conj_model_synonymy amod_heads_matching nn_heads_noun nn_heads_phrase appos_synonymy_heads cc_model_and amod_name_proper conj_model_name amod_name_matching ' +o,5605,'Most recently , Yarowsky used an unsupervised learning procedure to perform WSD , although this is only tested on disambiguating words into binary , coarse sense distinction ',Yarowsky,'advmod_recently_Most advmod_procedure_recently nsubj_procedure_Yarowsky dep_procedure_used det_procedure_an amod_procedure_unsupervised nn_procedure_learning aux_perform_to xcomp_procedure_perform nn__WSD dobj_perform_ mark_tested_although nsubjpass_tested_this auxpass_tested_is advmod_tested_only advcl_procedure_tested prep_tested_on pcomp_on_disambiguating dobj_disambiguating_words prep_disambiguating_into amod_distinction_binary amod_distinction_coarse nn_distinction_sense pobj_into_distinction ' +o,5606,'For each word in LDV , three existing thesauri are consulted : Rogets Thesaurus , Collins COBUILD Thesaurus , and WordNet ',Lin,'prep_consulted_For det_word_each pobj_For_word prep_word_in pobj_in_LDV num_thesauri_three amod_thesauri_existing nsubjpass_consulted_thesauri auxpass_consulted_are nn__Rogets nn__Thesaurus dobj_consulted_ nn__Collins nn__COBUILD nn__Thesaurus conj__ cc__and nn__WordNet conj__ ' +p,5607,'The efficient block alignment algorithm in Section 4 is related to the inversion transduction grammar approach to bilingual parsing described in : in both cases the number of alignments is drastically reduced by introducing appropriate re-ordering restrictions ',Wu,'det_algorithm_The amod_algorithm_efficient nn_algorithm_block nn_algorithm_alignment nsubjpass_related_algorithm prep_algorithm_in pobj_in_Section num_Section_4 auxpass_related_is prep_related_to det_approach_the nn_approach_inversion nn_approach_transduction nn_approach_grammar pobj_to_approach prep_related_to amod_parsing_bilingual pobj_to_parsing partmod_parsing_described prep_described_in pobj_in_ prep_reduced_in det_cases_both pobj_in_cases det_number_the nsubjpass_reduced_number prep_number_of pobj_of_alignments auxpass_reduced_is advmod_reduced_drastically parataxis_related_reduced prep_reduced_by pcomp_by_introducing amod_restrictions_appropriate amod_restrictions_re-ordering dobj_introducing_restrictions ' +p,5608,'3 System Overview 31 Translation model The system developed for this years shared task is a state-of-the-art , two-pass phrase-based statistical machine translation system based on a log-linear translation model ',Koehn,'num_model_3 nn_model_System nn_model_Overview num_model_31 nn_model_Translation det_system_The nsubj_developed_system dep_model_developed mark_shared_for det_years_this nsubj_shared_years advcl_developed_shared nsubj_system_task cop_system_is det_system_a amod_system_state-of-the-art amod_system_two-pass amod_system_phrase-based amod_system_statistical nn_system_machine nn_system_translation ccomp_shared_system prep_system_based dep_based_on det_model_a amod_model_log-linear nn_model_translation pobj_on_model ' +o,5609,'((:I (:Q DET NAMED-ENTITY) ENTER[V] (:Q THE ROOM[N])) (:I (:Q DET FEMALE-INDIVIDUAL) HAVE[V] (:Q DET ROOM[N])) (:I (:Q DET FEMALE-INDIVIDUAL) SLEEP[V]) (:I (:Q DET FEMALE-INDIVIDUAL) HAVE[V] (:Q DET (:F PLUR CLOTHE[N]))) (:I (:Q DET (:F PLUR CLOTHE[N])) WASHED[A])) Here the upper-case sentences are automatically generated verbalizations of the abstracted LFs shown beneath them.1 The initial development of KNEXT was based on the hand-constructed parse trees in the Penn Treebank version of the Brown corpus, but subsequently Schubert and collaborators refined and extended the system to work with parse trees obtained with statistical parsers (e.g., that of Collins (1997) or Charniak (2000)) applied to larger corpora, such as the British National Corpus (BNC), a 100 million-word, mixed genre collection, along with Web corpora of comparable size (see work of Van Durme et al.',Collins,'' +o,5610,'This merging of contexts is different than clustering words , but is applicable , as word clustering relies on knowing which contexts identify the same category ',Brown,'det_merging_This nsubj_different_merging prep_merging_of pobj_of_contexts cop_different_is mark__than amod_words_clustering nsubj__words advcl_different_ cc__but cop_applicable_is conj__applicable mark_relies_as nn_clustering_word nsubj_relies_clustering advcl_applicable_relies prep_relies_on pcomp_on_knowing nsubj_contexts_which ccomp_knowing_contexts xcomp_contexts_identify det_category_the amod_category_same dobj_identify_category ' +o,5611,' , various classification models and linguistic features have been proposed to improve the classification performance ',Pang,'amod_models_ amod_models_various nn_models_classification nsubjpass_proposed_models cc_models_and amod_features_linguistic conj_models_features aux_proposed_have auxpass_proposed_been aux_improve_to xcomp_proposed_improve det_performance_the nn_performance_classification dobj_improve_performance ' +o,5612,'Like baseNP chunking , content chunk parsing is also a kind of shallow parsing ',Ramshaw,'prep_kind_Like nn__baseNP nn__chunking pobj_Like_ nn_parsing_content nn_parsing_chunk nsubj_kind_parsing cop_kind_is advmod_kind_also det_kind_a prep_kind_of nn_parsing_shallow pobj_of_parsing ' +o,5613,'The weights 1 , , M are typically learned to directly minimize a standard evaluation criterion on development data -LRB- eg , the BLEU score ; Papineni et al , -RRB- using numerical search ',Och,'det_M_The amod_M_weights tmod_weights_1 nsubjpass_learned_M auxpass_learned_are advmod_learned_typically aux_minimize_to advmod_minimize_directly xcomp_learned_minimize det_criterion_a amod_criterion_standard nn_criterion_evaluation dobj_minimize_criterion prep_criterion_on nn_data_development pobj_on_data dep_data_eg det_score_the amod_score_BLEU appos_eg_score appos_eg_Papineni cc_Papineni_et conj_Papineni_al appos_eg_ partmod_data_using amod_search_numerical dobj_using_search ' +o,5614,'Adwait estimates a probability distribution for tagging using a maximum entropy approach ',Ratnaparkhi,'nn__Adwait nsubj_estimates_ det_distribution_a nn_distribution_probability dobj_estimates_distribution prep_estimates_for pcomp_for_tagging xcomp_tagging_using det_approach_a amod_approach_maximum nn_approach_entropy dobj_using_approach ' +p,5615,'Introduction Automatic word alignment is a vital component of all statistical machine translation -LRB- SMT -RRB- approaches ',Brown,'nn_Automatic_Introduction nsubj_approaches_Automatic nn__word nn__alignment nsubj_component_ cop_component_is det_component_a amod_component_vital rcmod_Automatic_component prep_component_of det_translation_all amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_Automatic_SMT ' +o,5616,'We obtained word alignments of training data by first running GIZA + + and then applying the refinement rule grow-diag-final-and ',Och,'nsubj_obtained_We nn_alignments_word dobj_obtained_alignments prep_alignments_of nn_data_training pobj_of_data prep_obtained_by advmod_running_first dep__running nsubj_+_GIZA xcomp_running_+ cc_+_+ conj_+_ cc_running_and advmod_running_then conj_running_applying det_rule_the amod_rule_refinement dobj_applying_rule amod__grow-diag-final-and pobj_by_ ' +o,5617,'Charniak gives a thorough explanation of the equations for an HMM model , and Kupiec describes an HMM tagging system in detail ',Marcus,'nn__Charniak nsubj_gives_ det_explanation_a amod_explanation_thorough dobj_gives_explanation prep_explanation_of det_equations_the pobj_of_equations prep_equations_for det_model_an nn_model_HMM pobj_for_model cc_gives_and nn__Kupiec nsubj_describes_ conj_gives_describes det_system_an nn_system_HMM amod_system_tagging dobj_describes_system prep_system_in pobj_in_detail ' +o,5618,'1 Introduction Various papers use phrase-based translation systems that have shown to improve translation quality over single-word based translation systems introduced in ',Brown,'num_papers_1 nn_papers_Introduction amod_papers_Various nsubj_use_papers amod_systems_phrase-based nn_systems_translation nsubj__systems ccomp_use_ complm_shown_that aux_shown_have ccomp__shown aux_improve_to xcomp_shown_improve nn_quality_translation dobj_improve_quality prep_improve_over pobj_over_single-word prep_improve_based nn_systems_translation dep_based_systems partmod_systems_introduced prep_introduced_in ' +p,5619,'4 Data and Evaluation For the CoNLL shared task , we have chosen to work with the same sections of the Penn Treebank as the widely used data set for base noun phrase recognition : WSJ sections 15-18 of the Penn Treebank as training material and section 20 as test material 3 ',Ramshaw,'num_Data_4 tmod_chosen_Data dep_Data_and dep_Data_Evaluation prep_chosen_For det_task_the nn_task_CoNLL amod_task_shared pobj_For_task nsubj_chosen_we aux_chosen_have aux_work_to xcomp_chosen_work prep_work_with det_sections_the amod_sections_same pobj_with_sections prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank prep_work_as det__the advmod_used_widely amod__used nn__data amod__set prep_set_for nn_noun_base pobj_for_noun nn__phrase nn__recognition pobj_as_ nn_15-18_WSJ nn_15-18_sections dep__15-18 prep_15-18_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank prep_Treebank_as nn_material_training pobj_as_material cc_Treebank_and conj_Treebank_section num_section_20 prep_section_as nn_material_test pobj_as_material num_material_3 ' +o,5620,'The various extraction measures have been discussed in great detail in the literature , their performance has been compared , and the methods have been combined to improve overall performance ',Dunning,'det_measures_The amod_measures_various nn_measures_extraction nsubjpass_discussed_measures aux_discussed_have auxpass_discussed_been prep_discussed_in amod_detail_great pobj_in_detail prep_detail_in det__the nn__literature pobj_in_ poss_performance_their nsubjpass_compared_performance aux_compared_has auxpass_compared_been ccomp_discussed_compared dep_compared_ cc_discussed_and det_methods_the nsubjpass_combined_methods aux_combined_have auxpass_combined_been conj_discussed_combined aux_improve_to xcomp_combined_improve amod__overall nn__performance dobj_improve_ ' +o,5621,'42 Approximated BLEU We used the BLEU score as the loss function computed by : BLEU -LRB- E ; E -RRB- = exp 1N Nsummationdisplay n = 1 log pn -LRB- E , E -RRB- BP -LRB- E , E -RRB- -LRB- 7 -RRB- where pn -LRB- -RRB- is the n-gram precision of hypothesized translations E = -LCB- et -RCB- Tt = 1 given reference translations E = -LCB- et -RCB- Tt = 1 and BP -LRB- -RRB- 1 is a brevity penalty ',Papineni,'num_BLEU_42 nn_BLEU_Approximated nsubj_used_We dep_BLEU_used det__the amod__BLEU nn__score dobj_used_ prep_used_as det_function_the nn_function_loss pobj_as_function dep_BLEU_computed prep_computed_by parataxis_computed_BLEU nn_E_E nsubj_=_E ccomp_BLEU_= nn_n_exp num_n_1N nn_n_Nsummationdisplay dobj_=_n dep_n_= iobj_=_1 nn_BP_log nn_BP_pn appos_BP_E dep_E_E dobj_=_BP appos_BP_E dep_E_E appos_BP_7 advmod_penalty_where nsubj_penalty_pn cop_precision_is det_precision_the amod_precision_n-gram parataxis_penalty_precision prep_precision_of amod_translations_hypothesized pobj_of_translations nsubj_=_E rcmod_precision_= nn_Tt_et nsubj_=_Tt ccomp_=_= dobj_=_1 partmod_1_given nn_translations_reference iobj_given_translations nn_Tt_E nn_Tt_= appos_Tt_et dobj_given_Tt dep_Tt_= dobj_=_1 cc_1_and nn_-LRB-_BP nsubj_penalty_1 cop_penalty_is det_penalty_a amod_penalty_brevity advcl_BLEU_penalty ' +o,5622,'Table 2 summarizes the characteristics of the training corpus used for training the parameters of Model 4 proposed in ',Brown,'number_2_Table num_summarizes_2 det_characteristics_the nsubj_proposed_characteristics prep_characteristics_of det_corpus_the nn_corpus_training pobj_of_corpus partmod_corpus_used prep_used_for pcomp_for_training det_parameters_the dobj_training_parameters prep_parameters_of pobj_of_Model num_Model_4 dep_summarizes_proposed prt_proposed_in ' +o,5623,'Most of researchers focus on how to extract useful textual features -LRB- lexical , syntactic , punctuation , etc -RRB- for determining the semantic orientation of the sentences using machine learning algorithm ',Pang,'nsubj_focus_Most prep_Most_of pobj_of_researchers prep_focus_on advmod_extract_how aux_extract_to pcomp_on_extract amod_features_useful dep_useful_textual dobj_extract_features dep_features_lexical appos_lexical_syntactic appos_lexical_punctuation dep_lexical_etc prep_extract_for pcomp_for_determining det_orientation_the amod_orientation_semantic dobj_determining_orientation prep_orientation_of det_sentences_the pobj_of_sentences partmod_sentences_using dobj_using_machine xcomp_using_learning nn__algorithm dobj_learning_ ' +o,5624,' searches with the full model , but makes assumptions about the the amount of reordering the language model can trigger in order to limit exploration ',Huang,'nsubj_searches_ prep_searches_with det_model_the amod_model_full pobj_with_model cc_searches_but conj_searches_makes dobj_makes_assumptions prep_makes_about det_amount_the dep_amount_the pobj_about_amount prep_amount_of pobj_of_reordering det_model_the nn_model_language nsubj_trigger_model aux_trigger_can dep_searches_trigger mark_limit_in dep_limit_order aux_limit_to purpcl_trigger_limit dobj_limit_exploration ' +o,5625,'Methods for doing so , for stochastic parser output , are described by Johnson and ',Cahill,'nsubjpass_described_Methods prep_Methods_for pcomp_for_doing advmod_doing_so prep_doing_for amod_output_stochastic nn_output_parser pobj_for_output auxpass_described_are prep_described_by nn_and_Johnson nn_and_ pobj_by_and ' +o,5626,'1 Introduction Summarizing spoken documents has been extensively studied over the past several years ',Penn,'num_Summarizing_1 nn_Summarizing_Introduction dep_spoken_Summarizing nsubjpass_studied_documents aux_studied_has auxpass_studied_been advmod_studied_extensively ccomp_spoken_studied prep_studied_over det_years_the amod_years_past amod_years_several pobj_over_years ' +o,5627,'1 Introduction Previous corpus-based sense disambiguation methods require substantial amounts of sense-tagged training data or aligned bilingual corpora ',Brown,'num_sense_1 amod_sense_Introduction amod_sense_Previous amod_sense_corpus-based nn_methods_disambiguation nsubj_require_methods dep_sense_require amod_amounts_substantial dobj_require_amounts prep_amounts_of amod_data_sense-tagged nn_data_training pobj_of_data partmod_data_ cc__or conj__aligned amod_corpora_bilingual dobj__corpora ' +o,5628,'Table 8 compares the F1 results of our baseline model with Nakagawa and Uchimoto and on CTB 30 ',Zhang,'nsubj_compares_Table num_Table_8 det_results_the nn_results_F1 dobj_compares_results prep_results_of poss_model_our nn_model_baseline pobj_of_model prep_compares_with nn__Nakagawa cc_Nakagawa_and conj_Nakagawa_Uchimoto pobj_with_ cc_compares_and conj_compares_ prep__on pobj_on_CTB num_CTB_30 ' +n,5629,'In the thriving area of research on automatic analysis and processing of product reviews , little attention has been paid to the important task studied here assessing review helpfulness ',Turney,'prep_paid_In det_area_the amod_area_thriving pobj_In_area prep_area_of pobj_of_research prep_research_on amod_analysis_automatic pobj_on_analysis cc_analysis_and conj_analysis_processing prep_research_of nn__product nn__reviews pobj_of_ amod_attention_little nsubjpass_paid_attention aux_paid_has auxpass_paid_been prep_paid_to det_task_the amod_task_important pobj_to_task partmod_task_studied advmod_studied_here xcomp_studied_assessing nn_helpfulness_review dobj_assessing_helpfulness ' +o,5630,'Section 4 concludes the paper with a critical assessment of the proposed approach and a discussion of the prospects for application in the construction of corpora comparable in size and quality to existing treebanks -LRB- such as , for example , the Penn Treebank for English or the TIGER Treebank for German -RRB- ',Marcus,'nsubj_concludes_Section num_Section_4 det_paper_the dobj_concludes_paper prep_concludes_with det_assessment_a amod_assessment_critical pobj_with_assessment prep_assessment_of det_approach_the amod_approach_proposed pobj_of_approach cc_approach_and det_discussion_a conj_approach_discussion prep_discussion_of det_prospects_the pobj_of_prospects prep_prospects_for pobj_for_application prep_application_in det_construction_the pobj_in_construction prep_construction_of pobj_of_corpora amod_discussion_comparable prep_comparable_in pobj_in_size cc_size_and conj_size_quality prep_comparable_to amod_treebanks_existing pobj_to_treebanks dep_as_such dep_discussion_as prep_Treebank_for pobj_for_example det_Treebank_the nn_Treebank_Penn pobj_as_Treebank prep_Treebank_for nn__English pobj_for_ cc_Treebank_or det_Treebank_the nn_Treebank_TIGER conj_Treebank_Treebank prep_Treebank_for amod__German pobj_for_ ' +o,5631,'For a more detailed introduction to maximum entropy estimation see ',Berger,'prep_see_For det_introduction_a advmod_detailed_more amod_introduction_detailed pobj_For_introduction prep_introduction_to amod_entropy_maximum pobj_to_entropy nsubj_see_estimation ' +o,5632,'34 -RRB- 31 Probabilistic model In the probabilistic formulation , the task of learning taxonomies from a corpus is seen as a probability maximization problem ',Snow,'dep_seen_34 num_model_31 nn_model_Probabilistic dep_In_model prep_seen_In det__the amod__probabilistic nn__formulation pobj_In_ det_task_the nsubjpass_seen_task prep_task_of pcomp_of_learning dobj_learning_taxonomies prep_learning_from det_corpus_a pobj_from_corpus auxpass_seen_is prep_seen_as det_problem_a nn_problem_probability nn_problem_maximization pobj_as_problem ' +o,5633,'We augment Collins head-driven model 2 to incorporate a semantic label on each internal node ',Collins,'nsubj_augment_We nn__Collins nn__head-driven nn__model num__2 dobj_augment_ aux_incorporate_to xcomp_augment_incorporate det_label_a amod_label_semantic dobj_incorporate_label prep_incorporate_on det_node_each amod_node_internal pobj_on_node ' +o,5634,'Such metrics have been introduced in other fields , including PARADISE for spoken dialogue systems , BLEU for machine translation ,1 and ROUGE for summarisation ',Papineni,'amod_metrics_Such nsubjpass_introduced_metrics aux_introduced_have auxpass_introduced_been prep_introduced_in amod_fields_other pobj_in_fields prep_introduced_including nn__PARADISE pobj_including_ prep__for amod_systems_spoken nn_systems_dialogue pobj_for_systems nn__BLEU conj__ prep__for nn_translation_machine pobj_for_translation num_translation_,1 cc__and nn__ROUGE conj__ prep__for pobj_for_summarisation ' +o,5635,'It was later applied by as a way to determine if a sequence of N words -LRB- Ngram -RRB- came from an independently distributed sample ',Dunning,'nsubjpass_applied_It auxpass_applied_was advmod_applied_later prep_applied_by pobj_by_ prep_applied_as det_way_a pobj_as_way aux_determine_to infmod_way_determine mark_came_if det_sequence_a nsubj_came_sequence prep_sequence_of nn_words_N pobj_of_words appos_sequence_Ngram advcl_determine_came prep_came_from det_sample_an amod_sample_independently amod_sample_distributed pobj_from_sample ' +o,5636,'Co-Training has been used before in applications like word-sense disambiguation , web-page classification and namedentity identification ',Yarowsky,'nsubjpass_used_Co-Training aux_used_has auxpass_used_been dep_in_before prep_used_in pobj_in_applications prep_applications_like amod__word-sense nn__disambiguation pobj_like_ amod__web-page nn__classification conj__ cc__and amod__namedentity nn__identification conj__ ' +o,5637,'The IBM models , together with a Hidden Markov Model -LRB- HMM -RRB- , form a class of generative models that are based on a lexical translation model P -LRB- fj ei -RRB- where each word fj in the foreign sentence fm1 is generated by precisely one word ei in the sentence el1 , independently of the other translation decisions ',Brown,'det_models_The nn_models_IBM nsubj_form_models rcmod_models_together dep_together_with det_Model_a nn_Model_Hidden nn_Model_Markov pobj_with_Model abbrev_Model_HMM det_class_a dobj_form_class prep_class_of amod_models_generative pobj_of_models nsubjpass_based_that auxpass_based_are rcmod_models_based prep_based_on det_P_a amod_P_lexical nn_P_translation nn_P_model pobj_on_P nn_ei_fj appos_P_ei advmod_generated_where det_fj_each nn_fj_word nsubjpass_generated_fj prep_fj_in det_fm1_the amod_fm1_foreign nn_fm1_sentence pobj_in_fm1 auxpass_generated_is rcmod_P_generated prep_generated_by advmod_ei_precisely nn_ei_one nn_ei_word pobj_by_ei prep_ei_in det_el1_the nn_el1_sentence pobj_in_el1 advmod_generated_independently prep_class_of det_decisions_the amod_decisions_other nn_decisions_translation pobj_of_decisions ' +o,5638,'Experiment Implementation : We apply SVM algorithm to construct our classifiers which has been shown to perform better than many other classification algorithms ',Pang,'nn_Implementation_Experiment nsubj_apply_We dep_Implementation_apply nn_algorithm_SVM nsubj_construct_algorithm aux_construct_to xcomp_apply_construct poss_classifiers_our dobj_construct_classifiers nsubjpass_shown_which aux_shown_has auxpass_shown_been rcmod_classifiers_shown aux_perform_to xcomp_shown_perform advmod_perform_better prep_perform_than amod_algorithms_many amod_algorithms_other nn_algorithms_classification pobj_than_algorithms ' +o,5639,'2 WordNet-based semantic relatedness measures 21 Basic measures Two similarity\\/distance measures from the Perl package WordNet-Similarity written by are used ',Pedersen,'num_relatedness_2 amod_relatedness_WordNet-based amod_relatedness_semantic nsubj_measures_relatedness num_measures_21 nn_measures_Basic dobj_measures_measures num_measures_Two amod_measures_similarity\\/distance nsubjpass_used_measures prep_measures_from det_WordNet-Similarity_the nn_WordNet-Similarity_Perl nn_WordNet-Similarity_package pobj_from_WordNet-Similarity partmod_WordNet-Similarity_written prep_written_by pobj_by_ auxpass_used_are dep_measures_used ' +o,5640,'The third exploits automatic subjectivity analysis in applications such as review classification -LRB- eg , -RRB- , mining texts for product reviews -LRB- eg , -RRB- , summarization -LRB- eg , -RRB- , information extraction -LRB- eg , -RRB- , 1Note that sentiment , the focus of much recent work in the area , is a type of subjectivity , specifically involving positive or negative opinion , emotion , or evaluation ',Pang,'det_exploits_The amod_exploits_third amod_analysis_automatic nn_analysis_subjectivity dep_mining_analysis prep_analysis_in pobj_in_applications dep_as_such prep_applications_as pobj_as_review dep_mining_classification appos_classification_eg dep_eg_ dep_exploits_mining dobj_mining_texts prep_texts_for nn_reviews_product pobj_for_reviews appos_reviews_eg dep_eg_ appos_reviews_summarization appos_summarization_eg dep_eg_ nn_extraction_information appos_reviews_extraction appos_extraction_eg dep_eg_ appos_reviews_1Note complm_type_that nsubj_type_sentiment det_focus_the appos_sentiment_focus prep_focus_of amod_work_much amod_work_recent pobj_of_work prep_work_in det_area_the pobj_in_area cop_type_is det_type_a dep_1Note_type prep_type_of pobj_of_subjectivity advmod_involving_specifically partmod_reviews_involving amod_opinion_positive cc_positive_or conj_positive_negative dobj_involving_opinion conj_reviews_emotion cc_reviews_or conj_reviews_evaluation ' +o,5641,' and Hogan et al ',Cahill,'cc__and conj__Hogan cc_Hogan_et conj_Hogan_al ' +o,5642,'For the correct identification of phrases in a Korean query , it would help to identify the lexical relations and produce statistical information on pairs of words in a text corpus as in ',Smadja,'prep_help_For det_identification_the amod_identification_correct pobj_For_identification prep_identification_of pobj_of_phrases prep_phrases_in det_query_a amod_query_Korean pobj_in_query nsubj_help_it aux_help_would aux_identify_to xcomp_help_identify det_relations_the amod_relations_lexical dobj_identify_relations cc_identify_and conj_identify_produce amod_information_statistical dobj_produce_information prep_produce_on pobj_on_pairs prep_pairs_of pobj_of_words prep_words_in det_corpus_a nn_corpus_text pobj_in_corpus prep_produce_as dep_as_in ' +o,5643,'For example , the sentence I went to California last May would be marked for base NPs as : I went to California last May I 0 0 I B I indicating that the NPs are I , California and last May This approach has been studied in ',Ramshaw,'prep_marked_For pobj_For_example det_sentence_the nsubjpass_marked_sentence nsubj_went_I rcmod_sentence_went prep_went_to pobj_to_California amod_May_last tmod_went_May aux_marked_would auxpass_marked_be prep_marked_for nn_NPs_base pobj_for_NPs prep_marked_as nsubj_went_I parataxis_marked_went prep_went_to pobj_to_California amod_May_last tmod_went_May num_May_I num_I_0 num_I_0 num_I_I nn_I_B dobj_went_I partmod_I_indicating complm_I_that det_NPs_the nsubj_I_NPs cop_I_are ccomp_indicating_I attr_I_California cc_marked_and amod_May_last tmod_studied_May det_approach_This nsubjpass_studied_approach aux_studied_has auxpass_studied_been conj_marked_studied prep_studied_in ' +o,5644,'In the context of statistical machine translation , we may interpretE as an English sentence , F its translation in French , and A a representation of how the words correspond to each other in the two sentences ',Brown,'prep_interpretE_In det_context_the pobj_In_context prep_context_of amod__statistical nn__machine nn__translation pobj_of_ nsubj_interpretE_we aux_interpretE_may prep_interpretE_as det_sentence_an amod_sentence_English pobj_as_sentence nn_translation_F poss_translation_its conj_sentence_translation prep_translation_in pobj_in_French cc_sentence_and det_representation_A dep_representation_a conj_sentence_representation prep_representation_of advmod_correspond_how det_words_the nsubj_correspond_words pcomp_of_correspond prep_correspond_to det_other_each pobj_to_other prep_other_in det_sentences_the num_sentences_two pobj_in_sentences ' +p,5645,'Many efficient techniques exist to extract multiword expressions , collocations , lexical units and idioms ',Smadja,'amod_techniques_Many amod_techniques_efficient nsubj_exist_techniques aux_extract_to xcomp_exist_extract amod_units_multiword nn_units_expressions conj_units_collocations amod_units_lexical dobj_extract_units cc_units_and conj_units_idioms ' +o,5646,'Among them , have proposed a way to exploit bilingual dictionnaries at training time ',Brown,'prep_proposed_Among pobj_Among_them nsubj_proposed_ aux_proposed_have det_way_a dobj_proposed_way aux_exploit_to infmod_way_exploit amod_dictionnaries_bilingual dobj_exploit_dictionnaries prep_exploit_at nn_time_training pobj_at_time ' +o,5647,'While recent proposals for evaluation of MT systems have involved multi-parallel corpora , statistical MT algorithms typically only use one-parallel data ',Papineni,'mark_involved_While amod_proposals_recent nsubj_involved_proposals prep_proposals_for pobj_for_evaluation prep_evaluation_of nn_systems_MT pobj_of_systems aux_involved_have advcl_use_involved amod__multi-parallel nn__corpora dobj_involved_ amod_algorithms_statistical nn_algorithms_MT nsubj_use_algorithms advmod_use_typically advmod_use_only amod_data_one-parallel dobj_use_data ' +o,5648,'Of course , many applications require smoothing of the estimated distributionsthis problem also has known solutions in MapReduce ',Brants,'prep_require_Of pobj_Of_course amod_applications_many nsubj_require_applications nsubj_known_smoothing prep_smoothing_of det_problem_the amod_problem_estimated nn_problem_distributionsthis pobj_of_problem advmod_known_also aux_known_has ccomp_require_known dobj_known_solutions prep_solutions_in pobj_in_MapReduce ' +o,5649,'203 Estimating the parameters for these models is more difficult (and more computationally expensive) than with the models considered in the previous section: rather than simply being able to count the word pairs and alignment relationships and estimate the models directly, we must use an existing model to compute the expected counts for all possible alignments, and then use these counts to update the new model.7 This training strategy is referred to as expectationmaximization (EM) and is guaranteed to always improve the quality of the prior model at each iteration (Brown et al., 1993; Dempster et al., 1977).',Brown,'' +o,5650,'The best prosodic label sequence is then , L = argmax L nproductdisplay i P -LRB- li -RRB- -LRB- 6 -RRB- To estimate the conditional distribution P -LRB- li -RRB- we use the general technique of choosing the maximum entropy -LRB- maxent -RRB- distribution that estimates the average of each feature over the training data ',Berger,'det_sequence_The dep_prosodic_best amod_sequence_prosodic nn_sequence_label nsubj_is_sequence advmod_is_then nsubj_estimate_L dep_L_= nn_L_argmax iobj_=_L amod_P_nproductdisplay amod_P_i dobj_=_P appos_P_li appos_P_6 aux_estimate_To xcomp_is_estimate det_P_the amod_P_conditional nn_P_distribution dobj_estimate_P appos_P_li nsubj_use_we dep_is_use det_technique_the amod_technique_general dobj_use_technique prep_technique_of pcomp_of_choosing det_distribution_the amod_distribution_maximum nn_distribution_entropy appos_distribution_maxent dobj_choosing_distribution nsubj_estimates_that rcmod_technique_estimates det_average_the dobj_estimates_average prep_average_of det_feature_each pobj_of_feature prep_feature_over det_data_the nn_data_training pobj_over_data ' +o,5651,'Other scores for the word arc are set as in ',Rosti,'amod_scores_Other nsubjpass_set_scores prep_scores_for det_arc_the nn_arc_word pobj_for_arc auxpass_set_are prep_set_as dep_as_in ' +n,5652,'In such tasks , feature calculation is also very expensive in terms of time required ; huge sets of extracted rules must be sorted in two directions for relative frequency calculation of such features as the translation probability p -LRB- f e -RRB- and reverse translation probability p -LRB- e f -RRB- ',Koehn,'prep_expensive_In amod_tasks_such pobj_In_tasks nn_calculation_feature nsubj_expensive_calculation cop_expensive_is advmod_expensive_also advmod_expensive_very prep_expensive_in pobj_in_terms prep_terms_of pobj_of_time partmod_time_required amod_sets_huge nsubjpass_sorted_sets prep_sets_of amod_rules_extracted pobj_of_rules aux_sorted_must auxpass_sorted_be parataxis_expensive_sorted prep_sorted_in num_directions_two pobj_in_directions prep_directions_for amod_calculation_relative nn_calculation_frequency pobj_for_calculation prep_calculation_of amod_features_such pobj_of_features prep_features_as det_p_the nn_p_translation nn_p_probability pobj_as_p dep_-LRB-_f dep_f_e cc_sorted_and conj_sorted_reverse nn_p_translation nn_p_probability dobj_reverse_p dep_f_e dep_-LRB-_f ' +o,5653,'Model interpolation in this case perSystem Training Heldout LR LP MAP Brown ; T Brown ; H 760 754 MAP Brown ; T WSJ ; 24 769 771 Gildea WSJ ; 2-21 861 866 MAP WSJ ; 2-21 WSJ ; 24 869 871 Charniak WSJ ; 2-21 WSJ ; 24 867 866 Ratnaparkhi WSJ ; 2-21 863 875 WSJ ; 2-21 881 883 Charniak -LRB- 2000 -RRB- WSJ ; 2-21 WSJ ; 24 896 895 -LRB- 2000 -RRB- WSJ ; 2-21 896 899 Table 4 : Parser performance on WSJ ; 23 , baselines ',Collins,'dep_Model_interpolation prep_interpolation_in det_Brown_this nn_Brown_case nn_Brown_perSystem nn_Brown_Training nn_Brown_Heldout nn_Brown_LR nn_Brown_LP nn_Brown_MAP pobj_in_Brown dep_interpolation_T dep_T_Brown dep_interpolation_H num_Brown_760 num_Brown_754 nn_Brown_MAP dep_H_Brown dep_interpolation_T dep_T_WSJ number_769_24 dep_interpolation_769 num_WSJ_771 nn_WSJ_Gildea dep_769_WSJ num_WSJ_2-21 num_WSJ_861 num_WSJ_866 nn_WSJ_MAP dep_interpolation_WSJ nn_WSJ_2-21 dep_WSJ_WSJ num_869_24 dep_interpolation_869 nn_WSJ_871 nn_WSJ_Charniak nn_WSJ_ dep_869_WSJ dep_interpolation_2-21 dep_2-21_WSJ number_867_24 dep_interpolation_867 nn_WSJ_866 nn_WSJ_Ratnaparkhi nn_WSJ_ dep_867_WSJ number_863_2-21 dep_interpolation_863 nn_WSJ_875 nn_WSJ_ nn_WSJ_ dep_863_WSJ num_881_2-21 dep_interpolation_881 nn_WSJ_883 nn_WSJ_Charniak appos_WSJ_2000 dep_881_WSJ dep_interpolation_2-21 dep_2-21_WSJ num_896_24 dep_interpolation_896 nn_WSJ_895 nn_WSJ_ appos_WSJ_2000 dep_896_WSJ number_896_2-21 dep_interpolation_896 num_Table_899 dep_896_Table num_Table_4 nn_performance_Parser dep_interpolation_performance prep_performance_on pobj_on_WSJ dep_interpolation_23 appos_23_baselines ' +o,5654,'The learning algorithm , which is illustrated in , proceeds as follows ',Collins,'det_algorithm_The nn_algorithm_learning nsubj_proceeds_algorithm nsubjpass_illustrated_which auxpass_illustrated_is rcmod_algorithm_illustrated prep_illustrated_in pobj_in_ mark_follows_as advcl_proceeds_follows ' +o,5655,'POS tag the text using ',Ratnaparkhi,'nsubj_tag_POS det_text_the dobj_tag_text partmod_text_using ' +p,5656,'1 Introduction The maximum entropy model has attained great popularity in the NLP field due to its power , robustness , and successful performance in various NLP tasks ',Berger,'num_Introduction_1 det__The amod__maximum amod__entropy nn__model nsubj_attained_ aux_attained_has rcmod_Introduction_attained amod_popularity_great dobj_attained_popularity prep_attained_in det_field_the nn_field_NLP pobj_in_field amod__due prep_due_to poss_power_its pobj_to_power conj_power_robustness cc_power_and amod_performance_successful conj_power_performance prep_performance_in amod_tasks_various nn_tasks_NLP pobj_in_tasks dep_Introduction_ ' +o,5657,'One is to use a stochastic gradient descent -LRB- SGD -RRB- or Perceptron like online learning algorithm to optimize the weights of these features directly for MT ',Tillmann,'nsubj_is_One aux_use_to xcomp_is_use det_descent_a amod_descent_stochastic nn_descent_gradient dobj_use_descent abbrev_descent_SGD cc_descent_or conj_descent_Perceptron prep_Perceptron_like amod_algorithm_online nn_algorithm_learning pobj_like_algorithm aux_optimize_to xcomp_use_optimize det_weights_the dobj_optimize_weights prep_weights_of det_features_these pobj_of_features dep_for_directly prep_optimize_for pobj_for_MT ' +p,5658,'This was recently followed by who introduce state-of-the-art nearly unlexicalized PCFG parsers ',Matsuzaki,'nsubjpass_followed_This auxpass_followed_was advmod_followed_recently prep_followed_by pobj_by_ nsubj_introduce_who rcmod__introduce amod_parsers_state-of-the-art advmod_unlexicalized_nearly amod_parsers_unlexicalized nn_parsers_PCFG dobj_introduce_parsers ' +o,5659,'Second , phrase translation pairs are extracted from the word aligned corpus ',Koehn,'advmod_extracted_Second nn_pairs_phrase nn_pairs_translation nsubjpass_extracted_pairs auxpass_extracted_are prep_extracted_from det_word_the pobj_from_word partmod_word_aligned dobj_aligned_corpus ' +o,5660,'62 Translation Results For the translation experiments , we report the two accuracy measures BLEU and NIST as well as the two error rates word error rate -LRB- WER -RRB- and positionindependent word error rate -LRB- PER -RRB- ',Papineni,'num_Results_62 nn_Results_Translation prep_report_For det_experiments_the nn_experiments_translation pobj_For_experiments nsubj_report_we dep_Results_report det_rate_the num__two nn__accuracy nn__measures nn__BLEU dep_rate_ cc__and nn__NIST conj__ dep_well_as cc_rate_well dep_well_as det_rates_the num_rates_two nn_rates_error conj_rate_rates dep_rate_word dep_rate_error nn_rate_rate dep_rate_WER cc_rate_and amod_word_positionindependent conj_rate_word nn_rate_error dobj_report_rate abbrev_rate_PER ' +p,5661,'Recent comparisons of approaches that can be trained on corpora have shown that in most cases statistical aproaches yield better results than finite-state , rule-based , or memory-based taggers ',Ratnaparkhi,'amod_comparisons_Recent nsubj_shown_comparisons prep_comparisons_of pobj_of_approaches nsubjpass_trained_that aux_trained_can auxpass_trained_be rcmod_approaches_trained prep_trained_on nn__corpora pobj_on_ aux_shown_have complm__that prep__in amod_cases_most pobj_in_cases amod_aproaches_statistical nsubj__aproaches ccomp_shown_ xcomp__yield amod_results_better dobj_yield_results prep_yield_than amod__finite-state conj_finite-state_rule-based cc_finite-state_or conj_finite-state_memory-based dep_finite-state_taggers pobj_than_ ' +p,5662,'Nowadays , most state-of-the-art SMT systems are based on bilingual phrases ',Koehn,'advmod_based_Nowadays advmod_systems_most amod_systems_state-of-the-art nn_systems_SMT nsubjpass_based_systems auxpass_based_are prep_based_on amod_phrases_bilingual pobj_on_phrases ' +o,5663,'1 Introduction Statistical language modeling has been widely used in natural language processing applications such as Automatic Speech Recognition -LRB- ASR -RRB- , Statistical Machine Translation -LRB- SMT -RRB- and Information Retrieval -LRB- IR -RRB- ',Brown,'num_modeling_1 nn_modeling_Introduction nn_modeling_Statistical nn_modeling_language nsubjpass_used_modeling aux_used_has auxpass_used_been advmod_used_widely prep_used_in amod_applications_natural nn_applications_language nn_applications_processing pobj_in_applications dep_as_such prep_applications_as nn_Recognition_Automatic nn_Recognition_Speech pobj_as_Recognition abbrev_Recognition_ASR nn__Statistical nn__Machine nn__Translation abbrev__SMT conj_Recognition_ cc_Recognition_and nn__Information nn__Retrieval abbrev__IR conj_Recognition_ ' +p,5664,'52 Results We use a Maximum Entropy -LRB- ME -RRB- classi er which allows an e cient combination of many overlapping features ',Manning,'num_Results_52 nsubj_use_We dep_Results_use det__a nn_Entropy_Maximum nn__Entropy abbrev_Entropy_ME nn__classi nn__er dobj_use_ nsubj_allows_which rcmod__allows det_combination_an amod_combination_e nn_combination_cient dobj_allows_combination prep_combination_of amod_features_many amod_features_overlapping pobj_of_features ' +o,5665,'22 The Translation Model We adapted Model 1 to our purposes ',Brown,'det_Translation_The nsubj_Model_Translation dep_22_Model nsubj_adapted_We ccomp_Model_adapted nn__Model num__1 dobj_adapted_ prep_adapted_to poss_purposes_our pobj_to_purposes ' +o,5666,'These lists are rescored with the following models : -LRB- a -RRB- the different models used in the decoder which are described above , -LRB- b -RRB- two different features based on IBM Model 1 , -LRB- c -RRB- posterior probabilities for words , phrases , n-grams , and sentence length , all calculated over the Nbest list and using the sentence probabilities which the baseline system assigns to the translation hypotheses ',Brown,'det_lists_These nsubjpass_rescored_lists auxpass_rescored_are prep_rescored_with det_models_the amod_models_following pobj_with_models dep_-RRB-_a det_models_the amod_models_different nsubj_hypotheses_models partmod_models_used prep_used_in det_decoder_the pobj_in_decoder nsubjpass_described_which auxpass_described_are rcmod_decoder_described prep_described_above dep_-RRB-_b num_features_two amod_features_different appos_models_features prep_features_based dep_based_on nn__IBM nn__Model num__1 pobj_on_ dep_-RRB-_c nn_probabilities_posterior appos__probabilities prep_probabilities_for pobj_for_words conj_words_phrases conj_words_n-grams cc_words_and nn__sentence nn__length conj_words_ nsubj_calculated_all rcmod_models_calculated prep_calculated_over det_list_the nn_list_Nbest pobj_over_list cc_calculated_and conj_calculated_using det_probabilities_the nn_probabilities_sentence dobj_using_probabilities rel_assigns_which det_system_the nn_system_baseline nsubj_assigns_system rcmod_probabilities_assigns prep_assigns_to det_translation_the pobj_to_translation parataxis_rescored_hypotheses ' +o,5667,'The tensor has been adapted with a straightforward extension of pointwise mutual information for three-way cooccurrences , following equation 4 ',Church,'det_tensor_The nsubjpass_adapted_tensor aux_adapted_has auxpass_adapted_been prep_adapted_with det_extension_a amod_extension_straightforward pobj_with_extension prep_extension_of amod__pointwise amod__mutual nn__information pobj_of_ prep__for amod_cooccurrences_three-way pobj_for_cooccurrences prep_extension_following pobj_following_equation num_equation_4 ' +o,5668,'1 proposes a method for word sense -LRB- translation -RRB- disambiguation that is based on a bootstrapping technique , which we refer to here as Monolingual Bootstrapping -LRB- MB -RRB- ',Yarowsky,'num__1 nsubj_proposes_ det_method_a dobj_proposes_method prep_method_for nn_sense_word pobj_for_sense appos_method_translation dep_method_disambiguation nsubjpass_based_that auxpass_based_is rcmod_disambiguation_based prep_based_on det_technique_a amod_technique_bootstrapping pobj_on_technique rel_refer_which nsubj_refer_we rcmod_technique_refer prep_refer_to pobj_to_here prep_refer_as nn_Bootstrapping_Monolingual pobj_as_Bootstrapping abbrev_Bootstrapping_MB ' +p,5669,'1 Introduction Syntactic methods are an increasingly promising approach to statistical machine translation , being both algorithmically appealing and empirically successful ',Wu,'num_methods_1 nn_methods_Introduction nn_methods_Syntactic cop_approach_are det_approach_an advmod_approach_increasingly amod_approach_promising rcmod_methods_approach prep_approach_to amod_translation_statistical nn_translation_machine pobj_to_translation cop_appealing_being dep_appealing_both advmod_appealing_algorithmically xcomp_approach_appealing amod__ cc__and advmod_successful_empirically conj__successful dep_methods_ ' +o,5670,'Starting from a word-based alignment for each pair of sentences , the training for the algorithm accepts all contiguous bilingual phrase pairs -LRB- up to a predetermined maximum length -RRB- whose words are only aligned with each other ',Koehn,'prep_accepts_Starting dep_Starting_from det_alignment_a amod_alignment_word-based pobj_from_alignment prep_alignment_for det_pair_each pobj_for_pair prep_pair_of pobj_of_sentences det_training_the appos_sentences_training prep_training_for det_algorithm_the pobj_for_algorithm nsubj_contiguous_all xcomp_accepts_contiguous amod_pairs_bilingual nn_pairs_phrase nsubj_accepts_pairs dep_pairs_up dep_up_to det_length_a amod_length_predetermined amod_length_maximum pobj_to_length poss_aligned_whose nsubjpass_aligned_words auxpass_aligned_are advmod_aligned_only dep_pairs_aligned prep_aligned_with det_other_each pobj_with_other ' +o,5671,'HMMs have been used many times for POS tagging and chunking , in supervised , semisupervised , and in unsupervised settings ',Johnson,'nsubjpass_used_HMMs aux_used_have auxpass_used_been quantmod_times_many dobj_used_times prep_times_for pobj_for_POS amod_times_tagging cc_tagging_and conj_tagging_chunking prep_times_in pobj_in_supervised conj_supervised_semisupervised cc_supervised_and conj_supervised_in amod_settings_unsupervised pobj_in_settings ' +o,5672,'Collocations were extracted according to the method described in by moving a window on texts ',Church,'nsubjpass_extracted_Collocations auxpass_extracted_were prep_extracted_according dep_according_to det_method_the pobj_to_method partmod_method_described prep_described_in pobj_in_ prep_described_by pcomp_by_moving det_window_a dobj_moving_window prep_moving_on pobj_on_texts ' +o,5673,'Such approaches have shown promise in applications such as web page classification , named entity classification , parsing , and machine translation ',McClosky,'amod_approaches_Such nsubj_shown_approaches aux_shown_have dobj_shown_promise prep_shown_in pobj_in_applications dep_as_such prep_applications_as nn__web nn__page nn__classification pobj_as_ partmod__named nn__entity nn__classification dobj_named_ amod__parsing conj__ cc__and nn__machine nn__translation conj__ ' +o,5674,'The techniques examined are Structural Correspondence Learning -LRB- SCL -RRB- and Self-training ',McClosky,'det_techniques_The nsubj_Structural_techniques partmod_techniques_examined aux_Structural_are nn_Learning_Correspondence dobj_Structural_Learning abbrev_Learning_SCL amod_Learning_ cc__and conj__Self-training ' +o,5675,'Also , adding a constituent size\\/distance effect , as described by Schubert and as used by some researchers in parsing -LRB- eg Lesmo and Torasso and -RRB- would almost certainly improve parsing ',Collins,'advmod_improve_Also dep_improve_adding det_effect_a amod_effect_constituent nn_effect_size\\/distance dobj_adding_effect advmod_described_as amod_researchers_described prep_described_by nn__Schubert pobj_by_ cc_described_and advmod_used_as conj_described_used prep_used_by pobj_by_some nsubj_improve_researchers prep_researchers_in pobj_in_parsing nn_Lesmo_eg dep_parsing_Lesmo cc_Lesmo_and nn__Torasso conj_Lesmo_ cc__and conj__ aux_improve_would advmod_certainly_almost advmod_improve_certainly dobj_improve_parsing ' +n,5676,'Other statistical systems that address word classification probleans do not emphasize the use of linguistic knowledge and do not deal with a specific word class , or do not exploit as much linguistic knowledge as we do ',Brown,'amod_systems_Other amod_systems_statistical nsubj__systems complm_emphasize_that nn_probleans_address nn_probleans_word nn_probleans_classification nsubj_emphasize_probleans aux_emphasize_do neg_emphasize_not dep_systems_emphasize det_use_the dobj_emphasize_use prep_use_of amod_knowledge_linguistic pobj_of_knowledge cc_emphasize_and aux_deal_do neg_deal_not conj_emphasize_deal prep_deal_with det_class_a amod_class_specific nn_class_word pobj_with_class cc__or aux_exploit_do neg_exploit_not conj__exploit prep_exploit_as amod_knowledge_much amod_knowledge_linguistic pobj_as_knowledge mark__as nsubj__we dep__do advcl_exploit_ ' +o,5677,'One such model is the IBM Model 1 ',Brown,'num_model_One amod_model_such nsubj_Model_model cop_Model_is det_Model_the nn_Model_IBM dep_Model_1 ' +o,5678,'Forest reranking with a language model can be performed over this n-ary forest using the cube growing algorithm of ',Huang,'nsubjpass_performed_Forest partmod_Forest_reranking prep_reranking_with det_model_a nn_model_language pobj_with_model aux_performed_can auxpass_performed_be prep_performed_over det_forest_this amod_forest_n-ary pobj_over_forest partmod_forest_using det_algorithm_the nn_algorithm_cube amod_algorithm_growing dobj_using_algorithm prep_algorithm_of ' +o,5679,'First , we show how one can use an existing statistical translation model in order to automatically derive a statistical TMEM ',Brown,'advmod_show_First nsubj_show_we advmod_use_how nsubj_use_one aux_use_can ccomp_show_use det__an amod__existing amod__statistical nn__translation nn__model dobj_use_ mark_derive_in dep_derive_order aux_derive_to advmod_derive_automatically purpcl_use_derive det_TMEM_a amod_TMEM_statistical dobj_derive_TMEM ' +o,5680,'We set our space usage to match the 308 bytes per n-gram reported in and held out just over 1M unseen n-grams to test the error rates of our models ',Talbot,'nsubj_set_We poss_usage_our nn_usage_space dobj_set_usage aux_match_to xcomp_set_match det_bytes_the num_bytes_308 dobj_match_bytes prep_bytes_per pobj_per_n-gram dep_n-gram_reported prep_reported_in pobj_in_ cc_reported_and conj_reported_held prep_held_out advmod_over_just quantmod_unseen_over number_unseen_1M num_n-grams_unseen pobj_out_n-grams aux_test_to xcomp_held_test det_rates_the nn_rates_error dobj_test_rates prep_rates_of poss_models_our pobj_of_models ' +o,5681,'These results were achieved using the statistical alignments provided by model 5 and smoothed 11-grams and 6-grams , respectively ',Brown,'det_results_These nsubjpass_achieved_results auxpass_achieved_were xcomp_achieved_using det_alignments_the amod_alignments_statistical dobj_using_alignments prep_using_provided dep_provided_by pobj_by_model num_model_5 partmod_model_ cc__and conj__smoothed dobj__11-grams cc_11-grams_and conj_11-grams_6-grams advmod__respectively ' +o,5682,'157 ena or the linguist \'s abstraction capabilities -LRB- eg knowledge about what is relevant in the context -RRB- , they tend to reach a 95-97 \% accuracy in the analysis of several languages , in particular English ',Cutting,'num_ena_157 cc_ena_or det_linguist_the poss_capabilities_linguist possessive_linguist_\'s nn_capabilities_abstraction conj_ena_capabilities nn_knowledge_eg dep_capabilities_knowledge prep_knowledge_about nsubj_relevant_what cop_relevant_is pcomp_about_relevant prep_relevant_in det_context_the pobj_in_context nsubj_tend_they dep_ena_tend aux_reach_to xcomp_tend_reach det_accuracy_a num_accuracy_95-97 nn_accuracy_\% dobj_reach_accuracy prep_accuracy_in det_analysis_the pobj_in_analysis prep_analysis_of amod_languages_several pobj_of_languages prep_ena_in amod_English_particular pobj_in_English ' +o,5683,'This score measures the precision of unigrams , bigrams , trigrams and fourgrams with respect to a reference translation with a penalty for too short sentences ',Papineni,'det_score_This nsubj_measures_score det_precision_the dobj_measures_precision prep_precision_of pobj_of_unigrams conj_unigrams_bigrams conj_unigrams_trigrams cc_unigrams_and conj_unigrams_fourgrams prep_measures_with pobj_with_respect prep_measures_to det_translation_a nn_translation_reference pobj_to_translation prep_translation_with det_penalty_a pobj_with_penalty prep_penalty_for advmod_short_too amod_sentences_short pobj_for_sentences ' +p,5684,'3 Perceptron Reranking As observes , perceptron training involves a simple , on-line algorithm , with few iterations typically required to achieve good performance ',Collins,'num_Reranking_3 nn_Reranking_Perceptron mark_observes_As nsubj_observes_ advcl_involves_observes nn_training_perceptron nsubj_involves_training dep_Reranking_involves det_algorithm_a amod_algorithm_simple amod_algorithm_on-line dobj_involves_algorithm prep_involves_with amod_iterations_few pobj_with_iterations advmod_required_typically partmod_iterations_required aux_achieve_to xcomp_required_achieve amod_performance_good dobj_achieve_performance ' +o,5685,'It acquires a set of synchronous lexical entries by running the IBM alignment model and learns a log-linear model to weight parses ',Brown,'nsubj_acquires_It det_set_a dobj_acquires_set prep_set_of amod_entries_synchronous amod_entries_lexical pobj_of_entries prep_acquires_by pcomp_by_running det_model_the nn_model_IBM nn_model_alignment nsubj__model ccomp_running_ cc__and conj__learns det_model_a amod_model_log-linear dobj__model prep__to nn_parses_weight pobj_to_parses ' +o,5686,'Most related to our approach , used inversion transduction grammarsa synchronous context-free formalism for this task ',Wu,'advmod_related_Most dep_used_related prep_related_to poss_approach_our pobj_to_approach advmod_used_ amod_grammarsa_inversion nn_grammarsa_transduction dobj_used_grammarsa amod__synchronous amod__context-free nn__formalism nsubj_used_ prep__for det_task_this pobj_for_task ' +o,5687,'In this paper we extend this work to represent sets of situation-specific events not unlike scripts , caseframes , and FrameNet frames ',Bean,'prep_extend_In det_paper_this pobj_In_paper nsubj_extend_we det_work_this nsubj_represent_work aux_represent_to xcomp_extend_represent dobj_represent_sets prep_sets_of amod_events_situation-specific pobj_of_events dep_unlike_not prep_represent_unlike pobj_unlike_scripts amod__caseframes conj_scripts_ cc_scripts_and nn__FrameNet nn__frames conj_scripts_ ' +o,5688,'Previous research in this area includes several models which incorporate hidden variables ',Koo,'amod_research_Previous nsubj_includes_research prep_research_in det_area_this pobj_in_area amod_models_several dobj_includes_models nsubj_incorporate_which rcmod_models_incorporate amod_variables_hidden dobj_incorporate_variables ' +o,5689,'In order to build models that perform well in new -LRB- target -RRB- domains we usually find two settings : In the semi-supervised setting the goal is to improve the system trained on the source domain using unlabeled data from the target domain , and the baseline is that of the system c2008 ','Daume III','mark_build_In dep_build_order aux_build_to dep_find_build dobj_build_models nsubj_perform_that rcmod_models_perform advmod_perform_well prep_perform_in amod_domains_new appos_domains_target pobj_in_domains nsubj_find_we advmod_find_usually num_settings_two measure__settings acomp_find_ prep_is_In det_setting_the amod_setting_semi-supervised pobj_In_setting det_goal_the nsubj_is_goal parataxis_find_is aux_improve_to xcomp_is_improve det_system_the dobj_improve_system partmod_system_trained prep_trained_on det_domain_the nn_domain_source pobj_on_domain xcomp_trained_using amod_data_unlabeled dobj_using_data prep_using_from det_domain_the nn_domain_target pobj_from_domain cc_find_and det_baseline_the nsubj_that_baseline cop_that_is conj_find_that prep_that_of det_c2008_the nn_c2008_system pobj_of_c2008 ' +o,5690,'Based on this assumption , stored all bigrams of words along with their relative position , p -LRB- -5 -LRB- p _ ~ 5 -RRB- ',Smadja,'prep_stored_Based dep_Based_on det_assumption_this pobj_on_assumption nsubj_stored_ det_bigrams_all dobj_stored_bigrams prep_bigrams_of pobj_of_words advmod_words_along dep_along_with poss_p_their amod_p_relative nn_p_position pobj_with_p dep_p_-5 nn_~_p amod_~__ appos_p_~ num_~_5 ' +o,5691,' used this method for word sense disambiguation ',Yarowsky,'nsubj_used_ det_method_this dobj_used_method prep_used_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +o,5692,'Generation of paraphrase examples was also investigated ',Barzilay,'nsubjpass_investigated_Generation prep_Generation_of amod_examples_paraphrase pobj_of_examples auxpass_investigated_was advmod_investigated_also ' +o,5693,' -RRB- and on speech repair detection and correction -LRB- eg ',Brown,'dobj__and prep__on nn_detection_speech nn_detection_repair pobj_on_detection cc_detection_and nn_eg_correction conj_detection_eg ' +o,5694,'As in other work , we collapsed AI -RRB- VP and Pl ? Jl \' to the same label when calculating these scores ',Collins,'prep_collapsed_As dep_As_in amod_work_other pobj_in_work nsubj_collapsed_we ccomp_collapsed_AI dobj_AI_VP cc_VP_and nn_Jl_Pl conj_VP_Jl prep_AI_to det_label_the amod_label_same pobj_to_label advmod_calculating_when advcl_AI_calculating det_scores_these dobj_calculating_scores ' +o,5695,' and claim that fine-grained semantic distinctions are unlikely to be of practical value for many applications ',Dolan,'nsubj_claim_ cc__and conj__ complm_unlikely_that amod_distinctions_fine-grained amod_distinctions_semantic nsubj_unlikely_distinctions cop_unlikely_are ccomp_claim_unlikely aux_be_to xcomp_unlikely_be prep_be_of amod_value_practical pobj_of_value prep_value_for amod_applications_many pobj_for_applications ' +o,5696,'ther background on this method of hypothesis testing the reader is referred to ',Dunning,'nn_background_ther nsubjpass_referred_background prep_background_on det_method_this pobj_on_method prep_method_of pobj_of_hypothesis partmod_hypothesis_testing det_reader_the dobj_testing_reader auxpass_referred_is xcomp_referred_to ' +p,5697,'In order to estimate the conditional distributions shown in Table 1 , we use the general technique of choosing the MaxEnt distribution that properly estimates the average of each feature over the training data ',Berger,'mark_estimate_In dep_estimate_order aux_estimate_to dep_use_estimate det_distributions_the amod_distributions_conditional dobj_estimate_distributions partmod_distributions_shown prep_shown_in pobj_in_Table num_Table_1 nsubj_use_we det_technique_the amod_technique_general dobj_use_technique prep_technique_of pcomp_of_choosing det_distribution_the nn_distribution_MaxEnt dobj_choosing_distribution nsubj_estimates_that advmod_estimates_properly rcmod_distribution_estimates det_average_the dobj_estimates_average prep_average_of det_feature_each pobj_of_feature prep_feature_over det_data_the nn_data_training pobj_over_data ' +p,5698,'41 Features We used a dependency structure as the context for words because it is the most widely used and one of the best performing contextual information in the past studies ',Lin,'num_Features_41 nsubj_structure_We dep_structure_used det_structure_a amod_structure_dependency dep_Features_structure prep_structure_as det_context_the pobj_as_context prep_context_for pobj_for_words mark_used_because nsubj_used_it cop_used_is dep_used_the advmod_used_most advmod_used_widely advcl_structure_used cc_used_and conj_used_one prep_one_of det_information_the dep_performing_best amod_information_performing amod_information_contextual pobj_of_information prep_information_in det_studies_the amod_studies_past pobj_in_studies ' +o,5699,'Our method is based on a decision list proposed by Yarowsky ',Yarowsky,'poss_method_Our nsubjpass_based_method auxpass_based_is prep_based_on det_list_a nn_list_decision pobj_on_list partmod_list_proposed prep_proposed_by pobj_by_Yarowsky ' +o,5700,'As in much recent empirical work in discourse processing , we performed an intercoder reliability study investigating agreement in annotating the times ',Carletta,'prep_performed_As dep_As_in advmod_work_much amod_work_recent amod_work_empirical pobj_in_work prep_work_in nn__discourse nn__processing pobj_in_ nsubj_performed_we det_study_an nn_study_intercoder nn_study_reliability dobj_performed_study partmod_study_investigating dobj_investigating_agreement prep_investigating_in pcomp_in_annotating det_times_the dobj_annotating_times ' +o,5701,'Different optimization techniques are available , like the Simplex algorithm or the special Minimum Error Training as described in ',Och,'amod_techniques_Different nn_techniques_optimization nsubj_available_techniques cop_available_are prep_available_like det_algorithm_the nn_algorithm_Simplex pobj_like_algorithm cc_algorithm_or det_Training_the amod_Training_special nn_Training_Minimum nn_Training_Error conj_algorithm_Training mark_described_as ccomp_available_described prt_described_in ' +o,5702,'41 Experimental Setup We use the whole Penn Treebank corpus as our data set ',Marcus,'num_Setup_41 nn_Setup_Experimental nsubj_use_We dep_Setup_use det__the amod__whole nn__Penn nn__Treebank nn__corpus dobj_use_ mark_set_as poss_data_our nsubj_set_data advcl_use_set ' +o,5703,' , -LRB- 3 -RRB- thesaurus categories , -LRB- 4 -RRB- translation in another language , -LRB- 5 -RRB- automatically induced clusters with sublexical representation , and -LRB- 6 -RRB- hand-crafted lexicons ',Yarowsky,'advmod__ dep__3 nn_categories_thesaurus nsubj__categories ccomp_automatically_ dep__4 nsubj__translation prep_translation_in det_language_another pobj_in_language dep__ dep_automatically_5 amod_clusters_induced dobj_automatically_clusters prep_clusters_with amod__sublexical nn__representation pobj_with_ cc_clusters_and appos_clusters_6 amod_lexicons_hand-crafted conj_clusters_lexicons nsubj_automatically_ ' +o,5704,'Following the phrase extraction phase in PHARAOH , we eliminate word gaps by incorporating unaligned words as part of the extracted NL phrases ',Koehn,'prep_eliminate_Following det_phase_the nn_phase_phrase nn_phase_extraction pobj_Following_phase prep_phase_in pobj_in_PHARAOH nsubj_eliminate_we nn_gaps_word dobj_eliminate_gaps prep_eliminate_by pcomp_by_incorporating amod_words_unaligned dobj_incorporating_words prep_incorporating_as pobj_as_part prep_part_of det_phrases_the nn_phrases_extracted nn_phrases_NL pobj_of_phrases ' +o,5705,'Many probabilistic evaluation models have been published inspired by one or more of these feature types , but discrepancies between training sets , algorithms , and hardware environments make it difficult , if not impossible , to compare the models objectively ',Collins,'amod_models_Many amod_models_probabilistic nn_models_evaluation nsubjpass_published_models aux_published_have auxpass_published_been ccomp_published_inspired prep_inspired_by pobj_by_one cc_one_or conj_one_more prep_one_of det__these nn__feature nn__types nn__ nn__ nn__ nn__ nn__ nn__ pobj_of_ cc_published_but nsubj_sets_discrepancies prep_discrepancies_between pobj_between_training conj_published_sets dep_sets_algorithms cc_published_and nn_environments_hardware nsubj_make_environments conj_published_make nsubj_difficult_it xcomp_make_difficult dep_not_if cc_difficult_not conj_difficult_impossible aux_compare_to ccomp_difficult_compare det_objectively_the nn_objectively_models dobj_compare_objectively ' +o,5706,' proposed a Perceptron like learning algorithm to solve sequence classification in the traditional left-to-right order ',Collins,'nsubj_proposed_ det_Perceptron_a dobj_proposed_Perceptron prep_proposed_like pcomp_like_learning dobj_learning_algorithm aux_solve_to xcomp_learning_solve nn_classification_sequence dobj_solve_classification prep_solve_in det_order_the amod_order_traditional amod_order_left-to-right pobj_in_order ' +o,5707,'Recently , many works combined a MRD and a corpus for word sense disambiguatio ',Yarowsky,'advmod_combined_Recently amod_works_many nsubj_combined_works det_MRD_a dobj_combined_MRD cc_MRD_and det_corpus_a conj_MRD_corpus prep_combined_for nn_disambiguatio_word nn_disambiguatio_sense pobj_for_disambiguatio ' +n,5708,'As one can see in Table 4 , the resulting parser ranks among the best lexicalized parsers , beating those of Collins and Charniak and Johnson 8 Its F1 performance is a 27 \% reduction in error over et al ',Matsuzaki,'mark_see_As nsubj_see_one aux_see_can advcl_ranks_see prep_see_in pobj_in_Table num_Table_4 det_parser_the amod_parser_resulting nsubj_ranks_parser prep_ranks_among det_parsers_the amod_parsers_best amod_parsers_lexicalized pobj_among_parsers xcomp_ranks_beating dobj_beating_those prep_those_of nn__Collins pobj_of_ cc__and nn__Charniak cc_Charniak_and conj_Charniak_Johnson conj__ num__8 poss_performance_Its nn_performance_F1 nsubj_reduction_performance cop_reduction_is det_reduction_a dep_\%_27 amod_reduction_\% dep_ranks_reduction prep_reduction_in pobj_in_error prep_error_over pobj_over_ cc__et conj__al ' +p,5709,'This corpus-based information typically concerns sequences of 1-3 tags or words ',Cutting,'det_information_This amod_information_corpus-based nsubj_concerns_information advmod_concerns_typically dobj_concerns_sequences prep_sequences_of num_tags_1-3 pobj_of_tags cc_tags_or conj_tags_words ' +p,5710,'It has been argued that METEOR correlates better with human judgment due to higher weight on recall than precision ',Banerjee,'nsubjpass_argued_It aux_argued_has auxpass_argued_been complm_correlates_that nsubj_correlates_METEOR ccomp_argued_correlates dep_correlates_better prep_better_with amod_judgment_human pobj_with_judgment dep_to_due prep_better_to amod_weight_higher pobj_to_weight prep_weight_on pobj_on_recall prep_better_than pobj_than_precision ' +o,5711,'5 Augmenting the corpus with an extracted dictionary Previous research has shown that including word aligned data during training can improve translation results ',Fraser,'nsubj_improve_5 partmod_5_Augmenting det_corpus_the dobj_Augmenting_corpus prep_Augmenting_with det_dictionary_an amod_dictionary_extracted pobj_with_dictionary amod__Previous nn__research nsubj_shown_ aux_shown_has rcmod_dictionary_shown dobj_shown_that prep_that_including pobj_including_word partmod_word_aligned dobj_aligned_data prep_aligned_during pobj_during_training aux_improve_can nn_results_translation dobj_improve_results ' +o,5712,'2 Statistical Word Alignment Model According to the IBM models , the statistical word alignment model can be generally represented as in equation -LRB- 1 -RRB- ',Brown,'nsubj_Model_2 nn_Model_Statistical nn_Model_Word nn_Model_Alignment prep_represented_According dep_According_to det__the nn__IBM nn__models pobj_to_ det_model_the amod_model_statistical nn_model_word nn_model_alignment nsubjpass_represented_model aux_represented_can auxpass_represented_be advmod_represented_generally rcmod_Model_represented prep_represented_as dep_as_in pobj_in_equation dep_Model_1 ' +o,5713,'Our experimental results display that our SDB model achieves a substantial improvement over the baseline and significantly outperforms XP + according to the BLEU metric ',Papineni,'poss_results_Our amod_results_experimental nsubj_display_results complm_achieves_that poss_model_our nn_model_SDB nsubj_achieves_model ccomp_display_achieves det_improvement_a amod_improvement_substantial dobj_achieves_improvement prep_achieves_over det_baseline_the pobj_over_baseline cc_achieves_and advmod_outperforms_significantly conj_achieves_outperforms nn_+_XP dobj_outperforms_+ prep_outperforms_according dep_according_to det_metric_the amod_metric_BLEU pobj_to_metric ' +o,5714,'We evaluate this method over the part of speech tagged portion of the Penn Treebank corpus ',Marcus,'nsubj_evaluate_We det_method_this dobj_evaluate_method prep_evaluate_over det_part_the pobj_over_part prep_part_of pobj_of_speech partmod_speech_tagged dobj_tagged_portion prep_portion_of det_corpus_the nn_corpus_Penn nn_corpus_Treebank pobj_of_corpus ' +o,5715,'1 Introduction Bilingual word alignment is first introduced as an intermediate result in statistical machine translation -LRB- SMT -RRB- ',Brown,'num_alignment_1 nn_alignment_Introduction nn_alignment_Bilingual nn_alignment_word nsubjpass_introduced_alignment auxpass_introduced_is advmod_introduced_first prep_introduced_as det_result_an amod_result_intermediate pobj_as_result prep_result_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT ' +o,5716,'OHara and Wiebe also make use of high level features , in their case the Penn Treebank and FrameNet to classify prepositions ',Marcus,'nsubj__OHara cc_OHara_and conj_OHara_Wiebe advmod__also dep__make dobj_make_use prep_use_of amod_features_high nn_features_level pobj_of_features prep_make_in poss_case_their pobj_in_case det__the nn__Penn nn__Treebank nsubj_classify_ cc__and nn__FrameNet conj__ aux_classify_to infmod_case_classify dobj_classify_prepositions ' +o,5717,'Our approach to STC uses a thesaurus based on corpus statistics for real-valued similarity calculation ',Lin,'poss_approach_Our nsubj_uses_approach prep_approach_to pobj_to_STC det_thesaurus_a dobj_uses_thesaurus prep_uses_based dep_based_on nn_statistics_corpus pobj_on_statistics partmod_statistics_ prep__for amod_calculation_real-valued nn_calculation_similarity pobj_for_calculation ' +o,5718,'We use a bidirectional search strategy , and our algorithm is based on Perceptron learning ',Collins,'nsubj_use_We det__a amod__bidirectional nn__search nn__strategy dobj_use_ cc_use_and poss_algorithm_our nsubjpass_based_algorithm auxpass_based_is conj_use_based prep_based_on pobj_on_Perceptron partmod_Perceptron_learning ' +o,5719,'In recent years , many researchers build alignment links with bilingual corpora ',Wu,'prep_build_In amod_years_recent pobj_In_years amod_researchers_many nsubj_build_researchers amod_links_alignment dobj_build_links prep_build_with amod_corpora_bilingual pobj_with_corpora ' +o,5720,'Related Work The first application of log-linear models to parsing is the work of Ratnaparkhi and colleagues ',Ratnaparkhi,'nn_Work_Related det_application_The amod_application_first nsubj_work_application prep_application_of amod_models_log-linear pobj_of_models prep_models_to pobj_to_parsing cop_work_is det_work_the dep_Work_work prep_work_of pobj_of_Ratnaparkhi cc_Ratnaparkhi_and conj_Ratnaparkhi_colleagues ' +o,5721,'Beam-search parsing using an unnormalized discriminative model , as in , requires a slightly different search strategy than the original generative model described in Roark ',Collins,'nsubj_requires_Beam-search partmod_Beam-search_parsing xcomp_parsing_using det_model_an amod_model_unnormalized nn_model_discriminative dobj_using_model prep_parsing_as dep_as_in pobj_in_ det_strategy_a advmod_different_slightly amod_strategy_different nn_strategy_search dobj_requires_strategy prep_requires_than det_model_the amod_model_original nn_model_generative pobj_than_model partmod_model_described prep_described_in nn__Roark pobj_in_ ' +p,5722,'2 Treebanking The Penn Treebank is annotated with information to make predicate-argument structure easy to decode , including function tags and markers of empty categories that represent displaced constituents ',Marcus,'num_Treebanking_2 det__The nn__Penn nn__Treebank nsubjpass_annotated_ auxpass_annotated_is dep_Treebanking_annotated prep_annotated_with pobj_with_information aux_make_to xcomp_annotated_make amod_structure_predicate-argument nsubj_easy_structure xcomp_make_easy aux_decode_to xcomp_easy_decode prep_annotated_including nn_tags_function pobj_including_tags cc_tags_and conj_tags_markers prep_tags_of amod_categories_empty pobj_of_categories nsubj_represent_that rcmod_tags_represent amod_constituents_displaced dobj_represent_constituents ' +o,5723,'Following the evaluation methodology of Wong and Mooney , we performed 4 runs of the standard 10-fold cross validation and report the averaged performance in this section using the standard automatic evaluation metric BLEU and NIST 2 ',Papineni,'prep_performed_Following det_methodology_the nn_methodology_evaluation pobj_Following_methodology prep_methodology_of nn__Wong cc_Wong_and conj_Wong_Mooney pobj_of_ nsubj_performed_we num_runs_4 dobj_performed_runs prep_runs_of det_validation_the amod_validation_standard amod_validation_10-fold nn_validation_cross pobj_of_validation cc_performed_and conj_performed_report det_performance_the amod_performance_averaged dobj_report_performance prep_report_in det_section_this pobj_in_section partmod_section_using det_evaluation_the amod_evaluation_standard amod_evaluation_automatic dobj_using_evaluation amod_BLEU_metric nsubj__BLEU rcmod_evaluation_ cc__and conj__NIST num_2_ dobj__2 ' +o,5724,'The resulting memory limitations alone can prevent the practical learning of highly split grammars ',Matsuzaki,'det_memory_The amod_memory_resulting nsubj_prevent_limitations advmod_prevent_alone aux_prevent_can dep_memory_prevent det_learning_the amod_learning_practical dobj_prevent_learning prep_learning_of advmod_grammars_highly nn_grammars_split pobj_of_grammars ' +o,5725,'The algorithm is essentially the same as the one introduced in ',Collins,'det_algorithm_The nsubj_same_algorithm cop_same_is advmod_same_essentially det_same_the prep_same_as det_one_the pobj_as_one partmod_one_introduced prep_introduced_in ' +o,5726,'Models of this type include : , which use semantic word clustering , and , which uses variablelength context ',Brown,'nsubj_include_Models prep_Models_of det_type_this pobj_of_type dobj_include_ nsubj_use_which rcmod__use amod_clustering_semantic nn_clustering_word dobj_use_clustering cc__and conj__ nsubj_uses_which rcmod__uses amod_context_variablelength dobj_uses_context ' +o,5727,'In another generation approach , look for pairs of slotted word lattices that share many common slot fillers ; the lattices are generated by applying a multiplesequence alignment algorithm to a corpus of multiple news articles about the same events ',Barzilay,'prep_look_In det_approach_another nn_approach_generation pobj_In_approach nsubj_look_ prep_look_for pobj_for_pairs prep_pairs_of amod_lattices_slotted nn_lattices_word pobj_of_lattices det_share_that dep_lattices_share amod_fillers_many amod_fillers_common nn_fillers_slot dobj_look_fillers det_lattices_the nsubjpass_generated_lattices auxpass_generated_are dep_fillers_generated prep_generated_by pcomp_by_applying det_algorithm_a amod_algorithm_multiplesequence nn_algorithm_alignment dobj_applying_algorithm prep_applying_to det_corpus_a pobj_to_corpus prep_corpus_of amod_articles_multiple nn_articles_news pobj_of_articles prep_articles_about det_events_the amod_events_same pobj_about_events ' +o,5728,'43 Baselines 431 Word Alignment We used the GIZA + + implementation of IBM word alignment model 4 for word alignment , and the heuristics described in to derive the intersection and refined alignment ',Brown,'num_Baselines_43 measure_431_Baselines amod_Alignment_431 nn_Alignment_Word nsubj_used_We dep_Alignment_used det_implementation_the amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dobj_used_implementation prep_implementation_of nn_model_IBM nn_model_word nn_model_alignment pobj_of_model dep__4 amod_model_ prep_used_for nn_alignment_word pobj_for_alignment cc_used_and det_heuristics_the nsubj_described_heuristics conj_used_described prep_described_in pobj_in_ aux_derive_to xcomp_described_derive det_intersection_the dobj_derive_intersection cc_intersection_and amod_alignment_refined conj_intersection_alignment ' +o,5729,'Tbest = argmax T P -LRB- T F -RRB- -LRB- 1 -RRB- P -LRB- T F -RRB- = productdisplay X Y in T Feats = -LCB- ai ai -LRB- X -RRB- -RCB- P -LRB- X Y X , Feats -RRB- -LRB- 2 -RRB- 3 Disambiguation Models The basic generation model presented in used simple probabilistic context-free grammars ',Cahill,'nsubj_=_Tbest nn_P_argmax nn_P_T dobj_=_P nn_F_T appos_P_F appos_P_1 partmod_P_P nn_F_T nsubj_=_F ccomp_P_= nn_Y_productdisplay nn_Y_X dobj_=_Y prep_Y_in nn_=_T nn_=_Feats nn_P_= nn_ai_ai dep_=_ai appos_=_X pobj_in_P nn_X_X nn_X_Y appos_P_X dep_X_Feats dep_presented_2 num_Models_3 nn_Models_Disambiguation nsubj_presented_Models det_model_The amod_model_basic nn_model_generation dep_Models_model dep_=_presented prep_presented_in advmod_used_ amod_grammars_used amod_grammars_simple amod_grammars_probabilistic amod_grammars_context-free pobj_in_grammars ' +o,5730,'glish -LRB- previously used for self-training of parsers -RRB- ',McClosky,'advmod_used_previously dep_glish_used prep_used_for pobj_for_self-training prep_self-training_of amod__parsers pobj_of_ ' +o,5731,'A monotonic segmentation copes with monotonic alignments , that is , j -LRB- k ? ? aj -LRB- ak following the notation of ',Brown,'det_segmentation_A amod_segmentation_monotonic nsubj_aj_segmentation dep_segmentation_copes dep_copes_with amod_alignments_monotonic pobj_with_alignments nsubj_is_that rcmod_alignments_is nn_k_j appos_alignments_k dobj_aj_ak prep_aj_following det_notation_the pobj_following_notation prep_notation_of ' +n,5732,'Our experiments on the Canadian Hansards show that our unsupervised technique is significantly more effective than picking seeds by hand , which in turn is known to rival supervised methods ',Yarowsky,'poss_experiments_Our nsubj_show_experiments prep_experiments_on det_Hansards_the amod_Hansards_Canadian pobj_on_Hansards dep_known_show complm_effective_that poss_technique_our amod_technique_unsupervised nsubj_effective_technique cop_effective_is advmod_effective_significantly advmod_effective_more ccomp_show_effective prep_effective_than pcomp_than_picking dobj_picking_seeds prep_picking_by nn__hand pobj_by_ nsubjpass_known_which prep_which_in pobj_in_turn auxpass_known_is prep_known_to amod_methods_rival amod_methods_supervised pobj_to_methods ' +o,5733,'Numerous experiments have shown parallel bilingual corpora to provide a rich source of constraints for statistical analysis ',Brown,'amod_experiments_Numerous nsubj_shown_experiments aux_shown_have amod_corpora_parallel amod_corpora_bilingual dobj_shown_corpora aux_provide_to xcomp_shown_provide det_source_a amod_source_rich dobj_provide_source prep_source_of pobj_of_constraints prep_constraints_for amod_analysis_statistical pobj_for_analysis ' +o,5734,'In particular , this method has been used for word sense disambiguation and thesaurus construction ',Lin,'prep_used_In pobj_In_particular det_method_this nsubjpass_used_method aux_used_has auxpass_used_been prep_used_for nn__word nn__sense nn__disambiguation pobj_for_ cc__and nn_construction_thesaurus conj__construction ' +o,5735,'Most previous work exploiting unsupervised training data for inferring POS tagging models has focused on semi-supervised methods in the in which the learner is provided with a lexicon specifying the possible tags for each word or a small number of prototypes for each POS ',Haghighi,'amod_work_Most amod_work_previous nsubj_focused_work partmod_work_exploiting amod_data_unsupervised nn_data_training dobj_exploiting_data prep_exploiting_for amod_models_inferring dep_models_POS amod_models_tagging pobj_for_models aux_focused_has prep_focused_on amod_methods_semi-supervised pobj_on_methods rel_provided_in dep_in_the pobj_in_in pobj_in_which det_learner_the nsubjpass_provided_learner auxpass_provided_is rcmod_methods_provided prep_provided_with det_lexicon_a pobj_with_lexicon partmod_lexicon_specifying det_tags_the amod_tags_possible dobj_specifying_tags prep_tags_for det__each nn__word pobj_for_ cc_tags_or det_number_a amod_number_small conj_tags_number prep_number_of pobj_of_prototypes prep_prototypes_for det_POS_each pobj_for_POS ' +o,5736,'Since then this idea has been applied to several tasks , including word sense disambiguation and named-entity recognition ',Yarowsky,'prep_applied_Since pobj_Since_then det_idea_this nsubjpass_applied_idea aux_applied_has auxpass_applied_been prep_applied_to amod_tasks_several pobj_to_tasks prep_applied_including nn__word nn__sense nn__disambiguation pobj_including_ cc__and amod_recognition_named-entity conj__recognition ' +o,5737,'The theory has been applied in probabilistic language modeling , natural language processing , as well as computational vision ',Berger,'det_theory_The nsubjpass_applied_theory aux_applied_has auxpass_applied_been prep_applied_in amod_modeling_probabilistic nn_modeling_language pobj_in_modeling amod__ amod__natural nn__language nn__processing dep_modeling_ dep_well_as cc_modeling_well dep_well_as amod__computational nn__vision conj_modeling_ ' +o,5738,'Finally , since non-projective constructions often involve long-distance dependencies , the problem is closely related to the recovery of empty categories and non-local dependencies in constituency-based parsing ',Cahill,'advmod_related_Finally mark_involve_since amod_constructions_non-projective nsubj_involve_constructions advmod_involve_often dep_related_involve amod_dependencies_long-distance dobj_involve_dependencies det_problem_the nsubjpass_related_problem auxpass_related_is advmod_related_closely prep_related_to det_recovery_the pobj_to_recovery prep_recovery_of amod_categories_empty pobj_of_categories cc_recovery_and amod_dependencies_non-local conj_recovery_dependencies prep_dependencies_in amod_parsing_constituency-based pobj_in_parsing ' +o,5739,'However , in the experiments described here , we focus on alignment at the level of sentences , this for a number of reasons : First , sentence alignments have so far proven their usefulness in a number of applications , eg bilingual lexicography , automatic translation verification and the automatic acquisition of knowledge about translation ',Brown,'advmod_focus_However prep_focus_in det_experiments_the pobj_in_experiments partmod_experiments_described advmod_described_here nsubj_focus_we prep_focus_on pobj_on_alignment prep_focus_at det_level_the pobj_at_level prep_level_of pobj_of_sentences appos_sentences_this prep_this_for det_number_a pobj_for_number prep_number_of pobj_of_reasons advmod_proven_First nn_alignments_sentence nsubj_proven_alignments aux_proven_have advmod_far_so advmod_proven_far parataxis_focus_proven poss_usefulness_their dobj_proven_usefulness prep_proven_in det_number_a pobj_in_number prep_number_of pobj_of_applications amod__eg amod__bilingual amod__lexicography conj_applications_ amod__automatic nn__translation nn__verification conj_applications_ cc_applications_and det_acquisition_the amod_acquisition_automatic conj_applications_acquisition prep_acquisition_of pobj_of_knowledge prep_proven_about pobj_about_translation ' +n,5740,'2 Related Work One of the major problems with the IBM models and the HMM models is that they are restricted to the alignment of each source-language word to at most one targetlanguage word ',Brown,'num_Work_2 nn_Work_Related nsubj_is_Work num_Work_One prep_Work_of det_problems_the amod_problems_major pobj_of_problems prep_problems_with det__the nn__IBM nn__models pobj_with_ cc__and det__the nn__HMM nn__models conj__ complm_restricted_that nsubjpass_restricted_they auxpass_restricted_are ccomp_is_restricted prep_restricted_to det_alignment_the pobj_to_alignment prep_alignment_of det_word_each amod_word_source-language pobj_of_word prep_restricted_to quantmod_one_at dep_at_most num_word_one nn_word_targetlanguage pobj_to_word ' +p,5741,'For example , factored translation models retain the simplicity of phrase-based SMT while adding the ability to incorporate additional features ',Koehn,'prep__For pobj_For_example amod_models_factored nn_models_translation nsubj__models xcomp__retain det_simplicity_the dobj_retain_simplicity prep_simplicity_of amod_SMT_phrase-based pobj_of_SMT prep_retain_while pcomp_while_adding det_ability_the dobj_adding_ability aux_incorporate_to infmod_ability_incorporate amod_features_additional dobj_incorporate_features ' +o,5742,' demonstrates the case of binary SCFG parsing , where six string boundary variables , three for each language as in monolingual CFG parsing , interact with each other , yielding an O -LRB- N6 -RRB- dynamic programming algorithm , where N is the string length , assuming the two paired strings are comparable in length ',Wu,'advmod_demonstrates_ det_case_the nsubj_demonstrates_case prep_case_of amod_parsing_binary nn_parsing_SCFG pobj_of_parsing advmod_interact_where num_variables_six nn_variables_string nn_variables_boundary nsubj_interact_variables appos_variables_three prep_three_for det_language_each pobj_for_language dep_in_as prep_language_in amod_parsing_monolingual nn_parsing_CFG pobj_in_parsing rcmod_case_interact prep_interact_with det_algorithm_each amod_algorithm_other amod_algorithm_yielding det_O_an dep_yielding_O appos_O_N6 amod_algorithm_dynamic nn_algorithm_programming pobj_with_algorithm advmod_length_where nsubj_length_N cop_length_is det_length_the nn_length_string rcmod_algorithm_length xcomp_length_assuming det_strings_the num_strings_two amod_strings_paired nsubj_comparable_strings cop_comparable_are ccomp_assuming_comparable prep_comparable_in pobj_in_length ' +o,5743,'Similar to , each word in the confusion network is associated with a word posterior probability ',Rosti,'aux__to xcomp_Similar_ det_word_each nsubjpass_associated_word prep_word_in det_network_the nn_network_confusion pobj_in_network auxpass_associated_is ccomp__associated prep_associated_with det_probability_a nn_probability_word nn_probability_posterior pobj_with_probability ' +o,5744,'accuracy Training data Turney 66 \% unsupervised 8715 \% supervised Aue & Gamon -LRB- 2005 -RRB- 914 \% supervised SO 7395 \% unsupervised SM+SO to increase seed words , then SO 7485 \% weakly supervised Table 7 : Classification accuracy on the movie review domain Turney achieves 66 \% accuracy on the movie review domain using the PMI-IR algorithm to gather association scores from the web ',Pang,'nn_data_accuracy nn_data_Training nn__Turney nsubj_supervised_ num_\%_66 nsubj_supervised_\% partmod_\%_unsupervised xcomp_unsupervised_ num_\%_8715 dobj__\% rcmod__supervised nsubj_supervised_Aue cc_Aue_& conj_Aue_Gamon appos_Aue_2005 dep_\%_914 amod_Aue_\% ccomp_supervised_supervised nn_SM+SO_SO num_\%_7395 measure_unsupervised_\% amod_SM+SO_unsupervised dobj_supervised_SM+SO aux_increase_to xcomp_supervised_increase nn_words_seed dobj_increase_words advmod_increase_then dep_increase_SO num_\%_7485 dobj_SO_\% advmod_supervised_weakly dep_data_supervised dobj_supervised_Table num_Table_7 amod_accuracy_Classification dep_Table_accuracy prep_accuracy_on det_domain_the nn_domain_movie nn_domain_review pobj_on_domain nn__Turney nsubj_achieves_ dep_data_achieves dep_\%_66 amod_accuracy_\% dobj_achieves_accuracy prep_achieves_on det_domain_the nn_domain_movie nn_domain_review pobj_on_domain partmod_domain_using det_algorithm_the nn_algorithm_PMI-IR dobj_using_algorithm aux_gather_to xcomp_using_gather nn_scores_association dobj_gather_scores prep_gather_from det_web_the pobj_from_web ' +o,5745,'We ran GIZA + + on the training corpus in both directions using its default setting , and then applied the refinement rule diagand described in to obtain a single many-to-many word alignment for each sentence pair ',Och,'nsubj_ran_We nsubj_using_GIZA cc__+ advmod__+ dep_using_ prep__on det_corpus_the nn_corpus_training pobj_on_corpus prep__in det_directions_both pobj_in_directions xcomp_ran_using poss_setting_its nn_setting_default dobj_using_setting cc_ran_and advmod_applied_then conj_ran_applied det_diagand_the amod_diagand_refinement nn_diagand_rule dobj_applied_diagand partmod_diagand_described prep_described_in pobj_in_ aux_obtain_to xcomp_described_obtain det_alignment_a amod_alignment_single amod_alignment_many-to-many nn_alignment_word dobj_obtain_alignment prep_alignment_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,5746,'2 Related Work ThisworkbuildsuponthatofMcCarthyetal which acquires predominant senses for target words from a large sample of text using distributional similarity to provide evidence for predominance ',Lin,'num__2 amod__Related nn__Work nn__ThisworkbuildsuponthatofMcCarthyetal nsubj_acquires_which rcmod__acquires amod_senses_predominant dobj_acquires_senses prep_senses_for nn_words_target pobj_for_words prep_acquires_from det_sample_a amod_sample_large pobj_from_sample prep__of pobj_of_text partmod_text_using amod__distributional nn__similarity dobj_using_ aux_provide_to xcomp_using_provide dobj_provide_evidence prep_provide_for pobj_for_predominance ' +o,5747,'The measures2 Mutual Information -LRB- a0a2a1 -RRB- , the log-likelihood ratio test , two statistical tests : t-test and a3a5a4 - test , and co-occurrence frequency are applied to two sets of data : adjective-noun -LRB- AdjN -RRB- pairs and preposition-noun-verb -LRB- PNV -RRB- triples , where the AMs are applied to -LRB- PN , V -RRB- pairs ',Dunning,'det_measures2_The nn__Mutual nn__Information appos__a0a2a1 nsubj_test_ det__the amod__log-likelihood nn__ratio nn__test appos__ num_tests_two amod_tests_statistical appos__tests dep_tests_t-test cc_t-test_and conj_t-test_a3a5a4 dep_measures2_test cc_test_and amod_frequency_co-occurrence nsubjpass_applied_frequency auxpass_applied_are conj_test_applied prep_applied_to num_sets_two pobj_to_sets prep_sets_of pobj_of_data amod_pairs_adjective-noun appos_pairs_AdjN dobj_applied_pairs cc_applied_and dep_triples_preposition-noun-verb abbrev_triples_PNV conj_applied_triples advmod_applied_where det_AMs_the nsubjpass_applied_AMs auxpass_applied_are advcl_triples_applied prep_applied_to appos_pairs_PN dep_PN_V pobj_to_pairs ' +p,5748,'Many machine learning techniques have been successfully applied to chunking tasks , such as Regularized Winnow , SVMs , CRFs , Maximum Entropy Model , Memory Based Learning and SNoW ',Collins,'amod_techniques_Many nn_techniques_machine nn_techniques_learning nsubjpass_applied_techniques aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_to amod_tasks_chunking pobj_to_tasks dep_as_such prep_tasks_as nn__Regularized nn__Winnow pobj_as_ nn__SVMs conj__ nn__CRFs conj__ nn__Maximum nn__Entropy nn__Model conj__ nn__Memory nn__Based nn__Learning conj__ cc__and nn__SNoW conj__ ' +o,5749,'234 ADV Non-specific adverbial BNF Benefemtive CLF It-cleft CLR \'Closely related\' DIR Direction DTV Dative EXT Extent HLN Headline LGS Logical subject L0C Location MNI~ Manner N0M Nominal PRD Predicate PRP Purpose PUT Locative complement of \'put\' SBJ Subject TMP Temporal TPC Topic TTL Title V0C Vocative Grammatical DTV 0.48\% LGS 3.0\% PRD 18.\% PUT 0.26\% SBJ 78.\% v0c 0.025\% Figure 1: Penn treebank function tags 53.\% Form/Function 37.\% Topicalisation 2.2\% 0.25\% NOM 6.8\% 2.5\% TPC 100\% 2.2\% 1.5\% ADV 11.\% 4.2\% 9.3\% BN\'F 0.072\% 0.026\% 0.13\% DIR 8.3\% 3.0\% 41.\% EXT 3.2\% 1.2\% 0.013\% LOC 25.\% 9.2\% MNR 6.2\% 2.3\% PI~ 5.2\% 1.9\% 33.\% 12.\% Miscellaneous 9.5\% CLR 94.\% 8.8\% CLF 0.34\% 0.03\% HLN 2.6\% 0.25\% TTL 3.1\% 0.29\% Figure 2: Categories of function tags and their relative frequencies one project that used them at all: (Collins, 1997) defines certain constituents as complements based on a combination of label and function tag information.',Collins,'' +o,5750,'More recently , use the distinction between pronouns , nominals and proper nouns 660 in their unsupervised , generative model for coreference resolution ; for their model , this is absolutely critical for achieving better accuracy ',Haghighi,'advmod_recently_More advmod_use_recently nsubj_use_ det_distinction_the dobj_use_distinction prep_distinction_between nn_nominals_pronouns pobj_between_nominals cc_nominals_and amod_nouns_proper conj_nominals_nouns tmod_use_660 prep_660_in poss_model_their amod_model_unsupervised amod_model_generative pobj_in_model prep_model_for amod_resolution_coreference pobj_for_resolution prep_critical_for poss_model_their pobj_for_model nsubj_critical_this cop_critical_is advmod_critical_absolutely parataxis_use_critical prep_critical_for pcomp_for_achieving amod_accuracy_better dobj_achieving_accuracy ' +o,5751,'The most frequently used resource for synonym extraction is large monolingual corpora ',Hindle,'det_resource_The advmod_frequently_most advmod_used_frequently amod_resource_used nsubj_corpora_resource prep_resource_for nn_extraction_synonym pobj_for_extraction cop_corpora_is amod_corpora_large amod_corpora_monolingual ' +o,5752,'32 Translation quality Table 2 presents the impact of parse quality on a treelet translation system , measured using BLEU ',Papineni,'num_quality_32 nn_quality_Translation nsubj_presents_Table num_Table_2 dep_quality_presents det_impact_the dobj_presents_impact prep_impact_of amod_quality_parse pobj_of_quality prep_presents_on det_system_a amod_system_treelet nn_system_translation pobj_on_system partmod_system_measured xcomp_measured_using dobj_using_BLEU ' +o,5753,'The feasibility of such post-parse deepening -LRB- for a statistical parser -RRB- is demonstrated by ',Cahill,'det_feasibility_The nsubjpass_demonstrated_feasibility prep_feasibility_of amod_post-parse_such pobj_of_post-parse amod_post-parse_deepening dep_post-parse_for det_parser_a amod_parser_statistical pobj_for_parser auxpass_demonstrated_is prep_demonstrated_by ' +o,5754,'While transfer learning was proposed more than a decade ago , its application in natural language processing is still a relatively new territory , and its application in relation extraction is still unexplored ',Blitzer,'mark_proposed_While nn_learning_transfer nsubjpass_proposed_learning auxpass_proposed_was advcl__proposed dep_than_more quantmod_a_than num_decade_a measure_ago_decade advmod_proposed_ago dep_ago_ poss_application_its nsubj__application prep_application_in amod_processing_natural nn_processing_language pobj_in_processing cop__is advmod__still det__a advmod_new_relatively amod__new nn__territory cc__and poss_application_its nsubjpass_unexplored_application prep_application_in nn_extraction_relation pobj_in_extraction auxpass_unexplored_is advmod_unexplored_still conj__unexplored ' +o,5755,'Tillmann and Zhang and et al ',Liang,'nn__Tillmann cc_Tillmann_and conj_Tillmann_Zhang cc__and conj__ cc__et conj__al ' +o,5756,'Such text-oriented ranking methods can be applied to tasks ranging from automated extraction of keyphrases , to extractive summarization and word sense disambiguation ',Mihalcea,'amod_methods_Such amod_methods_text-oriented amod_methods_ranking nsubjpass_applied_methods aux_applied_can auxpass_applied_be prep_applied_to pobj_to_tasks partmod_tasks_ranging prep_ranging_from amod_extraction_automated pobj_from_extraction prep_extraction_of pobj_of_keyphrases aux_extractive_to xcomp_ranging_extractive nn_disambiguation_summarization cc_summarization_and conj_summarization_word nn_disambiguation_sense dobj_extractive_disambiguation ' +o,5757,'We run the decoder with its default settings and then use Koehn \'s implementation of minimum error rate training to tune the feature weights on the development set ',Och,'nsubj_run_We det_decoder_the dobj_run_decoder prep_run_with poss_settings_its nn_settings_default pobj_with_settings cc_run_and advmod_run_then conj_run_use poss_implementation_Koehn possessive_Koehn_\'s dobj_use_implementation prep_implementation_of amod__minimum nn__error nn__rate nn__training pobj_of_ prep_use_to pobj_to_tune det_weights_the nn_weights_feature dobj_run_weights prep_run_on det_set_the nn_set_development pobj_on_set ' +o,5758,'Next we use the conclusions from two psycholinguistic experiments on ranking the Cf-list , the salience of discourse entities in prepended phrases and the ordering of possessor and possessed in complex NPs , to try to improve the performance of LRC ',Marcus,'nsubj_use_we dep_Next_use det_conclusions_the dobj_use_conclusions prep_use_from num_experiments_two amod_experiments_psycholinguistic pobj_from_experiments prep_experiments_on amod_Cf-list_ranking det_Cf-list_the pobj_on_Cf-list det_salience_the conj_experiments_salience prep_salience_of nn_entities_discourse pobj_of_entities prep_entities_in amod__prepended nn__phrases pobj_in_ cc_experiments_and det_ordering_the conj_experiments_ordering prep_ordering_of pobj_of_possessor cc_use_and conj_use_possessed prep_possessed_in amod__complex nn__NPs pobj_in_ aux_try_to xcomp_possessed_try aux_improve_to xcomp_try_improve det_performance_the dobj_improve_performance prep_performance_of pobj_of_LRC ' +o,5759,'No pretagged text is necessary for Hidden Markov Models ',Cutting,'det_text_No amod_text_pretagged nsubj_necessary_text cop_necessary_is prep_necessary_for nn_Models_Hidden nn_Models_Markov pobj_for_Models ' +n,5760,'This implies that the complexity of structure divergence between two languages is higher than suggested in literature ',Galley,'nsubj_implies_This complm_suggested_that det_complexity_the nsubjpass_suggested_complexity prep_complexity_of nn_divergence_structure pobj_of_divergence prep_divergence_between num_languages_two pobj_between_languages auxpass_suggested_is advmod_suggested_higher dep_higher_than ccomp_implies_suggested prep_suggested_in pobj_in_literature ' +o,5761,' used the averaged perceptron ',Collins,'nsubj_used_ det_perceptron_the amod_perceptron_averaged dobj_used_perceptron ' +o,5762,'For handling word identities , one could follow the approach used for handling the POS tags and view the POS tags and word identities as two separate sources of information ',Brown,'prep_follow_For pcomp_For_handling nn_identities_word dobj_handling_identities nsubj_follow_one aux_follow_could det_approach_the dobj_follow_approach partmod_approach_used prep_used_for pcomp_for_handling det__the dep__POS nn__tags dobj_handling_ cc_follow_and conj_follow_view det_tags_the dep_tags_POS dobj_view_tags cc_tags_and nn_identities_word conj_tags_identities prep_view_as num_sources_two amod_sources_separate pobj_as_sources prep_sources_of pobj_of_information ' +o,5763,'22 Using Log-Likelihood-Ratios to Estimate Word Translation Probabilities Our method for computing the probabilistic translation lexicon LLR-Lex is based on the the Log2http : \\/ \\/ wwwfjochcom\\/GIZA + + html Likelihood-Ratio -LRB- LLR -RRB- statistic , which has also been used by Moore -LRB- 2004a ; 2004b -RRB- and Melamed -LRB- 2000 -RRB- as a measure of word association ',Dunning,'partmod_22_Using dobj_Using_Log-Likelihood-Ratios prep_Using_to nn_Probabilities_Estimate nn_Probabilities_Word nn_Probabilities_Translation pobj_to_Probabilities poss_method_Our nsubjpass_based_method prep_method_for pcomp_for_computing det_LLR-Lex_the amod_LLR-Lex_probabilistic nn_LLR-Lex_translation nn_LLR-Lex_lexicon dobj_computing_LLR-Lex auxpass_based_is dep_22_based prep_based_on det_Log2http_the dep_Log2http_the pobj_on_Log2http number_\\/_\\/ num__\\/ amod__wwwfjochcom\\/GIZA amod__+ cc_+_+ conj_+_html amod__Likelihood-Ratio abbrev__LLR nn__statistic dep_Log2http_ nsubjpass_used_which aux_used_has advmod_used_also auxpass_used_been rcmod__used prep_used_by pobj_by_Moore nn_2004b_2004a appos_Moore_2004b cc_Moore_and conj_Moore_Melamed appos_Moore_2000 prep_used_as det_measure_a pobj_as_measure prep_measure_of nn_association_word pobj_of_association ' +o,5764,': ~ The difl ` erent kinds of noun chunks covered by our grmnmar are listed below and illustrated with exmnples : a combination of a non-obligatory deternfiner , optional adjectives or cardinals and the noun 1Other types of lexicalised PCFGs have been -LRB- h ! scrib -LRB- \'d in , , , -LRB- Chcll -RRB- a and lelinek , 1998 -RRB- mid ',Collins,'parataxis_been_~ det_kinds_The amod_kinds_difl nn_kinds_erent nsubjpass_listed_kinds prep_kinds_of nn_chunks_noun pobj_of_chunks partmod_chunks_covered prep_covered_by poss_grmnmar_our pobj_by_grmnmar auxpass_listed_are ccomp_~_listed advmod_listed_below cc_listed_and conj_listed_illustrated prep_illustrated_with pobj_with_exmnples det_combination_a dep_exmnples_combination prep_combination_of det_deternfiner_a amod_deternfiner_non-obligatory pobj_of_deternfiner amod_adjectives_optional nsubj_been_adjectives cc_adjectives_or conj_adjectives_cardinals cc_adjectives_and det_types_the amod_types_noun amod_types_1Other conj_adjectives_types prep_types_of amod_PCFGs_lexicalised pobj_of_PCFGs aux_been_have dep__h nn_\'d_scrib dep_h_\'d prep_\'d_in nn__ appos__ pobj_in_ appos__Chcll dep__a cc_a_and conj_a_lelinek appos_a_1998 amod__mid dobj_been_ ' +o,5765,'Much research has been done to improve tagging accuracy using several different models and methods , including : hidden Markov models -LRB- HMMs -RRB- , ; rule-based systems , ; memory-based systems ; maximum-entropy systems ; path voting constraint systems ; linear separator systems ; and majority voting systems ',Ratnaparkhi,'nsubjpass_done_research aux_done_has auxpass_done_been dep_Much_done aux_improve_to xcomp_done_improve amod_accuracy_tagging dobj_improve_accuracy partmod_accuracy_using amod_models_several amod_models_different dobj_using_models cc_models_and conj_models_methods prep_done_including amod_models_hidden nn_models_Markov pobj_including_models abbrev_models_HMMs amod_models_ dep__ amod_systems_rule-based nsubj__systems parataxis_done_ advmod__ amod_systems_memory-based nsubj__systems parataxis_done_ amod_systems_maximum-entropy nsubj__systems parataxis_done_ nn_systems_path nn_systems_voting nn_systems_constraint nsubj__systems parataxis_done_ amod_systems_linear nn_systems_separator nsubj__systems parataxis_done_ cc_done_and nn_systems_majority nn_systems_voting nsubj__systems conj_done_ ' +o,5766,'3 Model 1 and Model 2 l ~ cl -RRB- lacing the -LRB- l -LRB- ~ , t -RRB- endence on aj-l in the HMM alignment mo -LRB- M I -RRB- y a del -RRB- endence on j , we olltain a model wlfich -LRB- : an lie seen as a zero-order Hid -LRB- l -LRB- mMarkov Model which is similar to Model 2 1 -RRB- rot -RRB- ose -LRB- t t\\/y ',Brown,'num_Model_3 num_Model_1 cc_Model_and nn_cl_Model num_cl_2 nn_cl_l amod_cl_~ conj_Model_cl dep_olltain_lacing det_endence_the amod_endence_l appos_endence_~ dep_~_t dep_a_endence prep_endence_on amod_y_aj-l prep_aj-l_in det_alignment_the nn_alignment_HMM pobj_in_alignment nn_y_mo dep_y_M dep_M_I pobj_on_y dep_endence_a dep_a_del dobj_lacing_endence prep_lacing_on pobj_on_j nsubj_olltain_we dep_Model_olltain det_wlfich_a nn_wlfich_model dobj_olltain_wlfich det_lie_an dep_seen_lie dep_wlfich_seen prep_seen_as det_zero-order_a pobj_as_zero-order nn_l_Hid dep_zero-order_l nn_Model_mMarkov dep_seen_Model nsubj_similar_which cop_similar_is rcmod_Model_similar aux_Model_to xcomp_similar_Model number_1_2 dobj_Model_1 prep_seen_rot dep_olltain_ose dobj_ose_t advmod_ose_t\\/y ' +o,5767,'In particular , we use a randomly-selected corpus the first five columns as information-like consisting of a 67 million word subset of the TREC Similarly , since the last four columns share databases ',Dunning,'prep_use_In pobj_In_particular nsubj_use_we det_corpus_a amod_corpus_randomly-selected dobj_use_corpus det_columns_the amod_columns_first num_columns_five dep_as_columns prep_use_as pobj_as_information-like partmod_information-like_consisting prep_consisting_of det_subset_a number_million_67 num_subset_million nn_subset_word pobj_of_subset prep_subset_of det_TREC_the pobj_of_TREC advmod_consisting_Similarly prep_consisting_since det_columns_the amod_columns_last num_columns_four pobj_since_columns nn_databases_share nsubj__databases dep_use_ ' +o,5768,'The model scaling factors M1 are trained with respect to the final translation quality measured by an error criterion ',Och,'det_M1_The nn_M1_model nn_M1_scaling nn_M1_factors nsubjpass_trained_M1 auxpass_trained_are prep_trained_with pobj_with_respect prep_trained_to det_quality_the amod_quality_final nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_by det_criterion_an nn_criterion_error pobj_by_criterion ' +o,5769,'Some research into factored machine translation has been published by ',Koehn,'det_research_Some nsubjpass_published_research prep_research_into amod_translation_factored nn_translation_machine pobj_into_translation aux_published_has auxpass_published_been prep_published_by ' +o,5770,'The feature weights for the overall translation models were trained using minimum-error-rate training procedure ',Och,'det_weights_The nn_weights_feature nsubjpass_trained_weights prep_weights_for det_models_the amod_models_overall nn_models_translation pobj_for_models auxpass_trained_were xcomp_trained_using num_procedure_ amod_procedure_minimum-error-rate nn_procedure_training dobj_using_procedure ' +o,5771,'The standard method to overcome this problem to use the model in both directions -LRB- interchanging the source and target languages -RRB- and applying heuristic-based combination techniques to produce a refined alignment henceforth referred to as RA Several researchers have proposed algorithms for improving word alignment systems by injecting additional knowledge or combining different alignment models ',Koehn,'det_method_The amod_method_standard aux_overcome_to dep_method_overcome det_problem_this dobj_overcome_problem aux_use_to infmod_problem_use det_model_the dobj_use_model prep_use_in det_directions_both pobj_in_directions parataxis_use_interchanging det_languages_the nn_languages_source cc_source_and conj_source_target dobj_interchanging_languages cc_overcome_and csubj_referred_applying amod_techniques_heuristic-based nn_techniques_combination dobj_applying_techniques aux_produce_to xcomp_applying_produce det_henceforth_a amod_henceforth_refined nn_henceforth_alignment nn_henceforth_ dobj_produce_henceforth conj_overcome_referred prep_referred_to quantmod_RA_as pobj_to_RA amod_researchers_Several nsubj_proposed_researchers aux_proposed_have dep_method_proposed dobj_proposed_algorithms prep_proposed_for pcomp_for_improving nn_systems_word nn_systems_alignment dobj_improving_systems prep_improving_by pcomp_by_injecting amod_knowledge_additional dobj_injecting_knowledge cc_improving_or conj_improving_combining amod_models_different nn_models_alignment dobj_combining_models ' +o,5772,'This tagging scheme is the IOB scheme originally put forward by ',Ramshaw,'det_scheme_This amod_scheme_tagging nsubj_scheme_scheme cop_scheme_is det_scheme_the nn_scheme_IOB advmod_put_originally partmod_scheme_put advmod_put_forward prep_put_by ' +o,5773,'212 Research on Syntax-Based SMT A number of researchers have proposed models where the translation process involves syntactic representations of the source and\\/or target languages ',Berger,'nn_Research_212 dep_number_Research prep_Research_on pobj_on_Syntax-Based dep_number_SMT det_number_A dep_have_number prep_number_of pobj_of_researchers dep_have_ amod_models_proposed dobj_have_models advmod_involves_where det_process_the nn_process_translation nsubj_involves_process advcl_have_involves amod_representations_syntactic dobj_involves_representations prep_representations_of det_source_the pobj_of_source cc_have_and\\/or conj_have_target dobj_target_languages ' +o,5774,' , 2004 -RRB- and lscript22 regularization ',Goodman,'dep__ amod__2004 cc__and amod_regularization_lscript22 conj__regularization ' +o,5775,'Lexical collocation functions , especially those determined statistically , have recently attracted considerable attention in computational linguistics mainly , though not exclusively , for use in disambiguation ',Church,'amod_functions_Lexical nn_functions_collocation nsubj_attracted_functions advmod_statistically_especially det_statistically_those amod_statistically_determined appos_functions_statistically aux_attracted_have advmod_attracted_recently amod_attention_considerable dobj_attracted_attention prep_attention_in amod_linguistics_computational pobj_in_linguistics partmod_linguistics_ advmod__mainly mark_exclusively_though neg_exclusively_not parataxis__exclusively prep_attracted_for pobj_for_use prep_use_in pobj_in_disambiguation ' +p,5776,'Nowadays , most state-of-the-art SMT systems are based on bilingual phrases ',Och,'advmod_based_Nowadays advmod_systems_most amod_systems_state-of-the-art nn_systems_SMT nsubjpass_based_systems auxpass_based_are prep_based_on amod_phrases_bilingual pobj_on_phrases ' +o,5777,'Previous research in automatic acquisition focuses primarily on the use of statistical techniques , such as bilingual alignment , or extraction of syntactic constructions from online dictionaries and corpora ',Church,'amod_research_Previous nsubj_focuses_research prep_research_in amod_acquisition_automatic pobj_in_acquisition advmod_focuses_primarily prep_focuses_on det_use_the pobj_on_use prep_use_of amod_techniques_statistical pobj_of_techniques dep_as_such conj_use_as amod__bilingual nn__alignment pobj_as_ cc_use_or conj_use_extraction prep_extraction_of amod_constructions_syntactic pobj_of_constructions prep_constructions_from amod_dictionaries_online pobj_from_dictionaries cc_dictionaries_and nn__corpora conj_dictionaries_ ' +o,5778,'However , found that it is actually harmful to restrict phrases to constituents in parse trees , because the restriction would cause the system to miss many reliable translations , such as the correspondence between there is in English and es gibt -LRB- it gives -RRB- in German ',Koehn,'advmod_found_However nsubj_found_ complm_harmful_that nsubj_harmful_it cop_harmful_is advmod_harmful_actually ccomp_found_harmful aux_restrict_to xcomp_harmful_restrict dobj_restrict_phrases prep_restrict_to pobj_to_constituents prep_constituents_in amod_trees_parse pobj_in_trees mark_cause_because det_restriction_the nsubj_cause_restriction aux_cause_would advcl_harmful_cause det_system_the nsubj_miss_system aux_miss_to xcomp_cause_miss amod_translations_many amod_translations_reliable dobj_miss_translations dep_as_such prep_translations_as det_correspondence_the pobj_as_correspondence prep_correspondence_between expl_is_there pcomp_between_is prep_is_in pobj_in_English cc_is_and dep_gibt_es conj_is_gibt nsubj_gives_it parataxis_cause_gives prep_cause_in pobj_in_German ' +p,5779,'3 Probabilistic Parsing Models 31 Probabilistic Context-Free Grammars Lexicalization has been shown to improve parsing performance for the Penn Treebank ',Collins,'dep_Models_3 nn_Parsing_Probabilistic nsubj_Models_Parsing dobj_Models_31 nn_Lexicalization_Probabilistic nn_Lexicalization_Context-Free nn_Lexicalization_Grammars nsubjpass_shown_Lexicalization aux_shown_has auxpass_shown_been dep_Models_shown aux_improve_to xcomp_shown_improve amod_performance_parsing dobj_improve_performance prep_performance_for det_Treebank_the nn_Treebank_Penn pobj_for_Treebank ' +o,5780,'Pivots are features occurring frequently and behaving similarly in both domains ',Blitzer,'nsubj_features_Pivots cop_features_are dep_features_occurring advmod_occurring_frequently cc_occurring_and conj_occurring_behaving advmod_behaving_similarly prep_behaving_in det_domains_both pobj_in_domains ' +o,5781,'A very impor232 Author Best Hindle and Rooth 800 \% Resnik and Hearst 839 \% WN Resnik and Hearst 750 \% Ratnaparkhi et al ',Marcus,'det_Hindle_A advmod_impor232_very amod_Hindle_impor232 nn_Hindle_Author nn_Hindle_Best dep_\%_Hindle cc_Hindle_and nn__Rooth conj_Hindle_ dep_\%_800 amod__\% nn__Resnik cc_Resnik_and conj_Resnik_Hearst dep_\%_ num_\%_839 nn_Resnik_WN dep_\%_Resnik cc_\%_and nn__Hearst conj_\%_ dep_\%_750 amod__\% nn_al_Ratnaparkhi nn_al_et dep_\%_al ' +n,5782,' report better perplexity results on the Verbmobil Corpus with their HMMbased alignment model in comparison to Model 2 of ',Brown,'amod_report_ amod_results_better nn_results_perplexity dep_report_results prep_results_on det_Corpus_the nn_Corpus_Verbmobil pobj_on_Corpus prep_Corpus_with poss_model_their nn_model_HMMbased nn_model_alignment pobj_with_model prep_model_in pobj_in_comparison prep_comparison_to pobj_to_Model num_Model_2 prep_results_of ' +o,5783,'Most systems extract co-occurrence and syntactic information from the words surrounding the target term , which is then converted into a vector-space representation of the contexts that each target term appears in ',Brown,'amod_systems_Most nsubj_extract_systems amod_information_co-occurrence cc_co-occurrence_and conj_co-occurrence_syntactic dobj_extract_information prep_extract_from det_words_the pobj_from_words partmod_words_surrounding det_term_the nn_term_target dobj_surrounding_term nsubjpass_converted_which auxpass_converted_is advmod_converted_then rcmod_term_converted prep_converted_into det_representation_a amod_representation_vector-space pobj_into_representation prep_representation_of det_contexts_the pobj_of_contexts complm_appears_that det_term_each nn_term_target nsubj_appears_term ccomp_converted_appears prt_appears_in ' +o,5784,'Table 1 shows the impact of increasing reordering window length on translation quality for the ? dev06 ? ? data2 Increasing the reordering window past 2 has minimal impact on translation quality , implying that most of the reordering effects across Spanish and English are well modeled at the local or phrase level ',Koehn,'nsubj_shows_Table num_Table_1 det_impact_the dobj_shows_impact prep_impact_of amod_length_increasing nn_length_reordering nn_length_window pobj_of_length partmod_length_ prep__on nn_quality_translation pobj_on_quality prep__for dep_for_the dep_the_dev06 nsubj_has_data2 partmod_data2_Increasing det_window_the nn_window_reordering dobj_Increasing_window prep_Increasing_past pobj_past_2 dep_shows_has amod_impact_minimal dobj_has_impact prep_impact_on nn_quality_translation pobj_on_quality partmod_impact_implying complm_modeled_that nsubjpass_modeled_most prep_most_of det_effects_the nn_effects_reordering pobj_of_effects prep_effects_across pobj_across_Spanish cc_Spanish_and conj_Spanish_English auxpass_modeled_are advmod_modeled_well ccomp_implying_modeled prep_modeled_at det_level_the amod_level_local cc_local_or conj_local_phrase pobj_at_level ' +o,5785,'thresholding (DeNero and Klein, 2007).',DeNero,'' +o,5786,'Parse selection constitutes an important part of many parsing systems ',Blitzer,'amod_selection_Parse nsubj_constitutes_selection det_part_an amod_part_important dobj_constitutes_part prep_part_of amod__many amod__parsing nn__systems pobj_of_ ' +o,5787,' used skip-chain Conditional Random Fields to model pragmatic dependencies between paired meeting utterances -LRB- eg QUESTION-ANSWER relations -RRB- , and used a combination of lexical , prosodic , structural and discourse features to rank utterances by importance ',Galley,'nsubj_used_ amod_Fields_skip-chain nn_Fields_Conditional nn_Fields_Random dobj_used_Fields aux_model_to xcomp_used_model amod_dependencies_pragmatic dobj_model_dependencies prep_dependencies_between amod_utterances_paired nn_utterances_meeting pobj_between_utterances nn_relations_eg nn_relations_QUESTION-ANSWER abbrev_utterances_relations cc_used_and conj_used_used det_combination_a dobj_used_combination prep_combination_of amod_features_lexical conj_lexical_prosodic conj_lexical_structural cc_lexical_and conj_lexical_discourse pobj_of_features aux_rank_to xcomp_used_rank dobj_rank_utterances prep_rank_by pobj_by_importance ' +o,5788,'Since it is not feasible to maximize the likelihood of the observations directly , we maximize the expected log likelihood by considering the EM auxiliary function , in a similar manner to that used for modelling contextual variations of phones for ASR ',Berger,'mark_feasible_Since nsubj_feasible_it cop_feasible_is neg_feasible_not advcl_maximize_feasible aux_maximize_to xcomp_feasible_maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_of det_observations_the pobj_of_observations advmod_maximize_directly nsubj_maximize_we det_likelihood_the amod_likelihood_expected nn_likelihood_log dobj_maximize_likelihood prep_maximize_by pcomp_by_considering det_function_the nn_function_EM amod_function_auxiliary dobj_considering_function prep_maximize_in det_manner_a amod_manner_similar pobj_in_manner prep_manner_to pobj_to_that partmod_that_used prep_used_for pcomp_for_modelling amod_variations_contextual dobj_modelling_variations prep_variations_of pobj_of_phones prep_phones_for nn__ASR pobj_for_ ' +o,5789,'Recall that the log likelihood of our model is : d parenleftBigg Lorig -LRB- Dd ; d -RRB- i -LRB- d , i , i -RRB- 2 2 2d parenrightBigg i -LRB- , i -RRB- 2 2 2 We now introduce a new variable d = d , and plug it into the equation for log likelihood : d parenleftBigg Lorig -LRB- Dd ; d + -RRB- i -LRB- d , i -RRB- 2 2 2d parenrightBigg i -LRB- , i -RRB- 2 2 2 The result is the model of , where the d are the domain-specific feature weights , and d are the domain-independent feature weights ','Daume III','complm_is_that det_likelihood_the nn_likelihood_log nsubj_is_likelihood prep_likelihood_of poss_model_our pobj_of_model ccomp_Recall_is nsubj_parenleftBigg_d parataxis_Recall_parenleftBigg nn_Dd_Lorig dobj_parenleftBigg_Dd dep_model_d nsubj_model_i dep_2_d dep_2_i dep_2_i dep_i_2 number_2d_2 num_parenrightBigg_2d prep_2_parenrightBigg dep_parenrightBigg_i dep_,_i number_2_2 num_2_2 dep_parenrightBigg_2 nsubj_introduce_We advmod_introduce_now rcmod_2_introduce det_d_a amod_d_new amod_d_variable nsubj_d_d dep_d_= ccomp_introduce_d cc_introduce_and conj_introduce_plug dobj_plug_it prep_plug_into det_equation_the pobj_into_equation prep_equation_for nn_likelihood_log pobj_for_likelihood dep_i_d partmod_d_parenleftBigg xcomp_parenleftBigg_Lorig nsubj_i_Dd nn_+_d dep_Dd_+ ccomp_Lorig_i dep_i_d appos_d_i number_2_2 num_i_2 amod_i_2d nn_i_parenrightBigg dobj_i_i dep_,_i dep_i_2 number_2_2 num_i_2 det_result_The dep_i_result cop_model_is det_model_the dep_Dd_model prep_model_of pobj_of_ advmod_weights_where det_d_the nsubj_weights_d cop_weights_are det_weights_the amod_weights_domain-specific nn_weights_feature rcmod__weights cc_model_and nsubj_weights_d cop_weights_are det_weights_the amod_weights_domain-independent nn_weights_feature conj_model_weights ' +p,5790,'The Perceptron style for natural language processing problems as initially proposed by can provide state of the art results on various domains including text chunking , syntactic parsing , etc The main drawback of the Perceptron style algorithm is that it does not have a mechanism for attaining the maximize margin of the training data ',Collins,'det_style_The amod_style_Perceptron nsubj_provide_style prep_style_for amod_problems_natural nn_problems_language nn_problems_processing pobj_for_problems advmod_initially_as advmod_proposed_initially partmod_problems_proposed prep_proposed_by pobj_by_ aux_provide_can dobj_provide_state prep_state_of det_results_the nn_results_art pobj_of_results prep_provide_on amod_domains_various pobj_on_domains prep_domains_including nn_chunking_text pobj_including_chunking amod_parsing_syntactic appos_chunking_parsing dep_chunking_etc det_drawback_The amod_drawback_main nsubj_is_drawback prep_drawback_of det_algorithm_the nn_algorithm_Perceptron nn_algorithm_style pobj_of_algorithm dep_provide_is complm_have_that nsubj_have_it aux_have_does neg_have_not ccomp_is_have det_mechanism_a dobj_have_mechanism prep_mechanism_for pcomp_for_attaining det_margin_the amod_margin_maximize dobj_attaining_margin prep_margin_of det_data_the nn_data_training pobj_of_data ' +o,5791,'Agreement among annotators was measured using the K statistic ',Carletta,'nsubjpass_measured_Agreement prep_Agreement_among pobj_among_annotators auxpass_measured_was xcomp_measured_using det_statistic_the nn_statistic_K dobj_using_statistic ' +o,5792,'We divided these case roles into four types by location in the article as in (Iida et al., 2006), i) the case role depends on the predicate or the predicate depends on the case role in the intra-sentence (dependency relations), ii) the case role does not depend on the predicate and the predicate does not depend on the case role in the intra-sentence (zeroanaphoric (intra-sentential)), iii) the case role is not in the sentence containing the predicate (zeroanaphoric (inter-sentential)), and iv) the case role and the predicate are in the same phrase (in same phrase).',Iida,'' +o,5793,'21 Heuristic Grammar Induction Grammar based SMT models almost exclusively follow the same two-stage approach to grammar induction developed for phrase-based methods ',Koehn,'num_Grammar_21 amod_Grammar_Heuristic nn_Grammar_Grammar nn_Grammar_Induction dep_based_Grammar nn_models_SMT nsubj_follow_models advmod_exclusively_almost advmod_follow_exclusively ccomp_based_follow det_approach_the amod_approach_same amod_approach_two-stage dobj_follow_approach prep_approach_to nn_induction_grammar pobj_to_induction partmod_induction_developed prep_developed_for amod_methods_phrase-based pobj_for_methods ' +o,5794,'Line 4 and 5 are similar to the phrase extraction algorithm by ',Och,'nsubj_similar_Line num_Line_4 cc_Line_and conj_Line_5 cop_similar_are prep_similar_to det_algorithm_the nn_algorithm_phrase nn_algorithm_extraction pobj_to_algorithm prep_similar_by ' +o,5795,'External information such as the discourse or domain dependency of each word sense is expected to lead to system improvement ',Yarowsky,'amod_information_External nsubjpass_expected_information dep_as_such prep_information_as det_dependency_the nn_dependency_discourse cc_discourse_or conj_discourse_domain pobj_as_dependency prep_dependency_of det__each nn__word nn__sense pobj_of_ auxpass_expected_is aux_lead_to xcomp_expected_lead prep_lead_to nn_improvement_system pobj_to_improvement ' +o,5796,'With IOB2 representation , the problem of Chinese chunking can be regarded as a sequence labeling task ',Ramshaw,'prep_regarded_With nn__IOB2 nn__representation pobj_With_ det_problem_the nsubjpass_regarded_problem prep_problem_of amod_chunking_Chinese pobj_of_chunking aux_regarded_can auxpass_regarded_be prep_regarded_as det_task_a nn_task_sequence nn_task_labeling pobj_as_task ' +o,5797,'This contrasts with alternative alignment models such as those of Melamed and , which impose a one-to-one constraint on alignments ',Wu,'nsubj_contrasts_This prep_contrasts_with amod_models_alternative nn_models_alignment pobj_with_models dep_as_such prep_models_as pobj_as_those prep_those_of nn__Melamed pobj_of_ cc__and conj__ nsubj_impose_which rcmod_models_impose det_constraint_a amod_constraint_one-to-one dobj_impose_constraint prep_impose_on pobj_on_alignments ' +o,5798,'For each feature function , there is a model parameter i The best word segmentation W \\* is determined by the decision rule as = = = M i ii W M W WSfWSScoreW 0 0 \\* -RRB- , -LRB- maxarg -RRB- , , -LRB- maxarg -LRB- 2 -RRB- Below we describe how to optimize s Our method is a discriminative approach inspired by the Minimum Error Rate Training method proposed in ',Och,'prep_is_For det_function_each nn_function_feature pobj_For_function expl_is_there det_parameter_a nn_parameter_model nsubj_is_parameter dep_segmentation_i det_segmentation_The amod_segmentation_best nn_segmentation_word dep_parameter_segmentation nn_\\*_W nsubjpass_determined_\\* auxpass_determined_is rcmod_parameter_determined prep_determined_by det_decision_the nsubj_rule_decision dep_by_rule prep_rule_as nn_\\*_= nn_\\*_= nn_\\*_= nn_\\*_M nn_\\*_i nn_\\*_ii nn_\\*_W nn_\\*_M nn_\\*_W nn_\\*_WSfWSScoreW nn_\\*_0 nn_\\*_0 pobj_as_\\* appos_\\*_maxarg dep_rule_maxarg dobj_maxarg_2 pobj_by_Below nsubj_describe_we dep_is_describe advmod_optimize_how aux_optimize_to ccomp_describe_optimize dobj_optimize_s poss_method_Our nsubj_approach_method cop_approach_is det_approach_a amod_approach_discriminative rcmod_s_approach partmod_approach_inspired prep_inspired_by det_method_the nn_method_Minimum nn_method_Error nn_method_Rate nn_method_Training pobj_by_method partmod_method_proposed prep_proposed_in ' +o,5799,'2 Word-to-Word Bitext Alignment We will study the problem of aligning an English sentence to a French sentence and we will use the word alignment of the IBM statistical translation models ',Brown,'dep_Word-to-Word_2 amod_Alignment_Word-to-Word nn_Alignment_Bitext nsubj_study_We aux_study_will dep_Alignment_study det_problem_the dobj_study_problem prep_problem_of pcomp_of_aligning det_sentence_an amod_sentence_English dobj_aligning_sentence prep_aligning_to det_sentence_a amod_sentence_French pobj_to_sentence cc_study_and nsubj_use_we aux_use_will conj_study_use det_alignment_the nn_alignment_word dobj_use_alignment prep_alignment_of det_models_the nn_models_IBM amod_models_statistical nn_models_translation pobj_of_models ' +o,5800,'This incremental process can be iterated to the point that the system 1 It is not just a matter of time , but also of required linguistic skills -LRB- see for example -RRB- ',Marcus,'det_process_This amod_process_incremental aux_iterated_can auxpass_iterated_be rcmod_process_iterated prep_iterated_to det_point_the pobj_to_point complm_1_that det_system_the nsubj_1_system ccomp_iterated_1 nsubj_matter_It cop_matter_is neg_matter_not advmod_matter_just det_matter_a rcmod_1_matter prep_matter_of pobj_of_time dep_also_but dep_1_also prep_process_of amod_skills_required amod_skills_linguistic pobj_of_skills dep_process_see prep_see_for nn__example pobj_for_ ' +o,5801,' use cascaded processing for full parsing with good results ',Collins,'amod_use_ nsubj_cascaded_use dobj_cascaded_processing prep_cascaded_for amod_parsing_full pobj_for_parsing prep_cascaded_with amod_results_good pobj_with_results ' +o,5802,'predict correctly the label of a test instance xN +1 is bounded by 2N +1 EN +1 bracketleftbigd + D bracketrightbig2 where D = D -LRB- w , , -RRB- = radicalBigsummationtext N i = 12i This result is used to explain the convergence of weighted or voted perceptron algorithms ',Collins,'advmod_radicalBigsummationtext_correctly det_label_the nsubjpass_bounded_label prep_label_of det_xN_a nn_xN_test nn_xN_instance pobj_of_xN num_xN_+1 auxpass_bounded_is dep_radicalBigsummationtext_bounded prep_bounded_by number_+1_2N num_bracketrightbig2_+1 nn_bracketrightbig2_EN num_bracketrightbig2_+1 amod_bracketrightbig2_bracketleftbigd cc_bracketleftbigd_+ conj_bracketleftbigd_D pobj_by_bracketrightbig2 advmod_D_where nsubj_D_D dep_D_= rcmod_bracketrightbig2_D dep_bracketrightbig2_w advmod_radicalBigsummationtext_= acomp_predict_radicalBigsummationtext nn_i_N nsubj_=_i ccomp_radicalBigsummationtext_= dobj_=_12i det_result_This nsubjpass_used_result auxpass_used_is rcmod_12i_used aux_explain_to xcomp_used_explain det_convergence_the dobj_explain_convergence prep_convergence_of pobj_of_weighted cc_used_or conj_used_voted nn_algorithms_perceptron dobj_voted_algorithms ' +o,5803,'10Both Pharoah and our system have weights trained using MERT on sentences of length 30 words or less , to ensure that training and test conditions are matched ',Och,'nn_Pharoah_10Both nsubj_have_Pharoah cc_Pharoah_and poss_system_our conj_Pharoah_system dobj_have_weights partmod_weights_trained xcomp_trained_using nn__MERT dobj_using_ prep__on pobj_on_sentences prep_sentences_of pobj_of_length num_words_30 dep_using_words cc_words_or conj_words_less aux_ensure_to xcomp_have_ensure complm_matched_that nn_conditions_training cc_training_and conj_training_test nsubjpass_matched_conditions auxpass_matched_are ccomp_ensure_matched ' +o,5804,'These features are calculated by mining the parse trees , and then could be used for resolution by using manually designed rules , or using machine-learning methods ',Luo,'det_features_These nsubjpass_calculated_features auxpass_calculated_are prep_calculated_by pcomp_by_mining det_trees_the amod_trees_parse dobj_mining_trees cc_calculated_and nsubjpass_used_then aux_used_could auxpass_used_be conj_calculated_used prep_used_for pobj_for_resolution prep_used_by pcomp_by_using advmod_using_manually ccomp_using_designed nsubj__rules xcomp_designed_ cc_using_or conj_using_using amod_methods_machine-learning dobj_using_methods ' +o,5805,'5 Effectiveness Comparison 51 English-Chinese ATIS Models Both the transfer and transducer systems were trained and evaluated on English-to-Mandarin Chinese translation of transcribed utterances from the ATIS corpus ',Brown,'num_Comparison_5 nn_Comparison_Effectiveness dep_transfer_Comparison num_Models_51 amod_Models_English-Chinese nn_Models_ATIS dep_Comparison_Models dep_transfer_Both det_transfer_the cc_transfer_and nn_systems_transducer conj_transfer_systems auxpass_trained_were dep_transfer_trained cc_trained_and conj_trained_evaluated prep_trained_on amod_translation_English-to-Mandarin amod_translation_Chinese pobj_on_translation prep_translation_of amod_utterances_transcribed pobj_of_utterances prep_utterances_from det__the nn__ATIS nn__corpus pobj_from_ ' +o,5806,'It is possible to recognize a common structure of these works , based on a typical bootstrap schema : Step 1 : Initial unsupervised categorization ',Yarowsky,'nsubj_possible_It cop_possible_is aux_recognize_to xcomp_possible_recognize det_structure_a amod_structure_common dobj_recognize_structure prep_structure_of det_works_these pobj_of_works prep_recognize_based dep_based_on det_bootstrap_a amod_bootstrap_typical dep_on_bootstrap nn__schema dep_bootstrap_ dep_bootstrap_Step num_Step_1 amod_categorization_Initial amod_categorization_unsupervised dep_Step_categorization ' +o,5807,'It has been shown that both Nave Bayes and SVMs perform with similar accuracy on different sentiment tagging tasks ',Pang,'nsubjpass_shown_It aux_shown_has auxpass_shown_been complm_perform_that preconj_Bayes_both amod_Bayes_Nave nsubj_perform_Bayes cc_Bayes_and conj_Bayes_SVMs ccomp_shown_perform prep_perform_with amod_accuracy_similar pobj_with_accuracy prep_accuracy_on amod_tasks_different nn_tasks_sentiment nn_tasks_tagging pobj_on_tasks ' +o,5808,'Note that the translation direction is inverted from what would be normally expected ; correspondingly the models built around this equation are often called invertedtranslationmodels ',Brown,'complm_inverted_that det_direction_the nn_direction_translation nsubjpass_inverted_direction auxpass_inverted_is ccomp_Note_inverted prep_inverted_from nsubjpass_expected_what aux_expected_would auxpass_expected_be advmod_expected_normally dep_correspondingly_expected pobj_from_correspondingly det_models_the nsubjpass_called_models partmod_models_built prep_built_around det_equation_this pobj_around_equation auxpass_called_are advmod_called_often rcmod_correspondingly_called dobj_called_invertedtranslationmodels ' +o,5809,'1 Introduction The research presented in this paper forms part of an ongoing effort to develop methods to induce wide-coverage multilingual LexicalFunctional Grammar -LRB- LFG -RRB- resources from treebanks by means of automatically associating LFG f-structure information with constituency trees produced by probabilistic parsers ',Cahill,'num_research_1 nn_research_Introduction nn_research_The dep_presented_research prep_presented_in det_part_this nn_part_paper nn_part_forms pobj_in_part prep_part_of det_effort_an amod_effort_ongoing pobj_of_effort aux_develop_to xcomp_presented_develop nsubj_induce_methods aux_induce_to xcomp_develop_induce amod_Grammar_wide-coverage nn_Grammar_multilingual nn_Grammar_LexicalFunctional dobj_induce_Grammar dep_induce_LFG amod_resources_ dobj_induce_resources prep_resources_from pobj_from_treebanks prep_induce_by pobj_by_means prep_means_of advmod_associating_automatically pcomp_of_associating nn_information_LFG amod_information_f-structure dobj_associating_information prep_associating_with nn_trees_constituency pobj_with_trees partmod_trees_produced prep_produced_by amod_parsers_probabilistic pobj_by_parsers ' +o,5810,'(DeRose 1988; Cutting et al 1992; Merialdo 1994).',Cutting,'' +o,5811,'We trained log linear models with theperceptronalgorithm usingfea746 Markov order Classification Task 0 1 2 S1 -LRB- no multi-word constituent start -RRB- 967 969 969 E1 -LRB- no multi-word constituent end -RRB- 973 973 973 Table 2 : Classification accuracy on development set for binary classes S1 and E1 , for various Markov orders ',Collins,'nsubj_trained_We amod_models_log amod_models_linear dobj_trained_models prep_models_with pobj_with_theperceptronalgorithm nn_Task_ nn_Task_usingfea746 nn_Task_Markov nn_Task_order nn_Task_Classification nsubj_0_Task rcmod_models_0 number_2_1 num_S1_2 iobj_0_S1 det_start_no amod_start_multi-word nn_start_constituent appos_S1_start number_969_967 dobj_0_969 num_E1_969 dep_models_E1 det_end_no amod_end_multi-word nn_end_constituent appos_E1_end number_973_973 dep_models_973 num_Table_973 dep_973_Table num_Table_2 amod_accuracy_Classification dep_973_accuracy prep_accuracy_on pobj_on_development partmod_development_set prep_set_for amod_S1_binary nn_S1_classes pobj_for_S1 cc_S1_and conj_S1_E1 prep_models_for amod_orders_various amod_orders_Markov pobj_for_orders ' +o,5812,'Other models -LRB- , Xiong et al ',Wu,'amod_models_Other dep_models_ dep__Xiong cc_Xiong_et conj_Xiong_al ' +o,5813,'1 Introduction Shallow parsing has received a reasonable amount of attention in the last few years -LRB- for example -RRB- ',Ramshaw,'num_parsing_1 nn_parsing_Introduction nn_parsing_Shallow nsubj_received_parsing aux_received_has det_amount_a amod_amount_reasonable dobj_received_amount prep_received_of pobj_of_attention prep_attention_in det_years_the amod_years_last amod_years_few pobj_in_years dep_received_for nn__example pobj_for_ ' +o,5814,'The best accuracies are observed when the labelsarecreatedfromdistributionallysimilarwords using dependency-based similarity measure -LRB- Depend -RRB- ',Lin,'det_accuracies_The amod_accuracies_best nsubjpass_observed_accuracies auxpass_observed_are advmod_using_when det_labelsarecreatedfromdistributionallysimilarwords_the nsubj_using_labelsarecreatedfromdistributionallysimilarwords advcl_observed_using amod_measure_ amod_measure_dependency-based nn_measure_similarity dobj_using_measure appos_measure_Depend ' +o,5815,'While they train the parameters using a maximum a posteriori estimator , we extend the MERT algorithm to take the evaluation metric into account ',Och,'mark_train_While nsubj_train_they advcl_extend_train det_parameters_the nsubj_using_parameters dep_train_using det_maximum_a nsubj_estimator_maximum det_estimator_a nn_estimator_posteriori xcomp_using_estimator nsubj_extend_we det__the nn__MERT nn__algorithm nsubj_take_ aux_take_to xcomp_extend_take det_metric_the nn_metric_evaluation dobj_take_metric prep_take_into pobj_into_account ' +o,5816,'For classi cation , we use a maximum entropy model , from the logistic regression package in Weka , with all default parameter settings ',Berger,'prep_use_For nn_cation_classi pobj_For_cation nsubj_use_we det__a amod__maximum amod__entropy nn__model dobj_use_ prep_use_from det_package_the amod_package_logistic nn_package_regression pobj_from_package prep_package_in nn__Weka pobj_in_ prep_use_with det_settings_all nn_settings_default nn_settings_parameter pobj_with_settings ' +o,5817,'Parametertuningwasdonewithminimum error rate training , which was used to maximize BLEU ',Och,'nn_rate_error dobj_Parametertuningwasdonewithminimum_rate nn__training nsubj_Parametertuningwasdonewithminimum_ nsubjpass_used_which auxpass_used_was rcmod__used aux_maximize_to xcomp_used_maximize amod__BLEU dobj_maximize_ ' +o,5818,'Alignment models to structure the translation model are introduced in ',Brown,'amod_models_Alignment nsubjpass_introduced_models aux_structure_to infmod_models_structure det_model_the nn_model_translation dobj_structure_model auxpass_introduced_are prep_introduced_in ' +p,5819,'The current state-of-the-art is to use minimum error rate training -LRB- MERT -RRB- as described in ',Och,'det_state-of-the-art_The amod_state-of-the-art_current nsubj_is_state-of-the-art aux_use_to xcomp_is_use amod_training_minimum nn_training_error nn_training_rate dobj_use_training abbrev_training_MERT advmod_described_as amod_training_described prep_use_in ' +p,5820,'In this work , we propose two models that can be categorized as extensions of standard word lexicons : A discriminative word lexicon that uses global , ie sentence-level source information to predict the target words using a statistical classifier and a trigger-based lexicon model that extends the well-known IBM model 1 with a second trigger , allowing for a more finegrained lexical choice of target words ',Brown,'prep_propose_In det_work_this pobj_In_work nsubj_propose_we num_models_two dobj_propose_models nsubjpass_categorized_that aux_categorized_can auxpass_categorized_be rcmod_models_categorized prep_categorized_as pobj_as_extensions prep_extensions_of amod_lexicons_standard nn_lexicons_word pobj_of_lexicons det_lexicon_A amod_lexicon_discriminative nn_lexicon_word dep_models_lexicon nsubj_uses_that rcmod_lexicon_uses amod_information_global amod_information_ie nn_information_sentence-level nn_information_source dobj_uses_information aux_predict_to xcomp_uses_predict det_words_the nn_words_target dobj_predict_words xcomp_predict_using det_classifier_a amod_classifier_statistical dobj_using_classifier cc_models_and det_model_a amod_model_trigger-based nn_model_lexicon conj_models_model nsubj_extends_that rcmod_model_extends det__the amod__well-known nn__IBM nn__model num__1 dobj_extends_ prep__with det_trigger_a amod_trigger_second pobj_with_trigger xcomp_extends_allowing prep_allowing_for det_choice_a amod_choice_more amod_choice_finegrained amod_choice_lexical pobj_for_choice prep_choice_of nn_words_target pobj_of_words ' +o,5821,'As the training data from DVDs is much more similar to books than that from kitchen , we should give the data from DVDs a higher weight ',Blitzer,'mark_similar_As det_data_the nn_data_training nsubj_similar_data prep_data_from pobj_from_DVDs cop_similar_is advmod_similar_much advmod_similar_more advcl_give_similar prep_similar_to pobj_to_books prep_books_than pobj_than_that prep_that_from nn__kitchen pobj_from_ nsubj_give_we aux_give_should det_data_the iobj_give_data prep_data_from pobj_from_DVDs det_weight_a amod_weight_higher dobj_give_weight ' +o,5822,'Since Odds = P/(1 P), we multiply both sides of Definition 3 by (1P(U|E))1 to obtain, P(U|E) 1P(U|E) = P(E|U)P(U) P(E)(1P(U|E)) (7) By substituting Equation 6 in Equation 7 and later, applying the multiplication rule P(U|E)P(E) = P(E|U)P(U) to it, we will obtain: P(U|E) P(U|E) = P(E|U)P(U) P(E|U)P(U) (8) We proceed to take the log of the odds in Equation 8 (i.e. logit) to get: log P(E|U)P(E|U) = log P(U|E)P(U|E) log P(U)P(U) (9) While it is obvious that certain words tend to cooccur more frequently than others (i.e. idioms and collocations), such phenomena are largely arbitrary (Smadja, 1993).',Smadja,'' +o,5823,'Indeed , the prominent approach for evaluating the quality of rule acquisition algorithms is by human judgment of the learned rules ',Barzilay,'advmod_is_Indeed det_approach_the amod_approach_prominent nsubj_is_approach prep_approach_for pcomp_for_evaluating det_quality_the dobj_evaluating_quality prep_quality_of nn_algorithms_rule nn_algorithms_acquisition pobj_of_algorithms prep_is_by amod_judgment_human pobj_by_judgment prep_judgment_of det_rules_the amod_rules_learned pobj_of_rules ' +o,5824,'We present two approaches to SMT-based query expansion , both of which are implemented in the framework of phrase-based SMT ',Och,'nsubj_present_We num_approaches_two dobj_present_approaches prep_present_to amod_expansion_SMT-based nn_expansion_query pobj_to_expansion dep_of_both nsubjpass_implemented_of pobj_of_which auxpass_implemented_are rcmod_expansion_implemented prep_implemented_in det_framework_the pobj_in_framework prep_framework_of amod_SMT_phrase-based pobj_of_SMT ' +p,5825,' describe a more efficient algorithm that can compute all edge expectations in O -LRB- n3 -RRB- time using the inverse of the Kirchoff matrix K1 ',Smith,'nsubj_describe_ det_algorithm_a advmod_efficient_more amod_algorithm_efficient dobj_describe_algorithm nsubj_compute_that aux_compute_can rcmod_algorithm_compute det_expectations_all nn_expectations_edge dobj_compute_expectations prep_expectations_in nn_time_O appos_time_n3 pobj_in_time xcomp_compute_using det_inverse_the dobj_using_inverse prep_inverse_of det_K1_the nn_K1_Kirchoff nn_K1_matrix pobj_of_K1 ' +o,5826,'In particular , most of the work on parsing with kernel methods has focussed on kernels over parse trees ',Collins,'prep_focussed_In pobj_In_particular nsubj_focussed_most prep_most_of det_work_the pobj_of_work prep_work_on pcomp_on_parsing prep_parsing_with nn_methods_kernel pobj_with_methods aux_focussed_has prep_focussed_on pobj_on_kernels prep_kernels_over amod_trees_parse pobj_over_trees ' +p,5827,'Both , and Tillmann and Zhang -LRB- 2006 -RRB- report on effective machine translation -LRB- MT -RRB- models involving large numbers of features with discriminatively trained weights ',Liang,'det__Both nsubj_report_ cc__and conj__Tillmann cc_Tillmann_and conj_Tillmann_Zhang appos__2006 prep_report_on amod_models_effective nn_models_machine nn_models_translation abbrev_models_MT pobj_on_models prep_models_involving amod_numbers_large pobj_involving_numbers prep_numbers_of pobj_of_features prep_features_with advmod_trained_discriminatively amod_weights_trained pobj_with_weights ' +o,5828,'Most current approaches emphasize within-sentence dependencies such as the distortion in , the dependency of alignment in HMM , and syntax mappings in ',Brown,'amod_approaches_Most amod_approaches_current nsubj_emphasize_approaches amod_dependencies_within-sentence dobj_emphasize_dependencies dep_as_such prep_dependencies_as det_distortion_the pobj_as_distortion prep_distortion_in pobj_in_ det_dependency_the conj_dependencies_dependency prep_dependency_of pobj_of_alignment prep_alignment_in nn__HMM pobj_in_ cc_dependencies_and nn_mappings_syntax conj_dependencies_mappings prep_mappings_in pobj_in_ ' +o,5829,'At the same time , grammar theoreticians have proposed various generative synchronous grammar formalisms for MT , such as Synchronous Context Free Grammars -LRB- S-CFG -RRB- or Synchronous Tree Adjoining Grammars -LRB- S-TAG -RRB- ',Wu,'prep_proposed_At det_time_the amod_time_same pobj_At_time nn_theoreticians_grammar nsubj_proposed_theoreticians aux_proposed_have amod_formalisms_various amod_formalisms_generative amod_formalisms_synchronous nn_formalisms_grammar dobj_proposed_formalisms prep_formalisms_for pobj_for_MT dep_as_such prep_formalisms_as nn__Synchronous nn__Context nn__Free nn__Grammars appos__S-CFG pobj_as_ cc__or nn__Synchronous nn__Tree nn__Adjoining nn__Grammars appos__S-TAG conj__ ' +o,5830,'However , Moores Law , the driving force of change in computing since then , has opened the way for recent progress in the field , such as Statistical Machine Translation -LRB- SMT -RRB- ',Koehn,'advmod_opened_However nn_Law_Moores nsubj_opened_Law det_force_the amod_force_driving appos_Law_force prep_force_of pobj_of_change prep_force_in pobj_in_computing prep_force_since pobj_since_then aux_opened_has det_way_the dobj_opened_way prep_way_for amod_progress_recent pobj_for_progress prep_progress_in det_field_the pobj_in_field dep_as_such prep_progress_as nn_Translation_Statistical nn_Translation_Machine pobj_as_Translation abbrev_Translation_SMT ' +o,5831,'Its still possible to use MSA if , for example , the input is pre-clustered to have the same constituent ordering -LRB- -RRB- ',Barzilay,'poss_MSA_Its advmod_possible_still amod_MSA_possible prep_possible_to pobj_to_use mark_pre-clustered_if prep_pre-clustered_for pobj_for_example det_input_the nsubj_pre-clustered_input cop_pre-clustered_is dep_MSA_pre-clustered aux_have_to xcomp_pre-clustered_have det_constituent_the amod_constituent_same dobj_have_constituent partmod_constituent_ordering dep_MSA_ ' +o,5832,'ROUGE is an evaluation metric designed to evaluate automatically generated summaries ',Lin,'nn__ROUGE nsubj_metric_ cop_metric_is det_metric_an nn_metric_evaluation partmod_metric_designed aux_evaluate_to xcomp_designed_evaluate advmod_evaluate_automatically dep_evaluate_generated dobj_generated_summaries ' +o,5833,'In TAC 2008 Summarization track , all submitted runs were scored with the ROUGE and Basic Elements -LRB- BE -RRB- metrics ',Lin,'prep_submitted_In nn_track_TAC num_track_2008 nn_track_Summarization pobj_In_track nsubj_submitted_all nsubjpass_scored_runs auxpass_scored_were ccomp_submitted_scored prep_scored_with det__the nn__ROUGE pobj_with_ cc__and nn_metrics_Basic nn_metrics_Elements abbrev_metrics_BE conj__metrics dep_with_ ' +o,5834,'5 The Experimental Results We used the Penn Treebank WSJ corpus to perform empirical experiments on the proposed parsing models ',Marcus,'num_Results_5 nn_Results_The nn_Results_Experimental nsubj_used_We dep_Results_used det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_WSJ nsubj__corpus ccomp_used_ aux_perform_to xcomp__perform amod_experiments_empirical dobj_perform_experiments prep_perform_on det_models_the amod_models_proposed nn_models_parsing pobj_on_models ' +o,5835,'Since word senses are often associated with domains , word senses can be consequently distinguished by way of determining the domain of each description ',Yarowsky,'mark_associated_Since nn_senses_word nsubjpass_associated_senses auxpass_associated_are advmod_associated_often advcl_distinguished_associated prep_associated_with nn__domains pobj_with_ nn_senses_word nsubj_distinguished_senses aux_distinguished_can cop_distinguished_be advmod_distinguished_consequently prep_distinguished_by pobj_by_way prep_way_of pcomp_of_determining det_domain_the dobj_determining_domain prep_domain_of det_description_each pobj_of_description ' +o,5836,'First , we can construct an infinite number of more specialized PCFGs by splitting or refining the PCFGs nonterminals into increasingly finer states ; this leads to the iPCFG or infinite PCFG ',Liang,'advmod_construct_First nsubj_construct_we aux_construct_can det_number_an amod_number_infinite dobj_construct_number prep_number_of amod_PCFGs_more amod_PCFGs_specialized pobj_of_PCFGs prep_construct_by pcomp_by_splitting cc_splitting_or conj_splitting_refining det_nonterminals_the nn_nonterminals_PCFGs dobj_splitting_nonterminals prep_splitting_into advmod_finer_increasingly amod_states_finer pobj_into_states nsubj_leads_this parataxis_construct_leads prep_leads_to det_PCFG_the amod_PCFG_iPCFG cc_iPCFG_or conj_iPCFG_infinite pobj_to_PCFG ' +p,5837,'32 Comparison between SVM , Bootstrapping and LP For WSD , SVM is one of the state of the art supervised learning algorithms , while bootstrapping is one of the state of the art semi-supervised learning algorithms ',Yarowsky,'num_Comparison_32 nsubj_one_Comparison prep_Comparison_between pobj_between_SVM conj_SVM_Bootstrapping cc_SVM_and conj_SVM_LP prep_one_For pobj_For_WSD nsubj_one_SVM cop_one_is prep_one_of det_state_the pobj_of_state prep_state_of det_art_the pobj_of_art partmod_art_supervised xcomp_supervised_learning amod__algorithms dobj_learning_ mark_one_while nsubj_one_bootstrapping cop_one_is advcl_supervised_one prep_one_of det_state_the pobj_of_state prep_state_of det_art_the pobj_of_art partmod_art_semi-supervised xcomp_semi-supervised_learning dobj_learning_algorithms ' +o,5838,'Machine translation based on a deeper analysis of the syntactic structure of a sentence has long been identified as a desirable objective in principle -LRB- consider -RRB- ',Wu,'nn_translation_Machine partmod_translation_based prep_based_on det_analysis_a amod_analysis_deeper pobj_on_analysis prep_analysis_of det_structure_the amod_structure_syntactic pobj_of_structure prep_structure_of det_sentence_a pobj_of_sentence aux_identified_has advmod_identified_long auxpass_identified_been rcmod_sentence_identified prep_identified_as det_objective_a amod_objective_desirable pobj_as_objective prep_objective_in pobj_in_principle dep_translation_consider acomp_consider_ ' +o,5839,'W -LRB- S , T -RRB- = summationdisplay uS , vT w -LRB- u , v -RRB- Globally optimal minimum cuts can be found in polynomial time and near-linear running time in practice , using the maximum flow algorithm ',Pang,'nsubj_summationdisplay_S dep_S_T dep_summationdisplay_= ccomp_W_summationdisplay nsubj_W_uS nn_w_vT appos_uS_w appos_w_u dep_u_v nn_cuts_Globally amod_cuts_optimal nn_cuts_minimum nsubjpass_found_cuts aux_found_can auxpass_found_be rcmod_w_found prep_found_in amod_time_polynomial pobj_in_time cc_time_and conj_time_near-linear ccomp_found_running dobj_running_time prep_time_in pobj_in_practice dep_running_using det_algorithm_the amod_algorithm_maximum nn_algorithm_flow dobj_using_algorithm ' +o,5840,'The words with the highest association probabilities are chosen as acquired words for entity e 41 Base Model I Using the translation model I , where each word is equally likely to be aligned with each entity , we have p -LRB- w e -RRB- = 1 -LRB- l + 1 -RRB- m mproductdisplay j = 1 lsummationdisplay i = 0 p -LRB- wj ei -RRB- -LRB- 1 -RRB- where l and m are the lengths of entity and word sequences respectively ',Brown,'det_words_The nsubjpass_chosen_words prep_words_with det_probabilities_the amod_probabilities_highest nn_probabilities_association pobj_with_probabilities auxpass_chosen_are prep_chosen_as amod_words_acquired pobj_as_words prep_words_for pobj_for_entity dep_41_e dep_Model_41 nsubj_Model_Base dep_have_Model nsubj_Using_I dep_Model_Using det_model_the nn_model_translation dobj_Using_model nsubj__I rcmod_model_ advmod_likely_where det_word_each nsubj_likely_word cop_likely_is advmod_likely_equally advcl_Model_likely aux_aligned_to auxpass_aligned_be xcomp_likely_aligned prep_aligned_with det_entity_each pobj_with_entity nsubj_have_we dep_chosen_have dobj_have_p number_e_w dep_p_e number_1_= num_=_1 dep_=_l number_l_+ number_l_1 nn_=_m nn_=_mproductdisplay nn_=_j nsubj_=_= dep_lsummationdisplay_1 amod_=_lsummationdisplay dep_lsummationdisplay_i rcmod_p_= num_p_0 dobj_=_p nn_ei_wj appos_p_ei appos_p_1 advmod_lengths_where nsubj_lengths_l cc_l_and conj_l_m cop_lengths_are det_lengths_the rcmod_p_lengths prep_lengths_of nn_sequences_entity cc_entity_and conj_entity_word pobj_of_sequences advmod_lengths_respectively ' +o,5841,'The alignment a J 1 that has the highest probability -LRB- under a certain model -RRB- is also called the Viterbi alignment -LRB- of that model -RRB- : a J 1 = argmax a J 1 p -LRB- f J 1 , a J 1 e I 1 -RRB- -LRB- 8 -RRB- A detailed comparison of the quality of these Viterbi alignments for various statistical alignment models compared to human-made word alignments can be found in ',Och,'det_alignment_The det_J_a nsubjpass_called_J num_J_1 nsubj_has_that rcmod_J_has det_probability_the amod_probability_highest dobj_has_probability dep_has_under det_model_a amod_model_certain pobj_under_model auxpass_called_is advmod_called_also dep_alignment_called det_alignment_the nn_alignment_Viterbi dobj_called_alignment dep_alignment_of det_model_that pobj_of_model det_J_a dep_alignment_J num_J_1 dep_J_= iobj_=_argmax det_p_a nn_p_J num_p_1 dobj_=_p dep_J_f dep_-LRB-_J dep_J_1 det_e_a nn_e_J num_e_1 appos_1_e num_1_I dep_J_1 appos_p_8 det_comparison_A amod_comparison_detailed nsubjpass_found_comparison prep_comparison_of det_quality_the pobj_of_quality prep_quality_of det_alignments_these nn_alignments_Viterbi pobj_of_alignments prep_alignments_for amod_models_various amod_models_statistical nn_models_alignment pobj_for_models prep_models_compared dep_compared_to amod_alignments_human-made nn_alignments_word pobj_to_alignments aux_found_can auxpass_found_be dep_alignment_found prep_found_in ' +o,5842,'This approach has also been used by -LRB- Dagan and Itai , 1994 ; Gale et al , 1992 ; Shiitze , 1992 ; Gale et al , 1993 ; Yarowsky , 1995 ; Gale and Church , 1Lunar is not an unknown word in English , Yeltsin finds its translation in the 4-th candidate ',Brown,'det_approach_This nsubjpass_used_approach aux_used_has advmod_used_also auxpass_used_been prep_used_by pobj_by_Dagan cc_Dagan_and conj_Dagan_Itai nsubj_finds_1994 dep_1994_Gale cc_Gale_et conj_Gale_al appos_Gale_1992 dep_Gale_Shiitze appos_Shiitze_1992 dep_Gale_Gale cc_Gale_et conj_Gale_al appos_Gale_1993 dep_Gale_Yarowsky appos_Yarowsky_1995 dep_Gale_Gale cc_Gale_and conj_Gale_Church nsubj_word_1Lunar cop_word_is neg_word_not det_word_an amod_word_unknown dep_1994_word prep_word_in pobj_in_English dep_1994_Yeltsin rcmod_Dagan_finds poss_translation_its dobj_finds_translation prep_finds_in det_candidate_the amod_candidate_4-th pobj_in_candidate ' +n,5843,' have implemented a dependency parser with good accuracy -LRB- it is almost as good at dependency parsing as Charniak -RRB- and very impressive speed -LRB- it is about ten times faster than and four times faster than Charniak -RRB- ',Collins,'nsubj_implemented_ aux_implemented_have det_parser_a amod_parser_dependency dobj_implemented_parser prep_parser_with amod_accuracy_good pobj_with_accuracy nsubj_good_it cop_good_is advmod_good_almost advmod_good_as dep_parser_good prep_good_at advmod_parsing_dependency pcomp_at_parsing prep_parsing_as nn__Charniak pobj_as_ cc_parser_and advmod_impressive_very amod_speed_impressive conj_parser_speed nsubj_faster_it cop_faster_is quantmod_ten_about num_times_ten measure_faster_times dep_speed_faster prep_faster_than pobj_than_ cc_faster_and num_times_four measure_faster_times conj_faster_faster prep_faster_than nn__Charniak pobj_than_ ' +p,5844,'Turning off the extensions to GIZA + + and training p0 as in produces a substantial increase in AER ',Brown,'prt_Turning_off det_extensions_the dobj_Turning_extensions prep_Turning_to nn_+_GIZA nn_+_+ pobj_to_+ cc_+_and amod_p0_training conj_+_p0 dep_in_as prep_p0_in pobj_in_ dep_Turning_produces det_increase_a amod_increase_substantial nsubj_Turning_increase prep_increase_in pobj_in_AER ' +o,5845,'Based on the proofs in and Li et al ',Collins,'dep_Based_on det_proofs_the pobj_on_proofs prep_proofs_in pobj_in_ cc__and conj__Li cc_Li_et conj_Li_al ' +n,5846,'Unlike minimum error rate training , our system is able to exploit large numbers of specific features in the same manner as static reranking systems ',Och,'prep_able_Unlike amod__minimum nn__error nn__rate nn__training pobj_Unlike_ poss_system_our nsubj_able_system cop_able_is aux_exploit_to xcomp_able_exploit amod_numbers_large dobj_exploit_numbers prep_numbers_of amod_features_specific pobj_of_features prep_exploit_in det_manner_the amod_manner_same pobj_in_manner prep_manner_as amod_systems_static amod_systems_reranking pobj_as_systems ' +o,5847,'corpora and corpus query tools has been particularly significant in the area of compiling and developing lexicographic materials and in the area of creating various kinds of lexical resources , such as WordNet and FrameNet ',Lin,'nn_tools_corpora cc_corpora_and conj_corpora_corpus nn_tools_query nsubj_significant_tools aux_significant_has dep_significant_been advmod_significant_particularly prep_significant_in det_area_the pobj_in_area prep_area_of pcomp_of_compiling cc_compiling_and conj_compiling_developing amod__lexicographic nn__materials dobj_compiling_ cc_in_and conj_in_in det_area_the pobj_in_area prep_area_of pcomp_of_creating amod_kinds_various dobj_creating_kinds prep_kinds_of amod_resources_lexical pobj_of_resources dep_as_such prep_kinds_as nn__WordNet pobj_as_ cc__and nn__FrameNet conj__ ' +o,5848,'Our experience suggests that disjunctive LFs are an important capability , especially as one seeks to make grammars reusable across applications , and to employ domain-specific , sentence-level paraphrases ',Barzilay,'poss_experience_Our nsubj_suggests_experience complm_capability_that amod_LFs_disjunctive nsubj_capability_LFs cop_capability_are det_capability_an amod_capability_important ccomp_suggests_capability advmod_seeks_especially dep_seeks_as nsubj_seeks_one dep_capability_seeks aux_make_to ccomp_seeks_make amod_reusable_grammars dobj_make_reusable prep_make_across pobj_across_applications cc_make_and aux_employ_to conj_make_employ amod_paraphrases_domain-specific amod_paraphrases_sentence-level dobj_employ_paraphrases ' +p,5849,'22 Maximum Entropy Models Maximum entropy -LRB- ME -RRB- models , also known as 928 log-linear and exponential learning models , provide a general purpose machine learning technique for classification and prediction which has been successfully applied to natural language processing including part of speech tagging , named entity recognition etc Maximum entropy models can integrate features from many heterogeneous information sources for classification ',Berger,'number_Maximum_22 num_models_Maximum nn_models_Entropy nn_models_Models nn_models_Maximum nn_models_entropy abbrev_models_ME nsubj_provide_ advmod_known_also partmod__known prep_known_as num_log-linear_928 pobj_as_log-linear cc_log-linear_and amod_models_exponential nn_models_learning conj_log-linear_models dep_models_provide det_machine_a amod_machine_general nn_machine_purpose dobj_provide_machine partmod_machine_learning dobj_learning_technique prep_technique_for pobj_for_classification cc_classification_and conj_classification_prediction nsubjpass_applied_which aux_applied_has auxpass_applied_been advmod_applied_successfully rcmod_technique_applied prep_applied_to amod_processing_natural nn_processing_language pobj_to_processing prep_processing_including pobj_including_part prep_part_of nn_tagging_speech pobj_of_tagging partmod_part_named nn_etc_entity nn_etc_recognition dobj_named_etc amod_models_Maximum nn_models_entropy nsubj_integrate_models aux_integrate_can rcmod_etc_integrate dobj_integrate_features prep_integrate_from amod_sources_many amod_sources_heterogeneous nn_sources_information pobj_from_sources prep_sources_for pobj_for_classification ' +o,5850,'This also makes our grammar weakly equivalent to an inversion transduction grammar , although the conversion would create a very large number of new nonterminal symbols ',Wu,'nsubj_makes_This advmod_makes_also poss_grammar_our nsubj_equivalent_grammar advmod_equivalent_weakly xcomp_makes_equivalent prep_equivalent_to det__an nn__inversion nn__transduction nn__grammar pobj_to_ mark_create_although det_conversion_the nsubj_create_conversion aux_create_would advcl_makes_create det_number_a advmod_large_very amod_number_large dobj_create_number prep_number_of amod_symbols_new amod_symbols_nonterminal pobj_of_symbols ' +o,5851,'The statistic is recast as : -LRB- fs , w -RRB- -LRB- sys , sys -RRB- = agr -LRB- fs , w -RRB- -LRB- sys , sys -RRB- P agr -LRB- fs , -RRB- -LRB- sys , sys -RRB- N P agr -LRB- fs , -RRB- -LRB- sys , sys -RRB- N In this modified form , -LRB- fs , w -RRB- represents the divergence in relative agreement wrt f s for target noun w , relative to the mean relative agreement wrt f s over all words ',Carletta,'det_statistic_The nsubj_recast_ cop_recast_is ccomp_statistic_recast advmod_recast_as appos_recast_fs dep_fs_w appos_recast_sys dep_sys_sys dep_recast_= dobj_=_agr appos_agr_fs dep_fs_w nn_P_sys dep_P_sys nsubj_agr_P dep_agr_agr dep_P_fs appos_P_sys dep_sys_sys nn_P_N dep_agr_P partmod_P_agr dobj_agr_fs appos_agr_sys dep_sys_sys nsubj_represents_N prep_N_In det_form_this amod_form_modified pobj_In_form appos_form_fs dep_fs_w rcmod_agr_represents det_divergence_the dobj_represents_divergence prep_divergence_in amod_wrt_relative nn_wrt_agreement pobj_in_wrt dep_for_f dep_for_s dep_statistic_for nn_w_target nn_w_noun pobj_for_w dep_for_relative dep_for_to det_wrt_the amod_wrt_mean amod_wrt_relative nn_wrt_agreement pobj_to_wrt nsubj_s_f rcmod_wrt_s prep_s_over det_words_all pobj_over_words ' +o,5852,'Clark reports results on a corpus containing 12 million terms , Schcurrency1utze on one containing 25 million terms , and on one containing 365 million terms ',Brown,'nn__Clark nsubj_reports_ dobj_reports_results prep_reports_on det_corpus_a pobj_on_corpus partmod_corpus_containing number_million_12 num_terms_million dobj_containing_terms nn__Schcurrency1utze nsubj_containing_ prep__on pobj_on_one conj_reports_containing number_million_25 num_terms_million dobj_containing_terms cc_reports_and nsubj_containing_ prep__on pobj_on_one conj_reports_containing number_million_365 num_terms_million dobj_containing_terms ' +o,5853,'They give a probabilistic formation of paraphrasing which naturally falls out of the fact that they use techniques from phrase-based statistical machine translation: e2 = argmax e2:e2negationslash=e1 p(e2|e1) (1) where p(e2|e1) = summationdisplay f p(f|e1)p(e2|f,e1) (2) summationdisplay f p(f|e1)p(e2|f) (3) Phrase translation probabilities p(f|e1) and p(e2|f) are commonly calculated using maximum likelihood estimation (Koehn et al., 2003): p(f|e) = count(e,f)summationtext f count(e,f) (4) where the counts are collected by enumerating all bilingual phrase pairs that are consistent with the 197 conseguido .opportunitiesequalcreatetofailedhasprojecteuropeanthe oportunidadesdeigualdadlahanoeuropeoproyectoel Figure 1: The interaction of the phrase extraction heuristic with unaligned English words means that the Spanish phrase la igualdad aligns with equal, create equal, and to create equal.',Koehn,'' +o,5854,'However , in yet unpublished work we found that at least for the computation of synonyms and related words neither syntactical analysis nor singular value decomposition lead to significantly better results than the approach described here when applied to the monolingual case , so we did not try to include these methods in our system ',Grefenstette,'advmod_try_However prep_try_in advmod_work_yet amod_work_unpublished pobj_in_work nsubj_found_we rcmod_work_found complm_lead_that dep_for_at dep_at_least prep_lead_for det_computation_the pobj_for_computation prep_computation_of amod_words_synonyms cc_synonyms_and conj_synonyms_related pobj_of_words preconj_analysis_neither amod_analysis_syntactical nsubj_lead_analysis cc_analysis_nor nn_decomposition_singular nn_decomposition_value conj_analysis_decomposition ccomp_found_lead prep_lead_to advmod_better_significantly amod_results_better pobj_to_results mark_described_than det_approach_the nsubj_described_approach advcl_lead_described advmod_described_here advmod_applied_when advcl_described_applied prep_applied_to det__the amod__monolingual nn__case pobj_to_ advmod_we_so nsubj_try_we aux_try_did neg_try_not aux_include_to xcomp_try_include det_methods_these dobj_include_methods prep_include_in poss_system_our pobj_in_system ' +o,5855,'These range from twoword to multi-word , with or without syntactic structure ',Smadja,'det_range_These prep_range_from pobj_from_twoword prep_range_to pobj_to_multi-word conj_to_with cc_to_or conj_to_without amod_structure_syntactic pobj_without_structure ' +o,5856,'HockenmaierandSteedman showedthat a CCG corpus could be created by adapting the Penn Treebank ',Marcus,'nn__HockenmaierandSteedman nsubj_showedthat_ det_corpus_a nn_corpus_CCG nsubjpass_created_corpus aux_created_could auxpass_created_be ccomp_showedthat_created prep_created_by pcomp_by_adapting det_Treebank_the nn_Treebank_Penn dobj_adapting_Treebank ' +o,5857,'2 Prior Work Statistical machine translation , as pioneered by IBM , is grounded in the noisy channel model ',Marcus,'num_translation_2 amod_translation_Prior nn_translation_Work nn_translation_Statistical nn_translation_machine nsubjpass_grounded_translation mark_pioneered_as dep_grounded_pioneered prep_pioneered_by nn__IBM pobj_by_ auxpass_grounded_is prep_grounded_in det_model_the amod_model_noisy nn_model_channel pobj_in_model ' +o,5858,'The notation will assume ChineseEnglish word alignment and ChineseEnglish MT Here we adopt a notation similar to ',Brown,'det_notation_The nsubj_assume_notation aux_assume_will amod_alignment_ChineseEnglish nn_alignment_word dobj_assume_alignment cc_alignment_and nn_MT_ChineseEnglish conj_alignment_MT advmod_adopt_Here nsubj_adopt_we dep_assume_adopt det_notation_a nsubj_similar_notation xcomp_adopt_similar prep_similar_to ' +n,5859,'However , to what extent that assumption holds is tested only on a small number of language pairs using hand aligned data ',Wellington,'dep_holds_However dep_holds_to pobj_to_what dep_what_extent det_assumption_that nsubj_holds_assumption auxpass_tested_is ccomp_holds_tested advmod_tested_only prep_tested_on det_number_a amod_number_small nsubj_using_number prep_number_of nn_pairs_language pobj_of_pairs pcomp_on_using dobj_using_hand partmod_hand_aligned dobj_aligned_data ' +o,5860,'The complexities of 15 restricted alignment problems in two very different synchronous grammar formalisms of syntax-based machine translation , inversion transduction grammars -LRB- ITGs -RRB- and a restricted form of range concatenation grammars -LRB- -LRB- 2,2 -RRB- - BRCGs -RRB- , are investigated ',Wu,'det_complexities_The nsubjpass_investigated_complexities prep_complexities_of num_problems_15 amod_problems_restricted nn_problems_alignment pobj_of_problems prep_problems_in num_formalisms_two advmod_different_very amod_formalisms_different amod_formalisms_synchronous nn_formalisms_grammar pobj_in_formalisms prep_complexities_of amod_translation_syntax-based nn_translation_machine pobj_of_translation nn_grammars_inversion nn_grammars_transduction appos_translation_grammars abbrev_translation_ITGs dep_translation_ cc__and det_form_a amod_form_restricted conj__form prep_complexities_of nn_grammars_range nn_grammars_concatenation pobj_of_grammars abbrev_complexities_2,2 dep_2,2_BRCGs dep_complexities_ auxpass_investigated_are ' +o,5861,'In all other respects , our work departs from previous research on broad -- coverage 16 I t I I I I I i ! I i I I I I I I I I I I I i I 1 , I I I I I i I 1 I I I I probabilistic parsing , which either attempts to learn to predict gr ~ rarn ~ tical structure of test data directly from a training treebank , or employs a grammar and sometimes a dictionary to capture linguistic expertise directly , but arguably at a less detailed and informative level than in the research reported here ',Marcus,'prep_departs_In det_respects_all amod_respects_other pobj_In_respects poss_work_our nsubj_departs_work prep_departs_from amod_research_previous pobj_from_research prep_departs_on amod_I_broad nn_t_coverage num_t_16 nn_t_I dep_--_t dep_i_I dep_i_I dep_i_I dep_i_I dep_i_I amod_t_i pobj_on_I partmod_I_i iobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_i dobj_i_I num_I_1 num_I_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_i dobj_i_I dobj_i_1 dobj_i_I dobj_i_I dobj_i_I dobj_i_I amod_parsing_probabilistic dep_from_parsing nsubj_attempts_which advmod_attempts_either rcmod_parsing_attempts aux_learn_to xcomp_attempts_learn aux_predict_to xcomp_learn_predict nn_structure_gr nn_structure_~ nn_structure_rarn nn_structure_~ nn_structure_tical dobj_predict_structure prep_structure_of nn_data_test pobj_of_data advmod_predict_directly prep_i_from det__a nn__training nn__treebank pobj_from_ cc_departs_or conj_departs_employs det_grammar_a dobj_employs_grammar cc_grammar_and advmod_grammar_sometimes det_dictionary_a conj_grammar_dictionary aux_capture_to xcomp_employs_capture amod_expertise_linguistic dobj_capture_expertise advmod__directly amod_expertise_ cc__but conj__arguably prep_arguably_at det_level_a advmod_detailed_less amod_level_detailed cc_detailed_and conj_detailed_informative pobj_at_level dep_in_than prep_capture_in det_research_the pobj_in_research partmod_research_reported advmod_reported_here ' +o,5862,'The algorithm is slightly different from other online training algorithms in that we keep and update oracle translations , which is a set of good translations reachable by a decoder according to a metric , ie BLEU ',Liang,'det_algorithm_The nsubj_different_algorithm cop_different_is advmod_different_slightly prep_different_from amod_algorithms_other amod_algorithms_online nn_algorithms_training pobj_from_algorithms partmod_algorithms_ prep__in complm_keep_that nsubj_keep_we pcomp_in_keep cc_keep_and conj_keep_update amod_translations_oracle dobj_keep_translations nsubj_set_which cop_set_is det_set_a rcmod_algorithms_set prep_set_of amod_translations_good pobj_of_translations partmod_translations_reachable prep_reachable_by det_decoder_a pobj_by_decoder prep_reachable_according dep_according_to det__a amod__metric amod__ie nn__BLEU pobj_to_ ' +o,5863,'For instance , several studies have shown that BLEU correlates with human ratings on machine translation quality ',Papineni,'prep_shown_For pobj_For_instance amod_studies_several nsubj_shown_studies aux_shown_have complm_correlates_that nsubj_correlates_BLEU ccomp_shown_correlates prep_correlates_with amod_ratings_human pobj_with_ratings prep_ratings_on nn_quality_machine nn_quality_translation pobj_on_quality ' +o,5864,'The SemEval-2010 task we present here builds on thework ofNakov , where NCs are paraphrased by combinations of verbs and prepositions ',Nakov,'det_task_The amod_task_SemEval-2010 nsubj_builds_we advmod_builds_present advmod_builds_here dep_task_builds prep_builds_on nn__thework nn__ofNakov pobj_on_ advmod_paraphrased_where nsubjpass_paraphrased_NCs auxpass_paraphrased_are rcmod__paraphrased prep_paraphrased_by pobj_by_combinations prep_combinations_of pobj_of_verbs cc_verbs_and conj_verbs_prepositions ' +o,5865,'In order to build models that perform well in new -LRB- target -RRB- domains we usually find two settings ','Daume III','mark_build_In dep_build_order aux_build_to dep_perform_build dobj_build_models nsubj_perform_that advmod_perform_well prep_perform_in amod_domains_new appos_domains_target pobj_in_domains nsubj_find_we advmod_find_usually dep_perform_find num_settings_two dobj_find_settings ' +o,5866,' , Tillmann -LRB- 2003 -RRB- , and Vogel et al ',Koehn,'nn_2003_Tillmann conj__2003 cc__and conj__Vogel cc_Vogel_et conj_Vogel_al ' +n,5867,'Although evaluated on a different test set , our method also outperforms the correlation with human scores reported in ',Liu,'mark_evaluated_Although advcl_outperforms_evaluated prep_evaluated_on det_set_a amod_set_different nn_set_test pobj_on_set poss_method_our nsubj_outperforms_method advmod_outperforms_also det_correlation_the dobj_outperforms_correlation prep_outperforms_with amod_scores_human pobj_with_scores partmod_scores_reported prep_reported_in ' +o,5868,'(Lin, 2004; Lin and Och, 2004).',Lin,'' +o,5869,'The two annotators agreed on the annotations of 385\\/453 turns , achieving 8499 \% agreement -LRB- Kappa = 068 -RRB- ',Carletta,'det_annotators_The num_annotators_two partmod_annotators_agreed prep_agreed_on det_annotations_the pobj_on_annotations prep_annotations_of num_turns_385\\/453 pobj_of_turns xcomp_agreed_achieving num_\%_8499 dobj_achieving_\% dep_annotators_agreement nsubj__Kappa dep__= num__068 dep_agreement_ ' +o,5870,'So , we pre-tagged the input to the Bikel parser using the MXPOST tagger ',Ratnaparkhi,'advmod_pre-tagged_So nsubj_pre-tagged_we det_input_the dobj_pre-tagged_input prep_pre-tagged_to det_parser_the nn_parser_Bikel pobj_to_parser xcomp_pre-tagged_using det_tagger_the amod_tagger_MXPOST dobj_using_tagger ' +p,5871,'Several recent real-world parsers have improved state-of-the-art parsing accuracy by relying on probabilistic or weighted versions of bilexical grammars ',Collins,'amod_parsers_Several amod_parsers_recent amod_parsers_real-world nsubj_improved_parsers aux_improved_have amod_accuracy_state-of-the-art nn_accuracy_parsing dobj_improved_accuracy prep_improved_by pcomp_by_relying prep_relying_on amod_versions_probabilistic cc_probabilistic_or conj_probabilistic_weighted pobj_on_versions prep_versions_of amod_grammars_bilexical pobj_of_grammars ' +p,5872,' has described an ef cient exact one-dimensional error minimization technique for a similar search problem in machine translation ',Och,'nsubj_described_ aux_described_has det_technique_an nn_technique_ef nn_technique_cient amod_technique_exact amod_technique_one-dimensional nn_technique_error nn_technique_minimization dobj_described_technique prep_described_for det_problem_a amod_problem_similar nn_problem_search pobj_for_problem prep_problem_in nn_translation_machine pobj_in_translation ' +p,5873,' do not achieve higher BLEU scores , but do score better according to human grammaticality judgments for in-coverage cases ',Riezler,'nsubj_achieve_ aux_achieve_do neg_achieve_not amod_scores_higher nn_scores_BLEU dobj_achieve_scores cc_achieve_but conj_achieve_do dep_better_score dobj_do_better prep_do_according dep_according_to amod_judgments_human nn_judgments_grammaticality pobj_to_judgments prep_judgments_for amod_cases_in-coverage pobj_for_cases ' +o,5874,'For instance , some approaches coarsely discriminate between biographical and non-biographical information , whileothersgobeyondbinary distinction by identifying atomic events eg , occupation and marital status that are typically included in a biography ',Biadsy,'prep_approaches_For pobj_For_instance nsubj_approaches_some advmod_discriminate_coarsely amod_eg_discriminate prep_discriminate_between amod_information_biographical cc_biographical_and conj_biographical_non-biographical pobj_between_information amod_eg_ advmod_distinction_whileothersgobeyondbinary amod_eg_distinction prep_distinction_by pcomp_by_identifying amod_events_atomic dobj_identifying_events dobj_approaches_eg conj_eg_occupation cc_eg_and amod_status_marital conj_eg_status nsubjpass_included_that auxpass_included_are advmod_included_typically rcmod_status_included prep_included_in det__a nn__biography pobj_in_ ' +o,5875,'One of the popular statistical machine translation paradigms is the phrase-based model -LRB- PBSMT -RRB- ',Och,'nsubj_model_One prep_One_of det_paradigms_the amod_paradigms_popular amod_paradigms_statistical nn_paradigms_machine nn_paradigms_translation pobj_of_paradigms cop_model_is det_model_the amod_model_phrase-based abbrev_model_PBSMT ' +o,5876,'The assumptions we made were the following: a lexical token in one half of the translation unit (TU) corresponds to at most one non-empty lexical unit in the other half of the TU; this is the 1:1 mapping assumption which underlines the work of many other researchers (Ahrenberg et al (2000), Brew and McKelvie (1996), Hiemstra (1996), Kay and Rscheisen (1993), Tiedmann (1998), Melamed (2001) etc); a polysemous lexical token, if used several times in the same TU, is used with the same meaning; this assumption is explicitly used by Gale and Church (1991), Melamed (2001) and implicitly by all the previously mentioned authors; a lexical token in one part of a TU can be aligned to a lexical token in the other part of the TU only if the two tokens have compatible types (part-of-speech); in most cases, compatibility reduces to the same POS, but it is also possible to define other compatibility mappings (e.g. participles or gerunds in English are quite often translated as adjectives or nouns in Romanian and vice-versa); although the word order is not an invariant of translation, it is not random either (Ahrenberg et al (2000)); when two or more candidate translation pairs are equally scored, the one containing tokens which are closer in relative position are preferred.',Brown,'' +o,5877,'As with conventional smoothing methods , triangulation increases the robustness of phrase translation estimates ',Koehn,'prep_increases_As dep_As_with amod__conventional amod__smoothing nn__methods pobj_with_ nsubj_increases_triangulation det_robustness_the dobj_increases_robustness prep_robustness_of nn_estimates_phrase nn_estimates_translation pobj_of_estimates ' +o,5878,'More specifically , the latter system uses the IBM-1 lexical parameters for computing the translation probabilities of two possible new tuples : the one resulting when the null-aligned-word is attached to Table 6 Evaluation results for experiments on n-gram size incidence ',Brown,'advmod_specifically_More advmod_uses_specifically det_system_the amod_system_latter nsubj_uses_system det__the nn__IBM-1 amod__lexical nn__parameters dobj_uses_ prep_uses_for pcomp_for_computing det_probabilities_the nn_probabilities_translation dobj_computing_probabilities prep_probabilities_of num_tuples_two amod_tuples_possible amod_tuples_new pobj_of_tuples det_one_the dep_probabilities_one partmod_one_resulting advmod_attached_when det_null-aligned-word_the nsubjpass_attached_null-aligned-word auxpass_attached_is advcl_resulting_attached aux_Table_to xcomp_attached_Table num_results_6 nn_results_Evaluation dobj_Table_results prep_Table_for pobj_for_experiments prep_experiments_on amod_incidence_n-gram nn_incidence_size pobj_on_incidence ' +p,5879,'Although they obtained consistent and stable performance gains for MT , these were inferior to the gains yielded by Ochs procedure in ',Och,'mark_obtained_Although nsubj_obtained_they advcl_inferior_obtained amod_gains_consistent cc_consistent_and conj_consistent_stable nn_gains_performance dobj_obtained_gains prep_obtained_for pobj_for_MT nsubj_inferior_these cop_inferior_were prep_inferior_to det_gains_the pobj_to_gains partmod_gains_yielded prep_yielded_by nn_procedure_Ochs pobj_by_procedure prep_yielded_in ' +n,5880,'However , it seems unrealistic to expect a one-size-fits-all approach to be achieve uniformly high performance across varied languages , and , in fact , it doesnt Though the system presented in outperforms the best systems in the 2006 PASCAL challenge for Turkish and Finnish , it still does significantly worse on these languages than English -LRB- F-scores of 662 and 665 , compared to 794 -RRB- ',Dasgupta,'advmod_unrealistic_However nsubj_unrealistic_it cop_unrealistic_seems aux_expect_to xcomp_unrealistic_expect det_approach_a amod_approach_one-size-fits-all dobj_expect_approach aux_achieve_to aux_achieve_be xcomp_expect_achieve advmod_achieve_uniformly amod_performance_high dep_uniformly_performance prep_achieve_across amod_languages_varied pobj_across_languages cc_unrealistic_and prep_doesnt_in pobj_in_fact nsubj_doesnt_it conj_unrealistic_doesnt mark_outperforms_Though det_system_the nsubj_outperforms_system partmod_system_presented prep_presented_in pobj_in_ advcl_worse_outperforms det_systems_the amod_systems_best dobj_outperforms_systems prep_systems_in det_challenge_the num_challenge_2006 amod_challenge_PASCAL pobj_in_challenge prep_outperforms_for pobj_for_Turkish cc_Turkish_and conj_Turkish_Finnish nsubj_worse_it advmod_worse_still dep_worse_does advmod_worse_significantly ccomp_doesnt_worse prep_worse_on det_languages_these pobj_on_languages prep_languages_than pobj_than_English dep_worse_F-scores prep_F-scores_of pobj_of_662 cc_662_and conj_662_665 dep_F-scores_compared dep_compared_to pobj_to_794 ' +o,5881,'Step 3 -RRB- Answer Extraction : We select the top 5 ranked sentences and return them as , 1997 -RRB- , can be used to capture the binary dependencies between the head of each phrase ',Collins,'nn_Extraction_Step nn_Extraction_3 nn_Extraction_Answer nsubj_select_We dep_Extraction_select det_5_the amod_5_top nsubjpass_ranked_5 ccomp_select_ranked dobj_ranked_sentences cc_ranked_and conj_ranked_return dobj_return_them prep_return_as dep_as_ appos__1997 aux_used_can auxpass_used_be conj_ranked_used aux_capture_to purpcl_used_capture det_dependencies_the amod_dependencies_binary dobj_capture_dependencies prep_dependencies_between det_head_the pobj_between_head prep_head_of det_phrase_each pobj_of_phrase ' +o,5882,'2 Summary of approaches Given a source language sentence f, statistical machine translation defines the translation task as selecting the most likely target translation e under a model P(e|f), i.e.: e(f) = argmax e P(e|f) = argmax e msummationdisplay i=1 hi(e,f)i where the argmax operation denotes a search through a structured space of translation ouputs in the target language, hi(e,f) are bilingual features of e and f and monolingual features of e, and weights i are trained discriminitively to maximize translation quality (based on automatic metrics) on held out data (Och, 2003).',Och,'' +p,5883,'2 Evaluation Metrics Currently , the most widely used automatic MT evaluation metric is the NIST BLEU-4 ',Papineni,'num_Metrics_2 nn_Metrics_Evaluation nsubj_BLEU-4_Metrics advmod_Metrics_Currently det_metric_the advmod_used_most advmod_used_widely amod_metric_used amod_metric_automatic nn_metric_MT nn_metric_evaluation appos_Metrics_metric cop_BLEU-4_is det_BLEU-4_the amod_BLEU-4_NIST ' +o,5884,'On the other end of the spectrum , character-based bitext mapping algorithms are limited to language pairs where cognates are common ; in addition , they may easily be misled by superficial differences in formatting and page layout and must sacrifice precision to be computationally tractable ',Brown,'prep_limited_On det_end_the amod_end_other pobj_On_end prep_end_of det_spectrum_the pobj_of_spectrum amod_algorithms_character-based nn_algorithms_bitext nn_algorithms_mapping nsubjpass_limited_algorithms advmod_limited_ auxpass_limited_are prep_limited_to nn_pairs_language pobj_to_pairs advmod_common_where nsubj_common_cognates cop_common_are advcl_limited_common prep_misled_in pobj_in_addition nsubjpass_misled_they aux_misled_may advmod_misled_easily auxpass_misled_be parataxis_limited_misled prep_misled_by amod_differences_superficial pobj_by_differences prep_differences_in nn_layout_formatting cc_formatting_and conj_formatting_page pobj_in_layout cc_misled_and aux_sacrifice_must conj_misled_sacrifice dobj_sacrifice_precision aux_tractable_to cop_tractable_be advmod_tractable_computationally xcomp_sacrifice_tractable ' +p,5885,'Hiero Search Refinements offer several refinements to cube pruning to improve translation speed ',Huang,'nn__Hiero nn__Search nn__Refinements nsubj_offer_ amod_refinements_several dobj_offer_refinements prep_offer_to nn_pruning_cube pobj_to_pruning aux_improve_to xcomp_offer_improve nn_speed_translation dobj_improve_speed ' +o,5886,'Freund and Schapire originally proposed the averaged parameter method ; it was shown to give substantial improvements in accuracy for tagging tasks in ',Collins,'nn__Freund cc_Freund_and conj_Freund_Schapire nsubj_proposed_ advmod_proposed_originally det_method_the amod_method_averaged nn_method_parameter dobj_proposed_method nsubjpass_shown_it auxpass_shown_was parataxis_proposed_shown aux_give_to xcomp_shown_give amod_improvements_substantial dobj_give_improvements prep_improvements_in pobj_in_accuracy prep_give_for pcomp_for_tagging dobj_tagging_tasks prep_tagging_in ' +o,5887,'Hw6 : Implement beam search and reduplicate the POS tagger described in ',Ratnaparkhi,'amod_search_Implement nn_search_beam dep_Hw6_search cc_search_and conj_search_reduplicate det_tagger_the dep_tagger_POS nsubj_described_tagger rcmod_search_described prt_described_in ' +o,5888,'1 Introduction Supervised statistical parsers attempt to capture patterns of syntactic structure from a labeled set of examples for the purpose of annotating new sentences with their structure ',Titov,'num_parsers_1 nn_parsers_Introduction nn_parsers_Supervised amod_parsers_statistical nsubj_attempt_parsers aux_capture_to xcomp_attempt_capture dobj_capture_patterns prep_patterns_of amod_structure_syntactic pobj_of_structure prep_capture_from det_set_a amod_set_labeled pobj_from_set prep_set_of pobj_of_examples prep_examples_for det_purpose_the pobj_for_purpose prep_purpose_of pcomp_of_annotating amod_sentences_new dobj_annotating_sentences prep_annotating_with poss_structure_their pobj_with_structure ' +o,5889,'Thus , we can compute the source dependency LM score in the same way we compute the target side score , using a procedure described in ',He,'advmod_compute_Thus nsubj_compute_we aux_compute_can det_score_the nn_score_source nn_score_dependency nn_score_LM dobj_compute_score prep_compute_in det_way_the amod_way_same pobj_in_way nsubj_compute_we dep_compute_compute det_score_the nn_score_target nn_score_side dobj_compute_score xcomp_compute_using det_procedure_a dobj_using_procedure partmod_procedure_described prep_described_in pobj_in_ ' +o,5890,'A number of other re532 searchers have described previous work on preprocessing methods ',Berger,'det_number_A nsubj_described_number prep_number_of amod__other amod__re532 nn__searchers pobj_of_ aux_described_have amod_work_previous dobj_described_work prep_described_on amod_methods_preprocessing pobj_on_methods ' +o,5891,'For our studies here , the parser employed was that of applied to the sentences of the British National Corpus ',Collins,'prep_that_For poss_studies_our pobj_For_studies advmod_that_here det_parser_the nsubj_that_parser partmod_parser_employed cop_that_was prep_that_of pobj_of_ partmod__applied prep_applied_to det_sentences_the pobj_to_sentences prep_sentences_of det__the nn__British nn__National nn__Corpus pobj_of_ ' +o,5892,'The most commonly used automatic evaluation metrics , BLEU and NIST , are based on the assumption that The closer a machine translation is to a professional human translation , the better it is ',Papineni,'det_metrics_The advmod_used_most advmod_used_commonly amod_metrics_used amod_metrics_automatic nn_metrics_evaluation nsubjpass_based_metrics nn__BLEU appos_metrics_ cc__and nn__NIST conj__ auxpass_based_are prep_based_on det_assumption_the pobj_on_assumption complm_is_that dep_closer_The dep_is_closer det_translation_a nn_translation_machine nsubj_is_translation dep_is_is prep_is_to det_translation_a amod_translation_professional amod_translation_human pobj_to_translation dep_better_the dep_is_better nsubj_is_it ccomp_based_is ' +o,5893,'As with many dependency parsers , we handle non-projective -LRB- ie crossing -RRB- arcs by transforming them into noncrossing arcs with augmented labels1 Because our syntactic derivations are equivalent to those of , we use their HEAD methods to projectivise the syntactic dependencies ',Titov,'prep_handle_As dep_As_with amod__many amod__dependency nn__parsers pobj_with_ nsubj_handle_we amod_arcs_non-projective number_crossing_ie dep_arcs_crossing dobj_handle_arcs prep_handle_by pcomp_by_transforming dobj_transforming_them prep_transforming_into pcomp_into_noncrossing dobj_noncrossing_arcs prep_noncrossing_with amod_labels1_augmented pobj_with_labels1 mark_equivalent_Because poss_derivations_our amod_derivations_syntactic nsubj_equivalent_derivations cop_equivalent_are advcl_use_equivalent prep_equivalent_to pobj_to_those prep_those_of pobj_of_ nsubj_use_we dep_handle_use poss_methods_their amod_methods_HEAD dobj_use_methods aux_projectivise_to xcomp_use_projectivise det_dependencies_the amod_dependencies_syntactic dobj_projectivise_dependencies ' +o,5894,'4 Semi-Supervised Training for Word Alignments Intuitively , in approximate EM training for Model 4 , the E-step corresponds to calculating the probability of all alignments according to the current model estimate , while the M-step is the creation of a new model estimate given a probability distribution over alignments -LRB- calculated in the E-step -RRB- ',Brown,'num_Training_4 amod_Training_Semi-Supervised nsubj_corresponds_Training prep_Training_for nn_Intuitively_Word nn_Intuitively_Alignments pobj_for_Intuitively prep_corresponds_in amod_training_approximate nn_training_EM pobj_in_training prep_training_for nn__Model num__4 pobj_for_ det_E-step_the nsubj_corresponds_E-step prep_corresponds_to pcomp_to_calculating det_probability_the dobj_calculating_probability prep_probability_of det_alignments_all pobj_of_alignments prep_calculating_according dep_according_to det_estimate_the amod_estimate_current nn_estimate_model pobj_to_estimate mark_creation_while det_M-step_the nsubj_creation_M-step cop_creation_is det_creation_the dep_estimate_creation prep_creation_of det_estimate_a amod_estimate_new nn_estimate_model pobj_of_estimate partmod_estimate_given det_probability_a iobj_given_probability dobj_given_distribution prep_distribution_over pobj_over_alignments dep_alignments_calculated prep_calculated_in det_E-step_the pobj_in_E-step ' +o,5895,'The kappa statistic has become the de facto standard to assess inter-annotator agreement ',Carletta,'det__The nn__kappa nn__statistic nsubj_standard_ aux_standard_has cop_standard_become det_standard_the amod_standard_de dep_de_facto aux_assess_to xcomp_standard_assess amod_agreement_inter-annotator dobj_assess_agreement ' +o,5896,'However , they make different types of errors , which can be seen as a reflection of their theoretical differences ',McDonald,'advmod_make_However nsubj_make_they amod_types_different dobj_make_types prep_types_of pobj_of_errors nsubjpass_seen_which aux_seen_can auxpass_seen_be rcmod_errors_seen prep_seen_as det_reflection_a pobj_as_reflection prep_reflection_of poss_differences_their amod_differences_theoretical pobj_of_differences ' +o,5897,'We just assign these rules a constant score trained using our implementation of Minimum Error Rate Training , which is 07 in our system ',Och,'nsubj_assign_We advmod_assign_just det_rules_these dobj_assign_rules det_score_a amod_score_constant nsubj_trained_score dep_assign_trained xcomp_trained_using poss_implementation_our dobj_using_implementation prep_implementation_of nn__Minimum nn__Error nn__Rate nn__Training pobj_of_ nsubj_07_which cop_07_is rcmod__07 prep_07_in poss_system_our pobj_in_system ' +o,5898,'3 Monolingual comparable corpus : Similar to the methods in , we construct a corpus of comparable documents from a large corpus D of news articles ',Barzilay,'num_corpus_3 amod_corpus_Monolingual amod_corpus_comparable amod_articles_Similar prep_Similar_to det_methods_the pobj_to_methods prep_methods_in pobj_in_ nsubj_construct_we dep__construct det_corpus_a dobj_construct_corpus prep_corpus_of amod_documents_comparable pobj_of_documents prep_construct_from det_D_a amod_D_large nn_D_corpus pobj_from_D prep_D_of pobj_of_news dep_corpus_articles ' +o,5899,'We can stipulate the time line to be linearly ordered -LRB- although it is not in approaches that build ignorance of relative times into the representation of time nor in approaches using branching futures -RRB- , and we can stipulate it to be dense -LRB- although it is not in the situation calculus -RRB- ',Hobbs,'nsubj_stipulate_We aux_stipulate_can det_line_the nn_line_time nsubjpass_ordered_line aux_ordered_to auxpass_ordered_be advmod_ordered_linearly xcomp_stipulate_ordered mark_is_although nsubj_is_it parataxis_stipulate_is neg_is_not prep_is_in pobj_in_approaches nsubj_build_that rcmod_approaches_build dobj_build_ignorance prep_ignorance_of amod_times_relative pobj_of_times prep_build_into det_representation_the nn_approaches_representation prep_representation_of nn__time pobj_of_ cc_representation_nor conj_representation_in pobj_into_approaches partmod_approaches_using xcomp_using_branching nn__futures dobj_branching_ cc_stipulate_and nsubj_stipulate_we aux_stipulate_can conj_stipulate_stipulate nsubj_dense_it aux_dense_to cop_dense_be xcomp_stipulate_dense mark_is_although nsubj_is_it parataxis_dense_is neg_is_not prep_is_in det_calculus_the nn_calculus_situation pobj_in_calculus ' +o,5900,'In cases where the number of gold tags is different than the number of induced tags , some must necessarily remain unassigned ',Johnson,'prep_unassigned_In pobj_In_cases advmod_different_where det_number_the nsubj_different_number prep_number_of amod_tags_gold pobj_of_tags cop_different_is rcmod_cases_different prep_different_than det_number_the pobj_than_number prep_number_of amod_tags_induced pobj_of_tags nsubj_unassigned_some aux_unassigned_must advmod_unassigned_necessarily cop_unassigned_remain ' +o,5901,'Modulo more minor differences , these notions are close to the ideas of interpretation as abduction and generation as abduction -LRB- ltobbs et al -LRB- 1990:26 -28 -RRB- -RRB- , where we take abduction , in the former case for instance , to be a process returning a temporal-causal structure which can explain the utterance in context ',Hobbs,'dep_close_Modulo advmod_minor_more amod_differences_minor dobj_Modulo_differences det_notions_these nsubj_close_notions cop_close_are prep_close_to det_ideas_the pobj_to_ideas prep_ideas_of pobj_of_interpretation prep_interpretation_as nn__abduction pobj_as_ cc__and conj__generation prep_close_as pobj_as_abduction dep_abduction_ltobbs cc_ltobbs_et conj_ltobbs_al number_-28_1990:26 dep_abduction_-28 advmod_take_where nsubj_take_we rcmod_abduction_take dobj_take_abduction prep_take_in det_case_the amod_case_former pobj_in_case prep_case_for pobj_for_instance aux_process_to cop_process_be det_process_a xcomp_take_process partmod_process_returning det_structure_a amod_structure_temporal-causal dobj_returning_structure nsubj_explain_which aux_explain_can rcmod_structure_explain det_utterance_the dobj_explain_utterance prep_utterance_in pobj_in_context ' +o,5902,'Then P -LRB- eI1jfj1 -RRB- = summationtextaI 1 P -LRB- eI1 , aI1jfj1 -RRB- ',Brown,'advmod_P_Then nsubj_summationtextaI_eI1jfj1 dep_summationtextaI_= ccomp_P_summationtextaI num_P_1 nsubj_P_P appos_P_eI1 dep_eI1_aI1jfj1 ' +o,5903,'In the general language UPenn annotation efforts for the WSJ sections of the Penn Treebank , sentences are annotated with POS tags , parse trees , as well as discourse annotation from the Penn Discourse Treebank , while verbs and verb arguments are annotated with Propbank rolesets ',Marcus,'prep_annotated_In det_language_the amod_language_general pobj_In_language partmod_language_UPenn nn_efforts_annotation dobj_UPenn_efforts prep_UPenn_for det_sections_the nn_sections_WSJ pobj_for_sections prep_sections_of det__the nn__Penn nn__Treebank pobj_of_ nsubjpass_annotated_sentences auxpass_annotated_are prep_annotated_with dep_tags_POS pobj_with_tags amod_trees_parse appos_tags_trees dep_well_as cc_tags_well dep_well_as nn_annotation_discourse conj_tags_annotation prep_annotation_from det__the nn__Penn nn__Discourse nn__Treebank pobj_from_ mark_annotated_while nsubjpass_annotated_verbs cc_verbs_and nn_arguments_verb conj_verbs_arguments auxpass_annotated_are advcl_annotated_annotated prep_annotated_with nn__Propbank nn__rolesets pobj_with_ ' +o,5904,'These methods often involve using a statistic such as 2 or the log likelihood ratio to create a score to measure the strength of correlation between source and target words ',Dunning,'det_methods_These nsubj_involve_methods advmod_involve_often xcomp_involve_using det_statistic_a dobj_using_statistic dep_as_such prep_statistic_as num__2 pobj_as_ cc__or det_log_the conj__log nn_ratio_likelihood nsubj__ratio dep_involve_ aux_create_to xcomp__create det_score_a dobj_create_score aux_measure_to infmod_score_measure det_strength_the dobj_measure_strength prep_strength_of pobj_of_correlation prep_correlation_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,5905,'Some methods only extract paraphrase patternsusingnewsarticlesoncertaintopics , while some others need seeds as initial input ',Barzilay,'det_methods_Some nsubj_only_methods ccomp_only_extract amod_patternsusingnewsarticlesoncertaintopics_paraphrase dobj_extract_patternsusingnewsarticlesoncertaintopics advmod_extract_ mark_need_while det_others_some nsubj_need_others advcl_extract_need dobj_need_seeds prep_need_as amod__initial nn__input pobj_as_ ' +o,5906,'Again we used Mohammad and Hirsts method along with distributional measure to determine the distributional closeness of two thesaurus concepts ',Lin,'dep_Again_we partmod_we_used nn_method_Mohammad cc_Mohammad_and conj_Mohammad_Hirsts nn_method_ dobj_used_method advmod_method_along dep_along_with amod_measure_ amod_measure_distributional pobj_with_measure aux_determine_to xcomp_used_determine det_closeness_the amod_closeness_distributional dobj_determine_closeness prep_closeness_of num_concepts_two nn_concepts_thesaurus pobj_of_concepts ' +p,5907,'After a brief period following the introduction of generally accepted and widely used metrics, BLEU (Papineni et al., 2002) and NIST (Doddington, 2002), when it seemed that this persistent problem has finally been solved, the researchers active in the field of machine translation (MT) started to express their worries that although these metrics are simple, fast and able to provide consistent results for a particular system during its development, they are not sufficiently reliable for the comparison of different systems or different language pairs.',Papineni,'' +o,5908,'-LRB- Collins parser always predicts a flat NP for such configurations -RRB- ',Collins,'nn__Collins nn__parser nsubj_predicts_ advmod_predicts_always det_NP_a amod_NP_flat dobj_predicts_NP prep_NP_for amod_configurations_such pobj_for_configurations ' +o,5909,'Stage 2 processing is then free to assign to the compound any bracketing for which it 3The design of this level of Lucy is influenced by , which advocates a level of ` surfaey \' logical form with predicates close to actual English words and a structure similar to the syntactic structure of the sentence ',Hobbs,'num_processing_2 dobj_Stage_processing cop_free_is advmod_free_then dep_Stage_free aux_assign_to xcomp_free_assign prep_assign_to det_compound_the pobj_to_compound det_bracketing_any nsubj_Stage_bracketing rel_3The_for pobj_for_which nsubj_3The_it ccomp_bracketing_3The nsubjpass_influenced_design prep_design_of det_level_this pobj_of_level prep_level_of pobj_of_Lucy auxpass_influenced_is ccomp_3The_influenced prep_influenced_by pobj_by_ nsubj_advocates_which rcmod__advocates det_level_a dep_close_level prep_level_of amod_form_surfaey dep_surfaey_logical pobj_of_form prep_form_with pobj_with_predicates dep_advocates_close prep_close_to amod_words_actual amod_words_English pobj_to_words cc_close_and det_structure_a dep_similar_structure conj_close_similar prep_similar_to det_structure_the amod_structure_syntactic pobj_to_structure prep_structure_of det_sentence_the pobj_of_sentence ' +o,5910,'The feature weights are tuned by the modified Koehns MER trainer ',Och,'det_weights_The nn_weights_feature nsubjpass_tuned_weights auxpass_tuned_are prep_tuned_by det_trainer_the amod_trainer_modified nn_trainer_Koehns nn_trainer_MER amod_trainer_ pobj_by_trainer ' +o,5911,'SCL for Discriminative Parse Selection So far , pivot features on the word level were used ',Blitzer,'nsubjpass_used_SCL prep_SCL_for nn_Selection_Discriminative nn_Selection_Parse pobj_for_Selection advmod_far_So advmod_Selection_far amod_features_pivot nsubjpass_used_features prep_features_on det_level_the nn_level_word pobj_on_level auxpass_used_were ' +o,5912,'In particular , we adopt the approach of phrase-based statistical machine translation ',Koehn,'prep_adopt_In pobj_In_particular nsubj_adopt_we det_approach_the dobj_adopt_approach prep_approach_of amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_of_translation ' +o,5913,'This makes it suitable for discriminative SMT training , which is still a challenge for large parameter sets ',Liang,'nsubj_makes_This nsubj_suitable_it xcomp_makes_suitable prep_suitable_for amod_training_discriminative nn_training_SMT pobj_for_training nsubj_challenge_which cop_challenge_is advmod_challenge_still det_challenge_a rcmod_training_challenge prep_challenge_for amod_sets_large nn_sets_parameter pobj_for_sets ' +o,5914,'1409 cally , and experimentally ',Wu,'num_cally_1409 cc_cally_and conj_cally_experimentally ' +o,5915,'Unfortunately , as was shown by Fraser and Marcu AER can have weak correlation with translation performance as measured by BLEU score , when the alignments are used to train a phrase-based translation system ',Papineni,'dep_have_Unfortunately mark_have_as auxpass_shown_was csubj_have_shown prep_shown_by pobj_by_Fraser cc_Fraser_and nn_AER_Marcu nn_AER_ conj_Fraser_AER aux_have_can amod_correlation_weak dobj_have_correlation prep_have_with nn_performance_translation pobj_with_performance mark_measured_as advcl_have_measured prep_measured_by amod__BLEU nn__score pobj_by_ advmod_used_when det_alignments_the nsubjpass_used_alignments auxpass_used_are dep__used aux_train_to xcomp_used_train det_system_a amod_system_phrase-based nn_system_translation dobj_train_system ' +o,5916,'Decoding Conditions For tuning of the decoder \'s parameters , minimum error training with respect to the BLEU score using was conducted using the respective development corpus ',Och,'nn_Conditions_Decoding prep__For pcomp_For_tuning prep_tuning_of det_decoder_the poss_parameters_decoder possessive_decoder_\'s pobj_of_parameters amod_training_minimum nn_training_error nsubj__training dep_Conditions_ prep__with pobj_with_respect prep__to det_score_the amod_score_BLEU pobj_to_score csubjpass_conducted_using auxpass_conducted_was dep_Conditions_conducted xcomp_conducted_using det_corpus_the amod_corpus_respective nn_corpus_development dobj_using_corpus ' +o,5917,'We perform a statistical analysis that provides information that complements the information provided by Cohen \'s Kappa ',Carletta,'nsubj_perform_We det_analysis_a amod_analysis_statistical dobj_perform_analysis nsubj_provides_that rcmod_analysis_provides dobj_provides_information nsubj_complements_that rcmod_information_complements det_information_the dobj_complements_information prep_complements_provided dep_provided_by poss_Kappa_Cohen possessive_Cohen_\'s pobj_by_Kappa ' +p,5918,'However , in , the authors investigate minimum translation units -LRB- MTU -RRB- which is a refinement over a similar approach by to eliminate the overlap issue ',Quirk,'advmod_investigate_However prep_investigate_in pobj_in_ det_authors_the nsubj_investigate_authors amod_units_minimum nn_units_translation dobj_investigate_units abbrev_units_MTU nsubj_refinement_which cop_refinement_is det_refinement_a rcmod_units_refinement prep_refinement_over det_approach_a amod_approach_similar pobj_over_approach prep_approach_by pcomp_by_ aux_eliminate_to xcomp__eliminate det_issue_the amod_issue_overlap dobj_eliminate_issue ' +o,5919,'We would expect better performance with the more accurate approximation based on variational inference proposed and evaluated in ',Titov,'nsubj_expect_We aux_expect_would amod_performance_better dobj_expect_performance prep_expect_with det_approximation_the advmod_accurate_more amod_approximation_accurate pobj_with_approximation partmod_approximation_based prep_based_on amod_inference_variational pobj_on_inference partmod_inference_proposed cc_proposed_and conj_proposed_evaluated prep_proposed_in ' +o,5920,'The fertility for the null word is treated specially ',Brown,'det_fertility_The nsubjpass_treated_fertility prep_fertility_for det_word_the amod_word_null pobj_for_word auxpass_treated_is advmod_treated_specially ' +p,5921,'Because of this property , vector space models have been used successfully both in computational linguistics and in cognitive science ',Snow,'prep_used_Because dep_Because_of det_property_this pobj_of_property nn_models_vector nn_models_space nsubjpass_used_models aux_used_have auxpass_used_been advmod_used_successfully preconj_in_both prep_used_in amod__computational nn__linguistics pobj_in_ cc_in_and conj_in_in amod__cognitive nn__science pobj_in_ ' +o,5922,'(Berger et al. , 1996).',Berger,'' +o,5923,'The rules extracted from the training bitext have the following features: a114 P( | )andP( | ), the latter of which is not found in the noisy-channel model, but has been previously found to be a helpful feature (Och and Ney 2002); 210 Chiang Hierarchical Phrase-Based Translation a114 the lexical weights P w ( | )andP w ( | ), which estimate how well the words in translate the words in (Koehn, Och, and Marcu 2003); 4 a114 a penalty exp(1) for extracted rules, analogous to Koehns phrase penalty (Koehn 2003), which allows the model to learn a preference for longer or shorter derivations.',Och,'' +o,5924,'1 Introduction Recent work in machine translation has evolved from the traditional word and phrase based models to include hierarchical phrase models and bilingual synchronous grammars ',Brown,'num_work_1 amod_work_Introduction amod_work_Recent nsubj_evolved_work prep_work_in nn_translation_machine pobj_in_translation aux_evolved_has prep_evolved_from det__the amod__traditional nn__word pobj_from_ cc__and conj__phrase partmod__based amod_models_ dobj_based_models aux_include_to xcomp_based_include amod_models_hierarchical nn_models_phrase dep__models dep_include_ cc__and amod_grammars_bilingual amod_grammars_synchronous dep__grammars conj__ ' +o,5925,'The traditional estimation method for word 98 alignment models is the EM algorithm which iteratively updates parameters to maximize the likelihood of the data ',Brown,'det_method_The amod_method_traditional nn_method_estimation nsubj__method prep_method_for nn_models_word num_models_98 nn_models_alignment pobj_for_models cop__is det__the nn__EM nn__algorithm dobj_maximize_which amod_parameters_iteratively nn_parameters_updates nsubj_maximize_parameters aux_maximize_to infmod__maximize det_likelihood_the dobj_maximize_likelihood prep_likelihood_of det_data_the pobj_of_data ' +p,5926,'1 Introduction The field of machine translation has seen many advances in recent years , most notably the shift from word-based to phrasebased models which use token n-grams as translation units ',Koehn,'num_Introduction_1 det_field_The nsubj_seen_field prep_field_of nn_translation_machine pobj_of_translation aux_seen_has dep_Introduction_seen amod_advances_many dobj_seen_advances prep_seen_in amod_years_recent pobj_in_years advmod_notably_most advmod_years_notably det_shift_the dep_years_shift prep_shift_from amod__word-based pobj_from_ prep_seen_to amod_models_phrasebased pobj_to_models nsubj_use_which rcmod_models_use amod_n-grams_token dobj_use_n-grams prep_use_as nn_units_translation pobj_as_units ' +o,5927,'Such transformations are typically denoted as paraphrases in the literature , where a wealth of methods for their automatic acquisition were proposed ',Barzilay,'amod_transformations_Such nsubjpass_denoted_transformations auxpass_denoted_are advmod_denoted_typically prep_denoted_as pobj_as_paraphrases prep_paraphrases_in det_literature_the pobj_in_literature advmod_proposed_where det_wealth_a nsubjpass_proposed_wealth prep_wealth_of pobj_of_methods prep_methods_for poss_acquisition_their amod_acquisition_automatic pobj_for_acquisition auxpass_proposed_were rcmod_literature_proposed ' +n,5928,'For comparison purposes , three additional heuristically-induced alignments are generated for each system : -LRB- 1 -RRB- Intersection of both directions -LRB- Aligner -LRB- int -RRB- -RRB- ; -LRB- 2 -RRB- Union of both directions -LRB- Aligner -LRB- union -RRB- -RRB- ; and -LRB- 3 -RRB- The previously bestknown heuristic combination approach called growdiag-final -LRB- Aligner -LRB- gdf -RRB- -RRB- ',Koehn,'prep_generated_For nn_purposes_comparison pobj_For_purposes num_alignments_three amod_alignments_additional amod_alignments_heuristically-induced nsubjpass_generated_alignments auxpass_generated_are prep_generated_for det_system_each pobj_for_system dep_Intersection_1 dep_system_Intersection prep_Intersection_of det_directions_both pobj_of_directions dep_directions_Aligner appos_directions_int dep_Union_2 dep_directions_Union prep_Union_of det_directions_both pobj_of_directions dep_directions_Aligner appos_directions_union cc_directions_and dep_approach_3 det_approach_The advmod_bestknown_previously amod_approach_bestknown nn_approach_heuristic nn_approach_combination conj_directions_approach partmod_Union_called amod__growdiag-final dobj_called_ nn_gdf_Aligner dep_Intersection_gdf ' +o,5929,'Most of the early work in this area was based on postulating generative probability models of language that included parse structures ',Collins,'nsubjpass_based_Most prep_Most_of det_work_the amod_work_early pobj_of_work prep_work_in det_area_this pobj_in_area auxpass_based_was prep_based_on pcomp_on_postulating amod_models_generative nn_models_probability dobj_postulating_models prep_models_of pobj_of_language nsubj_included_that rcmod_models_included amod_structures_parse dobj_included_structures ' +o,5930,'Pure statistical machine translation mltst in principle recover the most probable alignment out of all possible alignments between the input and a translation ',Brown,'amod_mltst_Pure amod_mltst_statistical nn_mltst_machine nn_mltst_translation nn_mltst_ nsubj_recover_mltst prep_mltst_in pobj_in_principle det_alignment_the advmod_probable_most amod_alignment_probable dobj_recover_alignment dep_of_out prep_alignment_of det_alignments_all amod_alignments_possible pobj_of_alignments prep_alignments_between det_input_the pobj_between_input cc_input_and det_translation_a conj_input_translation ' +o,5931,'Method dev test Finkel et al , 2005 baseline CRF 8551 + non-local features 8686 Krishnan and Manning , 2006 baseline CRF 8529 + non-local features 8724 Table 5 : Summary of performance with POS\\/chunk tags by TagChunk ',Finkel,'nn_al_dev nn_al_test nn_al_Finkel nn_al_et dobj_Method_al num_CRF_2005 amod_CRF_ nn_CRF_baseline appos_al_CRF number_+_8551 num_features_+ amod_features_non-local dep_CRF_features nn_Krishnan_8686 nsubj_Method_Krishnan cc_Krishnan_and conj_Krishnan_Manning num_features_2006 amod_features_ amod_features_baseline nn_features_CRF num_features_8529 num_features_+ amod_features_non-local appos_Krishnan_features num_Table_8724 dep_features_Table num_Table_5 dep_features_Summary prep_Summary_of pobj_of_performance prep_performance_with nn_tags_POS\\/chunk pobj_with_tags prep_tags_by pobj_by_TagChunk ' +o,5932,'The subsequent construction of translation table was done in exactly the same way as explained 4 in ',Koehn,'det_construction_The amod_construction_subsequent nsubjpass_done_construction prep_construction_of nn_table_translation pobj_of_table auxpass_done_was prep_done_in advmod_way_exactly det_way_the amod_way_same pobj_in_way mark_explained_as dep_way_explained dobj_explained_4 prep_explained_in ' +o,5933,'16In fact , we have experimented with other tagger combinations and configurations as wellwith the TnT , MaxEnt and TreeTagger , with or without the Morce tagger in the pack ; see below for the winning combination ',Ratnaparkhi,'dep_experimented_16In dobj_16In_fact nsubj_experimented_we aux_experimented_have prep_experimented_with amod_combinations_other nn_combinations_tagger pobj_with_combinations cc_combinations_and conj_combinations_configurations mark_see_as csubj_see_wellwith det__the nn__TnT dobj_wellwith_ nn__MaxEnt appos__ cc__and nn__TreeTagger conj__ prep_wellwith_with cc_with_or conj_with_without det_tagger_the nn_tagger_Morce pobj_with_tagger prep_wellwith_in det_pack_the pobj_in_pack advcl_experimented_see advmod_see_below prep_see_for det_combination_the amod_combination_winning pobj_for_combination ' +o,5934,'The Attr cells summarize the performance of the 6 models on the wiki table that are based on attributional similarity only ',Turney,'det_cells_The nn_cells_Attr nsubj_summarize_cells det_performance_the dobj_summarize_performance prep_performance_of det_models_the num_models_6 pobj_of_models prep_summarize_on det_table_the nn_table_wiki pobj_on_table nsubjpass_based_that auxpass_based_are rcmod_table_based prep_based_on amod_similarity_attributional pobj_on_similarity advmod_based_only ' +o,5935,'Starting out with a chunking pipeline , which uses a classical combination of tagger and chunker , with the Stanford POS tagger , the YamCha chunker and the Stanford Named Entity Recognizer , the desire to use richer syntactic representations led to the development of a parsing pipeline , which uses Charniak and Johnsons reranking parser to assign POS tags and uses base NPs as chunk equivalents , while also providing syntactic trees that can be used by feature extractors ',Finkel,'prt_Starting_out prep_Starting_with det_pipeline_a amod_pipeline_chunking pobj_with_pipeline nsubj_uses_which rcmod_pipeline_uses det_combination_a amod_combination_classical dobj_uses_combination prep_combination_of pobj_of_tagger cc_tagger_and conj_tagger_chunker prep_uses_with det__the nn__Stanford dep__POS nn__tagger pobj_with_ det__the nn__YamCha nn__chunker conj__ cc__and det__the nn__Stanford nn__Named nn__Entity nn__Recognizer conj__ det_desire_the dobj_uses_desire aux_use_to infmod_desire_use amod_representations_richer amod_representations_syntactic dobj_use_representations dep_Starting_led prep_led_to det_development_the pobj_to_development prep_development_of det_pipeline_a amod_pipeline_parsing pobj_of_pipeline nsubj_uses_which rcmod_pipeline_uses nsubj_reranking_Charniak cc_Charniak_and conj_Charniak_Johnsons dep_uses_reranking nn__parser dobj_reranking_ aux_assign_to xcomp_reranking_assign dep_tags_POS dobj_assign_tags cc_uses_and conj_uses_uses nn_NPs_base nsubj_Starting_NPs prep_NPs_as nn_equivalents_chunk pobj_as_equivalents mark_providing_while advmod_providing_also dep_NPs_providing amod_trees_syntactic dobj_providing_trees nsubjpass_used_that aux_used_can auxpass_used_be rcmod_trees_used prep_used_by nn_extractors_feature pobj_by_extractors ' +o,5936,'2 Related Work The issue of MWE processing has attracted much attention from the Natural Language Processing -LRB- NLP -RRB- community , including Smadja , 1993 ; Dagan and Church , 1994 ; Daille , 1995 ; 1995 ; McEnery et al , 1997 ; , 1997 ; Michiels and Dufour , 1998 ; Maynard and Ananiadou , 2000 ; Merkel and Andersson , 2000 ; Piao and McEnery , 2001 ; Sag et al , 2001 ; Tanaka and Baldwin , 2003 ; Dias , 2003 ; Baldwin et al , 2003 ; Nivre and Nilsson , 2004 Pereira et al , ',Wu,'num_Work_2 amod_Work_Related det_issue_The nsubj_attracted_issue prep_issue_of nn_processing_MWE pobj_of_processing aux_attracted_has dep_Work_attracted amod_attention_much dobj_attracted_attention prep_attracted_from det_community_the nn_Processing_Natural nn_Processing_Language nn_community_Processing abbrev_Processing_NLP pobj_from_community prep_community_including pobj_including_Smadja appos_Smadja_1993 dep_Smadja_Dagan cc_Dagan_and conj_Dagan_Church appos_Dagan_1994 dep_Smadja_Daille appos_Daille_1995 dep_Smadja_1995 dep_1995_McEnery cc_McEnery_et conj_McEnery_al appos_McEnery_1997 dep_Smadja_ appos__1997 dep_Smadja_Michiels cc_Michiels_and conj_Michiels_Dufour appos_Michiels_1998 dep_Smadja_Maynard cc_Maynard_and conj_Maynard_Ananiadou appos_Maynard_2000 dep_Smadja_Merkel cc_Merkel_and conj_Merkel_Andersson appos_Merkel_2000 dep_Smadja_Piao cc_Piao_and conj_Piao_McEnery appos_Piao_2001 dep_Smadja_Sag cc_Sag_et conj_Sag_al appos_Sag_2001 dep_Smadja_Tanaka cc_Tanaka_and conj_Tanaka_Baldwin appos_Tanaka_2003 dep_Smadja_Dias appos_Dias_2003 dep_Smadja_Baldwin cc_Baldwin_et conj_Baldwin_al appos_Baldwin_2003 dep_Smadja_Nivre cc_Nivre_and conj_Nivre_Nilsson nn_Pereira_2004 appos_Smadja_Pereira cc_Pereira_et conj_Pereira_al ' +o,5937,'In the February 2004 version of the PropBank corpus , annotations are done on top of the Penn TreeBank II parse trees ',Marcus,'prep_done_In det_version_the nn_version_February num_version_2004 pobj_In_version prep_version_of det_corpus_the nn_corpus_PropBank pobj_of_corpus nsubjpass_done_annotations auxpass_done_are prep_done_on pobj_on_top prep_top_of det_trees_the nn_trees_Penn nn_trees_TreeBank nn_trees_II nn_trees_parse pobj_of_trees ' +o,5938,'We present two approaches to SMT-based query expansion , both of which are implemented in the framework of phrase-based SMT ',Koehn,'nsubj_present_We num_approaches_two dobj_present_approaches prep_present_to amod_expansion_SMT-based nn_expansion_query pobj_to_expansion dep_of_both nsubjpass_implemented_of pobj_of_which auxpass_implemented_are rcmod_expansion_implemented prep_implemented_in det_framework_the pobj_in_framework prep_framework_of amod_SMT_phrase-based pobj_of_SMT ' +o,5939,'Our work expands on the general approach taken by but arrives at insights similar to those of the most recent work , albeit in a completely different manner ',Zhang,'poss_work_Our nsubj_expands_work prep_expands_on det_approach_the amod_approach_general pobj_on_approach partmod_approach_taken prep_taken_by pobj_by_ cc_expands_but conj_expands_arrives prep_arrives_at pobj_at_insights amod_insights_similar prep_similar_to pobj_to_those prep_those_of det__the advmod__most amod__recent nn__work pobj_of_ dep_at_albeit prep_arrives_in det_manner_a advmod_different_completely amod_manner_different pobj_in_manner ' +o,5940,'For efficiency reasons we report results on sentences of length 30 words or less10 The syntax-based method gives a BLEU score of 2504 , a 046 BLEU point gain over Pharoah ',Papineni,'prep_report_For nn_reasons_efficiency pobj_For_reasons nsubj_report_we dobj_report_results prep_report_on pobj_on_sentences prep_sentences_of pobj_of_length num_words_30 nsubj_gives_words cc_words_or amod_method_less10 det_method_The amod_method_syntax-based conj_words_method dep_report_gives det__a amod__BLEU iobj_gives_ dobj_gives_score prep_score_of pobj_of_2504 det_gain_a number_BLEU_046 num_gain_BLEU nn_gain_point dobj_gives_gain prep_gain_over pobj_over_Pharoah ' +n,5941,'Brill \'s results demonstrate that this approach can outperform the Hidden Markov Model approaches that are frequently used for part-of-speech tagging , as well as showing promise for other applications ',Cutting,'poss_results_Brill possessive_Brill_\'s nsubj_demonstrate_results complm_outperform_that det_approach_this nsubj_outperform_approach aux_outperform_can ccomp_demonstrate_outperform det_approaches_the nn_approaches_Hidden nn_approaches_Markov nn_approaches_Model dobj_outperform_approaches nsubjpass_used_that auxpass_used_are advmod_used_frequently rcmod_approaches_used prep_used_for amod__part-of-speech amod__tagging pobj_for_ dep_well_as cc_approaches_well dep_well_as nn_promise_showing conj_approaches_promise prep_promise_for amod_applications_other pobj_for_applications ' +p,5942,'The remaining six entries were all fully automatic machine translation systems ; in fact , they were all phrase-based statistical machine translation system that had been trained on the same parallel corpus and most used Bleubased minimum error rate training to optimize the weights of their log linear models feature functions ',Och,'det_entries_The amod_entries_remaining num_entries_six nsubj_systems_entries cop_systems_were det_systems_all advmod_automatic_fully amod_systems_automatic nn_systems_machine nn_systems_translation prep_system_in pobj_in_fact nsubj_system_they cop_system_were det_system_all amod_system_phrase-based amod_system_statistical nn_system_machine nn_system_translation parataxis_systems_system nsubjpass_trained_that aux_trained_had auxpass_trained_been rcmod_system_trained prep_trained_on det_corpus_the amod_corpus_same amod_corpus_parallel pobj_on_corpus cc_corpus_and advmod_used_most amod_rate_used nn_rate_Bleubased amod_rate_minimum nn_rate_error conj_corpus_rate xcomp_trained_training amod_functions_ aux_optimize_to xcomp__optimize det_weights_the dobj_optimize_weights prep_weights_of poss_models_their amod_models_log amod_models_linear pobj_of_models nn_functions_feature dobj_training_functions ' +o,5943,'Others use sentence cohesion , agreement\\/disagreement between speakers , or structural adjacency ',Thomas,'nsubj_use_Others nn__sentence nn__cohesion dobj_use_ conj__agreement\\/disagreement prep_agreement\\/disagreement_between nn__speakers pobj_between_ cc__or amod_adjacency_structural conj__adjacency ' +o,5944,'Grammar rules were induced with the syntaxbased SMT system SAMT described in , which requires initial phrase alignments that we generated with GIZA + + , and syntactic parse trees of the target training sentences , generated by the Stanford Parser pre-trained on the Penn Treebank ',Koehn,'nn_rules_Grammar nsubjpass_induced_rules auxpass_induced_were prep_induced_with det_SAMT_the amod_SAMT_syntaxbased nn_SAMT_SMT nn_SAMT_system pobj_with_SAMT partmod_SAMT_described prep_described_in pobj_in_ nsubj_requires_which rcmod__requires amod_alignments_initial nn_alignments_phrase dobj_requires_alignments complm_generated_that nsubj_generated_we ccomp_requires_generated prep_generated_with pobj_with_GIZA amod_SAMT_+ cc_+_+ conj_+_ cc_SAMT_and amod_trees_syntactic nn_trees_parse conj_SAMT_trees prep_trees_of det_sentences_the nn_sentences_target nn_sentences_training pobj_of_sentences partmod_SAMT_generated prep_generated_by det__the nn__Stanford nn__Parser pobj_by_ partmod__pre-trained prep_pre-trained_on det_Treebank_the nn_Treebank_Penn pobj_on_Treebank ' +o,5945,'The latter group did an experiment early on in which they found that manual tagging took about twice as long as correcting -LRB- automated tagging -RRB- , with about twice the interannotator disagreement rate and an error rate that was about 50 \% higher ',Marcus,'det_group_The amod_group_latter nsubj_did_group det_experiment_an dep_early_experiment dep_in_early dep_in_on prep_did_in dep_found_which nsubj_found_they pcomp_in_found complm_took_that amod_tagging_manual nsubj_took_tagging ccomp_found_took quantmod_long_about quantmod_long_twice quantmod_long_as dobj_took_long prep_long_as pobj_as_correcting amod_tagging_automated appos_long_tagging prep_long_with advmod_rate_about advmod_rate_twice det_rate_the amod_rate_interannotator nn_rate_disagreement pobj_with_rate cc_rate_and det_rate_an nn_rate_error conj_rate_rate nsubj_higher_that cop_higher_was quantmod_50_about num_\%_50 measure_higher_\% rcmod_rate_higher ' +o,5946,'Stochastic taggers use both contextual and morphological information , and the model parameters are usually defined or updated automatically from tagged texts -LRB- Cerf-Danon and E1-Beze 1991 ; Church 1988 ; et al 1992 ; Dermatas and Kokkinakis 1988 , 1990 , 1993 , 1994 ; Garside , Leech , and Sampson 1987 ; Kupiec 1992 ; Maltese \\* Department of Electrical Engineering , Wire Communications Laboratory -LRB- WCL -RRB- , University of Patras , 265 00 Patras , Greece ',Cutting,'amod_taggers_Stochastic nsubj_use_taggers det_information_both amod_information_contextual cc_contextual_and conj_contextual_morphological dobj_use_information cc_use_and det_parameters_the nn_parameters_model nsubjpass_defined_parameters auxpass_defined_are advmod_defined_usually conj_use_defined cc_defined_or conj_defined_updated dep_from_automatically prep_defined_from amod_texts_tagged pobj_from_texts dep_texts_Cerf-Danon cc_Cerf-Danon_and conj_Cerf-Danon_E1-Beze num_E1-Beze_1991 dep_Cerf-Danon_Church dep_Church_1988 dep_Cerf-Danon_ cc__et conj__al dep__1992 dep_Cerf-Danon_Dermatas nn_Kokkinakis_and dep_Dermatas_Kokkinakis num_Kokkinakis_1988 num_Kokkinakis_1990 num_Kokkinakis_1993 num_Kokkinakis_1994 dep_Cerf-Danon_Garside conj_Garside_Leech cc_Garside_and conj_Garside_Sampson num_Sampson_1987 dep_Cerf-Danon_Kupiec dep_Kupiec_1992 nn_Department_Maltese nn_Department_\\* dep_Cerf-Danon_Department prep_Department_of amod_Engineering_Electrical pobj_of_Engineering nn_Laboratory_Wire nn_Laboratory_Communications appos_texts_Laboratory abbrev_Laboratory_WCL appos_Laboratory_University prep_University_of pobj_of_Patras number_00_265 num_Patras_00 appos_texts_Patras appos_Patras_Greece ' +o,5947,'SRILM can produce classes to maximize the mutual information between the classes I -LRB- C -LRB- wt -RRB- ; C -LRB- wt 1 -RRB- -RRB- , as described in ',Brown,'nn__SRILM nsubj_produce_ aux_produce_can nsubj_maximize_classes aux_maximize_to xcomp_produce_maximize det_information_the amod_information_mutual dobj_maximize_information prep_information_between det_I_the nn_I_classes pobj_between_I nn_wt_C dep_I_wt dep_wt_C number_1_wt dep_C_1 advmod_described_as amod_I_described prep_maximize_in ' +o,5948,' found that human summarization can be traced back to six cut-andpaste operations of a text and proposed a revision method consisting of sentence reduction and combination modules with a sentence extraction part ',Jing,'nsubj_found_ complm_traced_that amod_summarization_human nsubjpass_traced_summarization aux_traced_can auxpass_traced_be ccomp_found_traced prt_traced_back prep_traced_to num_operations_six amod_operations_cut-andpaste pobj_to_operations prep_operations_of det_text_a pobj_of_text cc_traced_and conj_traced_proposed det_method_a nn_method_revision dobj_proposed_method partmod_method_consisting prep_consisting_of nn_reduction_sentence pobj_of_reduction cc_reduction_and nn_modules_combination conj_reduction_modules prep_consisting_with det_part_a nn_part_sentence nn_part_extraction pobj_with_part ' +o,5949,'However , the fact that the DGSSN uses a large-vocabulary tagger as a preprocessing stage may compensate for its smaller vocabulary ',Ratnaparkhi,'advmod_compensate_However det_fact_the nsubj_compensate_fact complm_uses_that det_DGSSN_the nsubj_uses_DGSSN dep_fact_uses det__a amod__large-vocabulary nn__tagger dobj_uses_ prep_uses_as det_stage_a amod_stage_preprocessing pobj_as_stage aux_compensate_may prep_compensate_for poss_vocabulary_its amod_vocabulary_smaller pobj_for_vocabulary ' +o,5950,'In our framework , we employ a simple HMM-based tagger , where the most probable tag sequence , a29a30 , given the words , a31 , is output : a29 a30 a20a22a32a34a33a36a35a38a37a39a32a41a40 a42 a43a45a44 a30a47a46 a31a49a48a17a20a22a32a34a33a50a35a38a37a39a32a41a40 a42 a43a45a44 a31 a46a30 a48 a43a51a44 a30 a48 Since we do not have enough data which is manually tagged with part-of-speech tags for our applications , we used Penn Treebank as our training set ',Marcus,'prep_employ_In poss_framework_our pobj_In_framework nsubj_employ_we det_tagger_a amod_tagger_simple amod_tagger_HMM-based dobj_employ_tagger advmod__where det_sequence_the advmod_probable_most amod_sequence_probable nn_sequence_tag nsubj__sequence appos_sequence_a29a30 prep_sequence_given det_words_the dep_given_words appos_words_a31 cop__is nn__output rcmod_tagger_ amod_a48_a29 amod_a48_a30 amod_a48_a20a22a32a34a33a36a35a38a37a39a32a41a40 amod_a48_a42 amod_a48_a43a45a44 amod_a48_a30a47a46 amod_a48_a31a49a48a17a20a22a32a34a33a50a35a38a37a39a32a41a40 amod_a48_a42 amod_a48_a43a45a44 amod_a48_a31 amod_a48_a46a30 amod_a48_a48 amod_a48_a43a51a44 nn_a48_a30 dep__a48 mark_have_Since nsubj_have_we aux_have_do neg_have_not advcl_used_have amod_data_enough dobj_have_data nsubjpass_tagged_which auxpass_tagged_is advmod_tagged_manually rcmod_data_tagged prep_tagged_with amod_tags_part-of-speech pobj_with_tags prep_tags_for poss_applications_our pobj_for_applications nsubj_used_we rcmod_a48_used nn__Penn nn__Treebank dobj_used_ prep__as poss_set_our nn_set_training pobj_as_set ' +o,5951,'3 Candidates extraction on Suffix array Suffix array -LRB- also known as String PATarray -RRB- is a compact data structure to handle arbitrary-length strings and performs much powerful on-line string search operations such as the ones supported by PAT-tree , but has less space overhead ',Dunning,'num_Candidates_3 nsubj_structure_extraction prep_extraction_on nn_array_Suffix nn_array_array nn_array_Suffix pobj_on_array advmod_known_also dep_extraction_known prep_known_as nn_PATarray_String pobj_as_PATarray advmod_structure_ cop_structure_is det_structure_a amod_structure_compact nn_structure_data dep_Candidates_structure aux_handle_to xcomp_structure_handle amod_strings_arbitrary-length dobj_handle_strings cc_structure_and conj_structure_performs advmod_powerful_much acomp_performs_powerful amod_operations_on-line nn_operations_string nn_operations_search dobj_structure_operations dep_as_such prep_operations_as det_ones_the pobj_as_ones partmod_ones_supported prep_supported_by pobj_by_PAT-tree cc_structure_but conj_structure_has amod_overhead_less nn_overhead_space dobj_has_overhead ' +o,5952,'A CHECK move requests the partner to confirm information that the speaker has some reason to believe , but is not entirely sure about ',Carletta,'det_CHECK_A nsubj_move_CHECK iobj_move_requests det_partner_the dobj_move_partner aux_confirm_to infmod_partner_confirm dobj_confirm_information complm_has_that det_speaker_the nsubj_has_speaker ccomp_confirm_has det_reason_some dobj_has_reason aux_believe_to infmod_reason_believe cc_has_but cop_sure_is neg_sure_not advmod_sure_entirely conj_has_sure prep_sure_about ' +o,5953,'The reader is referred to Schmid and for details ',Collins,'det_reader_The nsubjpass_referred_reader auxpass_referred_is prep_referred_to nn__Schmid pobj_to_ cc__and conj__ prep_referred_for pobj_for_details ' +o,5954,'We propose using distributional similarity -LRB- using -RRB- as an approximation of semantic distancebetweenthewordsinthetwoglosses , rather than requiring an exact match ',Lin,'nsubj_propose_We ccomp_propose_using amod_similarity_distributional dobj_using_similarity parataxis_using_using dobj_using_ prep_using_as det_approximation_an pobj_as_approximation prep_approximation_of amod_distancebetweenthewordsinthetwoglosses_semantic pobj_of_distancebetweenthewordsinthetwoglosses cc_using_rather dep_rather_than conj_using_requiring det_match_an amod_match_exact dobj_requiring_match ' +o,5955,'For our experiments we used the following features , analogous to Pharaohs default feature set : P -LRB- -RRB- and P -LRB- -RRB- , the latter of which is not found in the noisy-channel model , but has been previously found to be a helpful feature ; the lexical weights Pw -LRB- -RRB- and Pw -LRB- -RRB- , which estimate how well the words in translate the words in ; 2 a phrase penalty exp -LRB- 1 -RRB- , which allows the model to learn a preference for longer or shorter derivations , analogous to Koehns phrase penalty ',Koehn,'prep_used_For poss_experiments_our pobj_For_experiments nsubj_used_we det_features_the amod_features_following dobj_used_features amod_feature_analogous prep_analogous_to pobj_to_Pharaohs nn_feature_default appos_features_feature partmod_feature_set dep_feature_P cc_-RRB-_and conj_-RRB-_P det__the amod__latter prep_latter_of nsubjpass_found_which auxpass_found_is neg_found_not dep_weights_found prep_found_in det_model_the amod_model_noisy-channel pobj_in_model cc_found_but aux_found_has auxpass_found_been advmod_found_previously conj_found_found aux__to cop__be det__a amod__helpful nn__feature xcomp_found_ det_weights_the amod_weights_lexical pobj_of_weights nn__Pw cc_-RRB-_and conj_-RRB-_Pw appos_feature_ nsubj_estimate_which rcmod_feature_estimate advmod_translate_how advmod_how_well det_words_the nsubj_translate_words advmod_translate_in ccomp_estimate_translate det_words_the dobj_translate_words prep_translate_in dep_exp_2 det_exp_a amod_exp_phrase nn_exp_penalty dep_features_exp dep_features_1 nsubj_allows_which dep_1_allows det_model_the dobj_allows_model aux_learn_to infmod_model_learn det_preference_a dobj_learn_preference prep_preference_for amod_derivations_longer cc_longer_or conj_longer_shorter pobj_for_derivations amod_features_analogous prep_analogous_to nn_penalty_Koehns nn_penalty_phrase pobj_to_penalty ' +o,5956,' presented an approach that significantly reduces the amount of labeled data needed for word sense disambiguation ',Yarowsky,'nsubj_presented_ det_approach_an dobj_presented_approach nsubj_reduces_that advmod_reduces_significantly rcmod_approach_reduces det_amount_the dobj_reduces_amount prep_amount_of amod_data_labeled pobj_of_data partmod_data_needed prep_needed_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +n,5957,'It has the advantage of naturally capturing local reorderings and is shown to outperform word-based machine translation ',Koehn,'nsubjpass_has_It det_advantage_the dobj_has_advantage prep_advantage_of advmod_capturing_naturally pcomp_of_capturing amod_reorderings_local dobj_capturing_reorderings cc_has_and auxpass_shown_is conj_has_shown aux_outperform_to xcomp_shown_outperform amod_translation_word-based nn_translation_machine dobj_outperform_translation ' +o,5958,'Pointwise mutual information was used to measure strength of selection restrictions for instance by ',Church,'amod__Pointwise amod__mutual nn__information nsubjpass_used_ auxpass_used_was aux_measure_to xcomp_used_measure dobj_measure_strength prep_strength_of nn_restrictions_selection pobj_of_restrictions prep_restrictions_for pobj_for_instance prep_measure_by ' +o,5959,'There has been a sizable amount of research on structure induction ranging fromlinearsegmentation tocontent modeling ',Lin,'expl_been_There aux_been_has det_amount_a amod_amount_sizable dobj_been_amount prep_amount_of pobj_of_research prep_research_on nn_induction_structure pobj_on_induction partmod_induction_ranging nn__fromlinearsegmentation nn__ nn__tocontent nn__modeling dobj_ranging_ ' +o,5960,'In another line of research , and have shown that it is possible to reduce the need for supervision with the help of large amounts of unannotated data ',Yarowsky,'prep_shown_In det_line_another pobj_In_line prep_line_of pobj_of_research nsubj_shown_ cc__and conj__ aux_shown_have complm_possible_that nsubj_possible_it cop_possible_is ccomp_shown_possible aux_reduce_to xcomp_possible_reduce det_need_the dobj_reduce_need prep_need_for pobj_for_supervision prep_reduce_with det_help_the pobj_with_help prep_help_of amod_amounts_large pobj_of_amounts prep_amounts_of amod_data_unannotated pobj_of_data ' +o,5961,'In modern lexicalized parsers , POS tagging is often interleaved with parsing proper instead of being a separate preprocessing module ',Ratnaparkhi,'prep_interleaved_In amod_parsers_modern amod_parsers_lexicalized pobj_In_parsers dep_tagging_POS nsubjpass_interleaved_tagging auxpass_interleaved_is advmod_interleaved_often prep_interleaved_with pcomp_with_parsing dobj_parsing_proper dep_of_instead prep_proper_of cop_module_being det_module_a amod_module_separate nn_module_preprocessing pcomp_of_module ' +o,5962,'3 Related Work Many methods have been developed for automatically identifying subjective -LRB- opinion , sentiment , attitude , affect-bearing , etc -RRB- words , eg , ',Turney,'num_methods_3 nn_methods_Related nn_methods_Work amod_methods_Many nsubjpass_developed_methods aux_developed_have auxpass_developed_been prep_developed_for advmod_identifying_automatically pcomp_for_identifying amod_words_subjective dep_subjective_opinion appos_opinion_sentiment appos_opinion_attitude appos_opinion_affect-bearing dep_opinion_etc dobj_identifying_words appos_words_eg ' +o,5963,'However , showed that for natural language and text processing tasks , conditional models are usually better than joint likelihood models ',Klein,'advmod_showed_However nsubj_showed_ complm_better_that prep_better_for amod_language_natural pobj_for_language cc_language_and nn_tasks_text nn_tasks_processing conj_language_tasks amod_models_conditional nsubj_better_models cop_better_are advmod_better_usually ccomp_showed_better prep_better_than amod_models_joint nn_models_likelihood pobj_than_models ' +o,5964,'al 2003b -RRB- 147 is -LRB- B -RRB- eginning , -LRB- I -RRB- nside or -LRB- O -RRB- utside of a chunk ',Ramshaw,'nn_147_al nn_147_2003b nsubj_B_147 cop_B_is dep_B_eginning dep_B_I amod_B_nside cc_nside_or dep_nside_O conj_nside_utside prep_B_of det_chunk_a pobj_of_chunk ' +o,5965,'However , with their system trained on the medical corpus and then tested on the Wall Street Journal corpus , they achieve an overall prediction accuracy of only 54 \% ',Marcus,'advmod_achieve_However prep_achieve_with poss_system_their pobj_with_system dep_system_trained prep_trained_on det_corpus_the amod_corpus_medical pobj_on_corpus cc_trained_and advmod_tested_then conj_trained_tested prep_tested_on det__the nn__Wall nn__Street nn__Journal nn__corpus pobj_on_ nsubj_achieve_they det_accuracy_an amod_accuracy_overall nn_accuracy_prediction dobj_achieve_accuracy prep_accuracy_of quantmod_54_only num_\%_54 pobj_of_\% ' +o,5966,'We solve this using the local search defined in ',Brown,'nsubj_solve_We nsubj_using_this dep_solve_using det_search_the amod_search_local dobj_using_search partmod_search_defined prep_defined_in ' +p,5967,' introduced the averaged perceptron , as a way of reducing overfitting , and it has been shown to perform better than the non-averaged version on a number of tasks ',Collins,'nsubj_introduced_ det_perceptron_the amod_perceptron_averaged dobj_introduced_perceptron prep_introduced_as det_way_a pobj_as_way prep_way_of pcomp_of_reducing dobj_reducing_overfitting cc_introduced_and nsubjpass_shown_it aux_shown_has auxpass_shown_been conj_introduced_shown aux_perform_to xcomp_shown_perform advmod_perform_better prep_perform_than det_version_the amod_version_non-averaged pobj_than_version prep_version_on det_number_a pobj_on_number prep_number_of pobj_of_tasks ' +o,5968,'The extraction procedure utilizes a head percolation table as introduced by Magerman in combination with a variation of approach to the differentiation between complement and adjunct ',Collins,'det_procedure_The nn_procedure_extraction nsubj_utilizes_procedure det_table_a nn_table_head nn_table_percolation dobj_utilizes_table prep_utilizes_as pcomp_as_introduced prep_introduced_by nn__Magerman pobj_by_ prep__in pobj_in_combination prep_introduced_with det_variation_a pobj_with_variation prep_variation_of amod_approach_ pobj_of_approach prep_introduced_to det_differentiation_the pobj_to_differentiation prep_differentiation_between pobj_between_complement cc_complement_and conj_complement_adjunct ' +o,5969,'We can incorporate each model into the system in turn , and rank the results on a test corpus using BLEU ',Papineni,'nsubj_incorporate_We aux_incorporate_can det_model_each dobj_incorporate_model prep_incorporate_into det_system_the pobj_into_system prep_system_in pobj_in_turn cc_incorporate_and conj_incorporate_rank det_results_the dobj_rank_results prep_rank_on det_corpus_a nn_corpus_test pobj_on_corpus partmod_corpus_using dobj_using_BLEU ' +o,5970,'Most systems for automatic role-semantic analysis have used constituent syntax as in the Penn Treebank , although there has also been much research on the use of shallow syntax in SRL ',Marcus,'amod_systems_Most nsubj_used_systems prep_systems_for amod_analysis_automatic amod_analysis_role-semantic pobj_for_analysis aux_used_have amod_syntax_constituent dobj_used_syntax advmod_used_as dep_as_in det__the nn__Penn nn__Treebank pobj_in_ mark_been_although expl_been_there aux_been_has advmod_been_also advcl_used_been amod_research_much dobj_been_research prep_research_on det_use_the pobj_on_use prep_use_of nn__shallow nn__syntax pobj_of_ prep_been_in pobj_in_SRL ' +o,5971,'Pereira , Curran and use syntactic features in the vector definition ',Lin,'nn__Pereira nn__Curran conj__ cc__and amod_use_ conj__use amod_features_syntactic dep__features prep_features_in det_definition_the nn_definition_vector pobj_in_definition ' +o,5972,'3 Hebrew Simple NP Chunks The standard definition of English base-NPs is any noun phrase that does not contain another noun phrase , with possessives treated as a special case , viewing the possessive marker as the first word of a new base-NP ',Ramshaw,'num_NP_3 nn_NP_Hebrew nn_NP_Simple nsubj_Chunks_NP det_definition_The amod_definition_standard nsubj_phrase_definition prep_definition_of amod_base-NPs_English pobj_of_base-NPs cop_phrase_is det_phrase_any nn_phrase_noun ccomp_Chunks_phrase nsubj_contain_that aux_contain_does neg_contain_not rcmod_phrase_contain det_phrase_another nn_phrase_noun dobj_contain_phrase prep_contain_with pobj_with_possessives partmod_possessives_treated prep_treated_as det_case_a amod_case_special pobj_as_case xcomp_contain_viewing det_marker_the amod_marker_possessive dobj_viewing_marker prep_viewing_as det_word_the amod_word_first pobj_as_word prep_word_of det_base-NP_a amod_base-NP_new pobj_of_base-NP ' +o,5973,'(1) Here has(h,x) is a binary function that returns true if the history h has feature x.Inour experiments, we focused on such information as whether or not a string is found in a dictionary, the length of the string, what types of characters are used in the string, and what part-of-speech the adjacent morpheme is. Given a set of features and some training data, the M.E. estimation process produces a model, which is represented as follows (Berger et al. , 1996; Ristad, 1997; Ristad, 1998): P(f|h)= producttext i g i (h,f) i Z (h) (2) Z (h)= summationdisplay f productdisplay i g i (h,f) i.',Berger,'' +o,5974,'Artificial ungrammaticalities have been used in various NLP tasks The idea of an automatically generated ungrammatical treebank was proposed by Foster ',Okanohara,'amod_ungrammaticalities_Artificial nsubjpass_used_ungrammaticalities aux_used_have auxpass_used_been prep_used_in amod__various nn__NLP nn__tasks pobj_in_ det_idea_The nsubj_generated_idea prep_idea_of det_automatically_an pobj_of_automatically dep_used_generated amod_treebank_ungrammatical nsubjpass_proposed_treebank auxpass_proposed_was ccomp_generated_proposed prep_proposed_by nn__Foster pobj_by_ ' +o,5975,'Other authors have applied this approach to language modeling ',Berger,'amod_authors_Other nsubj_applied_authors aux_applied_have det_approach_this dobj_applied_approach prep_applied_to nn__language nn__modeling pobj_to_ ' +o,5976,'A quick search in the Penn Treebank shows that about 17 \% of all sentences contain parentheticals or other sentence fragments , interjections , or unbracketable constituents ',Marcus,'det_search_A amod_search_quick nsubj_shows_search prep_search_in det__the nn__Penn nn__Treebank pobj_in_ complm_contain_that quantmod_17_about num_\%_17 nsubj_contain_\% prep_\%_of det_sentences_all pobj_of_sentences ccomp_shows_contain dobj_contain_parentheticals cc_parentheticals_or amod_fragments_other nn_fragments_sentence conj_parentheticals_fragments conj_fragments_interjections cc_fragments_or amod_constituents_unbracketable conj_fragments_constituents ' +p,5977,'This averaging effect has been shown to reduce overfitting and produce much more stable results ',Collins,'det_effect_This nn_effect_averaging nsubjpass_shown_effect aux_shown_has auxpass_shown_been aux_reduce_to xcomp_shown_reduce dobj_reduce_overfitting cc_reduce_and conj_reduce_produce advmod_stable_much advmod_stable_more amod_results_stable dobj_produce_results ' +o,5978,'2 F 1 - score Maximization Training of LRM We first review the F 1 - score maximization training method for linear models using a logistic function described in ',Jansche,'num_F_2 num_F_1 nn_Training_score nn_Training_Maximization dep_F_Training prep_Training_of pobj_of_LRM nsubj_review_We advmod_review_first rcmod_Training_review det_F_the dobj_review_F num_F_1 nn_method_score nn_method_maximization nn_method_training dobj_review_method prep_method_for amod_models_linear pobj_for_models partmod_models_using det_function_a amod_function_logistic dobj_using_function partmod_function_described prep_described_in ' +o,5979,'In this respect , it resembles bilingual bracketing , but our model has more lexical items in the blocks with many-to-many word alignment freedom in both inner and outer parts ',Wu,'prep__In det_respect_this pobj_In_respect nsubj__it cop__resembles amod__bilingual amod__bracketing cc__but poss_model_our nsubj_has_model conj__has advmod_lexical_more amod_items_lexical dobj_has_items prep_items_in det_blocks_the pobj_in_blocks prep_blocks_with amod_freedom_many-to-many nn_freedom_word nn_freedom_alignment pobj_with_freedom prep_freedom_in det_parts_both amod_parts_inner cc_inner_and conj_inner_outer pobj_in_parts ' +o,5980,'41 Experimental Setup Like several previous work -LRB- eg , Mullen and Collier , and Lee , Whitelaw et al ',Pang,'nn_Setup_Experimental dep_41_Setup prep_Setup_Like amod_work_several amod_work_previous pobj_Like_work dep_work_eg dep_eg_Mullen cc_work_and conj_work_Collier num_Collier_ num_Collier_ cc_Collier_and nn__Lee conj_Collier_ conj__Whitelaw cc__et conj__al ' +o,5981,'Much research has been done to improve tagging accuracy using several different models and methods , including : hidden Markov models -LRB- HMMs -RRB- , ; rule-based systems , ; memory-based systems ; maximum-entropy systems ; path voting constraint systems ; linear separator systems ; and majority voting systems ',Marcus,'nsubjpass_done_research aux_done_has auxpass_done_been dep_Much_done aux_improve_to xcomp_done_improve amod_accuracy_tagging dobj_improve_accuracy partmod_accuracy_using amod_models_several amod_models_different dobj_using_models cc_models_and conj_models_methods prep_done_including amod_models_hidden nn_models_Markov pobj_including_models abbrev_models_HMMs amod_models_ dep__ amod_systems_rule-based nsubj__systems parataxis_done_ advmod__ amod_systems_memory-based nsubj__systems parataxis_done_ amod_systems_maximum-entropy nsubj__systems parataxis_done_ nn_systems_path nn_systems_voting nn_systems_constraint nsubj__systems parataxis_done_ amod_systems_linear nn_systems_separator nsubj__systems parataxis_done_ cc_done_and nn_systems_majority nn_systems_voting nsubj__systems conj_done_ ' +o,5982,'In recent years , many researchers have tried to automatically evaluate the quality of MT and improve the performance of automatic MT evaluations because improving the performance of automatic MT evaluation is expected to enable us to use and improve MT systems efficiently ',Lin,'prep_tried_In amod_years_recent pobj_In_years amod_researchers_many nsubj_tried_researchers aux_tried_have aux_evaluate_to advmod_evaluate_automatically xcomp_tried_evaluate det_quality_the dobj_evaluate_quality prep_quality_of pobj_of_MT cc_evaluate_and conj_evaluate_improve det_performance_the dobj_improve_performance prep_performance_of amod_evaluations_automatic nn_evaluations_MT pobj_of_evaluations xcomp_improve_ mark_expected_because csubjpass_expected_improving det_performance_the dobj_improving_performance prep_performance_of amod_evaluation_automatic nn_evaluation_MT pobj_of_evaluation auxpass_expected_is advcl__expected aux_enable_to xcomp_expected_enable nsubj_use_us aux_use_to xcomp_enable_use cc_use_and conj_use_improve nn_systems_MT dobj_use_systems advmod_use_efficiently ' +o,5983,'While we have observed reasonable results with both G 2 and Fisher \'s exact test , we have not yet discussed how these results compare to the results that can be obtained with a technique commonly used in corpus linguistics based on the mutual information -LRB- MI -RRB- measure : I -LRB- x , y -RRB- -- log 2 P -LRB- x , y -RRB- -LRB- 4 -RRB- P -LRB- x -RRB- P -LRB- y -RRB- In -LRB- 4 -RRB- , y is the seed term and x a potential target word ',Church,'mark_observed_While nsubj_observed_we aux_observed_have advcl_discussed_observed amod_results_reasonable dobj_observed_results prep_observed_with det_G_both pobj_with_G num_G_2 cc_G_and poss_test_Fisher possessive_Fisher_\'s amod_test_exact conj_G_test nsubj_discussed_we aux_discussed_have neg_discussed_not advmod_discussed_yet advmod_compare_how det_results_these nsubj_compare_results ccomp_discussed_compare prep_compare_to det_results_the pobj_to_results nsubjpass_obtained_that aux_obtained_can auxpass_obtained_be rcmod_results_obtained prep_obtained_with det_technique_a pobj_with_technique advmod_used_commonly partmod_technique_used prep_used_in nn_linguistics_corpus pobj_in_linguistics prep_used_based dep_based_on det_information_the amod_information_mutual pobj_on_information abbrev_information_MI nn__measure nsubj_log_ dep__I dep_I_x appos_x_y dep_discussed_log num_P_2 dobj_log_P dep_P_x appos_x_y appos_P_4 dep_P_P dep_P_x amod_P_P appos_P_y prep_log_In dep_y_4 pobj_In_y cop_term_is det_term_the nn_term_seed rcmod_y_term cc_y_and dep_word_x det_word_a amod_word_potential nn_word_target conj_y_word ' +p,5984,'Probably the most widely used association weight function is -LRB- point-wise -RRB- Mutual Information -LRB- MI -RRB- , , , , defined by : -RRB- -LRB- -RRB- -LRB- -RRB- , -LRB- log -RRB- , -LRB- 2 fPwP fwPfwMI = A known weakness of MI is its tendency to assign high weights for rare features ',Hindle,'advmod_used_Probably det_most_the nsubj_used_most advmod_used_widely dep_=_used nn_function_association nn_function_weight dobj_used_function cop_Information_is appos_Information_point-wise nn_Information_Mutual rcmod_function_Information abbrev_Information_MI advmod_Information_ appos_function_ appos__ appos__ partmod__defined prep_defined_by parataxis_used_log num_fPwP_2 nsubj_=_fPwP advmod_=_fwPfwMI det_weakness_A amod_weakness_known nsubj_tendency_weakness prep_weakness_of pobj_of_MI cop_tendency_is poss_tendency_its ccomp_=_tendency aux_assign_to infmod_tendency_assign amod_weights_high dobj_assign_weights prep_assign_for amod_features_rare pobj_for_features ' +o,5985,'The weighting parameters of these features were optimized in terms of BLEU by the approach of minimum error rate training ',Och,'det_parameters_The nn_parameters_weighting nsubjpass_optimized_parameters prep_parameters_of det_features_these pobj_of_features auxpass_optimized_were prep_optimized_in pobj_in_terms prep_terms_of pobj_of_BLEU prep_optimized_by det_approach_the pobj_by_approach prep_approach_of amod_training_minimum nn_training_error nn_training_rate pobj_of_training ' +o,5986,'Concluding Remarks Formalisms for finite-state and context-free transduction have a long history , and such formalisms have been applied to the machine translation problem , both in the finite-state case and the context-free case ',Wu,'nn_Formalisms_Remarks dobj_Concluding_Formalisms prep_Formalisms_for amod_transduction_finite-state cc_finite-state_and conj_finite-state_context-free pobj_for_transduction dep__have det_history_a amod_history_long dobj__history dep_Concluding_ cc__and amod_formalisms_such conj__formalisms aux_applied_have auxpass_applied_been rcmod_formalisms_applied prep_applied_to det_translation_the nn_translation_machine pobj_to_translation nsubj_Concluding_problem appos_problem_both prep_both_in det__the amod__finite-state nn__case pobj_in_ cc__and det_case_the amod_case_context-free conj__case ' +o,5987,'We accordingly introduce approaches which attempt to include semantic information into the coreference models from a variety of knowledge sources , eg WordNet , Wikipedia and automatically harvested patterns ',Ponzetto,'nsubj_introduce_We advmod_introduce_accordingly ccomp__introduce dobj_introduce_approaches nsubj_attempt_which rcmod_approaches_attempt aux_include_to xcomp_attempt_include amod_information_semantic dobj_include_information prep_include_into det_models_the amod_models_coreference pobj_into_models prep_include_from det_variety_a pobj_from_variety prep_variety_of nn_sources_knowledge pobj_of_sources nn__eg nn__WordNet nsubj__ nn__Wikipedia nsubj__ cc__and advmod_harvested_automatically amod_patterns_harvested conj__patterns ' +o,5988,'For evaluation , we use IBMs BLEU score to measure the performance of the SMS normalization ',Papineni,'prep_use_For pobj_For_evaluation nsubj_use_we nn_BLEU_IBMs nsubj_score_BLEU ccomp_use_score acomp_score_ aux_measure_to xcomp__measure det_performance_the dobj_measure_performance prep_performance_of det_normalization_the nn_normalization_SMS pobj_of_normalization ' +o,5989,'They compare two data representations and report that a representation with bracket structures outperforms the IOB tagging representation introduced by ',Ramshaw,'nsubj_compare_They num_representations_two nn_representations_data dobj_compare_representations cc_compare_and conj_compare_report complm_outperforms_that det_representation_a nsubj_outperforms_representation prep_representation_with nn_structures_bracket pobj_with_structures ccomp_report_outperforms det_representation_the nn_representation_IOB amod_representation_tagging dobj_outperforms_representation partmod_representation_introduced prep_introduced_by ' +o,5990,'Learning in this context consisted of estimating the parameters of the model with simple likelihood based techniques , but incorporating various smoothing and back-off estimation tricks to cope with the sparse data problems ',Collins,'nsubj_consisted_Learning prep_Learning_in det_context_this pobj_in_context prep_consisted_of pcomp_of_estimating det_parameters_the dobj_estimating_parameters prep_parameters_of det_model_the pobj_of_model prep_estimating_with amod_likelihood_simple pobj_with_likelihood partmod_likelihood_based dobj_based_techniques cc_estimating_but conj_estimating_incorporating amod_smoothing_various dobj_incorporating_smoothing cc_smoothing_and amod_tricks_back-off nn_tricks_estimation conj_smoothing_tricks aux_cope_to xcomp_incorporating_cope prep_cope_with det_problems_the amod_problems_sparse nn_problems_data pobj_with_problems ' +o,5991,'Doing inference with representations close to natural language has also been advocated by Jerry Hobbs , as in ',Hobbs,'amod_inference_Doing nsubjpass_advocated_inference prep_inference_with pobj_with_representations advmod_representations_close dep_close_to amod_language_natural pobj_to_language aux_advocated_has advmod_advocated_also auxpass_advocated_been prep_advocated_by nn_Hobbs_Jerry pobj_by_Hobbs dep_by_as dep_by_in ' +o,5992,'In Hirschberg and Nakatani , average reliability of segmentinitial labels among 3 coders on 9 monologues produced by the same speaker , labeled using text and speech , is8 or above for both read and spontaneous speech ; values of at least 8 are typically viewed as representing high reliability -LRB- see Section 32 -RRB- ',Carletta,'prep_coders_In nn__Hirschberg cc_Hirschberg_and conj_Hirschberg_Nakatani pobj_In_ amod__average nn__reliability nsubj_coders_ prep__of amod_labels_segmentinitial pobj_of_labels prep_labels_among pobj_among_3 prep_coders_on num_monologues_9 pobj_on_monologues dep_monologues_produced prep_produced_by det_speaker_the amod_speaker_same pobj_by_speaker dep_produced_labeled xcomp_labeled_using dobj_using_text cc_text_and conj_text_speech dobj_produced_is8 cc_on_or dep_for_above conj_on_for preconj_speech_both nn_speech_read cc_read_and conj_read_spontaneous pobj_for_speech nsubjpass_viewed_values prep_values_of quantmod_8_at dep_at_least pobj_of_8 auxpass_viewed_are advmod_viewed_typically parataxis_coders_viewed prep_viewed_as pcomp_as_representing amod_reliability_high dobj_representing_reliability dep_viewed_see dobj_see_Section num_Section_32 ' +n,5993,'Again the best result was obtained with IOB1 which is an imI -RRB- rovement of the best reported F , ~ = 1 rate for this data set -LRB- : 9203 -RRB- ',Ramshaw,'det_result_the amod_result_best dep_Again_result auxpass_obtained_was rcmod_result_obtained prep_obtained_with nn__IOB1 pobj_with_ nsubj_rovement_which cop_rovement_is det_rovement_an nn_rovement_imI rcmod__rovement prep_rovement_of det_F_the dep_reported_best amod_F_reported pobj_of_F advmod_=_~ dep__= num_rate_1 dobj_=_rate prep_rate_for det_set_this nn_set_data pobj_for_set dep_set_ dep__9203 ' +o,5994,'In all other respects , our work departs from previous research on broad -- coverage 16 I t I I I I I i ! I i I I I I I I I I I I I i I 1 , I I I I I i I 1 I I I I probabilistic parsing , which either attempts to learn to predict gr ~ rarn ~ tical structure of test data directly from a training treebank , or employs a grammar and sometimes a dictionary to capture linguistic expertise directly , but arguably at a less detailed and informative level than in the research reported here ',Brown,'prep_departs_In det_respects_all amod_respects_other pobj_In_respects poss_work_our nsubj_departs_work prep_departs_from amod_research_previous pobj_from_research prep_departs_on amod_I_broad nn_t_coverage num_t_16 nn_t_I dep_--_t dep_i_I dep_i_I dep_i_I dep_i_I dep_i_I amod_t_i pobj_on_I partmod_I_i iobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_i dobj_i_I num_I_1 num_I_I dobj_i_I dobj_i_I dobj_i_I dobj_i_I dobj_i_i dobj_i_I dobj_i_1 dobj_i_I dobj_i_I dobj_i_I dobj_i_I amod_parsing_probabilistic dep_from_parsing nsubj_attempts_which advmod_attempts_either rcmod_parsing_attempts aux_learn_to xcomp_attempts_learn aux_predict_to xcomp_learn_predict nn_structure_gr nn_structure_~ nn_structure_rarn nn_structure_~ nn_structure_tical dobj_predict_structure prep_structure_of nn_data_test pobj_of_data advmod_predict_directly prep_i_from det__a nn__training nn__treebank pobj_from_ cc_departs_or conj_departs_employs det_grammar_a dobj_employs_grammar cc_grammar_and advmod_grammar_sometimes det_dictionary_a conj_grammar_dictionary aux_capture_to xcomp_employs_capture amod_expertise_linguistic dobj_capture_expertise advmod__directly amod_expertise_ cc__but conj__arguably prep_arguably_at det_level_a advmod_detailed_less amod_level_detailed cc_detailed_and conj_detailed_informative pobj_at_level dep_in_than prep_capture_in det_research_the pobj_in_research partmod_research_reported advmod_reported_here ' +o,5995,'We want to avoid training a metric that as5Or , in a less adversarial setting , a system may be performing minimum error-rate training signs a higher than deserving score to a sentence that just happens to have many n-gram matches against the target-language reference corpus ',Och,'nsubj_want_We aux_avoid_to xcomp_want_avoid xcomp_avoid_training det_metric_a dobj_training_metric complm_performing_that nsubj_performing_as5Or prep_performing_in det_setting_a advmod_setting_less amod_setting_adversarial pobj_in_setting det_system_a nsubj_performing_system aux_performing_may aux_performing_be ccomp_avoid_performing amod_signs_minimum amod_signs_error-rate nn_signs_training nn_signs_ dobj_performing_signs det_score_a dep_deserving_higher dep_deserving_than amod_score_deserving nsubj_matches_score prep_score_to det_sentence_a pobj_to_sentence nsubj_happens_that advmod_happens_just rcmod_sentence_happens aux_have_to xcomp_happens_have amod_n-gram_many dobj_have_n-gram rcmod_signs_matches prep_matches_against det_corpus_the amod_corpus_target-language nn_corpus_reference pobj_against_corpus ' +o,5996,'Similarly , Murdock and Croft adopted a simple translation model from IBM model 1 and applied it to QA ',Brown,'advmod_adopted_Similarly nn__Murdock cc_Murdock_and conj_Murdock_Croft nsubj_adopted_ det_model_a amod_model_simple nn_model_translation dobj_adopted_model prep_adopted_from nn__IBM nn__model num__1 pobj_from_ cc_adopted_and conj_adopted_applied dobj_applied_it prep_applied_to pobj_to_QA ' +p,5997,' -RRB- , and Basque , which pose quite different and in the end less severe problems , there have been attempts at solving this problem for some of the highly inflectional European languages , such as , -LRB- Slovenian -RRB- , , -LRB- Czech -RRB- and -LRB- five Central and Eastern European languages -RRB- , but so far no system has reached in the absolute terms a performance comparable to English tagging -LRB- such as -RRB- , which stands around or above 97 \% ',Ratnaparkhi,'nn__and nn__Basque dep_quite_ dep__which nn_quite_pose dobj__quite amod_quite_different cc_quite_and conj_quite_in det_end_the pobj_in_end prep_in_less amod_problems_severe dep_less_problems expl_been_there aux_been_have dep__been dobj_been_attempts prep_attempts_at pcomp_at_solving det_problem_this dobj_solving_problem prep_solving_for pobj_for_some prep_some_of det_languages_the advmod_inflectional_highly amod_languages_inflectional amod_languages_European pobj_of_languages dep_as_such prep_some_as nn__ pobj_as_ appos__Slovenian nn__ conj__ appos__Czech cc__and conj__ num_Central_five dep__Central cc_Central_and amod_languages_Eastern amod_languages_European conj_Central_languages cc_been_but advmod_far_so advmod_reached_far det_system_no nsubj_reached_system aux_reached_has conj_been_reached prep_reached_in det_terms_the amod_terms_absolute pobj_in_terms det_performance_a dep_terms_performance amod_terms_comparable prep_comparable_to nsubj_tagging_English pcomp_to_tagging dep_as_such dep_tagging_as pobj_as_ nsubj_stands_which rcmod_terms_stands prt_stands_around cc__or conj__above num_\%_97 pobj_above_\% ' +o,5998,'These relations are then used for various tasks , ranging from the interpretation of a noun sequence or a prepositional phrase , to resolving structural ambiguity , to merging dictionary senses for WSD ',Dolan,'det_relations_These nsubjpass_used_relations auxpass_used_are advmod_used_then prep_used_for amod_tasks_various pobj_for_tasks partmod_tasks_ranging prep_ranging_from det_interpretation_the pobj_from_interpretation prep_interpretation_of det__a nn__noun nn__sequence pobj_of_ cc__or det__a amod__prepositional nn__phrase conj__ prep_used_to pcomp_to_resolving amod__structural amod__ambiguity dobj_resolving_ prep_used_to pcomp_to_merging nn_senses_dictionary dobj_merging_senses prep_merging_for pobj_for_WSD ' +o,5999,'2 Extracting paraphrases Much previous work on extracting paraphrases has focused on finding identifying contexts within aligned monolingual sentences from which divergent text can be extracted , and treated as paraphrases ',Barzilay,'dep_paraphrases_2 nsubj_paraphrases_Extracting amod_work_Much amod_work_previous dobj_paraphrases_work prep_paraphrases_on pcomp_on_extracting dobj_extracting_paraphrases nsubj_focused_ aux_focused_has dep_paraphrases_focused prep_focused_on pcomp_on_finding xcomp_finding_identifying dobj_identifying_contexts prep_identifying_within amod_sentences_aligned amod_sentences_monolingual pobj_within_sentences prep_identifying_from dobj_extracted_which amod_text_divergent nsubjpass_extracted_text aux_extracted_can auxpass_extracted_be pcomp_from_extracted cc_extracted_and conj_extracted_treated prep_treated_as pobj_as_paraphrases ' +o,6000,'The training of IBM model 4 was implemented by the GIZA + + package ',Och,'det_training_The nsubjpass_implemented_training prep_training_of pobj_of_IBM dep_IBM_model num_model_4 auxpass_implemented_was prep_implemented_by det_package_the amod_package_GIZA cc_GIZA_+ conj_GIZA_+ pobj_by_package ' +o,6001,'In the geometric interpolation above , the weight n controls the relative veto power of the n-gram approximation and can be tuned using MERT or a minimum risk procedure ',Och,'prep_controls_In det_interpolation_the amod_interpolation_geometric pobj_In_interpolation advmod_controls_above det_n_the nn_n_weight nsubj_controls_n det_power_the amod_power_relative nn_power_veto dobj_controls_power prep_power_of det_approximation_the amod_approximation_n-gram pobj_of_approximation cc_controls_and aux_tuned_can auxpass_tuned_be conj_controls_tuned dep_tuned_using nn__MERT dobj_using_ cc__or det__a amod__minimum nn__risk nn__procedure conj__ ' +o,6002,'In this work , we study a method for obtaining word phrases that is based on Stochastic Inversion Transduction Grammars that was proposed in ',Wu,'prep_study_In det_work_this pobj_In_work nsubj_study_we det_method_a dobj_study_method prep_study_for pcomp_for_obtaining nn_phrases_word dobj_obtaining_phrases nsubjpass_based_that auxpass_based_is rcmod_phrases_based prep_based_on nn_Grammars_Stochastic nn_Grammars_Inversion nn_Grammars_Transduction pobj_on_Grammars nsubjpass_proposed_that auxpass_proposed_was rcmod_Grammars_proposed prep_proposed_in ' +p,6003,'In order to overcome this problem , we look to the bootstrapping method outlined in ',Yarowsky,'mark_overcome_In dep_overcome_order aux_overcome_to dep_look_overcome det_problem_this dobj_overcome_problem nsubj_look_we prep_look_to det_method_the amod_method_bootstrapping pobj_to_method partmod_method_outlined prep_outlined_in ' +o,6004,'Pattern-based IE approaches employ seed data to learn useful patterns to pinpoint required fields values ',Jiao,'amod_approaches_Pattern-based nn_approaches_IE nsubj_employ_approaches nn_data_seed nsubj_learn_data aux_learn_to xcomp_employ_learn amod_patterns_useful dobj_learn_patterns aux_pinpoint_to xcomp_learn_pinpoint amod_values_required nn_values_fields dobj_pinpoint_values advmod_pinpoint_ ' +n,6005,'This is well illustrated by the Collins parser , scrutinized by Bikel -LRB- 2004 -RRB- , where several transformations are applied in order to improve the analysis of noun phrases , coordination and punctuation ',Collins,'nsubj_illustrated_This aux_illustrated_is advmod_illustrated_well prep_illustrated_by det__the nn__Collins nn__parser pobj_by_ dep_illustrated_scrutinized prep_scrutinized_by pobj_by_Bikel appos_Bikel_2004 advmod_applied_where amod_transformations_several nsubjpass_applied_transformations auxpass_applied_are advcl_illustrated_applied prep_applied_in pobj_in_order aux_improve_to xcomp_applied_improve det_analysis_the dobj_improve_analysis prep_analysis_of amod_phrases_noun pobj_of_phrases conj_phrases_coordination cc_phrases_and conj_phrases_punctuation ' +o,6006,'Of particular relevance are class-based language models -LRB- eg , -RRB- ',Brown,'prep_are_Of amod_relevance_particular pobj_Of_relevance amod_models_class-based nn_models_language nsubj_are_models appos_models_eg dep_eg_ ' +o,6007,'The first constraints are based on inversion transduction grammars -LRB- ITG -RRB- ',Wu,'det_constraints_The amod_constraints_first nsubjpass_based_constraints auxpass_based_are prep_based_on nn_grammars_inversion nn_grammars_transduction pobj_on_grammars abbrev_grammars_ITG ' +o,6008,'When the value of Ilw, r,w\'ll is unknown, we assume that A and C are conditionally independent given B. The probability of A, B and C cooccurring is estimated by PMLE( B ) PMLE( A[B ) PMLE( C[B ), where PMLE is the maximum likelihood estimation of a probability distribution and P.LE(B) = II*,*,*ll\' P. ,~E(AIB ) = II*,~,*ll \' P, LE(CIB) = When the value of Hw, r, w~H is known, we can obtain PMLE(A, B, C) directly: PMLE(A, B, C) = [[w, r, wll/[[*, *, *H Let I(w,r,w ~) denote the amount information contained in Hw, r,w~]]=c. Its value can be corn769 simgindZe(Wl, W2) = ~\'~(r,w)eTCwl)NTCw2)Aresubj.of.obj-of} min(I(Wl, r, w), I(w2, r, w) ) simHindte, (Wl, W2) = ~,(r,w)eT(w,)nT(w2) min(I(wl, r, w), I(w2, r, w)) ]T(Wl)NT(w2)I simcosine(Wl,W2) = x/IZ(w~)llZ(w2)l 2x IT(wl)nZ(w2)l simDice(Wl, W2) = iT(wl)l+lT(w2) I simJacard (Wl, W2) = T(wl )OT(w2)l T(wl) + T(w2)l-IT(Wl)rlT(w2)l Figure 1: Other Similarity Measures puted as follows: I(w,r,w\') = _ Iog(PMLE(B)PMLE(A]B)PMLE(CIB)) --(-log PMLE(A, B, C)) log IIw,r,wflll*,r,*ll -IIw,r,*ll xll*,r,w\'ll It is worth noting that I(w,r,w\') is equal to the mutual information between w and w\' (Hindle, 1990).',Hindle,'' +o,6009,'We use the same set of binary features as in previous work on this dataset ',Pang,'nsubj_use_We det_set_the amod_set_same dobj_use_set prep_set_of amod_features_binary pobj_of_features dep_in_as prep_use_in amod_work_previous pobj_in_work prep_work_on det_dataset_this pobj_on_dataset ' +o,6010,'At each training-set size , a new copy of the network is trained under each of the following conditions : -LRB- 1 -RRB- using SULU , -LRB- 2 -RRB- using SULU but supplying only the labeled training examples to synthesize , -LRB- 3 -RRB- standard network training , -LRB- 4 -RRB- using a re-implementation of an algorithm proposed by , and -LRB- 5 -RRB- using standard network training but with all training examples labeled to establish an upper bound ',Yarowsky,'dep_proposed_At det_size_each amod_size_training-set pobj_At_size det_copy_a amod_copy_new nsubjpass_trained_copy prep_copy_of det_network_the pobj_of_network auxpass_trained_is dep_proposed_trained prep_trained_under pobj_under_each prep_each_of det_conditions_the amod_conditions_following pobj_of_conditions appos_each_1 ccomp_trained_using acomp_using_SULU dep_SULU_2 xcomp_using_using dobj_using_SULU cc_using_but conj_using_supplying advmod_examples_only det_examples_the nn_examples_labeled nn_examples_training dobj_supplying_examples aux_synthesize_to xcomp_supplying_synthesize dep_proposed_3 amod_network_standard nsubj_proposed_network dep_network_training appos_network_4 partmod_network_using det_re-implementation_a dobj_using_re-implementation prep_re-implementation_of det_algorithm_an pobj_of_algorithm prep_proposed_by pobj_by_ cc_proposed_and nsubj_labeled_5 dep_5_using amod_training_standard nn_training_network dobj_using_training cc_using_but conj_using_with det_examples_all amod_examples_training pobj_with_examples conj_proposed_labeled aux_establish_to xcomp_labeled_establish det_bound_an amod_bound_upper dobj_establish_bound ' +o,6011,' -RRB- , less prior work exists for bilingual acquisition of domain-specific translations ',Smadja,'dep_exists_less amod_work_prior dep_less_work dobj__exists prep_exists_for amod_acquisition_bilingual pobj_for_acquisition prep__of amod_translations_domain-specific pobj_of_translations ' +o,6012,'1 Introduction Word alignmentdetection of corresponding words between two sentences that are translations of each otheris usually an intermediate step of statistical machine translation -LRB- MT -RRB- , but also has been shown useful for other applications such as construction of bilingual lexicons , word-sense disambiguation , projection of resources , and crosslanguage information retrieval ',Brown,'num_alignmentdetection_1 nn_alignmentdetection_Introduction nn_alignmentdetection_Word prep_alignmentdetection_of amod_words_corresponding pobj_of_words prep_words_between num_sentences_two pobj_between_sentences nsubj_translations_that cop_translations_are rcmod_sentences_translations prep_translations_of det_otheris_each pobj_of_otheris advmod_translations_usually det_step_an amod_step_intermediate nsubjpass__step prep_step_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_MT dep_alignmentdetection_ cc__but advmod_shown_also aux_shown_has auxpass_shown_been conj__shown acomp_shown_useful prep_shown_for amod_applications_other pobj_for_applications dep_as_such prep_applications_as pobj_as_construction prep_construction_of amod_lexicons_bilingual pobj_of_lexicons amod_disambiguation_word-sense conj_lexicons_disambiguation conj_lexicons_projection prep_projection_of pobj_of_resources cc_lexicons_and amod_retrieval_crosslanguage nn_retrieval_information conj_lexicons_retrieval ' +o,6013,'1 Introduction Empty categories -LRB- also called null elements -RRB- are used in the annotation of the PENN treebank in order to represent syntactic phenomena like constituent movement -LRB- eg whextraction -RRB- , discontinuous constituents , and missing elements -LRB- PRO elements , empty complementizers and relative pronouns -RRB- ',Marcus,'num_categories_1 nn_categories_Introduction amod_categories_Empty nsubjpass_used_categories advmod_called_also dep_categories_called amod_elements_null dobj_called_elements auxpass_used_are prep_used_in det_annotation_the pobj_in_annotation prep_annotation_of det__the nn__PENN nn__treebank pobj_of_ mark_represent_in dep_represent_order aux_represent_to purpcl_used_represent amod_phenomena_syntactic dobj_represent_phenomena prep_represent_like amod_movement_constituent pobj_like_movement nn_whextraction_eg appos_movement_whextraction amod_constituents_discontinuous conj_movement_constituents cc_movement_and amod_elements_missing conj_movement_elements amod_elements_PRO dep_elements_elements amod_complementizers_empty conj_elements_complementizers cc_elements_and amod_pronouns_relative conj_elements_pronouns ' +o,6014,'Collocation Dictionary of Modern Chinese Lexical Words , Business Publisher , China Yuan Liu , et al 1993 ',Marcus,'nn_Dictionary_Collocation prep_Dictionary_of nn_Words_Modern nn_Words_Chinese nn_Words_Lexical pobj_of_Words nn_Publisher_Business conj_Words_Publisher nn_Liu_China nn_Liu_Yuan conj_Words_Liu cc_Words_et conj_Words_al num_al_1993 ' +o,6015,'The probability distributions of these binary classifiers are learned using maximum entropy model ',Berger,'det_distributions_The nn_distributions_probability nsubjpass_learned_distributions prep_distributions_of det_classifiers_these amod_classifiers_binary pobj_of_classifiers auxpass_learned_are xcomp_learned_using amod_model_maximum nn_model_entropy dobj_using_model ' +o,6016,'Weischedel \'s group examines unknown words in the context of part-of-speech tagging ',Marcus,'poss__Weischedel possessive_Weischedel_\'s nn__group nsubj_examines_ amod_words_unknown dobj_examines_words prep_examines_in det_context_the pobj_in_context prep_context_of amod_tagging_part-of-speech pobj_of_tagging ' +o,6017,'We use the publicly available ROUGE toolkit tocomputerecall , precision , andF-scorefor ROUGE-1 ',Lin,'nsubj_use_We det_toolkit_the advmod_available_publicly amod_toolkit_available nn_toolkit_ROUGE dobj_use_toolkit xcomp_use_ nn_ROUGE-1_tocomputerecall dep_ROUGE-1_precision amod_ROUGE-1_andF-scorefor dobj__ROUGE-1 ' +o,6018,'Recently , Wikipedia is emerging as a source for extracting semantic relationships ',Kazama,'advmod_emerging_Recently nsubj_emerging_Wikipedia aux_emerging_is prep_emerging_as det_source_a pobj_as_source prep_source_for pcomp_for_extracting amod_relationships_semantic dobj_extracting_relationships ' +o,6019,'The observation that shallow syntactic information can be extracted using local information by examining the pattern itself , its nearby context and the local part-of-speech information has motivated the use of learning methods to recognize these patterns ',Ramshaw,'det_observation_The nsubj_shallow_that rcmod_observation_shallow amod_information_syntactic nsubjpass_extracted_information aux_extracted_can auxpass_extracted_be ccomp_shallow_extracted purpcl_extracted_using amod_information_local dobj_using_information prep_using_by pcomp_by_examining det_pattern_the dobj_examining_pattern nsubj_motivated_itself poss_context_its amod_context_nearby conj_itself_context cc_itself_and det_information_the amod_information_local amod_information_part-of-speech conj_itself_information aux_motivated_has dep_observation_motivated det_use_the dobj_motivated_use prep_use_of pcomp_of_learning dobj_learning_methods aux_recognize_to xcomp_learning_recognize det_patterns_these dobj_recognize_patterns ' +o,6020,'We report case-insensitive scores for version 06 of METEOR with all modules enabled , version 104 of IBM-style BLEU , and version 5 of TER ',Papineni,'nsubj_report_We amod_scores_case-insensitive dobj_report_scores prep_scores_for pobj_for_version num_version_06 prep_version_of nn__METEOR pobj_of_ prep__with det_modules_all pobj_with_modules partmod_modules_enabled conj_scores_version num_version_104 prep_version_of nn__IBM-style nn__BLEU pobj_of_ cc_scores_and conj_scores_version num_version_5 prep_version_of nn__TER pobj_of_ ' +o,6021,'It is dubious whether SWD is useful regarding recall-oriented metrics like METEOR , since SWD removes information in source sentences ',Banerjee,'nsubj_dubious_It cop_dubious_is complm_useful_whether nsubj_useful_SWD cop_useful_is ccomp_dubious_useful xcomp_useful_regarding amod_metrics_recall-oriented dobj_regarding_metrics prep_regarding_like nn__METEOR pobj_like_ mark_removes_since nsubj_removes_SWD advcl_useful_removes dobj_removes_information prep_information_in nn_sentences_source pobj_in_sentences ' +o,6022,'All the feature weights -LRB- s -RRB- were trained using our implementation of Minimum Error Rate Training ',Och,'predet_weights_All det_weights_the nn_weights_feature nsubjpass_trained_weights dep_-LRB-_s auxpass_trained_were xcomp_trained_using poss_implementation_our dobj_using_implementation prep_implementation_of nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_of_Training ' +o,6023,'In the first approach , heuristic rules are used to find the dependencies or penalties for label inconsistency are required to handset ad-hoc ',Finkel,'prep_used_In det_approach_the amod_approach_first pobj_In_approach amod_rules_heuristic nsubjpass_used_rules auxpass_used_are aux_find_to xcomp_used_find det_dependencies_the dobj_find_dependencies nsubjpass_required_ cc__or conj__penalties prep__for nn_inconsistency_label pobj_for_inconsistency auxpass_required_are dep_used_required aux_handset_to xcomp_required_handset dobj_handset_ad-hoc ' +o,6024,'2 Statistical Translation Engine A word-based translation engine is used based on the so-called IBM-4 model ',Brown,'num_Translation_2 nn_Translation_Statistical nsubj_Engine_Translation det_engine_A amod_engine_word-based nn_engine_translation nsubjpass_used_engine auxpass_used_is ccomp_Engine_used prep_used_based dep_based_on det_model_the amod_model_so-called nn_model_IBM-4 pobj_on_model ' +o,6025,' uses a mutual-information based metric derived from the distribution of subject , verb and object in a large corpus to classify nouns ',Hindle,'nsubj_uses_ det_mutual-information_a dobj_uses_mutual-information prep_uses_based dep_based_metric partmod_metric_derived prep_derived_from det_distribution_the pobj_from_distribution prep_distribution_of pobj_of_subject conj_subject_verb cc_subject_and conj_subject_object prep_derived_in det_corpus_a amod_corpus_large pobj_in_corpus aux_classify_to xcomp_derived_classify dobj_classify_nouns ' +n,6026,'It is faster and more mnemonic than the one in ',Dunning,'nsubj_mnemonic_It cop_mnemonic_is advmod_mnemonic_faster cc_faster_and conj_faster_more prep_mnemonic_than det_one_the measure_in_one dep_than_in ' +o,6027,'In analyzing opinions , judging document-level subjectivity , and answering opinion questions , the output of a sentence-level subjectivity classification can be used without modification ',Turney,'prep_used_In pcomp_In_analyzing nn__opinions dobj_analyzing_ ccomp_used_judging amod__document-level nn__subjectivity dobj_judging_ cc_judging_and conj_judging_answering nn_questions_opinion dobj_answering_questions advmod_answering_ det_output_the nsubjpass_used_output prep_output_of det_classification_a amod_classification_sentence-level nn_classification_subjectivity pobj_of_classification aux_used_can auxpass_used_be prep_used_without pobj_without_modification ' +o,6028,'Theyalsoappliedself-training to domain adaptation of a constituency parser ',McClosky,'prep_Theyalsoappliedself-training_to nn_adaptation_domain pobj_to_adaptation prep_Theyalsoappliedself-training_of det_parser_a nn_parser_constituency pobj_of_parser ' +o,6029,'In addition to collocation translation , there is also some related work in acquiring phrase or term translations from parallel corpus ',Dunning,'prep_is_In pobj_In_addition prep_addition_to nn_translation_collocation pobj_to_translation expl_is_there advmod_is_also det_work_some amod_work_related nsubj_is_work prep_work_in pcomp_in_acquiring nn_translations_phrase cc_phrase_or conj_phrase_term dobj_acquiring_translations prep_acquiring_from amod__parallel nn__corpus pobj_from_ ' +o,6030,' applied an internet-based technique to the semantic orientation classification of phrases , which had originally been developed for word sentiment classification ',Turney,'nsubj_applied_ det_technique_an amod_technique_internet-based dobj_applied_technique prep_applied_to det_classification_the amod_classification_semantic nn_classification_orientation pobj_to_classification prep_classification_of pobj_of_phrases nsubjpass_developed_which aux_developed_had advmod_developed_originally auxpass_developed_been rcmod_phrases_developed prep_developed_for nn_classification_word nn_classification_sentiment pobj_for_classification ' +o,6031,'Hyponymy relations were extracted from definition sentences ',Kazama,'amod_relations_Hyponymy nsubjpass_extracted_relations auxpass_extracted_were prep_extracted_from nn_sentences_definition pobj_from_sentences ' +n,6032,'1 Introduction Translations tables in Phrase-based Statistical Machine Translation -LRB- SMT -RRB- are often built on the basis of Maximum-likelihood Estimation -LRB- MLE -RRB- , being one of the major limitations of this approach that the source sentence context in which phrases occur is completely ignored ',Koehn,'num_tables_1 nn_tables_Introduction nn_tables_Translations nsubjpass_built_tables prep_tables_in amod_Translation_Phrase-based nn_Translation_Statistical nn_Translation_Machine pobj_in_Translation abbrev_tables_SMT auxpass_built_are advmod_built_often prep_built_on det_basis_the pobj_on_basis prep_basis_of nn_Estimation_Maximum-likelihood pobj_of_Estimation abbrev_Estimation_MLE cop_one_being xcomp_built_one prep_one_of det_limitations_the amod_limitations_major pobj_of_limitations prep_limitations_of det_approach_this pobj_of_approach complm_ignored_that det_context_the nn_context_source nn_context_sentence nsubjpass_ignored_context rel_occur_in pobj_in_which nsubj_occur_phrases rcmod_context_occur auxpass_ignored_is advmod_ignored_completely dep_approach_ignored ' +o,6033,'The cube-pruning by Chiang and the lazy cube-pruning of and Chiang turn the computation of beam pruning of CYK decoders into a top-k selection problem given two columns of translation hypotheses that need to be combined ',Huang,'det_cube-pruning_The nsubj_turn_cube-pruning prep_cube-pruning_by nn__Chiang pobj_by_ cc__and det_cube-pruning_the amod_cube-pruning_lazy conj__cube-pruning prep_cube-pruning_of nn__ cc__and conj__Chiang pobj_of_ det_computation_the dobj_turn_computation prep_computation_of nn_pruning_beam pobj_of_pruning prep_pruning_of nn_decoders_CYK pobj_of_decoders prep_turn_into det_problem_a amod_problem_top-k nn_problem_selection pobj_into_problem partmod_problem_given num_columns_two dobj_given_columns prep_columns_of nn_hypotheses_translation pobj_of_hypotheses nsubj_need_that rcmod_columns_need aux_combined_to auxpass_combined_be xcomp_need_combined ' +p,6034,'Much later work relies on the use of extremely large corpora which allow very precise , but sparse features ',Cucerzan,'amod__Much amod__later nn__work nsubj_relies_ prep_relies_on det_use_the pobj_on_use prep_use_of advmod_large_extremely amod_corpora_large pobj_of_corpora nsubj_allow_which rcmod_corpora_allow advmod_precise_very amod_features_precise cc_precise_but conj_precise_sparse dobj_allow_features ' +o,6035,'In designing LEAF , we were also inspired by dependency-based alignment models ',Wu,'prep_inspired_In pcomp_In_designing dobj_designing_LEAF nsubjpass_inspired_we auxpass_inspired_were advmod_inspired_also prep_inspired_by amod_models_dependency-based nn_models_alignment pobj_by_models ' +p,6036,'The simplest one is the BIO representation scheme , where a B denotes the first item of an element and an I any non-initial item , and a syllable with tag O is not a part of any element ',Ramshaw,'det_one_The amod_one_simplest nsubj__one cop__is det__the nn__BIO nn__representation nn__scheme advmod_denotes_where det_B_a nsubj_denotes_B rcmod__denotes det_item_the amod_item_first dobj_denotes_item prep_item_of det_element_an pobj_of_element cc_item_and det_I_an conj_item_I det_item_any amod_item_non-initial dep_I_item cc__and det_O_a amod_O_syllable prep_syllable_with pobj_with_tag nsubj_part_O cop_part_is neg_part_not det_part_a conj__part prep_part_of det_element_any pobj_of_element ' +o,6037,'1 Introduction Since their appearance , string-based evaluation metrics such as BLEU and NIST have been the standard tools used for evaluating MT quality ',Papineni,'dep_been_1 dep_Since_Introduction prep_been_Since poss_appearance_their pobj_Since_appearance amod_metrics_string-based nn_metrics_evaluation nsubj_been_metrics dep_as_such prep_metrics_as nn__BLEU pobj_as_ cc__and nn__NIST conj__ aux_been_have det_tools_the amod_tools_standard dobj_been_tools partmod_tools_used prep_used_for pcomp_for_evaluating nn_quality_MT dobj_evaluating_quality ' +o,6038,'622 We also identified a length effect similar to that studied by for self-training -LRB- using a reranker and large seed , as detailed in Section 2 -RRB- ',McClosky,'nsubj_identified_We advmod_identified_also rcmod_622_identified det_effect_a nn_effect_length nsubj_similar_effect xcomp_identified_similar prep_similar_to pobj_to_that partmod_that_studied prep_studied_by pobj_by_ prep__for pobj_for_self-training dep_622_using det_seed_a amod_seed_reranker cc_reranker_and conj_reranker_large dobj_using_seed advmod_detailed_as amod_seed_detailed prep_using_in pobj_in_Section num_Section_2 ' +o,6039,'51 Pharaoh The baseline system we used for comparison was Pharaoh , a freely available decoder for phrase-based translation models : p -LRB- e f -RRB- = p -LRB- f e -RRB- pLM -LRB- e -RRB- LM pD -LRB- e , f -RRB- D length -LRB- e -RRB- W -LRB- e -RRB- -LRB- 10 -RRB- We ran GIZA + + on the training corpus in both directions using its default setting , and then applied the refinement rule diagand described in to obtain a single many-to-many word alignment for each sentence pair ',Koehn,'num_Pharaoh_51 det_system_The amod_system_baseline nsubj__system nsubj_used_we rcmod_system_used prep_used_for pobj_for_comparison cop__was nn__Pharaoh dep_Pharaoh_ det_decoder_a advmod_available_freely amod_decoder_available appos__decoder prep_decoder_for amod_models_phrase-based nn_models_translation pobj_for_models dep_decoder_p dep_f_e dep_-LRB-_f rcmod_p_= amod_W_p dep_-LRB-_f dep_f_e amod_W_pLM dep_-LRB-_e nn_W_LM amod_W_pD dep_W_e appos_e_f nn_W_D nn_W_length dep_-LRB-_e dobj_=_W dep_-LRB-_e appos_W_10 nsubj_ran_We rcmod_W_ran nsubj_+_GIZA xcomp_ran_+ cc_+_+ conj_+_ prep_=_on det_corpus_the nn_corpus_training pobj_on_corpus prep_corpus_in det_directions_both pobj_in_directions dep_Pharaoh_using poss_setting_its nn_setting_default dobj_using_setting cc_using_and advmod_using_then conj_using_applied det_diagand_the amod_diagand_refinement nn_diagand_rule dobj_applied_diagand partmod_diagand_described prep_described_in pobj_in_ aux_obtain_to xcomp_applied_obtain det_alignment_a amod_alignment_single amod_alignment_many-to-many nn_alignment_word dobj_obtain_alignment prep_alignment_for det_pair_each nn_pair_sentence pobj_for_pair ' +o,6040,'42 Data The data comes from the CoNLL 2000 shared task , which consists of sentences from the Penn Treebank Wall Street Journal corpus ',Marcus,'num_data_42 nn_data_Data det_data_The nsubj_comes_data prep_comes_from det__the nn__CoNLL num__2000 amod__shared nn__task pobj_from_ nsubj_consists_which rcmod__consists prep_consists_of pobj_of_sentences prep_sentences_from det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_from_corpus ' +o,6041,'Dredze et al also indicated that unlabeled dependency parsing is not robust to domain adaptation ',Dredze,'nsubj_indicated_Dredze cc_Dredze_et conj_Dredze_al advmod_indicated_also complm_robust_that amod_parsing_unlabeled amod_parsing_dependency nsubj_robust_parsing cop_robust_is neg_robust_not ccomp_indicated_robust prep_robust_to nn_adaptation_domain pobj_to_adaptation ' +o,6042,'In addition , we developed a word clustering procedure -LRB- based on a standard approach -RRB- that optimizes conditional word clusters ',Brown,'prep_developed_In pobj_In_addition nsubj_developed_we det_procedure_a nn_procedure_word amod_procedure_clustering dobj_developed_procedure dep_procedure_based prep_based_on det__a amod__standard nn__approach pobj_on_ nsubj_optimizes_that rcmod_procedure_optimizes amod_clusters_conditional nn_clusters_word dobj_optimizes_clusters ' +o,6043,'Specifically , three features are used to instantiate the templates : POS tags on both sides : We assign POS tags using the MXPOST tagger for English and Chinese , and Connexor for Spanish ',Ratnaparkhi,'advmod_used_Specifically num_features_three nsubjpass_used_features auxpass_used_are aux_instantiate_to xcomp_used_instantiate det_templates_the dobj_instantiate_templates dep_tags_POS dep_templates_tags prep_tags_on det_sides_both pobj_on_sides nsubj_assign_We parataxis_used_assign dep_tags_POS nsubj_using_tags dep_assign_using det__the amod__MXPOST nn__tagger dobj_using_ prep__for pobj_for_English cc_English_and conj_English_Chinese cc__and conj__Connexor prep_Connexor_for pobj_for_Spanish ' +o,6044,'Many 649 similarity measures and weighting functions have been proposed for distributional vectors ; comparative studies include Lee , Curran and ',Weeds,'amod_measures_Many num_measures_649 nn_measures_similarity nsubjpass_proposed_measures cc_measures_and nn_functions_weighting conj_measures_functions aux_proposed_have auxpass_proposed_been prep_proposed_for amod_vectors_distributional pobj_for_vectors amod_studies_comparative nsubj_include_studies parataxis_proposed_include nn__Lee dobj_include_ nn_and_Curran nn_and_ appos__and ' +o,6045,'Also , we chose to average each individual perceptron prior to Bayesian averaging ',Collins,'advmod_chose_Also nsubj_chose_we aux_average_to xcomp_chose_average det__each amod__individual nn__perceptron dobj_average_ advmod_average_prior dep_prior_to nn_averaging_Bayesian pobj_to_averaging ' +o,6046,'Wiebe uses style distributionally similar adjectives in a cluster-and-label process to generate sentiment lexicon of adjectives ',Lin,'nn__Wiebe nsubj_uses_ amod_adjectives_ nn_adjectives_style advmod_similar_distributionally amod_adjectives_similar dobj_uses_adjectives prep_uses_in det_process_a amod_process_cluster-and-label pobj_in_process aux_generate_to infmod_process_generate nn_lexicon_sentiment dobj_generate_lexicon prep_lexicon_of pobj_of_adjectives ' +p,6047,'Conditional Markov models -LRB- CMM -RRB- have been successfully used in sequence labeling tasks incorporating rich feature sets ',Ratnaparkhi,'nn_models_Conditional nn_models_Markov nn__models abbrev__CMM nsubjpass_used_ aux_used_have auxpass_used_been advmod_used_successfully prep_used_in nn_tasks_sequence nn_tasks_labeling pobj_in_tasks xcomp_used_incorporating amod_sets_rich nn_sets_feature dobj_incorporating_sets ' +o,6048,'Gildea and Jurafsky used a supervised learning method to learn both the identifier of the semantic roles defined in FrameNet such as theme , target , goal , and the boundaries of the roles ',Pang,'nn__Gildea cc_Gildea_and conj_Gildea_Jurafsky nsubj_used_ det_method_a amod_method_supervised nn_method_learning dobj_used_method aux_learn_to xcomp_used_learn predet_identifier_both det_identifier_the dobj_learn_identifier prep_identifier_of det_roles_the amod_roles_semantic pobj_of_roles partmod_roles_defined prep_defined_in pobj_in_FrameNet dep_as_such prep_FrameNet_as pobj_as_theme conj_theme_target conj_theme_goal cc_theme_and det_boundaries_the conj_theme_boundaries prep_boundaries_of det__the nn__roles pobj_of_ ' +o,6049,'To simplify , the plausibility of a detected esl is roughly inversely proportional to the number of mutually excluding syntactic structures in the text segment that generated the esl -LRB- see for details -RRB- ',Marcus,'aux_simplify_To dep_proportional_simplify det_plausibility_the nsubjpass_proportional_plausibility prep_plausibility_of det_esl_a amod_esl_detected pobj_of_esl auxpass_proportional_is advmod_proportional_roughly advmod_proportional_inversely prep_proportional_to det_number_the pobj_to_number prep_number_of pobj_of_mutually prep_mutually_excluding amod_structures_syntactic pobj_excluding_structures prep_structures_in det_segment_the nn_segment_text pobj_in_segment nsubj_generated_that rcmod_segment_generated det_esl_the dobj_generated_esl dep_esl_see dobj_see_ prep__for pobj_for_details ' +o,6050,'Alignment is often used in training both generative and discriminative models ',Liang,'nsubjpass_used_Alignment auxpass_used_is advmod_used_often prep_used_in pcomp_in_training preconj_generative_both amod_models_generative cc_generative_and conj_generative_discriminative dobj_training_models ' +o,6051,'In the past two or three years , this kind of verification has been attempted for other aspects of semantic interpretation : by Passonneau and Litman for segmentation and by Kowtko , Isard , and Doherty and Carletta et al ',Marcus,'prep_attempted_In det_two_the amod_two_past pobj_In_two cc_two_or num_years_three conj_two_years det_kind_this nsubjpass_attempted_kind prep_kind_of pobj_of_verification aux_attempted_has auxpass_attempted_been prep_attempted_for amod_aspects_other pobj_for_aspects prep_aspects_of amod_interpretation_semantic pobj_of_interpretation prep_attempted_by nn__Passonneau cc_Passonneau_and conj_Passonneau_Litman pobj_by_ prep__for pobj_for_segmentation cc_by_and conj_by_by nn_Isard_Kowtko pobj_by_Isard cc_Isard_and nn__Doherty conj_Isard_ cc_Isard_and conj_Isard_Carletta cc_Carletta_et conj_Carletta_al ' +o,6052,'We adopt the approach of , using a small set of patterns to build relation models , and extend their work by re ning the training and classi cation process using parameter optimization , topic segmentation and syntactic parsing ',Marcu,'nsubj_adopt_We det_approach_the dobj_adopt_approach prep_approach_of pobj_of_ xcomp_adopt_using det_set_a amod_set_small dobj_using_set prep_set_of pobj_of_patterns aux_build_to xcomp_using_build nn_models_relation dobj_build_models cc_build_and conj_build_extend poss_work_their dobj_extend_work prep_extend_by pobj_by_re partmod_re_ning det_training_the dobj_ning_training cc_training_and nn_process_classi nn_process_cation conj_training_process xcomp_ning_using nn_optimization_parameter dobj_using_optimization nn_segmentation_topic conj_optimization_segmentation cc_optimization_and amod_parsing_syntactic conj_optimization_parsing ' +o,6053,'960 12 Alignment with Mixture Distribution Several papers have discussed the first issue , especially the problem of word alignments for bilingual corpora , , , , ',Brown,'number_12_960 num_Alignment_12 nsubj_discussed_Alignment prep_Alignment_with nn_papers_Mixture nn_papers_Distribution amod_papers_Several pobj_with_papers aux_discussed_have det_issue_the amod_issue_first dobj_discussed_issue advmod_issue_especially det_problem_the dep_issue_problem prep_problem_of nn_alignments_word pobj_of_alignments prep_alignments_for amod__bilingual nn__corpora nn__ appos__ appos__ pobj_for_ advmod_discussed_ ' +o,6054,'We measure this association using pointwise Mutual Information -LRB- MI -RRB- ',Church,'nsubj_measure_We det_association_this dobj_measure_association partmod_association_using nn_Information_pointwise nn_Information_Mutual dobj_using_Information abbrev_Information_MI ' +p,6055,'To compare the output of their shallow parser with the output of the well-known parser , Li and Roth applied the chunklink conversion script to extract the shallow constituents from the output of the Collins parser on WSJ section 00 ',Collins,'aux_compare_To dep_applied_compare det_output_the dobj_compare_output prep_output_of poss_parser_their nn_parser_shallow pobj_of_parser prep_compare_with det_output_the pobj_with_output prep_output_of det_parser_the amod_parser_well-known amod_parser_ pobj_of_parser nsubj_applied_Li cc_Li_and conj_Li_Roth det_script_the nn_script_chunklink nn_script_conversion dobj_applied_script aux_extract_to xcomp_applied_extract det_constituents_the nn_constituents_shallow dobj_extract_constituents prep_extract_from det_output_the pobj_from_output prep_output_of det_parser_the nn_parser_Collins pobj_of_parser prep_extract_on pobj_on_WSJ punct_WSJ_section appos_WSJ_00 ' +o,6056,' and classified sentiment polarity of reviews at the document level ',Turney,'amod_polarity_ cc__and conj__ amod_polarity_ amod_polarity_classified nn_polarity_sentiment prep_polarity_of pobj_of_reviews prep_reviews_at det_level_the nn_level_document pobj_at_level ' +o,6057,'There are many method proposed to extract rigid expressions from corpora such as a method of focusing on the binding strength of two words ; the distance between words ; and the number of combined words and frequency of appearance ',Smadja,'expl_are_There amod_method_many nsubj_are_method partmod_method_proposed aux_extract_to xcomp_proposed_extract amod_expressions_rigid dobj_extract_expressions prep_extract_from pobj_from_corpora dep_as_such prep_corpora_as det_method_a pobj_as_method prep_method_of pcomp_of_focusing prep_focusing_on det_strength_the amod_strength_binding pobj_on_strength prep_strength_of num__two nn__words pobj_of_ det_distance_the dep_method_distance prep_distance_between nn__words pobj_between_ cc_method_and det_number_the conj_method_number prep_number_of amod_words_combined pobj_of_words cc_words_and conj_words_frequency prep_words_of nn__appearance pobj_of_ ' +o,6058,'Estimation of the parameters has been described elsewhere ',Brown,'nsubjpass_described_Estimation prep_Estimation_of det_parameters_the pobj_of_parameters aux_described_has auxpass_described_been advmod_described_elsewhere ' +o,6059,'It differs from the many approaches where -LRB- 1 -RRB- is defined by a stochastic synchronous grammar and from transfer-based systems defined by context-free grammars ',Wu,'nsubj_differs_It prep_differs_from det_approaches_the amod_approaches_many pobj_from_approaches advmod_defined_where nsubjpass_defined_1 auxpass_defined_is rcmod_approaches_defined prep_defined_by det__a amod__stochastic amod__synchronous nn__grammar pobj_by_ cc_from_and conj_from_from amod_systems_transfer-based pobj_from_systems partmod_systems_defined prep_defined_by amod__context-free nn__grammars pobj_by_ ' +o,6060,'This approach to minimally supervised classifier construction has been widely studied , especially in cases in which the features of interest are orthogonal in some sense ',Yarowsky,'det_approach_This nsubjpass_studied_approach prep_approach_to advmod_supervised_minimally amod_construction_supervised nn_construction_classifier pobj_to_construction aux_studied_has auxpass_studied_been advmod_studied_widely advmod_studied_ advmod_studied_especially prep_studied_in pobj_in_cases rel_orthogonal_in pobj_in_which det_features_the nsubj_orthogonal_features prep_features_of pobj_of_interest cop_orthogonal_are rcmod_cases_orthogonal prep_orthogonal_in det__some nn__sense pobj_in_ ' +o,6061,'Due to the lack of a good Arabic parser compatible with the Sakhr tokenization that we used on the source side , we did not test the source dependency LM for Arabic-to-English MT When extracting rules with source dependency structures , we applied the same well-formedness constraint on the source side as we did on the target side , using a procedure described by ',He,'dep_to_Due prep_test_to det_lack_the pobj_to_lack prep_lack_of det_parser_a amod_parser_good nn_parser_Arabic pobj_of_parser amod_parser_compatible prep_compatible_with det_tokenization_the nn_tokenization_Sakhr pobj_with_tokenization complm_used_that nsubj_used_we dep_tokenization_used prep_used_on det_side_the nn_side_source pobj_on_side nsubj_test_we aux_test_did neg_test_not det_LM_the nn_LM_source nn_LM_dependency dobj_test_LM prep_test_for nn_MT_Arabic-to-English pobj_for_MT advmod_applied_When dep_applied_extracting dobj_extracting_rules prep_extracting_with nn_structures_source nn_structures_dependency pobj_with_structures nsubj_applied_we dep_MT_applied det_constraint_the amod_constraint_same amod_constraint_well-formedness dobj_applied_constraint prep_applied_on det_side_the nn_side_source pobj_on_side mark_did_as nsubj_did_we advcl_applied_did prep_did_on det_side_the nn_side_target pobj_on_side xcomp_did_using det_procedure_a dobj_using_procedure partmod_procedure_described prep_described_by pobj_by_ ' +o,6062,'Its roots are the same as computational linguistics -LRB- CL -RRB- , but it has been largely ignored in CL until recently ',Dunning,'poss_roots_Its nsubj_linguistics_roots cop_linguistics_are det_linguistics_the amod_linguistics_same advmod_computational_as amod_linguistics_computational abbrev_linguistics_CL cc_linguistics_but nsubjpass_ignored_it aux_ignored_has auxpass_ignored_been advmod_ignored_largely conj_linguistics_ignored prep_ignored_in pobj_in_CL prep_ignored_until dep_until_recently ' +o,6063,'Unlabeled dependencies can be readily obtained by processing constituent trees , such as those in the Penn Treebank , with a set of rules to determine the lexical heads of constituents ',Marcus,'amod_dependencies_Unlabeled nsubjpass_obtained_dependencies aux_obtained_can auxpass_obtained_be advmod_obtained_readily prep_obtained_by nn_trees_processing nn_trees_constituent pobj_by_trees dep_as_such prep_trees_as pobj_as_those prep_those_in det__the nn__Penn nn__Treebank pobj_in_ prep_obtained_with det_set_a pobj_with_set prep_set_of pobj_of_rules aux_determine_to purpcl_obtained_determine det_heads_the amod_heads_lexical dobj_determine_heads prep_heads_of pobj_of_constituents ' +o,6064,'These models include a standard unlexicalized PCFG parser , a head-lexicalized parser , and a maximum-entropy inspired parser ',Collins,'det_models_These nsubj_include_models det_parser_a amod_parser_standard amod_parser_unlexicalized nn_parser_PCFG dobj_include_parser det__a amod__head-lexicalized nn__parser conj_parser_ cc_parser_and det__a amod__maximum-entropy amod__inspired nn__parser conj_parser_ ' +p,6065,'While the former is piecewise constant and thus can not be optimized using gradient techniques , provides an approach that performs such training efficiently ',Och,'mark_constant_While det_former_the nsubjpass_constant_former cop_constant_is advmod_constant_piecewise advcl_provides_constant cc_constant_and advmod_optimized_thus aux_optimized_can neg_optimized_not auxpass_optimized_be conj_constant_optimized purpcl_optimized_using amod_techniques_gradient dobj_using_techniques nsubj_provides_ det_approach_an dobj_provides_approach nsubj_performs_that rcmod_approach_performs amod_training_such dobj_performs_training advmod_performs_efficiently ' +o,6066,'For example , in the WSJ corpus , part of the Penn Treebank 3 release , the string in -LRB- 1 -RRB- is a variation 12-gram since off is a variation nucleus that in one corpus occurrence is tagged as a preposition -LRB- IN -RRB- , while in another it is tagged as a particle -LRB- RP -RRB- ',Marcus,'prep_nucleus_For pobj_For_example prep_nucleus_in det_corpus_the nn_corpus_WSJ pobj_in_corpus nsubj_nucleus_part prep_part_of det__the nn__Penn nn__Treebank num__3 nn__release pobj_of_ det_string_the appos__string prep_string_in pobj_in_1 cop_12-gram_is det_12-gram_a nn_12-gram_variation rcmod_1_12-gram dep_off_since advmod_12-gram_off cop_nucleus_is det_nucleus_a nn_nucleus_variation nsubjpass_tagged_that prep_tagged_in num_occurrence_one nn_occurrence_corpus pobj_in_occurrence auxpass_tagged_is rcmod_nucleus_tagged prep_tagged_as det_preposition_a pobj_as_preposition abbrev_preposition_IN mark_tagged_while prep_tagged_in pobj_in_another nsubjpass_tagged_it auxpass_tagged_is advcl_tagged_tagged prep_tagged_as det_particle_a pobj_as_particle abbrev_particle_RP ' +o,6067,'42 Features For our experiments , we use a feature set analogous to the default feature set of Pharaoh ',Och,'num_Features_42 prep_use_For poss_experiments_our pobj_For_experiments nsubj_use_we dep_Features_use det_feature_a nsubj_set_feature ccomp_use_set acomp_set_analogous prep_analogous_to det_set_the nn_set_default nn_set_feature pobj_to_set prep_set_of pobj_of_Pharaoh ' +o,6068,'For a comparison , we also include the ROUGE-1 Fscores of each system output against the human compressed sentences ',Lin,'prep_include_For det_comparison_a pobj_For_comparison nsubj_include_we advmod_include_also det__the nn__ROUGE-1 nn__Fscores dobj_include_ prep__of det_output_each nn_output_system pobj_of_output prep_output_against det_sentences_the amod_sentences_human num_sentences_compressed pobj_against_sentences ' +o,6069,' , Warnke et al ',Berger,'appos__Warnke cc_Warnke_et conj_Warnke_al ' +o,6070,'We measure translation performance by the BLEU and METEOR scores with multiple translation references ',Banerjee,'nsubj_measure_We nn_performance_translation dobj_measure_performance prep_measure_by det__the nn__BLEU pobj_by_ cc__and nn_scores_METEOR nn_scores_ conj__scores prep_measure_with amod_references_multiple nn_references_translation pobj_with_references ' +o,6071,'F-Measure with an appropriate setting of will be useful during the development process of new alignment models , or as a maximization criterion for discriminative training of alignment models ',Fraser,'prep_F-Measure_with det_setting_an amod_setting_appropriate pobj_with_setting prep_setting_of aux_useful_will cop_useful_be pcomp_of_useful prep_useful_during det_process_the nn_process_development pobj_during_process prep_process_of amod_models_new nn_models_alignment pobj_of_models cc_with_or conj_with_as det_criterion_a nn_criterion_maximization pobj_as_criterion prep_criterion_for amod_training_discriminative pobj_for_training prep_training_of amod_models_alignment pobj_of_models ' +o,6072,'The model does not use context-free rules , but generates the next category using zeroth order Markov chains -LRB- see Section 33 -RRB- , hence no information about the previous sisters is included ',Collins,'det_model_The amod_model_ nsubj_use_model aux_use_does neg_use_not amod_rules_context-free dobj_use_rules cc_use_but conj_use_generates det_category_the amod_category_next dobj_generates_category dep_category_using amod_chains_zeroth nn_chains_order nn_chains_Markov dobj_using_chains parataxis_using_see dobj_see_Section num_Section_33 advmod_information_hence det_information_no dobj_using_information prep_information_about det_sisters_the amod_sisters_previous pobj_about_sisters auxpass_included_is rcmod_sisters_included ' +o,6073,'7 This discussion could also be cast in an information theoretic framework using the notion of ` mutual information \' , estimating the variance of the degree of match in order to find a frequency-threshold ',Church,'det_discussion_This nsubjpass_cast_discussion aux_cast_could advmod_cast_also auxpass_cast_be dep_7_cast prep_cast_in det_information_an pobj_in_information amod_framework_theoretic nsubj_using_framework dep_7_using det_notion_the dobj_using_notion prep_notion_of amod_information_mutual nn__information pobj_of_ dep_using_estimating det_variance_the dobj_estimating_variance prep_variance_of det_degree_the pobj_of_degree prep_degree_of pobj_of_match prep_estimating_in pobj_in_order aux_find_to xcomp_estimating_find det_frequency-threshold_a dobj_find_frequency-threshold ' +o,6074,'In several papers , selection criteria for single word trigger pairs were studied ',Brown,'prep_studied_In amod__several nn__papers pobj_In_ nn_criteria_selection nsubjpass_studied_criteria prep_criteria_for amod_pairs_single nn_pairs_word nn_pairs_trigger pobj_for_pairs auxpass_studied_were ' +o,6075,'See for more details on this approach ',Collins,'dobj_See_ prep_See_for amod_details_more pobj_for_details prep_details_on det_approach_this pobj_on_approach ' +o,6076,'Word alignment was carried out by running Giza + + implementation of IBM Model 4 initialized with 5 iterations of Model 1 , 5 of the HMM aligner , and 3 iterations of Model 4 in both directions and then symmetrizing using the grow-diag-final-and heuristic ',Koehn,'amod_alignment_Word nsubjpass_carried_alignment auxpass_carried_was prt_carried_out prep_carried_by pcomp_by_running amod_implementation_Giza cc_Giza_+ conj_Giza_+ dobj_running_implementation prep_implementation_of nn__IBM nn__Model num__4 amod__initialized prep_initialized_with num_iterations_5 pobj_with_iterations prep_iterations_of pobj_of_Model num_Model_1 num_Model_5 prep_iterations_of det_aligner_the nn_aligner_HMM pobj_of_aligner cc_iterations_and num_iterations_3 conj_iterations_iterations prep_iterations_of pobj_of_Model num_Model_4 pobj_of_ prep__in det_directions_both pobj_in_directions cc_running_and advmod_running_then conj_running_symmetrizing xcomp_symmetrizing_using det_heuristic_the amod_heuristic_grow-diag-final-and dobj_using_heuristic ' +o,6077,'A common choice for the local probabilistic classifier is maximum entropy classifiers ',Berger,'det_choice_A amod_choice_common nsubj_classifiers_choice prep_choice_for det_classifier_the amod_classifier_local amod_classifier_probabilistic pobj_for_classifier cop_classifiers_is amod_classifiers_maximum nn_classifiers_entropy ' +o,6078,'CIT -RRB- ',Pang,'nn_-RRB-_CIT ' +o,6079,'Illustrative clusterings of this type can also be found in Pereira , Tishby , and Lee , , Kneser and Ney , and Brill et al ',Brown,'amod_clusterings_Illustrative nsubjpass_found_clusterings prep_clusterings_of det_type_this pobj_of_type aux_found_can advmod_found_also auxpass_found_be prep_found_in nn_Tishby_Pereira pobj_in_Tishby cc_Tishby_and nn__Lee conj_Tishby_ nn__ conj__Kneser cc__and conj__Ney conj_Tishby_ cc_Tishby_and conj_Tishby_Brill cc_Brill_et conj_Brill_al ' +o,6080,'The decision rule was based on the standard loglinear interpolation of several models , with weights tunedbyMERTonthedevelopmentse ',Och,'det_rule_The nn_rule_decision nsubjpass_based_rule auxpass_based_was prep_based_on det_interpolation_the amod_interpolation_standard nn_interpolation_loglinear pobj_on_interpolation prep_interpolation_of amod_models_several pobj_of_models prep_based_with nn_tunedbyMERTonthedevelopmentse_weights pobj_with_tunedbyMERTonthedevelopmentse ' +o,6081,'Since manual word alignment is an ambiguous task , we also explicitly allow for ambiguous alignments , ie the links are marked as sure -LRB- S -RRB- or possible -LRB- P -RRB- ',Och,'mark_task_Since amod_alignment_manual nn_alignment_word nsubj_task_alignment cop_task_is det_task_an amod_task_ambiguous advcl_allow_task nsubj_allow_we advmod_allow_also advmod_allow_explicitly prep_allow_for amod_alignments_ambiguous pobj_for_alignments appos_alignments_ie det_links_the nsubjpass_marked_links auxpass_marked_are dep_allow_marked advmod_sure_as dep_marked_sure dep_sure_S cc_sure_or conj_sure_possible dep_possible_P ' +o,6082,'Finally , we compare against the mapping from WordNet to the Oxford English Dictionary constructed in , equivalent to clustering based solely on the OED feature ',Navigli,'advmod_compare_Finally nsubj_compare_we prep_compare_against det_mapping_the pobj_against_mapping prep_mapping_from pobj_from_WordNet prep_WordNet_to det_Dictionary_the nn_Dictionary_Oxford nn_Dictionary_English pobj_to_Dictionary partmod_Dictionary_constructed prep_constructed_in pobj_in_ acomp_compare_equivalent prep_equivalent_to pcomp_to_clustering prep_clustering_based dep_on_solely dep_based_on det_feature_the nn_feature_OED pobj_on_feature ' +o,6083,'We have -LRB- 11 -RRB- Hypernym Patterns based on patterns proposed by and , -LRB- 12 -RRB- Sibling Patterns which are basically conjunctions , and -LRB- 13 -RRB- Part-of Patterns based on patterns proposed by and ',Brown,'nsubj_have_We appos_Patterns_11 nn_Patterns_Hypernym dobj_have_Patterns prep_Patterns_based dep_based_on pobj_on_patterns partmod_patterns_proposed prep_proposed_by pobj_by_ cc__and conj__ appos__12 nn_Patterns_Sibling dep__Patterns nsubj_conjunctions_which cop_conjunctions_are advmod_conjunctions_basically rcmod_Patterns_conjunctions cc__and appos_Patterns_13 nn_Patterns_Part-of conj__Patterns partmod_Patterns_based prep_based_on pobj_on_patterns partmod_patterns_proposed prep_proposed_by pobj_by_ cc__and conj__ ' +o,6084,'6 Related Work and Discussion There are several studies that used automatically extracted gazetteers for NER ',Kazama,'num_Work_6 amod_Work_Related nsubj_extracted_Work cc_Work_and conj_Work_Discussion expl_are_There rcmod_Discussion_are amod_studies_several nsubj_are_studies nsubj_used_that rcmod_studies_used advmod_used_automatically dobj_extracted_gazetteers prep_extracted_for pobj_for_NER ' +p,6085,'Successful discriminative parsers have relied on generative models to reduce training time and raise accuracy above generative baselines ',Collins,'amod_parsers_Successful nn_parsers_discriminative nsubj_relied_parsers aux_relied_have prep_relied_on amod_models_generative pobj_on_models aux_reduce_to xcomp_relied_reduce nn_time_training dobj_reduce_time cc_reduce_and conj_reduce_raise dobj_raise_accuracy prep_raise_above amod_baselines_generative pobj_above_baselines ' +o,6086,'In our decoder , we incorporate two pruning techniques described by ',Huang,'prep_incorporate_In poss_decoder_our pobj_In_decoder nsubj_incorporate_we num_techniques_two nn_techniques_pruning dobj_incorporate_techniques partmod_techniques_described prep_described_by ' +p,6087,'Ochs procedure is the most widely-used version of MERT for SMT ',Och,'nn_procedure_Ochs nsubj_version_procedure cop_version_is det_version_the advmod_widely-used_most amod_version_widely-used prep_version_of pobj_of_MERT prep_MERT_for pobj_for_SMT ' +o,6088,'2 Phrase-based statistical machine translation Phrase-based SMT uses a framework of log-linear models to integrate multiple features ',Och,'num_translation_2 amod_translation_Phrase-based amod_translation_statistical nn_translation_machine nn_SMT_Phrase-based nsubj_uses_SMT dep_translation_uses det_framework_a dobj_uses_framework prep_framework_of amod_models_log-linear pobj_of_models partmod_models_ aux_integrate_to xcomp__integrate amod_features_multiple dobj_integrate_features ' +o,6089,'And third , 1This baseNP data set is available via ftp : \\/ \\/ ftpcisupennedu\\/pub\\/chunker \\/ 2Software for generating the data is available from http://lcg-wwwuiaacbe/conl199/npb/ 50 with the FZ = I rate which is equal to -LRB- 2 \\* precision \\* recall -RRB- \\/ -LRB- precision + recall -RRB- ',Ramshaw,'cc_available_And amod_data_third amod_data_1This nn_data_ nn_data_baseNP nsubj_available_data partmod_data_set cop_available_is prep_available_via pobj_via_ftp nn_\\/_\\/ nsubj_ftpcisupennedu\\/pub\\/chunker_\\/ parataxis_available_ftpcisupennedu\\/pub\\/chunker amod_2Software_\\/ dobj_ftpcisupennedu\\/pub\\/chunker_2Software prep_ftpcisupennedu\\/pub\\/chunker_for pcomp_for_generating det_data_the nsubj_available_data cop_available_is ccomp_generating_available prep_available_from amod_50_http://lcg-wwwuiaacbe/conl199/npb/ pobj_from_50 prep_available_with det_rate_the nn_rate_FZ nn_rate_= nn_rate_I pobj_with_rate nsubj_equal_which cop_equal_is rcmod_rate_equal aux_\\/_to num_\\*_2 amod_\\*_\\* nn_\\*_precision nsubj_recall_\\* parataxis_\\/_recall xcomp_equal_\\/ nn_+_precision nsubj_recall_+ dep_ftpcisupennedu\\/pub\\/chunker_recall ' +o,6090,'In , finite-state machine translation is based on and is used for decoding the target language string ',Brown,'prep_based_In pobj_In_ amod_translation_finite-state nn_translation_machine nsubjpass_based_translation auxpass_based_is prep_based_on pobj_on_ cc_based_and auxpass_used_is conj_based_used prep_used_for pcomp_for_decoding det_string_the nn_string_target nn_string_language dobj_decoding_string ' +o,6091,'2 Related Research Several researchers have already proposed methods for binarizing synchronous grammars in the context of machine translation ',Zhang,'num__2 nn__Related nn__Research amod__Several nn__researchers nsubj_proposed_ aux_proposed_have advmod_proposed_already dobj_proposed_methods prep_proposed_for pcomp_for_binarizing amod_grammars_synchronous dobj_binarizing_grammars prep_binarizing_in det_context_the pobj_in_context prep_context_of nn_translation_machine pobj_of_translation ' +o,6092,'An automatic metric which uses base forms and synonyms of the words in order to correlate better to human judgements has been 1 proposed in ',Banerjee,'det_metric_An amod_metric_automatic nsubj_been_metric nsubj_uses_which rcmod_metric_uses nn_forms_base dobj_uses_forms cc_forms_and conj_forms_synonyms prep_forms_of det_words_the pobj_of_words mark_correlate_in dep_correlate_order aux_correlate_to purpcl_uses_correlate advmod_correlate_better prep_correlate_to amod_judgements_human pobj_to_judgements aux_been_has dobj_been_1 amod_1_proposed prep_1_in ' +o,6093,'4 Experiments We evaluate the accuracy of HPSG parsing with dependencyconstraintsontheHPSGTreebank , which is extracted from the Wall Street Journal portion of the Penn Treebank 1 ',Marcus,'num_Experiments_4 nsubj_evaluate_We dep_Experiments_evaluate det_accuracy_the dobj_evaluate_accuracy prep_accuracy_of pcomp_of_HPSG xcomp_HPSG_parsing prep_parsing_with nn__dependencyconstraintsontheHPSGTreebank pobj_with_ nsubjpass_extracted_which auxpass_extracted_is rcmod__extracted prep_extracted_from det_portion_the nn_portion_Wall nn_portion_Street nn_portion_Journal pobj_from_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ num__1 ' +p,6094,'Some tasks can thrive on a nearly pure diet of unlabeled data ',Yarowsky,'det_tasks_Some nsubj_thrive_tasks aux_thrive_can prep_thrive_on det_diet_a advmod_diet_nearly amod_diet_pure pobj_on_diet prep_diet_of amod_data_unlabeled pobj_of_data ' +o,6095,'This method is employed in ',Cutting,'det_method_This nsubjpass_employed_method auxpass_employed_is prep_employed_in ' +o,6096,'This resembles the re-ranking approach ',Collins,'nsubj_approach_This cop_approach_resembles det_approach_the amod_approach_re-ranking ' +o,6097,'However , recent progress in machine translation and the continuous improvement on evaluation metrics such as BLEU suggest that SMT systems are already very good at choosing correct word translations ',Papineni,'advmod_suggest_However amod_progress_recent nsubj_suggest_progress prep_progress_in nn_translation_machine pobj_in_translation cc_progress_and det_improvement_the amod_improvement_continuous conj_progress_improvement prep_improvement_on nn_metrics_evaluation pobj_on_metrics dep_as_such prep_metrics_as nn__BLEU pobj_as_ complm_good_that nn_systems_SMT nsubj_good_systems cop_good_are advmod_good_already advmod_good_very ccomp_suggest_good prep_good_at pcomp_at_choosing amod_translations_correct nn_translations_word dobj_choosing_translations ' +n,6098,'In what concerns the evaluation process , although ROUGE is the most common evaluation metric for the automatic evaluation of summarization , since our approach might introduce in the summary information that it is not present in the original input source , we found that a human evaluation was more adequate to assess the relevance of that additional information ',Lin,'dep_found_In nsubj_process_what dep_process_concerns det_process_the nn_process_evaluation pcomp_In_process mark_metric_although nn__ROUGE nsubj_metric_ cop_metric_is det_metric_the advmod_common_most amod_metric_common nn_metric_evaluation advcl_process_metric prep_metric_for det_evaluation_the amod_evaluation_automatic pobj_for_evaluation prep_evaluation_of pobj_of_summarization mark_introduce_since poss_approach_our nsubj_introduce_approach aux_introduce_might advcl_process_introduce prep_introduce_in det_information_the nn_information_summary pobj_in_information complm_present_that nsubj_present_it cop_present_is neg_present_not ccomp_introduce_present prep_present_in det_source_the amod_source_original nn_source_input pobj_in_source nsubj_found_we complm_adequate_that det_evaluation_a amod_evaluation_human nsubj_adequate_evaluation cop_adequate_was advmod_adequate_more ccomp_found_adequate aux_assess_to xcomp_adequate_assess det_relevance_the dobj_assess_relevance prep_relevance_of det_information_that amod_information_additional pobj_of_information ' +p,6099,'4 Machine Translation Experiments 41 Experimental Setting For our MT experiments , we used a reimplementation of Moses , a state-of-the-art phrase-based system ',Koehn,'num_Experiments_4 nn_Experiments_Machine nn_Experiments_Translation tmod_used_Experiments num_Setting_41 nn_Setting_Experimental dep_Experiments_Setting prep_used_For poss_experiments_our nn_experiments_MT pobj_For_experiments nsubj_used_we det_reimplementation_a dobj_used_reimplementation prep_reimplementation_of nn__Moses pobj_of_ det_system_a amod_system_state-of-the-art amod_system_phrase-based appos__system ' +o,6100,'Practically , the grammar relaxation is done via the introduction of non-standard CCG rules ',Zettlemoyer,'advmod_done_Practically det_relaxation_the nn_relaxation_grammar nsubjpass_done_relaxation auxpass_done_is prep_done_via det_introduction_the pobj_via_introduction prep_introduction_of amod_rules_non-standard nn_rules_CCG pobj_of_rules ' +o,6101,'For determining whether an opinion sentence is positive or negative , we have used seed words similar to those produced by and extended them to construct a much larger set of semantically oriented words with a method similar to that proposed by ',Turney,'prep_used_For pcomp_For_determining complm_positive_whether det_sentence_an nn_sentence_opinion nsubj_positive_sentence cop_positive_is ccomp_determining_positive cc_positive_or conj_positive_negative nsubj_used_we aux_used_have nn_words_seed nsubj_similar_words xcomp_used_similar prep_similar_to pobj_to_those dep_those_produced prep_produced_by pobj_by_ cc_produced_and conj_produced_extended nsubj_construct_them aux_construct_to xcomp_extended_construct det_set_a advmod_larger_much amod_set_larger dobj_construct_set prep_set_of advmod_oriented_semantically amod_words_oriented pobj_of_words prep_produced_with det_method_a pobj_with_method amod_method_similar prep_similar_to pobj_to_that partmod_that_proposed prep_proposed_by ' +o,6102,'Others proposed distributional similarity measures between words ',Lin,'amod_measures_proposed amod_measures_distributional nn_measures_similarity dep_Others_measures prep_measures_between pobj_between_words ' +o,6103,'2 Motivation Automatic subjectivity analysis methods have been used in a wide variety of text processing applications , such as tracking sentiment timelines in online forums and news , review classification , mining opinions from product reviews , automatic expressive text-to-speech synthesis , text semantic analysis , and question answering ',Pang,'num_methods_2 nn_methods_Motivation nn_methods_Automatic nn_methods_subjectivity nn_methods_analysis nsubjpass_used_methods aux_used_have auxpass_used_been prep_used_in det_variety_a amod_variety_wide nn__variety prep_variety_of nn_applications_text nn_applications_processing pobj_of_applications dep_as_such conj_variety_as nn_timelines_tracking nn_timelines_sentiment pobj_as_timelines prep_timelines_in amod_forums_online pobj_in_forums cc_variety_and nn__news conj_variety_ nn__review nn__classification appos__ nn_opinions_mining appos__opinions prep_opinions_from nn__product nn__reviews pobj_from_ amod__automatic amod__expressive amod__text-to-speech nn__synthesis appos__ nn_analysis_text amod_analysis_semantic appos__analysis pobj_in_ cc_used_and conj_used_question nn__answering dobj_question_ ' +o,6104,'If the input consists of sevWe also adopt the approximation that treats every sentence with its reference as a separate corpus so that ngram counts are not accumulated , and parallel processing of sentences becomes possible ',Tillmann,'mark_consists_If det_input_the nsubj_consists_input advcl_adopt_consists prep_consists_of pobj_of_sevWe advmod_adopt_also det_approximation_the dobj_adopt_approximation nsubj_sentence_that dep_sentence_treats det_sentence_every rcmod_approximation_sentence prep_sentence_with poss_reference_its pobj_with_reference prep_reference_as det__a amod__separate nn__corpus pobj_as_ advmod_accumulated_so dep_accumulated_that nn_counts_ngram nsubjpass_accumulated_counts auxpass_accumulated_are neg_accumulated_not dep_sentence_accumulated cc_adopt_and amod_processing_parallel nsubj_possible_processing prep_processing_of pobj_of_sentences cop_possible_becomes conj_adopt_possible ' +o,6105,'Feature weight tuning was carried out using minimum error rate training , maximizing BLEU scores on a held-out development set ',Och,'amod_tuning_Feature nn_tuning_weight nsubjpass_carried_tuning auxpass_carried_was prep_carried_out pcomp_out_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training dep_using_maximizing amod_scores_BLEU dobj_maximizing_scores prep_maximizing_on det_set_a amod_set_held-out nn_set_development pobj_on_set ' +o,6106,'Since we approach decoding as xR transduction , the process is identical to that of constituencybased algorithms ',Huang,'mark_approach_Since nsubj_approach_we advcl_identical_approach xcomp_approach_decoding prep_decoding_as nn_transduction_xR pobj_as_transduction det_process_the nsubj_identical_process cop_identical_is prep_identical_to pobj_to_that prep_that_of amod_algorithms_constituencybased pobj_of_algorithms ' +o,6107,'Others try to accommodate both syntactic and lexical differences between the candidate translation and the reference , like CDER , which employs a version of edit distance for word substitution and reordering ; or METEOR , which uses stemming and WordNet synonymy ',Banerjee,'nsubj_try_Others aux_accommodate_to xcomp_try_accommodate preconj_syntactic_both amod_differences_syntactic cc_syntactic_and conj_syntactic_lexical dobj_accommodate_differences prep_differences_between det_translation_the nn_translation_candidate pobj_between_translation cc_differences_and det_reference_the conj_differences_reference prep_reference_like nn__CDER pobj_like_ nsubj_employs_which rcmod__employs det_version_a dobj_employs_version prep_version_of amod_distance_edit pobj_of_distance prep_distance_for nn_substitution_word pobj_for_substitution cc_substitution_and conj_substitution_reordering cc_differences_or nn__METEOR conj_differences_ nsubj_uses_which rcmod__uses xcomp_uses_stemming cc_accommodate_and conj_accommodate_WordNet advmod_WordNet_synonymy ' +o,6108,'Many techniques which have been studied for the purpose of machine translation , such as word sense disambiguation , anaphora resolution , and automatic pattern extraction from corpora , can accelerate the further enhancement of sentiment analysis , or other NLP tasks ',Yarowsky,'amod_techniques_Many nsubj_accelerate_techniques nsubjpass_studied_which aux_studied_have auxpass_studied_been rcmod_techniques_studied prep_studied_for det_purpose_the pobj_for_purpose prep_purpose_of nn_translation_machine pobj_of_translation dep_as_such prep_purpose_as nn__word nn__sense nn__disambiguation pobj_as_ nn__anaphora nn__resolution conj__ cc__and amod_extraction_automatic nn_extraction_pattern conj__extraction prep_studied_from nn__corpora pobj_from_ aux_accelerate_can det_enhancement_the amod_enhancement_further dobj_accelerate_enhancement prep_enhancement_of nn_analysis_sentiment pobj_of_analysis cc_enhancement_or amod_tasks_other nn_tasks_NLP conj_enhancement_tasks ' +o,6109,'For phrase-based translation model training , we used the GIZA + + toolkit ',Och,'prep_used_For amod_training_phrase-based nn_training_translation nn_training_model pobj_For_training nsubj_used_we det_GIZA_the dobj_used_GIZA amod_GIZA_+ cc_+_+ conj_+_toolkit ' +o,6110,'andw2 iscomputedusinganassociationscorebased on pointwise mutual information , asdefinedbyFano and used for a similar purpose in , as well as in many other studies in corpus linguistics ',Church,'nsubj_iscomputedusinganassociationscorebased_andw2 prep_iscomputedusinganassociationscorebased_on amod_information_pointwise amod_information_mutual pobj_on_information amod__asdefinedbyFano appos_information_ cc_iscomputedusinganassociationscorebased_and conj_iscomputedusinganassociationscorebased_used prep_used_for det_purpose_a amod_purpose_similar pobj_for_purpose prep_purpose_in pobj_in_ dep_well_as cc_for_well dep_well_as conj_for_in amod_studies_many amod_studies_other pobj_in_studies prep_studies_in nn_linguistics_corpus pobj_in_linguistics ' +o,6111,'The traditional method of evaluating similarity in a semantic network by measuring the path length between two nodes also captures this , albeit indirectly , when the semantic network is just an IS-A hierarchy : if the minimal path of IS-A links between two nodes is long , that means it is necessary to go high in the taxonomy , to more abstract concepts , in order to find their least upper bound ',Marcus,'det_method_The amod_method_traditional nsubj_this_method prep_method_of pcomp_of_evaluating dobj_evaluating_similarity prep_similarity_in det_network_a amod_network_semantic pobj_in_network prep_evaluating_by pcomp_by_measuring det_length_the nn_length_path dobj_measuring_length prep_length_between num_nodes_two pobj_between_nodes advmod_this_ advmod_this_also dep_this_captures dep_indirectly_albeit advmod_this_indirectly advmod_hierarchy_when det_network_the amod_network_semantic nsubj_hierarchy_network cop_hierarchy_is advmod_hierarchy_just det_hierarchy_an amod_hierarchy_IS-A advcl_this_hierarchy mark_long_if det_path_the amod_path_minimal nsubj_long_path prep_path_of amod_links_IS-A pobj_of_links prep_links_between num_nodes_two pobj_between_nodes cop_long_is advcl_means_long nsubj_means_that parataxis_this_means nsubj_necessary_it cop_necessary_is ccomp_means_necessary aux_high_to dep_high_go xcomp_necessary_high prep_high_in det_taxonomy_the pobj_in_taxonomy prep_high_to advmod_abstract_more amod_concepts_abstract pobj_to_concepts mark_find_in dep_find_order aux_find_to purpcl_high_find poss_bound_their amod_bound_least amod_bound_upper dobj_find_bound ' +o,6112,'In this spirit , we introduce a generalization of the classic k-gram models , widely used for string processing , to the case of trees ',Brown,'prep_introduce_In det_spirit_this pobj_In_spirit nsubj_introduce_we det_generalization_a dobj_introduce_generalization prep_generalization_of det_models_the amod_models_classic amod_models_k-gram pobj_of_models advmod_used_widely dep_introduce_used prep_used_for nn__string nn__processing pobj_for_ prep_used_to det_case_the pobj_to_case prep_case_of pobj_of_trees ' +n,6113,'The program takes the output of char_align , a robust alternative to sentence-based alignment programs , and applies word-level constraints using a version of Brown el al \'s Model 2 , modified and extended to deal with robustness issues ',Brown,'det_program_The nsubj_takes_program det_output_the dobj_takes_output prep_output_of nn__char_align pobj_of_ det_alternative_a amod_alternative_robust appos__alternative prep_takes_to amod_programs_sentence-based nn_programs_alignment pobj_to_programs cc_takes_and conj_takes_applies amod_constraints_word-level dobj_applies_constraints partmod_constraints_using det_version_a dobj_using_version prep_version_of nn_al_Brown nn_al_el poss_Model_al possessive_al_\'s pobj_of_Model dep__2 amod_Model_ partmod_Model_modified cc_modified_and conj_modified_extended aux_deal_to xcomp_modified_deal prep_deal_with amod_issues_robustness pobj_with_issues ' +o,6114,'See Table 4 in for the detail ',Kanayama,'number_4_Table dobj_See_4 prep_See_in pobj_in_ prep_See_for det_detail_the pobj_for_detail ' +o,6115,'Detail of the Bakeoff data sets is in ',Levow,'nsubj_is_Detail prep_Detail_of det_sets_the nn_sets_Bakeoff nn_sets_data pobj_of_sets prep_is_in ' +o,6116,'This kind of smoothing has also been used in the generative parser of and has been shown to have a relatively good performance for language modeling ',Collins,'det_kind_This nsubjpass_used_kind prep_kind_of pobj_of_smoothing aux_used_has advmod_used_also auxpass_used_been prep_used_in det_parser_the amod_parser_generative pobj_in_parser prep_parser_of pobj_of_ cc_used_and aux_shown_has auxpass_shown_been conj_used_shown aux_have_to xcomp_shown_have det_performance_a advmod_good_relatively amod_performance_good dobj_have_performance prep_performance_for nn__language nn__modeling pobj_for_ ' +n,6117,'6 Conclusion Traditional approaches for devising parsing models , smoothing techniques and evaluation metrics are not well suited for MH , as they presuppose 13The lack of head marking , for instance , precludes the use of lexicalized models a la ',Collins,'num_approaches_6 nn_approaches_Conclusion amod_approaches_Traditional dep_for_approaches prep_suited_for pcomp_for_devising amod_models_parsing dobj_devising_models csubjpass_suited_smoothing dobj_smoothing_techniques cc_techniques_and nn_metrics_evaluation conj_techniques_metrics auxpass_suited_are neg_suited_not advmod_suited_well ccomp_precludes_suited prep_suited_for pobj_for_MH mark_presuppose_as nsubj_presuppose_they dep_precludes_presuppose amod_lack_13The dobj_presuppose_lack prep_lack_of pobj_of_head partmod_head_marking prep_marking_for pobj_for_instance det_use_the dobj_precludes_use prep_use_of amod_models_lexicalized pobj_of_models det_la_a nsubj_precludes_la ' +o,6118,'In , the authors use the transcripts of debates from the US Congress to automatically classify speeches as supporting or opposing a given topic by taking advantage of the voting records of the speakers ',Thomas,'prep_use_In pobj_In_ det_authors_the nsubj_use_authors det_transcripts_the nsubj_classify_transcripts prep_transcripts_of pobj_of_debates prep_debates_from det_Congress_the nn_Congress_US pobj_from_Congress aux_classify_to advmod_classify_automatically xcomp_use_classify dobj_classify_speeches prep_classify_as pcomp_as_supporting cc_supporting_or conj_supporting_opposing det_topic_a amod_topic_given dobj_opposing_topic prep_opposing_by pcomp_by_taking dobj_taking_advantage prep_advantage_of det_records_the amod_records_voting pobj_of_records prep_records_of det_speakers_the pobj_of_speakers ' +o,6119,'Training on about 40,000 sentences achieves a crossing brackets rate of 107 , a better value than our 163 value for regular parsing or the 113 value assuming perfect segmentation\\/tagging , but even for similar text types , comparisons across languages are of course problematic ',Collins,'prep_Training_on quantmod_40,000_about num_sentences_40,000 pobj_on_sentences advmod_Training_ dep_Training_achieves det_rate_a amod_rate_crossing nn_rate_brackets nsubj_Training_rate prep_rate_of pobj_of_107 det_value_a amod_value_better appos_rate_value prep_value_than poss_value_our num_value_163 pobj_than_value prep_value_for amod_parsing_regular pobj_for_parsing cc_parsing_or det_value_the num_value_113 conj_parsing_value partmod_value_assuming acomp_assuming_perfect acomp_assuming_segmentation\\/tagging cc_segmentation\\/tagging_but dep_for_even conj_segmentation\\/tagging_for amod_types_similar nn_types_text pobj_for_types appos_types_comparisons prep_comparisons_across pobj_across_languages dep_languages_are prep_value_of nn_problematic_course pobj_of_problematic ' +o,6120,' proposed using GIZA + + to align words between the backbone and hypothesis ',Och,'nsubj_proposed_ xcomp_proposed_using dobj_using_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_using_to nn_words_align pobj_to_words prep_words_between det_backbone_the pobj_between_backbone cc_backbone_and conj_backbone_hypothesis ' +n,6121,'The experimental results show that our method outperforms the synchronous binarization method with over 08 BLEU scores on both NIST 2005 and NIST 2008 Chinese-to-English evaluation data sets ',Zhang,'det_results_The amod_results_experimental nsubj_show_results complm_outperforms_that poss_method_our nsubj_outperforms_method ccomp_show_outperforms det_method_the amod_method_synchronous nn_method_binarization dobj_outperforms_method dep_over_ dep_over_with prep_outperforms_over num_scores_08 amod_scores_BLEU pobj_over_scores prep_scores_on det_NIST_both pobj_on_NIST nn_data_2005 cc_2005_and conj_2005_NIST num_data_2008 amod_data_Chinese-to-English nn_data_evaluation nsubj_sets_data dep_show_sets ' +o,6122,'Clustering-based approaches usually represent word contexts as vectors and cluster words based on similarities of the vectors ',Lin,'amod_approaches_Clustering-based nsubj_represent_approaches advmod_represent_usually nn_contexts_word dobj_represent_contexts prep_represent_as pobj_as_vectors cc_vectors_and nn_words_cluster conj_vectors_words partmod_vectors_based prep_based_on pobj_on_similarities prep_similarities_of det_vectors_the pobj_of_vectors ' +p,6123,'g2 2 Motivation The success of Statistical Machine Translation -LRB- SMT -RRB- has sparked a successful line of investigation that treats paraphrase acquisition and generation essentially as a monolingual machine translation problem ',Barzilay,'nn_Motivation_g2 num_Motivation_2 det_success_The nsubj_sparked_success prep_success_of nn_Translation_Statistical nn_Translation_Machine pobj_of_Translation abbrev_Translation_SMT aux_sparked_has dep_Motivation_sparked det_line_a amod_line_successful dobj_sparked_line prep_line_of pobj_of_investigation nsubj_treats_that rcmod_line_treats amod_acquisition_paraphrase dobj_treats_acquisition cc_acquisition_and conj_acquisition_generation advmod_treats_essentially prep_treats_as det_problem_a amod_problem_monolingual nn_problem_machine nn_problem_translation pobj_as_problem ' +o,6124,'In particular , this holds for the SCFG implementing Inversion 3For two sequences of numbers , the notation y -LRB- z stands for y y , z z : y -LRB- z Transduction Grammar ',Wu,'prep_holds_In pobj_In_particular nsubj_holds_this mark_y_for det_SCFG_the nsubj_y_SCFG partmod_SCFG_implementing amod_sequences_Inversion amod_sequences_3For num_sequences_two dobj_implementing_sequences prep_sequences_of pobj_of_numbers det_y_the nn_y_notation appos_numbers_y amod_stands_z dep_y_stands prep_stands_for nn_y_y pobj_for_y dep_sequences_z dep_y_z advcl_holds_y amod_Grammar_z nn_Grammar_Transduction xcomp_y_Grammar ' +o,6125,'Labeled data for one domain might be used to train a initial classifier for another -LRB- possibly related -RRB- domain , and then bootstrapping can be employed to learn new knowledge from the new domain ',Blitzer,'amod_data_Labeled nsubjpass_used_data prep_data_for num_domain_one pobj_for_domain aux_used_might auxpass_used_be aux_train_to purpcl_used_train det_classifier_a amod_classifier_initial dobj_train_classifier prep_classifier_for amod_domain_another dep_related_possibly dep_another_related pobj_for_domain cc_used_and advmod_employed_then nsubjpass_employed_bootstrapping aux_employed_can auxpass_employed_be conj_used_employed aux_learn_to purpcl_employed_learn amod_knowledge_new dobj_learn_knowledge prep_learn_from det_domain_the amod_domain_new pobj_from_domain ' +o,6126,'Proposals have recently been made for protocols for the collection of human discourse segmentation data and for how to evaluate the validity of judgments so obtained ',Carletta,'nsubjpass_made_Proposals aux_made_have advmod_made_recently auxpass_made_been prep_made_for pobj_for_protocols prep_protocols_for det_collection_the pobj_for_collection prep_collection_of amod__human nn__discourse nn__segmentation nn__data pobj_of_ cc_for_and conj_for_for advmod_evaluate_how aux_evaluate_to pcomp_for_evaluate det_validity_the dobj_evaluate_validity prep_validity_of pobj_of_judgments advmod_obtained_so partmod_judgments_obtained ' +o,6127,'12 Statistical modeling for translation Earlier work in statistical machine translation is based on the noisy-channel formulation where T = arg max T p -LRB- TjS -RRB- = argmax T p -LRB- T -RRB- p -LRB- SjT -RRB- -LRB- 1 -RRB- where the target language model p -LRB- T -RRB- is further decomposed as p -LRB- T -RRB- \\/ productdisplay i p -LRB- tijti1 , , tik +1 -RRB- where k is the order of the language model and the translation model p -LRB- SjT -RRB- has been modeled by a sequence of five models with increasing complexity ',Brown,'num_modeling_12 nn_modeling_Statistical nsubjpass_based_modeling prep_modeling_for num_work_translation amod_work_Earlier pobj_for_work prep_work_in amod__statistical nn__machine nn__translation pobj_in_ auxpass_based_is prep_based_on det_formulation_the amod_formulation_noisy-channel pobj_on_formulation advmod_=_where nsubj_=_T dep_p_= nn_p_arg nn_p_max nn_p_T rcmod_T_p appos_T_TjS rcmod_formulation_= amod_p_argmax nn_p_T nn_p_p appos_p_T dobj_=_p appos_p_SjT appos_p_1 advmod_decomposed_where det_p_the nn_p_target nn_p_language nn_p_model nsubjpass_decomposed_p appos_p_T auxpass_decomposed_is advmod_decomposed_further rcmod_p_decomposed prep_decomposed_as dep_\\/_p dep_\\/_T amod_p_\\/ amod_p_productdisplay amod_p_i pobj_as_p dep_p_tijti1 amod_+1_tik appos_tijti1_+1 advmod_order_where nsubj_order_k cop_order_is det_order_the advcl_decomposed_order prep_order_of det_model_the nn_model_language pobj_of_model cc_decomposed_and det_p_the nn_p_translation nn_p_model nsubjpass_modeled_p appos_p_SjT aux_modeled_has auxpass_modeled_been conj_decomposed_modeled prep_modeled_by det_sequence_a pobj_by_sequence prep_sequence_of num_models_five pobj_of_models prep_modeled_with amod_complexity_increasing pobj_with_complexity ' +o,6128,'The SENSEVAL \' ~ tan -LRB- lard is clearly beaten by the earlier results of -LRB- 965 \% precision -RRB- and Schiitze -LRB- 1992 -RRB- -LRB- 92 \% precision -RRB- ',Yarowsky,'det_SENSEVAL_The poss_~_SENSEVAL nsubj_tan_~ nsubjpass_beaten_lard auxpass_beaten_is advmod_beaten_clearly ccomp_tan_beaten prep_beaten_by det_results_the amod_results_earlier pobj_by_results prep_results_of pobj_of_ number_\%_965 num_precision_\% appos__precision cc__and conj__Schiitze dep_Schiitze_1992 num_\%_92 measure_precision_\% amod_Schiitze_precision ' +o,6129,'3The usefulness of position varies significantly in different genres ',Penn,'amod_usefulness_3The nsubj_varies_usefulness prep_usefulness_of pobj_of_position advmod_varies_significantly prep_varies_in amod_genres_different pobj_in_genres ' +n,6130,' produced a corpus of 4,000 questions annotated with syntactic trees , and obtained an improvement in parsing accuracy for Bikels reimplementation of the Collins parser by training a new parser model with a combination of newspaper and question data ',Collins,'nsubj_produced_ det_corpus_a dobj_produced_corpus prep_corpus_of num_questions_4,000 pobj_of_questions partmod_questions_annotated prep_annotated_with amod_trees_syntactic pobj_with_trees cc_produced_and conj_produced_obtained det_improvement_an dobj_obtained_improvement prep_obtained_in pcomp_in_parsing dobj_parsing_accuracy prep_accuracy_for nn_reimplementation_Bikels pobj_for_reimplementation prep_accuracy_of det__the nn__Collins nn__parser pobj_of_ prep_parsing_by pcomp_by_training det_model_a amod_model_new nn_model_parser dobj_training_model prep_training_with det_combination_a pobj_with_combination prep_combination_of nn_data_newspaper cc_newspaper_and conj_newspaper_question pobj_of_data ' +o,6131,'Then the words are tagged as inside a phrase -LRB- I -RRB- , outside a phrase -LRB- O -RRB- or beginning of a phrase -LRB- B -RRB- ',Ramshaw,'advmod_words_Then det_words_the nsubjpass_tagged_words auxpass_tagged_are prep_tagged_as dep__inside det_phrase_a pobj_inside_phrase dep_phrase_I conj_inside_outside det_phrase_a pobj_outside_phrase appos_phrase_O cc_inside_or conj_inside_beginning prep_beginning_of det_phrase_a pobj_of_phrase appos_phrase_B pobj_as_ ' +n,6132,'1 Introduction Statistical phrase-based systems have consistently delivered state-of-the-art performance in recent machine translation evaluations , yet these systems remain weak at handling word order changes ',Koehn,'num__1 nn__Introduction nn__Statistical amod__phrase-based nn__systems nsubj_delivered_ aux_delivered_have advmod_delivered_consistently amod_performance_state-of-the-art dobj_delivered_performance prep_delivered_in amod_evaluations_recent nn_evaluations_machine nn_evaluations_translation pobj_in_evaluations advmod_delivered_yet det_systems_these nsubj_weak_systems cop_weak_remain dep_delivered_weak prep_weak_at pcomp_at_handling nn_changes_word nn_changes_order dobj_handling_changes ' +o,6133,'In supervised domain adaptation , besides the labeled source data , we have access to a comparably small , but labeled amount of target data ','Daume III','prep_have_In amod__supervised nn__domain nn__adaptation pobj_In_ prep_have_besides det_data_the amod_data_labeled nn_data_source pobj_besides_data nsubj_have_we dobj_have_access prep_have_to det_amount_a advmod_small_comparably amod_amount_small cc_small_but conj_small_labeled pobj_to_amount prep_amount_of nn_data_target pobj_of_data ' +o,6134,'This was a difcult challenge as many participants in the task failed to obtain any meaningful gains from unlabeled data ','Daume III','nsubj_challenge_This cop_challenge_was det_challenge_a amod_challenge_difcult mark_failed_as amod_participants_many nsubj_failed_participants prep_participants_in det_task_the pobj_in_task advcl_challenge_failed aux_obtain_to xcomp_failed_obtain det_gains_any amod_gains_meaningful dobj_obtain_gains prep_obtain_from amod__unlabeled nn__data pobj_from_ ' +o,6135,'? ? Initial phrase pairs are identified following the procedure typically employed in phrase based systems ',Koehn,'amod_pairs_Initial nn_pairs_phrase nsubjpass_identified_pairs auxpass_identified_are prep_identified_following det_procedure_the pobj_following_procedure advmod_employed_typically partmod_procedure_employed prep_employed_in pobj_in_phrase partmod_phrase_based dobj_based_systems ' +o,6136,'For this purpose , we adopt the view of the ITG constraints as a bilingual grammar as , eg , in ',Wu,'prep_adopt_For det_purpose_this pobj_For_purpose nsubj_adopt_we det_view_the dobj_adopt_view prep_view_of det_constraints_the nn_constraints_ITG pobj_of_constraints prep_constraints_as det_grammar_a amod_grammar_bilingual pobj_as_grammar advmod_grammar_as appos_grammar_eg prt_adopt_in ' +o,6137,'This formulation is similar to the energy minimization framework , which is commonly used in image analysis and has been recently applied in natural language processing ',Pang,'det_formulation_This nsubj_similar_formulation cop_similar_is prep_similar_to det_framework_the nn_framework_energy nn_framework_minimization pobj_to_framework nsubjpass_used_which auxpass_used_is advmod_used_commonly rcmod_framework_used prep_used_in nn__image nn__analysis pobj_in_ cc_used_and aux_applied_has auxpass_applied_been advmod_applied_recently conj_used_applied prep_applied_in amod_processing_natural nn_processing_language pobj_in_processing ' +o,6138,'c2005 Association for Computational Linguistics Automatic identification of sentiment vocabulary : exploiting low association with known sentiment terms Michael Gamon Anthony Aue Natural Language Processing Group Natural Language Processing Group Microsoft Research Microsoft Research mgamon @ microsoftcom anthaue @ microsoftcom Abstract We describe an extension to the technique for the automatic identification and labeling of sentiment terms described in and Turney and Littman -LRB- 2002 -RRB- ',Turney,'nsubj_Association_c2005 prep_Association_for nn_identification_Computational nn_identification_Linguistics nn_identification_Automatic pobj_for_identification prep_identification_of nn_vocabulary_sentiment pobj_of_vocabulary dep_Association_exploiting amod_association_low dobj_exploiting_association prep_exploiting_with amod_terms_known nn_terms_sentiment dep_anthaue_terms nn_anthaue_Michael nn_anthaue_Gamon nn_anthaue_Anthony nn_anthaue_Aue nn_anthaue_Natural nn_anthaue_Language nn_anthaue_Processing nn_anthaue_Group nn_anthaue_Natural nn_anthaue_Language nn_anthaue_Processing nn_anthaue_Group nn_anthaue_Microsoft nn_anthaue_Research nn_anthaue_Microsoft nn_anthaue_Research nn_anthaue_mgamon nn_anthaue_@ nn_anthaue_microsoftcom pobj_with_anthaue dep_Abstract_@ amod_Abstract_microsoftcom dep_describe_Abstract nsubj_describe_We dep_anthaue_describe det_extension_an dobj_describe_extension prep_describe_to det_technique_the pobj_to_technique prep_technique_for det_identification_the amod_identification_automatic pobj_for_identification cc_identification_and conj_identification_labeling prep_identification_of nn_terms_sentiment pobj_of_terms partmod_terms_described prep_described_in pobj_in_ cc__and conj__Turney cc_anthaue_and conj_anthaue_Littman dep_Association_2002 ' +p,6139,'22 Phrase-based Chinese-to-English MT The MT system used in this paper is Moses , a stateof-the-art phrase-based system ',Koehn,'num_MT_22 amod_MT_Phrase-based amod_MT_Chinese-to-English det_system_The nn_system_MT nsubj_Moses_system partmod_system_used prep_used_in det_paper_this pobj_in_paper cop_Moses_is dep_MT_Moses det_system_a amod_system_stateof-the-art amod_system_phrase-based appos_Moses_system ' +p,6140,' reports a success rate of 96 \% disambiguating twelve words with two clear sense distinctions each one -RRB- ',Yarowsky,'nsubj_reports_ det_rate_a nn_rate_success dobj_reports_rate prep_rate_of num_\%_96 pobj_of_\% partmod_\%_disambiguating num_words_twelve dobj_disambiguating_words prep_disambiguating_with num_distinctions_two amod_distinctions_clear nn_distinctions_sense pobj_with_distinctions det_one_each dep_distinctions_one ' +o,6141,'While this technique has been successfully applied to parsing the ATIS portion in the Penn Treebank , it is extremely time consuming ',Marcus,'mark_applied_While det_technique_this nsubjpass_applied_technique aux_applied_has auxpass_applied_been advmod_applied_successfully advcl_consuming_applied prep_applied_to pcomp_to_parsing det_portion_the nn_portion_ATIS dobj_parsing_portion prep_parsing_in det__the nn__Penn nn__Treebank pobj_in_ nsubj_consuming_it cop_consuming_is advmod_consuming_extremely nn_consuming_time ' +o,6142,'As argued in , Kappa values of 08 or higher are desirable for detecting associations between several coded variables ; we were thus satisfied with the level of agreement achieved ',Carletta,'mark_argued_As advcl_desirable_argued prep_argued_in pobj_in_ nn_values_Kappa nsubj_desirable_values prep_values_of pobj_of_08 cc_08_or conj_08_higher cop_desirable_are prep_desirable_for pcomp_for_detecting dobj_detecting_associations prep_associations_between amod_variables_several amod_variables_coded pobj_between_variables nsubjpass_satisfied_we auxpass_satisfied_were advmod_satisfied_thus parataxis_desirable_satisfied prep_satisfied_with det_level_the pobj_with_level prep_level_of pobj_of_agreement partmod_agreement_achieved ' +o,6143,' described the use of minimum error training directly optimizing the error rate on automatic MT evaluation metrics such as BLEU ',Och,'nsubj_described_ det_use_the dobj_described_use prep_use_of amod_training_minimum nn_training_error pobj_of_training advmod_optimizing_directly partmod_training_optimizing det_rate_the nn_rate_error dobj_optimizing_rate prep_optimizing_on amod_metrics_automatic nn_metrics_MT nn_metrics_evaluation pobj_on_metrics dep_as_such prep_metrics_as pobj_as_BLEU ' +o,6144,'The most relevant to our work are Kazama and Torisawa , Toral and Muoz , and ',Cucerzan,'dep_relevant_The advmod_relevant_most prep_relevant_to poss_work_our pobj_to_work cop__are nn__Kazama cc_Kazama_and conj_Kazama_Torisawa dep_relevant_ nn__Toral cc_Toral_and conj_Toral_Muoz conj__ cc__and nn__ conj__ ' +o,6145,'Due to its popularity for unsupervised POS induction research and its often-used tagset , for our initial research , we use the Wall Street Journal -LRB- WSJ -RRB- portion of the Penn Treebank , with 36 tags -LRB- plus 9 punctuation tags -RRB- , and we use sections 00-18 , leaving held-out data for future experiments4 Defining frequent frames as those occurring at 4Even if we wanted child-directed speech , the CHILDES database uses coarse POS tags ',Goldberg,'dep_to_Due prep_use_to poss_popularity_its pobj_to_popularity prep_popularity_for amod__unsupervised dep__POS nn__induction nn__research pobj_for_ cc_popularity_and poss_tagset_its amod_tagset_often-used conj_popularity_tagset prep_tagset_for poss_research_our amod_research_initial pobj_for_research nsubj_use_we det_portion_the nn_Journal_Wall nn_Journal_Street nn_portion_Journal abbrev_Journal_WSJ dobj_use_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ prep_portion_with num_tags_36 pobj_with_tags dep_tags_plus num_tags_9 nn_tags_punctuation dep_plus_tags cc_use_and nsubj_use_we conj_use_use dobj_use_sections tmod_use_00-18 xcomp_use_leaving amod_data_held-out dobj_leaving_data prep_leaving_for amod_experiments4_future pobj_for_experiments4 partmod_experiments4_Defining amod_frames_frequent dobj_Defining_frames prep_Defining_as pobj_as_those partmod_those_occurring prep_occurring_at pobj_at_4Even mark_wanted_if nsubj_wanted_we advcl_uses_wanted amod_speech_child-directed dobj_wanted_speech det__the nn__CHILDES nn__database nsubj_uses_ dep_use_uses amod_tags_coarse dep_tags_POS dobj_uses_tags ' +p,6146,'It is an online training algorithm and has been successfully used in many NLP tasks , such as POS tagging , parsing , Chinese word segmentation , and so on ',Jiang,'nsubjpass_algorithm_It cop_algorithm_is det_algorithm_an amod_algorithm_online nn_algorithm_training cc_algorithm_and aux_used_has auxpass_used_been advmod_used_successfully conj_algorithm_used prep_used_in amod_tasks_many nn_tasks_NLP pobj_in_tasks dep_as_such dep_tasks_as dep__POS amod__tagging pobj_as_ partmod__parsing amod__ amod__Chinese nn__word nn__segmentation dobj_parsing_ cc_as_and mark_on_so conj_as_on ' +o,6147,'2 Related Work The issue of MWE processing has attracted much attention from the Natural Language Processing -LRB- NLP -RRB- community , including , 1993 ; Dagan and Church , 1994 ; Daille , 1995 ; 1995 ; McEnery et al , 1997 ; Wu , 1997 ; Michiels and Dufour , 1998 ; Maynard and Ananiadou , 2000 ; Merkel and Andersson , 2000 ; Piao and McEnery , 2001 ; Sag et al , 2001 ; Tanaka and Baldwin , 2003 ; Dias , 2003 ; Baldwin et al , 2003 ; Nivre and Nilsson , 2004 Pereira et al , ',Smadja,'num_Work_2 amod_Work_Related det_issue_The nsubj_attracted_issue prep_issue_of nn_processing_MWE pobj_of_processing aux_attracted_has dep_Work_attracted amod_attention_much dobj_attracted_attention prep_attracted_from det_community_the nn_Processing_Natural nn_Processing_Language nn_community_Processing abbrev_Processing_NLP pobj_from_community prep_community_including pobj_including_ appos__1993 dep__Dagan cc_Dagan_and conj_Dagan_Church appos_Dagan_1994 dep__Daille appos_Daille_1995 dep__1995 dep_1995_McEnery cc_McEnery_et conj_McEnery_al appos_McEnery_1997 dep__Wu appos_Wu_1997 dep__Michiels cc_Michiels_and conj_Michiels_Dufour appos_Michiels_1998 dep__Maynard cc_Maynard_and conj_Maynard_Ananiadou appos_Maynard_2000 dep__Merkel cc_Merkel_and conj_Merkel_Andersson appos_Merkel_2000 dep__Piao cc_Piao_and conj_Piao_McEnery appos_Piao_2001 dep__Sag cc_Sag_et conj_Sag_al appos_Sag_2001 dep__Tanaka cc_Tanaka_and conj_Tanaka_Baldwin appos_Tanaka_2003 dep__Dias appos_Dias_2003 dep__Baldwin cc_Baldwin_et conj_Baldwin_al appos_Baldwin_2003 dep__Nivre cc_Nivre_and conj_Nivre_Nilsson nn_Pereira_2004 appos__Pereira cc_Pereira_et conj_Pereira_al ' +o,6148,'Estimated clues are derived from the parallel data using , for example , measures of co-occurrence -LRB- eg the Dice coefficient -RRB- , statistical alignment models -LRB- eg IBM models from statistical machine translation -RRB- , or string similarity measures -LRB- eg the longest common sub-sequence ratio -RRB- ',Brown,'amod_clues_Estimated nsubjpass_derived_clues auxpass_derived_are prep_derived_from det_data_the amod_data_parallel pobj_from_data partmod_data_using prep_using_for pobj_for_example dobj_using_measures prep_measures_of pobj_of_co-occurrence dep_measures_eg det__the nn__Dice nn__coefficient dobj_eg_ amod_alignment_statistical conj_measures_alignment dep_alignment_models dep_measures_eg nn_models_IBM dobj_eg_models prep_eg_from amod__statistical nn__machine nn__translation pobj_from_ cc_measures_or nn_measures_string nn_measures_similarity conj_measures_measures dep_measures_eg det__the amod__longest amod__common nn__sub-sequence nn__ratio dep_eg_ ' +o,6149,'The tag propagation\\/elimination scheme is adopted from ',Yarowsky,'det_scheme_The nn_scheme_tag nn_scheme_propagation\\/elimination nsubjpass_adopted_scheme auxpass_adopted_is prep_adopted_from ' +o,6150,'The data consists of sections of the Wall Street Journal part of the Penn TreeBank , with information on predicate-argument structures extracted from the PropBank corpus ',Marcus,'det_data_The nsubj_consists_data prep_consists_of pobj_of_sections prep_sections_of det_part_the nn_part_Wall nn_part_Street nn_part_Journal pobj_of_part prep_part_of det__the nn__Penn nn__TreeBank pobj_of_ prep_consists_with pobj_with_information prep_information_on amod_structures_predicate-argument pobj_on_structures partmod_structures_extracted prep_extracted_from det__the nn__PropBank nn__corpus pobj_from_ ' +p,6151,'One major resource for corpus-based research is the treebanks available in many research organizations , which carry skeletal syntactic structures or ` brackets \' that have been manually verified ',Marcus,'num_resource_One amod_resource_major nsubj__resource prep_resource_for amod_research_corpus-based pobj_for_research cop__is det__the amod__treebanks amod__available prep_available_in amod_organizations_many nn_organizations_research pobj_in_organizations nsubj_carry_which rcmod__carry amod_structures_skeletal amod_structures_syntactic dobj_carry_structures cc_structures_or conj_structures_brackets nsubjpass_verified_that aux_verified_have auxpass_verified_been advmod_verified_manually rcmod_brackets_verified ' +o,6152,'Previous authors have used numerous HMM-based models and other types of networks including maximum entropy models , conditional Markov models , conditional random elds -LRB- CRF -RRB- , and cyclic dependency networks ',Collins,'amod_authors_Previous nsubj_used_authors aux_used_have amod_models_numerous amod_models_HMM-based dobj_used_models amod_types_ cc__and conj__other nsubj__types prep_types_of pobj_of_networks prep_networks_including amod_models_maximum nn_models_entropy pobj_including_models rcmod_models_ amod_models_conditional nn_models_Markov dep__models ccomp_used_ amod_elds_conditional amod_elds_random nsubj__elds abbrev_elds_CRF ccomp_used_ cc_used_and amod_networks_cyclic nn_networks_dependency nsubj__networks conj_used_ ' +o,6153,'As an example , consider the fiat NP structures that are in the Penn Treebank ',Marcus,'prep_consider_As det_example_an pobj_As_example det_structures_the nn_structures_fiat nn_structures_NP dobj_consider_structures nsubj_are_that rcmod_structures_are prep_are_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +n,6154,'With these linguistic annotations , we expect the LABTG to address two traditional issues of standard phrase-based SMT in a more effective manner ',Koehn,'prep_expect_With det_annotations_these amod_annotations_linguistic pobj_With_annotations nsubj_expect_we det_LABTG_the nsubj_address_LABTG aux_address_to xcomp_expect_address num_issues_two amod_issues_traditional dobj_address_issues prep_issues_of amod__standard amod__phrase-based nn__SMT pobj_of_ prep__in det_manner_a advmod_effective_more amod_manner_effective pobj_in_manner ' +o,6155,'On the other hand , statistical MT employing IBM models translates an input sentence by the combination of word transfer and word re-ordering ',Brown,'prep_translates_On det_hand_the amod_hand_other pobj_On_hand amod__statistical nn__MT nn__employing nn__IBM nn__models nsubj_translates_ det_sentence_an nn_sentence_input dobj_translates_sentence prep_translates_by det_combination_the pobj_by_combination prep_combination_of nn_transfer_word pobj_of_transfer cc_transfer_and nn_re-ordering_word conj_transfer_re-ordering ' +o,6156,'Note that apart from previous work we use complete skip-chain -LRB- contextanswer -RRB- edges in hc -LRB- x , y -RRB- ',Ding,'complm_use_that dep_from_apart advcl_use_from amod__previous nn__work pobj_from_ nsubj_use_we ccomp_Note_use xcomp_use_complete amod_edges_skip-chain appos_edges_contextanswer dobj_complete_edges prep_complete_in pobj_in_hc dep_hc_x appos_x_y ' +o,6157,'Six features from were used as baseline features ',Och,'num_features_Six nsubjpass_used_features prep_features_from pobj_from_ auxpass_used_were prep_used_as nn_features_baseline pobj_as_features ' +o,6158,'NER proves to be a knowledgeintensive task , and it was reassuring to observe that System Resources Used F1 + LBJ-NER Wikipedia , Nonlocal Features , Word-class Model 9080 Semi-supervised on 1Gword unlabeled data 8992 Semi-supervised on 27Mword unlabeled data 8931 Wikipedia 8802 Non-local Features 8724 Non-local Features 8717 + Non-local Features 8686 Table 7 : Results for CoNLL03 data reported in the literature ',Kazama,'nsubj_proves_NER aux_task_to cop_task_be det_task_a amod_task_knowledgeintensive xcomp_proves_task cc_proves_and nsubj_reassuring_it cop_reassuring_was conj_proves_reassuring aux_observe_to xcomp_reassuring_observe complm_reported_that nn_Resources_System nsubj_reported_Resources partmod_Resources_Used nn_Wikipedia_F1 nn_Wikipedia_+ nn_Wikipedia_LBJ-NER dobj_Used_Wikipedia nn_Features_Nonlocal appos_Wikipedia_Features nn_Model_Word-class appos_Wikipedia_Model num__9080 measure_Semi-supervised_ amod_Model_Semi-supervised prep_Semi-supervised_on amod_data_1Gword amod_data_unlabeled pobj_on_data num__8992 measure_Semi-supervised_ dep_on_Semi-supervised advmod_Model_on amod_data_27Mword amod_data_unlabeled pobj_on_data number__8931 num_Features_ nn_Features_Wikipedia nn_Features_8802 nn_Features_ nn_Features_Non-local nn_Features_Features nn_Features_8724 nn_Features_ nn_Features_Non-local nn_Features_Features nn_Features_8717 nn_Features_+ nn_Features_ nn_Features_Non-local dep_Model_Features num_Table_8686 dep_Features_Table num_Table_7 dep_Wikipedia_Results prep_Results_for nn_data_CoNLL03 pobj_for_data ccomp_observe_reported prep_reported_in det_literature_the pobj_in_literature ' +o,6159,'5 Related Work There has not been much previous work on graphical models for full parsing , although recently several latent variable models for parsing have been proposed ',Koo,'num_Work_5 amod_Work_Related expl_been_There aux_been_has neg_been_not dep_Work_been amod_work_much amod_work_previous dobj_been_work prep_work_on amod_models_graphical pobj_on_models prep_models_for amod_parsing_full pobj_for_parsing mark_proposed_although advmod_proposed_recently amod_models_several amod_models_latent amod_models_variable nsubjpass_proposed_models prep_models_for pobj_for_parsing aux_proposed_have auxpass_proposed_been advcl_been_proposed ' +o,6160,'Generalized Forward Backward Reestimation Generalization of the Forward and Viterbi Algorithm In English part of speech taggers , the maximization of Equation -LRB- 1 -RRB- to get the most likely tag sequence , is accomplished by the Viterbi algorithm , and the maximum likelihood estimates of the parameters of Equation -LRB- 2 -RRB- are obtained from untagged corpus by the ForwardBackward algorithm ',Cutting,'nn_Generalization_Generalized nn_Generalization_Forward nn_Generalization_Backward nn_Generalization_Reestimation prep_Generalization_of det_Algorithm_the nn_Algorithm_Forward cc_Forward_and conj_Forward_Viterbi pobj_of_Algorithm dep_Generalization_In amod_part_English pobj_In_part prep_part_of nn_taggers_speech nsubjpass_accomplished_taggers det_maximization_the appos_taggers_maximization prep_maximization_of pobj_of_Equation appos_Equation_1 aux_get_to infmod_maximization_get det_sequence_the advmod_likely_most amod_sequence_likely nn_sequence_tag dobj_get_sequence auxpass_accomplished_is pcomp_of_accomplished prep_accomplished_by det__the nn__Viterbi nn__algorithm pobj_by_ cc_accomplished_and det_estimates_the amod_estimates_maximum nn_estimates_likelihood nsubjpass_obtained_estimates prep_estimates_of det_parameters_the pobj_of_parameters prep_parameters_of pobj_of_Equation appos_estimates_2 auxpass_obtained_are conj_accomplished_obtained prep_obtained_from amod_corpus_untagged pobj_from_corpus prep_obtained_by det_algorithm_the nn_algorithm_ForwardBackward pobj_by_algorithm ' +p,6161,'Successful approaches aimed at trying to overcome the sparse data limitation include backoff , Turing-Good variants , interpolation , deleted estimation , similarity-based models , Pos-language models and decision tree models ',Dagan,'amod_approaches_Successful nsubj_include_approaches partmod_approaches_aimed prep_aimed_at pcomp_at_trying aux_overcome_to xcomp_trying_overcome det_limitation_the amod_limitation_sparse nn_limitation_data dobj_overcome_limitation nn__backoff dobj_include_ amod__Turing-Good nn__variants conj__ nn__interpolation conj__ amod__deleted nn__estimation conj__ amod__similarity-based nn__models conj__ amod__Pos-language nn__models conj__ cc__and nn__decision nn__tree nn__models conj__ ' +o,6162,'Of the several slightly different definitions of a base NP in the literature we use for the purposes of this work the definition presented in and used also by and others ',Ramshaw,'prep_use_Of det_definitions_the amod_definitions_several advmod_different_slightly amod_definitions_different pobj_Of_definitions prep_definitions_of det_NP_a nn_NP_base pobj_of_NP prep_NP_in det_literature_the pobj_in_literature nsubj_use_we prep_use_for det_purposes_the pobj_for_purposes prep_purposes_of det_work_this pobj_of_work det_definition_the dobj_use_definition dep_definition_presented prep_presented_in pobj_in_ cc_presented_and conj_presented_used advmod_used_also prep_used_by pobj_by_ cc__and conj__others ' +o,6163,'Furthermore , the BLEU score performance suggests that our model is not very powerful , but some interesting hints can be found in Table 3 when we compare our method with a 5-gram language model to a state-of-the-art system Moses based on various evaluation metrics , including BLEU score , NIST score , METEOR , TER , WER and PER ',Banerjee,'advmod_suggests_Furthermore det_performance_the amod_performance_BLEU nn_performance_score nsubj_suggests_performance complm_powerful_that poss_model_our nsubj_powerful_model cop_powerful_is neg_powerful_not advmod_powerful_very ccomp_suggests_powerful cc_suggests_but det_hints_some amod_hints_interesting nsubjpass_found_hints aux_found_can auxpass_found_be conj_suggests_found prep_found_in pobj_in_Table num_Table_3 advmod_compare_when nsubj_compare_we advcl_found_compare poss_method_our dobj_compare_method prep_compare_with det_model_a amod_model_5-gram nn_model_language pobj_with_model prep_model_to det_system_a amod_system_state-of-the-art pobj_to_system nn__Moses nsubj_based_ dep_suggests_based prep_based_on amod_metrics_various nn_metrics_evaluation pobj_on_metrics prep_based_including amod_score_BLEU pobj_including_score amod__NIST nn__score conj_score_ nn__METEOR conj_score_ nn__TER conj_score_ conj_score_WER cc_score_and conj_score_PER ' +o,6164,'Finally , the loglikelihood ratios test -LRB- henceforth LLR -RRB- is applied on each set of pairs ',Dunning,'advmod_applied_Finally det_test_the amod_test_loglikelihood nn_test_ratios nsubjpass_applied_test nn_LLR_henceforth abbrev_test_LLR partmod_test_ auxpass_applied_is prep_applied_on det_set_each pobj_on_set prep_set_of pobj_of_pairs ' +o,6165,' make a similar assumption ',Kim,'advmod_make_ det_assumption_a amod_assumption_similar dobj_make_assumption ' +o,6166,' proposes a method for word sense disambiguation , which is based on Monolingual Bootstrapping ',Yarowsky,'advmod_proposes_ det_method_a nsubj_proposes_method prep_method_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation nsubjpass_based_which auxpass_based_is rcmod_method_based prep_based_on nn_Bootstrapping_Monolingual pobj_on_Bootstrapping ' +o,6167,'This set of context vectors is then clustered into a predetermined number of coherent clusters or context groups using Buckshot , a combination of the EM algorithm and agglomerative clustering ',Brown,'det_set_This nsubjpass_clustered_set prep_set_of nn_vectors_context pobj_of_vectors auxpass_clustered_is advmod_clustered_then prep_clustered_into det_number_a amod_number_predetermined pobj_into_number prep_number_of amod_clusters_coherent pobj_of_clusters cc_clusters_or nn_groups_context conj_clusters_groups partmod_clusters_using nn__Buckshot dobj_using_ det_combination_a appos__combination prep_combination_of det_algorithm_the nn_algorithm_EM pobj_of_algorithm cc_algorithm_and nn_clustering_agglomerative conj_algorithm_clustering ' +o,6168,'A subst(req, cons(c, argo)) st ^ rel(c, z) s2 ~(i,k,=,;~z[p~(:) ^ ~(~)]) (Vi,j,w)n(i,j,w) D (3z)cn(i,j,z,w) (Vi,j, k, w, z, c, rel)prep(i, j, w) ^ np(j, k, x) A rel(c, z) In 3 ptXi, k,,~z[w(c, z)], , Req(w)) For example, the first axiom says that there is a sentence from point i to point k asserting eventuality e if there is a noun phrase from i to j referring to z and a verb phrase from j to k denoting predicate p with arguments arg8 and having an associated requirement req, and there is (or, for $3, can be assumed to be) an eventuality e of p\'s being true of , where c is related to or coercible from x (with an assumability cost of $20), and the requirement req associated with p can be proved or, for $10, assumed to hold of the arguments of p. The symbol c&el denotes the conjunction of eventualities e and el (See Hobbs (1985b), p. 35).',Hobbs,'' +n,6169,'At the same time , we believe our method has advantages over the approach developed initially at IBM for training translation systems automatically ',Brown,'prep_believe_At det_time_the amod_time_same pobj_At_time nsubj_believe_we poss_method_our nsubj_has_method ccomp_believe_has dobj_has_advantages prep_advantages_over det_approach_the pobj_over_approach partmod_approach_developed advmod_developed_initially prep_developed_at nn__IBM pobj_at_ prep__for nn_systems_training nn_systems_translation pobj_for_systems advmod_developed_automatically ' +o,6170,'31 The Corpus The systems are applied to examples from the Penn Treebank a corpus of over 45 million words of American English annotated with both part-of-speech and syntactic tree information ',Marcus,'det_systems_The nn_systems_Corpus det_systems_The nsubjpass_applied_systems auxpass_applied_are dep_31_applied prep_applied_to pobj_to_examples prep_examples_from det__the nn__Penn nn__Treebank pobj_from_ det_corpus_a nsubj_annotated_corpus prep_corpus_of quantmod_million_over number_million_45 num_words_million pobj_of_words prep_words_of nn_English_American pobj_of_English dep_31_annotated prep_annotated_with det_information_both amod_information_part-of-speech cc_part-of-speech_and conj_part-of-speech_syntactic nn_information_tree pobj_with_information ' +o,6171,'The alignment of sentences can be done sufficiently well using cues such as sentence length or cognates ',Brown,'det_alignment_The nsubjpass_done_alignment prep_alignment_of pobj_of_sentences aux_done_can auxpass_done_be advmod_well_sufficiently advmod_done_well purpcl_done_using dobj_using_cues dep_as_such prep_cues_as nn__sentence nn__length pobj_as_ cc__or nn__cognates conj__ ' +o,6172,'Some of this work focuses on classifying the semantic orientation of individual words or phrases , using linguistic heuristics or a pre-selected set of seed words ',Turney,'nsubj_focuses_Some prep_Some_of det_work_this pobj_of_work prep_focuses_on pcomp_on_classifying det_orientation_the amod_orientation_semantic dobj_classifying_orientation prep_orientation_of amod_words_individual pobj_of_words cc_words_or conj_words_phrases dep_classifying_using amod_heuristics_linguistic dobj_using_heuristics cc_heuristics_or det_set_a amod_set_pre-selected conj_heuristics_set prep_set_of nn_words_seed pobj_of_words ' +n,6173,'Despite relying on a the same concept , our approach outperforms BE in most comparisons , and it often achieves higher correlations with human judgments than the string-matching metric ROUGE ',Lin,'prep_BE_Despite pcomp_Despite_relying prep_relying_on det_concept_a dep_concept_the amod_concept_same pobj_on_concept poss_approach_our nsubj_BE_approach aux_BE_outperforms prep_BE_in amod_comparisons_most pobj_in_comparisons cc_BE_and nsubj_achieves_it advmod_achieves_often conj_BE_achieves amod_correlations_higher dobj_achieves_correlations prep_achieves_with amod_judgments_human pobj_with_judgments prep_judgments_than det_ROUGE_the amod_ROUGE_string-matching amod_ROUGE_metric pobj_than_ROUGE ' +o,6174,'As in phrasebased translation model estimation , ? also contains two lexical weights , counters for number of target terminals generated ',Koehn,'prep_contains_As dep_As_in dep_in_phrasebased dep_phrasebased_translation nsubj_estimation_model ccomp_translation_estimation advmod_contains_also num_weights_two amod_weights_lexical dobj_contains_weights nsubj_contains_ appos__counters prep_counters_for pobj_for_number prep_number_of nn_terminals_target pobj_of_terminals partmod_terminals_generated ' +o,6175,'Abduction has been applied to the solution of local pragmatics problems and to story understanding ',Hobbs,'nsubjpass_applied_Abduction aux_applied_has auxpass_applied_been prep_applied_to det_solution_the pobj_to_solution prep_solution_of amod__local nn__pragmatics nn__problems pobj_of_ cc_to_and conj_to_to nn__story nn__understanding pobj_to_ ' +o,6176,' argue that these restrictions reduce our ability to model translation equivalence effectively ',Wellington,'nsubj_argue_ complm_reduce_that det_restrictions_these nsubj_reduce_restrictions ccomp_argue_reduce poss_ability_our dobj_reduce_ability aux_model_to infmod_ability_model nn_equivalence_translation dobj_model_equivalence advmod_model_effectively ' +o,6177,'Word alignment is also a required first step in other algorithms such as for learning sub-sentential phrase pairs or the generation of parallel treebanks ',Zhechev,'amod_alignment_Word nsubj_step_alignment cop_step_is advmod_step_also det_step_a amod_step_required amod_step_first prep_step_in amod_algorithms_other pobj_in_algorithms prep_step_such amod__as prep_as_for pcomp_for_learning amod_pairs_sub-sentential nn_pairs_phrase dobj_learning_pairs dep_such_ cc__or det_generation_the conj__generation prep_generation_of amod_treebanks_parallel pobj_of_treebanks ' +o,6178,'Intuitively , if we are able to find good correspondences through linking pivots , then the augmented source data should transfer better to a target domain ',Blitzer,'advmod_transfer_Intuitively mark_able_if nsubj_able_we cop_able_are dep_transfer_able aux_find_to xcomp_able_find amod_correspondences_good dobj_find_correspondences prep_find_through pcomp_through_linking dobj_linking_pivots advmod_data_then det_data_the amod_data_augmented nn_data_source nsubj_transfer_data aux_transfer_should advmod_transfer_better prep_transfer_to det_domain_a nn_domain_target pobj_to_domain ' +o,6179,'Then we use both Moses decoder and its suppo We run the decoder with its d then use Moses \' implementation of minimum error rate training to tune the feature weights on the development set ',Och,'advmod_use_Then nsubj_use_we det_decoder_both nn_decoder_Moses dobj_use_decoder cc_decoder_and poss_suppo_its conj_decoder_suppo nsubj_run_We dep_use_run det_decoder_the nsubj_implementation_decoder prep_decoder_with poss_use_its dep_use_d dep_d_then pobj_with_use poss_implementation_Moses xcomp_run_implementation prep_implementation_of amod__minimum nn__error nn__rate nn__training pobj_of_ aux_tune_to infmod_implementation_tune det_weights_the nn_weights_feature dobj_tune_weights prep_tune_on det_set_the nn_set_development pobj_on_set ' +p,6180,'As aptly pointed out in Jean , agreement measures proposed so far in the computational linguistics literature has failed to ask an important question of whether results obtained using agreement data are in any way different from random data ',Carletta,'mark_pointed_As nsubj_pointed_aptly advcl_failed_pointed prt_pointed_out prep_pointed_in nn__Jean pobj_in_ nn_measures_agreement nsubj_failed_measures partmod_measures_proposed advmod_far_so advmod_proposed_far prep_proposed_in det_literature_the amod_literature_computational nn_literature_linguistics pobj_in_literature aux_failed_has aux_ask_to xcomp_failed_ask det_question_an amod_question_important dobj_ask_question prep_question_of complm_different_whether nsubj_different_results partmod_results_obtained xcomp_obtained_using nn_data_agreement dobj_using_data cop_different_are advmod_different_in det_way_any dep_in_way pcomp_of_different prep_different_from amod_data_random pobj_from_data ' +o,6181,'2 Inside-out alignments identified so-called inside-out alignments , two alignment configurations that can not be induced by binary synchronous context-free grammars ; these alignment configurations , while infrequent in language pairs such as EnglishFrench , have been argued to be frequent in other language pairs , incl ',Wu,'num__2 amod__Inside-out nn__alignments nsubj_identified_ amod_alignments_so-called amod_alignments_inside-out dobj_identified_alignments num_configurations_two amod_configurations_alignment appos_alignments_configurations nsubjpass_induced_that aux_induced_can neg_induced_not auxpass_induced_be rcmod_configurations_induced prep_induced_by advmod_synchronous_binary amod_grammars_synchronous amod_grammars_context-free pobj_by_grammars det_configurations_these nn_configurations_alignment nsubjpass_argued_configurations mark_infrequent_while parataxis_argued_infrequent prep_infrequent_in nn_pairs_language pobj_in_pairs dep_as_such prep_pairs_as nn__EnglishFrench pobj_as_ aux_argued_have auxpass_argued_been parataxis_identified_argued aux_frequent_to cop_frequent_be xcomp_argued_frequent prep_frequent_in amod_pairs_other nn_pairs_language pobj_in_pairs advmod_frequent_incl ' +p,6182,'Many previous studies have shown that the log-likelihood ratio is well suited for this purpose ',Dunning,'amod_studies_Many amod_studies_previous nsubj_shown_studies aux_shown_have complm_suited_that det_ratio_the amod_ratio_log-likelihood nsubjpass_suited_ratio auxpass_suited_is advmod_suited_well ccomp_shown_suited prep_suited_for det_purpose_this pobj_for_purpose ' +o,6183,'1 Data Data for 64 verbs -LRB- shown in Table 1 -RRB- was collected from three corpora ; The British National Corpus -LRB- BNC -RRB- -LRB- http ` J\\/infooxacuk\\/bnc \\/ indexhtml -RRB- , the Penn Treehank parsed version of the Brown Corpus -LRB- Brown -RRB- , and the Penn Treebank Wall Street Journal corpas -LRB- WSJ -RRB- ',Marcus,'dep_collected_1 nn_Data_Data nsubjpass_collected_Data prep_Data_for num_verbs_64 pobj_for_verbs dep_verbs_shown prep_shown_in pobj_in_Table num_Table_1 auxpass_collected_was prep_collected_from num_corpora_three pobj_from_corpora det_Corpus_The nn_Corpus_British nn_Corpus_National dep_corpora_Corpus abbrev_corpora_BNC dep_collected_http amod_indexhtml_J\\/infooxacuk\\/bnc amod_indexhtml_\\/ dobj_http_indexhtml det_Treehank_the nn_Treehank_Penn nsubj_parsed_Treehank ccomp_collected_parsed dobj_parsed_version prep_version_of det_Corpus_the nn_Corpus_Brown pobj_of_Corpus appos_Corpus_Brown cc_collected_and det_Journal_the nn_Journal_Penn nn_Journal_Treebank nn_Journal_Wall nn_Journal_Street nsubj_corpas_Journal conj_collected_corpas dep_corpas_WSJ ' +o,6184,' -LRB- Z&C 08 -RRB- generated CTB 30 from CTB 40 ',Zhang,'nsubj_generated_ appos__Z&C num_Z&C_08 dobj_generated_CTB num_CTB_30 prep_generated_from pobj_from_CTB num_CTB_40 ' +o,6185,'Analyze resulting findings to determine a progression of competence In we discuss the initial steps we took in this process , including the development of a list of error codes documented by a coding manual , the verification of our manual and coding scheme by testing inter-coder reliability in a subset of the corpus -LRB- where we achieved a Kappa agreement score of a0 a1a3a2a5a4a7a6 -RRB- 2 , and the subsequent tagging of the entire corpus ',Carletta,'dep_Analyze_resulting dobj_resulting_findings aux_determine_to xcomp_resulting_determine det_progression_a dobj_determine_progression prep_progression_of pobj_of_competence prep_determine_In pcomp_In_ nsubj_discuss_we ccomp__discuss det_steps_the amod_steps_initial dobj_discuss_steps nsubj_took_we rcmod_steps_took prep_took_in det_process_this pobj_in_process prep_took_including det_development_the pobj_including_development prep_development_of det_list_a pobj_of_list prep_list_of nn_codes_error pobj_of_codes partmod_codes_documented prep_documented_by det_manual_a amod_manual_coding pobj_by_manual det_verification_the appos_manual_verification prep_verification_of poss_manual_our pobj_of_manual cc_resulting_and conj_resulting_coding dobj_coding_scheme prep_coding_by amod_reliability_testing amod_reliability_inter-coder pobj_by_reliability prep_reliability_in det_subset_a pobj_in_subset prep_reliability_of det_corpus_the pobj_of_corpus advmod_achieved_where nsubj_achieved_we dep_reliability_achieved det__a nn__Kappa nn__agreement nn__score dobj_achieved_ prep__of amod_a1a3a2a5a4a7a6_a0 pobj_of_a1a3a2a5a4a7a6 amod_reliability_2 cc_reliability_and det_tagging_the amod_tagging_subsequent conj_reliability_tagging prep_tagging_of det_corpus_the amod_corpus_entire pobj_of_corpus ' +o,6186,'2 Related Work Given its potential usefulness in coreference resolution , anaphoricity determination has been studied fairly extensively in the literature and can be classified into three categories : heuristic rule-based , statistics-based and learning-based ',Bergsma,'num_Work_2 amod_Work_Related prep_studied_Given poss_usefulness_its amod_usefulness_potential dep_Given_usefulness prep_usefulness_in amod_resolution_coreference pobj_in_resolution amod_determination_anaphoricity nsubjpass_studied_determination aux_studied_has auxpass_studied_been dep_Work_studied advmod_studied_fairly dep_in_extensively prep_studied_in det_literature_the pobj_in_literature cc_studied_and aux_classified_can auxpass_classified_be conj_studied_classified prep_classified_into num_categories_three pobj_into_categories amod__heuristic amod__rule-based dep_categories_ amod__statistics-based conj__ cc__and amod__learning-based conj__ ' +o,6187,'This evaluation shows that our WIDL-based approach to generation is capable of obtaining headlines that compare favorably , in both content and fluency , with extractive , state-of-the-art results , while it outperforms a previously-proposed abstractive system by a wide margin ',Lin,'det_evaluation_This nsubj_shows_evaluation complm_capable_that poss_approach_our amod_approach_WIDL-based nsubj_capable_approach prep_approach_to pobj_to_generation cop_capable_is ccomp_shows_capable prep_capable_of pcomp_of_obtaining dobj_obtaining_headlines nsubj_compare_that rcmod_headlines_compare advmod_compare_favorably prep_compare_in preconj_content_both amod__content cc_content_and conj_content_fluency conj_content_with amod_results_extractive amod_results_state-of-the-art pobj_with_results pobj_in_ mark_outperforms_while nsubj_outperforms_it advcl_capable_outperforms det_system_a amod_system_previously-proposed nn_system_abstractive dobj_outperforms_system prep_outperforms_by det__a amod__wide nn__margin pobj_by_ ' +o,6188,'6 Concluding remarks Our work presents a set of improvements on previous state of the art of Grammar Association: first, by providing better language models to the original system described in (Vidal et al. , 1993); second, by setting the technique into a rigorous statistical framework, clarifying which kind of probabilities have to be estimated by association models; third, by developing a novel and especially adequate association model: Loco C. On the other hand, though experimental results are quite good, we find them particularly relevant for pointing out directions to follow for further improvement of the Grammar Association technique.',Brown,'' +o,6189,'We set all feature weights by optimizing Bleu directly using minimum error rate training -LRB- MERT -RRB- on the tuning part of the development set -LRB- dev-test2009a -RRB- ',Och,'nsubj_set_We advmod_set_all dep_set_feature dobj_feature_weights prep_feature_by pcomp_by_optimizing nn__Bleu nsubj_using_ advmod_using_directly ccomp_optimizing_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training abbrev_training_MERT partmod_training_ prep__on det_part_the amod_part_tuning pobj_on_part prep_part_of det_set_the nn_set_development pobj_of_set appos_training_dev-test2009a ' +o,6190,'We obtained word alignments of training data by first running GIZA + + and then applying the refinement rule grow-diag-final-and ',Koehn,'nsubj_obtained_We nn_alignments_word dobj_obtained_alignments prep_alignments_of nn_data_training pobj_of_data prep_obtained_by advmod_running_first pcomp_by_running dobj_running_GIZA amod_GIZA_+ cc_+_+ conj_+_ cc_running_and advmod_applying_then conj_running_applying det_grow-diag-final-and_the amod_grow-diag-final-and_refinement nn_grow-diag-final-and_rule dobj_applying_grow-diag-final-and ' +o,6191,'When we trained external Chinese models , we used the same unlabeled data set as , including the bilingual dictionary ',DeNero,'advmod_trained_When nsubj_trained_we dep_used_trained amod_models_external amod_models_Chinese dobj_trained_models nsubj_used_we det_data_the amod_data_same amod_data_unlabeled dobj_used_data partmod_data_set prep_set_as pobj_as_ prep_data_including det_dictionary_the amod_dictionary_bilingual pobj_including_dictionary ' +o,6192,'The training data is aligned using the LEAF technique ',Fraser,'det_data_The nn_data_training nsubjpass_aligned_data auxpass_aligned_is xcomp_aligned_using det_technique_the nn_technique_LEAF dobj_using_technique ' +o,6193,' et al 2006 -RRB- ',McClosky,'nn_2006_ nn_2006_et nn_2006_al ' +o,6194,'This is con rmed by the translation experiments in which the evaluation data sets were translated using the servers translation engines and the translation quality was evaluated using the standard automatic evaluation metrics BLEU and METEOR where scores range between 0 -LRB- worst -RRB- and 1 -LRB- best -RRB- ',Banerjee,'nsubj__This cop__is amod__con amod__rmed prep_rmed_by det_experiments_the nn_experiments_translation pobj_by_experiments rel_translated_in pobj_in_which det_sets_the nn_sets_evaluation nn_sets_data nsubjpass_translated_sets auxpass_translated_were rcmod_experiments_translated xcomp_translated_using det_engines_the amod_engines_servers nn_engines_translation dobj_using_engines cc_translated_and det_quality_the nn_quality_translation nsubjpass_evaluated_quality auxpass_evaluated_was conj_translated_evaluated xcomp_evaluated_using det_metrics_the amod_metrics_standard amod_metrics_automatic nn_metrics_evaluation dobj_using_metrics nn__BLEU cc__and nn__METEOR conj__ advmod_range_where nsubj_range_scores rcmod__range prep_range_between pobj_between_0 dep__worst cc__and conj__1 dep_1_best ' +p,6195,'Recent work , , has shown that adding many millions of words of machine parsed and reranked LA Times articles does , in fact , improve performance of the parser on the closely related WSJ data ',McClosky,'amod_work_Recent nsubj_shown_work appos_work_ aux_shown_has complm_parsed_that csubj_parsed_adding amod_millions_many dobj_adding_millions prep_millions_of pobj_of_words prep_words_of pobj_of_machine ccomp_shown_parsed cc_parsed_and conj_parsed_reranked nn_articles_LA nn_articles_Times nsubj_does_articles ccomp_reranked_does prep_does_in pobj_in_fact ccomp_does_improve dobj_improve_performance prep_performance_of det_parser_the pobj_of_parser prep_parser_on det_data_the advmod_related_closely amod_data_related nn_data_WSJ pobj_on_data ' +o,6196,'Given an input training corpus of such derivations D = d1 dn, a vector feature function on derivations vectorF(d), and an initial weight vector vectorw, the perceptron performs two steps for each training example di D: Decode: d = argmaxdD(src(di)) parenleftBig vectorw vectorF(d) parenrightBig Update: vectorw = vectorw + vectorF(di) vectorF(d) where D(src(d)) enumerates all possible derivations with the same source side as d. To improve generalization, the final feature vector is the average of all vectors found during learning (Collins, 2002).',Collins,'' +o,6197,'1 Introduction Given a source-language -LRB- eg , French -RRB- sentence f , the problem of machine translation is to automatically produce a target-language -LRB- eg , English -RRB- translation e The mathematics of the problem were formalized by , and re-formulated by in terms of the optimization e = arg maxe Msummationdisplay m = 1 mhm -LRB- e , f -RRB- -LRB- 1 -RRB- where fhm -LRB- e , f -RRB- g is a set of M feature functions and fmg a set of weights ',Brown,'num_Introduction_1 dep_Introduction_Given det_sentence_a amod_sentence_source-language dep_source-language_eg dep_eg_French dep_Given_sentence nsubj_is_f det_problem_the appos_f_problem prep_problem_of nn_translation_machine pobj_of_translation rcmod_sentence_is aux_translation_to advmod_translation_automatically dep_translation_produce det_translation_a amod_translation_target-language dep_target-language_eg dep_eg_English xcomp_is_translation dep_translation_e det_mathematics_The nsubjpass_formalized_mathematics prep_mathematics_of det_problem_the pobj_of_problem auxpass_formalized_were dep_translation_formalized prep_formalized_by pobj_by_ cc_formalized_and dep_=_re-formulated prep_re-formulated_by pobj_by_ dep_=_in pobj_in_terms prep_terms_of det_optimization_the pobj_of_optimization dep_=_e nn_m_= nn_m_arg nn_m_maxe nn_m_Msummationdisplay nsubj_=_m conj_formalized_= num_mhm_1 dobj_=_mhm dep_mhm_e appos_e_f appos_mhm_1 advmod_set_where amod_g_fhm dep_g_e appos_e_f nsubj_set_g cop_set_is det_set_a rcmod_mhm_set prep_set_of nn_functions_M nn_functions_feature pobj_of_functions cc_translation_and conj_translation_fmg det_set_a dobj_fmg_set prep_set_of pobj_of_weights ' +o,6198,'We also trained an HMM aligner as described in and used the posteriors of this model as features ',DeNero,'nsubj_trained_We advmod_trained_also det_aligner_an nn_aligner_HMM dobj_trained_aligner mark_described_as advcl_trained_described prep_described_in pobj_in_ cc_trained_and conj_trained_used det_posteriors_the dobj_used_posteriors prep_posteriors_of det_model_this pobj_of_model prep_used_as pobj_as_features ' +o,6199,'Rule-based taggers use POS-dependent constraints defined by experienced linguists ',Marcus,'amod__Rule-based nn__taggers nsubj_use_ amod_constraints_POS-dependent dobj_use_constraints partmod_constraints_defined prep_defined_by amod_linguists_experienced pobj_by_linguists ' +o,6200,'We use the similarity proposed by ',Lin,'nsubj_use_We det_similarity_the dobj_use_similarity partmod_similarity_proposed prep_proposed_by ' +p,6201,'In addition to the widely used BLEU and NIST scores , we also evaluate translation quality with the recently proposed Meteor and four edit-distance style metrics , Word Error Rate -LRB- WER -RRB- , Positionindependent word Error Rate -LRB- PER -RRB- , CDER , which allows block reordering , and Translation Edit Rate -LRB- TER -RRB- ',Papineni,'prep_evaluate_In dep_In_addition dep_In_to det__the advmod_used_widely amod__used nn__BLEU pobj_In_ cc__and amod_scores_NIST amod_scores_ conj__scores nsubj_evaluate_we advmod_evaluate_also nn_quality_translation dobj_evaluate_quality prep_evaluate_with det__the advmod_proposed_recently amod__proposed nn__Meteor pobj_with_ cc__and num_metrics_four amod_metrics_edit-distance nn_metrics_style conj__metrics nn_Rate_Word nn_Rate_Error appos_metrics_Rate abbrev_Rate_WER amod__Positionindependent nn__word nn__Error nn__Rate abbrev__PER conj_Rate_ conj_Rate_CDER nsubj_allows_which rcmod_CDER_allows xcomp_allows_block nn__reordering dobj_block_ cc_Rate_and nn__Translation nn__Edit nn__Rate abbrev__TER conj_Rate_ ' +p,6202,'Most stateof-the-art SMT systems treat grammatical elements in exactly the same way as content words , and rely on general-purpose phrasal translations and target language models to generate these elements ',Koehn,'advmod_systems_Most amod_systems_stateof-the-art nn_systems_SMT nsubj_treat_systems amod_elements_grammatical dobj_treat_elements prep_treat_in advmod_way_exactly det_way_the amod_way_same pobj_in_way prep_way_as nn_words_content pobj_as_words cc_treat_and conj_treat_rely prep_rely_on amod_translations_general-purpose nn_translations_phrasal pobj_on_translations cc_translations_and nn_models_target nn_models_language conj_translations_models aux_generate_to xcomp_rely_generate det_elements_these dobj_generate_elements ' +o,6203,'Note that all systems were optimized using a non-deterministic implementation of the Minimum Error Rate Training described in ',Och,'complm_optimized_that det_systems_all nsubjpass_optimized_systems auxpass_optimized_were ccomp_Note_optimized xcomp_optimized_using det_implementation_a amod_implementation_non-deterministic dobj_using_implementation prep_implementation_of det_Training_the nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_of_Training partmod_Training_described prep_described_in ' +o,6204,'2 Recap of BLEU , ROUGE-W and METEOR The most commonly used automatic evaluation metrics , BLEU and NIST , are based on the assumption that The closer a machine translation is to a promt1 : Life is like one nice chocolate in box ref : Life is just like a box of tasty chocolate ref : Life is just like a box of tasty chocolate mt2 : Life is of one nice chocolate in box Figure 1 : Alignment Example for ROUGE-W fessional human translation , the better it is ',Papineni,'num_Recap_2 nsubjpass_based_Recap prep_Recap_of pobj_of_BLEU conj_BLEU_ROUGE-W cc_BLEU_and conj_BLEU_METEOR det_metrics_The advmod_used_most advmod_used_commonly amod_metrics_used amod_metrics_automatic nn_metrics_evaluation dep_Recap_metrics nn__BLEU appos_Recap_ cc__and nn__NIST conj__ auxpass_based_are prep_based_on det_assumption_the pobj_on_assumption complm_is_that dep_closer_The dep_is_closer det_translation_a nn_translation_machine nsubj_is_translation dep_is_is prep_is_to det_promt1_a pobj_to_promt1 nsubj_is_Life parataxis_is_is prep_is_like num_chocolate_one amod_chocolate_nice pobj_like_chocolate prep_chocolate_in nn_ref_box pobj_in_ref nsubj_is_Life parataxis_is_is advmod_is_just prep_is_like det_box_a pobj_like_box prep_box_of amod_ref_tasty nn_ref_chocolate pobj_of_ref nsubj_is_Life parataxis_is_is advmod_is_just prep_is_like det_box_a pobj_like_box prep_box_of amod_mt2_tasty nn_mt2_chocolate pobj_of_mt2 nsubj_is_Life parataxis_is_is prep_is_of num_chocolate_one amod_chocolate_nice pobj_of_chocolate prep_chocolate_in nn_Figure_box pobj_in_Figure num_Figure_1 nn_Example_Alignment dep_chocolate_Example prep_Example_for amod_translation_ROUGE-W amod_translation_fessional amod_translation_human pobj_for_translation dep_better_the dep_is_better nsubj_is_it ccomp_based_is ' +o,6205,'21 Relationship Types There is a large body of related work that deals with discovery of basic relationship types represented in useful resources such as WordNet , including hypernymy , synonymy and meronymy ',Snow,'num_Types_21 nn_Types_Relationship expl_is_There dep_Types_is det_body_a amod_body_large nsubj_is_body prep_body_of amod_work_related pobj_of_work complm_represented_that nsubj_represented_deals prep_deals_with pobj_with_discovery prep_discovery_of amod_types_basic nn_types_relationship pobj_of_types ccomp_is_represented prep_represented_in amod_resources_useful pobj_in_resources dep_as_such prep_resources_as pobj_as_WordNet xcomp_represented_including advmod__hypernymy acomp_including_ advmod__synonymy conj__ cc__and advmod__meronymy conj__ ' +p,6206,'For instance , shows that a simple feature augmentation method for SVM is able to effectively use both labeled target and source data to provide the best domainadaptation results in a number of NLP tasks ','Daume III','prep_shows_For pobj_For_instance nsubj_shows_ complm_able_that det_method_a amod_method_simple nn_method_feature nn_method_augmentation nsubj_able_method prep_method_for pobj_for_SVM cop_able_is ccomp_shows_able aux_use_to advmod_use_effectively xcomp_able_use dobj_use_both partmod_both_labeled nn_data_target cc_target_and conj_target_source dobj_labeled_data aux_provide_to xcomp_labeled_provide det_results_the amod_results_best nn_results_domainadaptation dobj_provide_results prep_results_in det_number_a pobj_in_number prep_number_of nn_tasks_NLP pobj_of_tasks ' +p,6207,'For English , we use three state-of-the-art taggers : the taggers of and in Step 1 , and the SVM tagger in Step 3 ',Shen,'prep_use_For pobj_For_English nsubj_use_we num_taggers_three amod_taggers_state-of-the-art dobj_use_taggers det_taggers_the dep_taggers_taggers prep_taggers_of pobj_of_ cc__and conj__ prep_use_in number_1_Step pobj_in_1 cc_1_and det__the nn__SVM nn__tagger conj_1_ prep__in pobj_in_Step num_Step_3 ' +p,6208,'We used the average perceptron algorithm of in our experiments , a variation that has been proven to be more effective than the standard algorithm shown in Figure 2 ',Collins,'nsubj_used_We det_algorithm_the amod_algorithm_average nn_algorithm_perceptron dobj_used_algorithm prep_algorithm_of pobj_of_ prep_used_in poss_experiments_our pobj_in_experiments det_variation_a appos_experiments_variation nsubjpass_proven_that aux_proven_has auxpass_proven_been rcmod_variation_proven aux_effective_to cop_effective_be advmod_effective_more xcomp_proven_effective prep_effective_than det_algorithm_the amod_algorithm_standard pobj_than_algorithm partmod_algorithm_shown prep_shown_in pobj_in_Figure num_Figure_2 ' +o,6209,' proposes two approximate models based on the variational approach ',Titov,'nsubj_proposes_ num_models_two amod_models_approximate dobj_proposes_models prep_proposes_based dep_based_on det_approach_the amod_approach_variational pobj_on_approach ' +p,6210,'More recently , phrase-based models have been proposed as a highly successful alternative to the IBM models ',Och,'advmod_recently_More advmod_proposed_recently amod__phrase-based nn__models nsubjpass_proposed_ aux_proposed_have auxpass_proposed_been prep_proposed_as det_alternative_a advmod_successful_highly amod_alternative_successful pobj_as_alternative prep_alternative_to det_models_the nn_models_IBM pobj_to_models ' +o,6211,'1087 Model 3 of is a zero-order alignment model like Model 2 including in addition fertility paranmters ',Brown,'num_Model_1087 nsubj_model_Model num_Model_3 prep_Model_of pobj_of_ cop_model_is det_model_a amod_model_zero-order nn_model_alignment prep_model_like pobj_like_Model num_Model_2 prep_model_including dep_including_in nn_paranmters_addition nn_paranmters_fertility pobj_in_paranmters ' +o,6212,'Among these methods , SVM is shown to perform better than other methods ',Pang,'prep_shown_Among det_methods_these pobj_Among_methods nsubjpass_shown_SVM auxpass_shown_is aux_perform_to xcomp_shown_perform advmod_perform_better prep_perform_than amod_methods_other pobj_than_methods ' +o,6213,'3 Maximum Entropy ME models implement the intuition that the best model is the one that is consistent with the set of constraints imposed by the evidence , but otherwise is as uniform as possible ',Berger,'num_models_3 amod_models_Maximum nn_models_Entropy nn_models_ME nsubj_intuition_models dep_intuition_implement det_intuition_the complm_one_that det_model_the amod_model_best nsubj_one_model cop_one_is det_one_the ccomp_intuition_one nsubj_consistent_that cop_consistent_is rcmod_one_consistent prep_consistent_with det_set_the pobj_with_set prep_set_of pobj_of_constraints partmod_constraints_imposed prep_imposed_by det_evidence_the pobj_by_evidence cc_intuition_but advmod_uniform_otherwise cop_uniform_is advmod_uniform_as conj_intuition_uniform prep_uniform_as pobj_as_possible ' +o,6214,'31 Candidate NPs Noun phrases were extracted using -RRB- ',Ramshaw,'num_phrases_31 nn_phrases_Candidate nn_phrases_NPs nn_phrases_Noun nsubjpass_extracted_phrases auxpass_extracted_were xcomp_extracted_using dobj_using_ ' +o,6215,'Many statistical metrics have been proposed , including pointwise mutual information -LRB- MI -RRB- , mean and variance , hypothesis testing -LRB- t-test , chisquare test , etc -RRB- , log-likelihood ratio -LRB- LR -RRB- , statistic language model , and so on ',Dunning,'amod_metrics_Many amod_metrics_statistical nsubjpass_proposed_metrics aux_proposed_have auxpass_proposed_been prep_proposed_including amod_information_pointwise amod_information_mutual pobj_including_information abbrev_information_MI partmod_information_ parataxis__mean cc_mean_and conj_mean_variance nn_testing_hypothesis dobj__testing amod_test_t-test amod_test_chisquare dep_testing_test appos_test_etc amod_ratio_log-likelihood conj_testing_ratio abbrev_ratio_LR amod__ amod__statistic nn__language nn__model dep_ratio_ cc_testing_and advmod_on_so advmod_testing_on ' +o,6216,'766 System Beam Error \% 5 337 1 290 289 Guided Learning , feature B 3 285 all 285 284 276 Guided Learning , feature E 1 273 Guided Learning , feature E 3 267 Table 4 : Comparison with the previous works According to the experiments shown above , we build our best system by using feature set E with beam width B = 3 ',Collins,'num_Beam_766 nn_Beam_System num_\%_Error nsubj__\% dep_Beam_ number_337_5 dobj__337 nn_Learning_ num_Learning_1 num_Learning_290 num_Learning_ num_Learning_289 nn_Learning_Guided dep_337_Learning nn_B_feature appos_337_B number_285_3 num__285 dep_B_ det_Learning_all num_Learning_285 num_Learning_ num_Learning_284 num_Learning_ num_Learning_276 nn_Learning_Guided dep__Learning nn_E_feature appos_337_E number_273_1 num_Learning_273 nn_Learning_Guided dep_E_Learning nn_E_feature appos_337_E number_267_3 dep_E_267 dep_267_Table num_Table_4 dep_337_Comparison prep_Comparison_with det_works_the amod_works_previous pobj_with_works prep_build_According dep_According_to det_experiments_the pobj_to_experiments partmod_experiments_shown prep_shown_above nsubj_build_we dep_Beam_build poss_system_our amod_system_best dobj_build_system prep_build_by pcomp_by_using nn_E_feature nn_E_set dobj_using_E prep_E_with nn_B_beam amod_B_width pobj_with_B dep_3_= dep_B_3 ' +p,6217,'Models that can handle non-independent lexical features have given very good results both for part-of-speech and structural disambiguation ',Ratnaparkhi,'nsubj_given_Models nsubj_handle_that aux_handle_can rcmod_Models_handle amod_features_non-independent amod_features_lexical dobj_handle_features aux_given_have advmod_good_very amod_results_good iobj_given_results dobj_given_both prep_both_for amod_disambiguation_part-of-speech cc_part-of-speech_and conj_part-of-speech_structural pobj_for_disambiguation ' +o,6218,'We also have an additional held-out translation set , the development set , which is employed by the MT system to train the weights of its log-linear model to maximize BLEU ',Och,'nsubj_have_We advmod_have_also det_set_an amod_set_additional amod_set_held-out nn_set_translation dobj_have_set det_set_the nn_set_development appos_set_set nsubjpass_employed_which auxpass_employed_is rcmod_set_employed prep_employed_by det_system_the nn_system_MT pobj_by_system aux_train_to xcomp_employed_train det_weights_the dobj_train_weights prep_weights_of poss_model_its amod_model_log-linear pobj_of_model aux_maximize_to xcomp_train_maximize dobj_maximize_BLEU ' +o,6219,'The translation component is an analog of the IBM model 2 , with parameters that are optimized for use with the trigram ',Brown,'det_component_The nn_component_translation nsubj_analog_component cop_analog_is det_analog_an prep_analog_of det__the nn__IBM nn__model num__2 pobj_of_ prep__with pobj_with_parameters nsubjpass_optimized_that auxpass_optimized_are rcmod_parameters_optimized prep_optimized_for pobj_for_use prep_optimized_with det_trigram_the pobj_with_trigram ' +o,6220,'The toolkit also implements suffixarray grammar extraction and minimum error rate training ',Och,'det_toolkit_The nsubj_implements_toolkit advmod_implements_also amod__suffixarray nn__grammar nn__extraction dobj_implements_ cc__and nn_training_minimum nn_training_error nn_training_rate conj__training ' +o,6221,'Pattern-based approaches are known for their high accuracy in recognizing instances of relations if the patterns are carefully chosen , either manually or via automatic bootstrapping ',Brown,'amod_approaches_Pattern-based nsubjpass_known_approaches auxpass_known_are prep_known_for poss_accuracy_their amod_accuracy_high pobj_for_accuracy prep_known_in dep_instances_recognizing pcomp_in_instances prep_instances_of pobj_of_relations mark_chosen_if det_patterns_the nsubjpass_chosen_patterns auxpass_chosen_are advmod_chosen_carefully advcl_instances_chosen dep_chosen_either advmod_chosen_manually conj_chosen_ cc_in_or conj_in_via amod__automatic amod__bootstrapping pobj_via_ ' +o,6222,'Our model uses an exemplar memory that consists of 133566 verb-role-noun triples extracted from the Wall Street Journal and Brown parts of the Penn Treebank ',Marcus,'poss_model_Our nsubj_uses_model det_memory_an nn_memory_exemplar dobj_uses_memory nsubj_consists_that rcmod_memory_consists prep_consists_of num_triples_133566 amod_triples_verb-role-noun pobj_of_triples partmod_triples_extracted prep_extracted_from det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_from_Journal cc_Journal_and nn_parts_Brown conj_Journal_parts prep_Journal_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +o,6223,'Hypotheses for unknown words , both stochastic , and connectionist have been applied to unlimited vocabulary taggers ',Marcus,'nsubjpass_applied_Hypotheses prep_Hypotheses_for amod_words_unknown pobj_for_words det__both amod__stochastic conj_words_ cc_words_and amod__connectionist conj_words_ aux_applied_have auxpass_applied_been prep_applied_to amod_taggers_unlimited nn_taggers_vocabulary pobj_to_taggers ' +o,6224,'For example , if we make a mean-field assumption , with respect to hidden structure and weights , the variationalalgorithmforapproximatelyinferringthe distribution over and trees y resembles the traditional EM algorithm very closely ',Johnson,'prep_algorithm_For pobj_For_example mark_make_if nsubj_make_we dep_algorithm_make det_assumption_a amod_assumption_mean-field dobj_make_assumption prep_make_with pobj_with_respect prep_respect_to amod_structure_hidden pobj_to_structure cc_structure_and conj_structure_weights dep_distribution_the amod_distribution_variationalalgorithmforapproximatelyinferringthe nsubj_algorithm_distribution dep_distribution_over cc_distribution_and nn_y_trees conj_distribution_y cop_algorithm_resembles det_algorithm_the amod_algorithm_traditional nn_algorithm_EM advmod_closely_very advmod_algorithm_closely ' +o,6225,'2 Maximum Entropy In this bakeoff , our basic model is based on the framework described in the work of which was applied for English POS tagging ',Ratnaparkhi,'dep_based_2 nn_Entropy_Maximum dep_In_Entropy prep_based_In det_bakeoff_this pobj_In_bakeoff poss_model_our amod_model_basic nsubjpass_based_model auxpass_based_is prep_based_on det_framework_the pobj_on_framework partmod_framework_described prep_described_in det_work_the pobj_in_work prep_work_of pobj_of_ nsubjpass_applied_which auxpass_applied_was rcmod_work_applied prep_applied_for nn_POS_English pobj_for_POS partmod_POS_tagging ' +o,6226,'1 Introduction Previous work on sentiment categorization makes an implicit assumption that a single score can express the polarity of an opinion text ',Pang,'num_work_1 amod_work_Introduction amod_work_Previous nsubj_makes_work prep_work_on nn_categorization_sentiment pobj_on_categorization det_assumption_an amod_assumption_implicit dobj_makes_assumption complm_express_that det_score_a amod_score_single nsubj_express_score aux_express_can ccomp_makes_express det_polarity_the dobj_express_polarity prep_polarity_of det_text_an nn_text_opinion pobj_of_text ' +o,6227,'For examples , see ',Pang,'prep_see_For pobj_For_examples ' +o,6228,'Since the introduction of BLEU the basic n-gram precision idea has been augmented in a number of ways ',Papineni,'prep_augmented_Since det_introduction_the pobj_Since_introduction prep_introduction_of nn__BLEU pobj_of_ det_idea_the amod_idea_basic amod_idea_n-gram nn_idea_precision nsubjpass_augmented_idea aux_augmented_has auxpass_augmented_been prep_augmented_in det_number_a pobj_in_number prep_number_of pobj_of_ways ' +o,6229,'1 Introduction The best performing systems for many tasks in natural language processing are based on supervised training on annotated corpora such as the Penn Treebank and the prepositional phrase data set first described in ',Marcus,'num_Introduction_1 det_systems_The dep_performing_best amod_systems_performing nsubjpass_based_systems prep_systems_for amod_tasks_many pobj_for_tasks prep_systems_in amod_processing_natural nn_processing_language pobj_in_processing auxpass_based_are dep_Introduction_based prep_based_on amod_training_supervised pobj_on_training prep_based_on amod_corpora_annotated pobj_on_corpora dep_as_such prep_corpora_as det__the nn__Penn nn__Treebank pobj_as_ cc__and det_phrase_the amod_phrase_prepositional conj__phrase nsubj_described_data partmod_data_set advmod_set_first dep_Introduction_described prep_described_in pobj_in_ ' +o,6230,'However , as pointed out in , there is no reason to believe that the resulting parameters are optimal with respect to translation quality measured with the Bleu score ',Och,'advmod_is_However mark_pointed_as dep_is_pointed prt_pointed_out prep_pointed_in pobj_in_ expl_is_there det_reason_no nsubj_is_reason aux_believe_to infmod_reason_believe complm_optimal_that det_parameters_the amod_parameters_resulting nsubj_optimal_parameters aux_optimal_are ccomp_believe_optimal prep_optimal_with pobj_with_respect prep_optimal_to nn_quality_translation pobj_to_quality partmod_quality_measured prep_measured_with det_score_the nn_score_Bleu pobj_with_score ' +o,6231,'The initial state contains terminal items , whose labels are the POS tags given by ',Ratnaparkhi,'det_state_The amod_state_initial nsubj_contains_state amod_items_terminal dobj_contains_items poss_labels_whose nsubj_tags_labels cop_tags_are det_tags_the dep_tags_POS rcmod_items_tags prep_tags_given dep_given_by ' +o,6232,'imum error rate training -LRB- MERT -RRB- to maximize BLEU score ',Och,'nn_training_imum nn_training_error nn_training_rate nsubj__training abbrev_training_MERT aux_maximize_to xcomp__maximize amod__BLEU nn__score dobj_maximize_ ' +o,6233,'As an example of it s application , N-gram co-occurrence is used for evaluating machine translations ',Papineni,'mark_application_As det_example_an nsubj_application_example prep_example_of pobj_of_it dep_application_s advcl_used_application amod_co-occurrence_N-gram nsubjpass_used_co-occurrence auxpass_used_is prep_used_for pcomp_for_evaluating nn_translations_machine dobj_evaluating_translations ' +o,6234,'3 The Log-Likelihood-Ratio Association Measure We base all our association-based word-alignment methods on the log-likelihood-ratio -LRB- LLR -RRB- statistic introduced to the NLP community by ',Dunning,'dep_Measure_3 det_Measure_The amod_Measure_Log-Likelihood-Ratio nn_Measure_Association nsubj_base_We dep_Measure_base predet_methods_all poss_methods_our amod_methods_association-based nn_methods_word-alignment dobj_base_methods prep_base_on det_statistic_the amod_statistic_log-likelihood-ratio abbrev_statistic_LLR pobj_on_statistic partmod_statistic_introduced prep_introduced_to det_community_the nn_community_NLP pobj_to_community prep_introduced_by ' +o,6235,'Recently , some generic methods were proposed to handle context-sensitive inference , but these usually treat only a single aspect of context matching -LRB- see Section 6 -RRB- ',Downey,'advmod_proposed_Recently det_methods_some amod_methods_generic nsubjpass_proposed_methods auxpass_proposed_were aux_handle_to xcomp_proposed_handle amod__context-sensitive nn__inference dobj_handle_ cc_proposed_but nsubj_treat_these advmod_treat_usually conj_proposed_treat advmod_aspect_only det_aspect_a amod_aspect_single dobj_treat_aspect prep_aspect_of nn_matching_context pobj_of_matching dep_aspect_see dobj_see_Section num_Section_6 ' +o,6236,'The discrepancy between DEV performance and TEST performance is due to temporal distance from TRAIN and high variance in BLEU score11 We also compared our model with Pharaoh ',Koehn,'det_discrepancy_The nsubj_due_discrepancy prep_discrepancy_between nn_performance_DEV pobj_between_performance cc_performance_and amod_performance_TEST conj_performance_performance cop_due_is prep_due_to amod_distance_temporal pobj_to_distance prep_distance_from pobj_from_TRAIN cc_distance_and amod_variance_high conj_distance_variance prep_variance_in amod_score11_BLEU pobj_in_score11 nsubj_compared_We advmod_compared_also dep_due_compared poss_model_our dobj_compared_model prep_compared_with pobj_with_Pharaoh ' +o,6237,'51 Experimental setup The baseline model was Hiero with the following baseline features : two language models phrase translation probabilities p -LRB- f e -RRB- and p -LRB- e f -RRB- lexical weighting in both directions word penalty penalties for : automatically extracted rules identity rules -LRB- translating a word into itself -RRB- two classes of number\\/name translation rules glue rules The probability features are base-100 logprobabilities ',Koehn,'num_setup_51 nn_setup_Experimental det_model_The nn_model_baseline nsubj_Hiero_model cop_Hiero_was dep_setup_Hiero prep_Hiero_with det__the amod__following nn__baseline nn__features pobj_with_ num_models_two nn_models_language dep_Hiero_models nn_p_phrase nn_p_translation nn_p_probabilities dep_models_p dep_-LRB-_f dep_f_e cc_Hiero_and conj_Hiero_p dep_f_e dep_-LRB-_f amod_weighting_lexical nsubj__weighting prep_weighting_in det_directions_both pobj_in_directions dep_setup_ nn_penalties_word nn_penalties_penalty dobj__penalties prep__for advmod_extracted_automatically parataxis__extracted dobj_extracted_rules nn_rules_identity dep_rules_rules dep_rules_translating det_word_a dobj_translating_word prep_translating_into pobj_into_itself num_classes_two dep_rules_classes prep_classes_of amod_rules_number\\/name nn_rules_translation pobj_of_rules nn_rules_glue dep_extracted_rules det_features_The nn_features_probability nsubj_logprobabilities_features cop_logprobabilities_are amod_logprobabilities_base-100 rcmod_rules_logprobabilities ' +o,6238,'Finally we use Minimum Error Training -LRB- MET -RRB- to train log-linear scaling factors that are applied to the WFSTs in Equation 1 ',Och,'advmod_use_Finally nsubj_use_we nn_Training_Minimum nn_Training_Error nsubj__Training abbrev_Training_MET dep_use_ aux_train_to xcomp__train amod_factors_log-linear amod_factors_scaling dobj_train_factors nsubjpass_applied_that auxpass_applied_are rcmod_factors_applied prep_applied_to det_WFSTs_the pobj_to_WFSTs prep_applied_in pobj_in_Equation num_Equation_1 ' +o,6239,'5 Related Work Automatically finding sentences with the same meaning has been extensively studied in the field of automatic paraphrasing using parallel corpora and corporawith multiple descriptionsof the same events ',Barzilay,'num_sentences_5 amod_sentences_Related nn_sentences_Work nn_sentences_Automatically nn_sentences_finding nsubjpass_studied_sentences prep_sentences_with det_meaning_the amod_meaning_same pobj_with_meaning aux_studied_has auxpass_studied_been advmod_studied_extensively prep_studied_in det_field_the pobj_in_field prep_field_of amod_paraphrasing_automatic pobj_of_paraphrasing xcomp_studied_using amod_corpora_parallel dobj_using_corpora cc_corpora_and amod_descriptionsof_corporawith amod_descriptionsof_multiple conj_corpora_descriptionsof det_events_the amod_events_same dep_descriptionsof_events ' +o,6240,'As mentioned in Section 22 , there are words which have two or more candidate POS tags in the PTB corpus ',Marcus,'mark_mentioned_As advcl_are_mentioned prep_mentioned_in pobj_in_Section num_Section_22 expl_are_there nsubj_are_words nsubj_have_which rcmod_words_have num_tags_two cc_two_or conj_two_more nn_tags_candidate dep_tags_POS dobj_have_tags prep_have_in det_corpus_the nn_corpus_PTB pobj_in_corpus ' +o,6241,'Finally , the fourth and fifth feature functions corresponded to two lexicon models based on IBM Model 1 lexical parameters p -LRB- t s -RRB- ',Brown,'advmod_corresponded_Finally det_fourth_the nsubj_corresponded_fourth cc_fourth_and amod_functions_fifth nn_functions_feature conj_fourth_functions prep_corresponded_to num_models_two nn_models_lexicon pobj_to_models prep_corresponded_based dep_based_on nn_Model_IBM nn_p_Model num_parameters_1 amod_parameters_lexical dep_Model_parameters pobj_on_p nn_s_t appos_p_s ' +o,6242,'Using an Maximum Entropy approach to POS tagging , reports a tagging accuracy of 966 \% on the Wall Street Journal ',Ratnaparkhi,'dep_reports_Using det_approach_an nn_approach_Maximum nn_approach_Entropy dobj_Using_approach prep_Using_to dep_tagging_POS pobj_to_tagging nsubj_reports_ det_accuracy_a amod_accuracy_tagging dobj_reports_accuracy prep_accuracy_of num_\%_966 pobj_of_\% prep_\%_on det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_on_Journal ' +p,6243,'It is often straightforward to obtain large amounts of unlabeled data , making semi-supervised approaches appealing ; previous work on semisupervised methods for dependency parsing includes ',Smith,'nsubj_straightforward_It cop_straightforward_is advmod_straightforward_often aux_obtain_to xcomp_straightforward_obtain amod_amounts_large dobj_obtain_amounts prep_amounts_of amod_data_unlabeled pobj_of_data xcomp_obtain_making amod_approaches_semi-supervised nsubj_appealing_approaches xcomp_making_appealing amod_work_previous nsubj_includes_work prep_work_on amod_methods_semisupervised pobj_on_methods prep_methods_for amod_parsing_dependency pobj_for_parsing parataxis_straightforward_includes ' +o,6244,'2 Motivation Automatic subjectivity analysis methods have been used in a wide variety of text processing applications , such as tracking sentiment timelines in online forums and news , review classification , mining opinions from product reviews , automatic expressive text-to-speech synthesis , text semantic analysis , and question answering ',Turney,'num_methods_2 nn_methods_Motivation nn_methods_Automatic nn_methods_subjectivity nn_methods_analysis nsubjpass_used_methods aux_used_have auxpass_used_been prep_used_in det_variety_a amod_variety_wide nn__variety prep_variety_of nn_applications_text nn_applications_processing pobj_of_applications dep_as_such conj_variety_as nn_timelines_tracking nn_timelines_sentiment pobj_as_timelines prep_timelines_in amod_forums_online pobj_in_forums cc_variety_and nn__news conj_variety_ nn__review nn__classification appos__ nn_opinions_mining appos__opinions prep_opinions_from nn__product nn__reviews pobj_from_ amod__automatic amod__expressive amod__text-to-speech nn__synthesis appos__ nn_analysis_text amod_analysis_semantic appos__analysis pobj_in_ cc_used_and conj_used_question nn__answering dobj_question_ ' +o,6245,'Some studies exploit topically related articles derived from multiple news sources ',Barzilay,'det_studies_Some nsubj_exploit_studies advmod_related_topically amod_articles_related dobj_exploit_articles partmod_articles_derived prep_derived_from amod_sources_multiple nn_sources_news pobj_from_sources ' +o,6246,'We also note that found movie reviews to be the most 2Indeed , although our choice of title was completely independent of his , our selections were eerily similar ',Turney,'nsubj_note_We advmod_note_also complm_found_that nsubj_found_ ccomp_note_found nn_reviews_movie dobj_found_reviews aux_2Indeed_to cop_2Indeed_be det_2Indeed_the amod_2Indeed_most xcomp_found_2Indeed mark_independent_although poss_choice_our nsubj_independent_choice prep_choice_of pobj_of_title cop_independent_was advmod_independent_completely advcl_2Indeed_independent prep_similar_of pobj_of_his poss_selections_our nsubj_similar_selections cop_similar_were advmod_similar_eerily dep_note_similar ' +o,6247,'c2006 Association for Computational Linguistics Robust PCFG-Based Generation using Automatically Acquired LFG Approximations Aoife Cahill1 and Josef van Genabith1,2 1 National Centre for Language Technology (NCLT) School of Computing, Dublin City University, Dublin 9, Ireland 2 Center for Advanced Studies, IBM Dublin, Ireland {acahill,josef}@computing.dcu.ie Abstract We present a novel PCFG-based architecture for robust probabilistic generation based on wide-coverage LFG approximations (Cahill et al. , 2004) automatically extracted from treebanks, maximising the probability of a tree given an f-structure.',Cahill,'' +o,6248,'The model cleanly incorporates both syntax and lexical semantics using quasi-synchronous dependency grammars ',Smith,'det_model_The nsubj_incorporates_model advmod_incorporates_cleanly preconj_syntax_both amod_semantics_syntax cc_syntax_and conj_syntax_lexical dobj_incorporates_semantics xcomp_incorporates_using amod_grammars_quasi-synchronous amod_grammars_dependency dobj_using_grammars ' +o,6249,'Previous workthe generative models described in and the earlier version of these models described in Collins -LRB- 1997 -RRB- conditioned on punctuation as surface features of the string , treating it quite differently from lexical items ',Collins,'amod_models_Previous amod_models_workthe nn_models_generative nsubj_described_models prep_described_in pobj_in_ cc__and det_version_the amod_version_earlier conj__version prep_version_of det_models_these pobj_of_models partmod_models_described prep_described_in pobj_in_Collins appos_Collins_1997 partmod_Collins_conditioned prep_conditioned_on pobj_on_punctuation prep_described_as nn_features_surface pobj_as_features prep_features_of det_string_the pobj_of_string xcomp_described_treating dobj_treating_it advmod_differently_quite advmod_treating_differently prep_treating_from amod_items_lexical pobj_from_items ' +o,6250,'Dunning (1993) argues for the use of G 2 rather than X 2, based on the claim that the sampling distribution of G 2 approaches the true chi-square distribution quicker than the sampling distribution of X 2 . However, Agresti (1996, page 34) makes the opposite claim: The sampling distributions of X 2 and G 2 get closer to chi-squared as the sample size n increasesThe convergence is quicker for X 2 than G 2 . In addition, Pedersen (2001) questions whether one statistic should be preferred over the other for the bigram acquisition task and cites Cressie and Read (1984), who argue that there are some cases where the Pearson statistic is more reliable than the log-likelihood statistic.',Dunning,'' +o,6251,'A variety of synset similarity measures based on properties of WordNet itself have been proposed ; nine such measures are discussed in , including gloss-based heuristics , information-content based measures , and others ',Pedersen,'det_variety_A prep_variety_of amod_measures_synset nn_measures_similarity pobj_of_measures partmod_measures_based prep_based_on pobj_on_properties prep_properties_of pobj_of_WordNet nsubjpass_proposed_itself aux_proposed_have auxpass_proposed_been dep_variety_proposed num_measures_nine amod_measures_such nsubjpass_discussed_measures auxpass_discussed_are parataxis_proposed_discussed prep_discussed_in pobj_in_ prep_discussed_including amod_heuristics_gloss-based pobj_including_heuristics amod_heuristics_ dep__information-content partmod_heuristics_based nsubj__measures xcomp_based_ cc_heuristics_and conj_heuristics_others ' +o,6252,'This con rms s nding that in sentence level evaluation , long n-grams in BLEU are not bene cial ',Liu,'det__This amod__con nn__rms nsubj_nding_ aux_nding_s complm_cial_that prep_cial_in nn_evaluation_sentence nn_evaluation_level pobj_in_evaluation amod_n-grams_long nsubj_cial_n-grams prep_n-grams_in pobj_in_BLEU cop_cial_are neg_cial_not advmod_cial_bene ccomp_nding_cial ' +o,6253,'3 describes the cube growing algorithm in further detail , including the precise form of the successor function for derivations ',Huang,'num__3 nsubj_describes_ det_algorithm_the nn_algorithm_cube amod_algorithm_growing dobj_describes_algorithm prep_algorithm_in amod_detail_further pobj_in_detail prep_describes_including det_form_the amod_form_precise pobj_including_form prep_form_of det_function_the nn_function_successor pobj_of_function prep_function_for pobj_for_derivations ' +n,6254,'More recent work has considered methods for speeding up the feature selection methods described in , Ratnaparkhi -LRB- 1998 -RRB- , and Della Pietra , Della Pietra , and Lafferty ',Berger,'amod__More amod__recent nn__work nsubj_considered_ aux_considered_has dobj_considered_methods prep_considered_for pcomp_for_speeding prt_speeding_up det_feature_the dobj_speeding_feature nn_methods_selection nsubj_described_methods dep_considered_described prep_described_in nn_Ratnaparkhi_ pobj_in_Ratnaparkhi appos_Ratnaparkhi_1998 cc_Ratnaparkhi_and nn_Pietra_Della conj_Ratnaparkhi_Pietra nn_Pietra_Della conj_Pietra_Pietra cc_Pietra_and nn__Lafferty conj_Pietra_ ' +o,6255,'CIT -RRB- ',Matsuzaki,'nn_-RRB-_CIT ' +o,6256,'Below is an example of the initial-state tagging of a sentence from the Penn Treebank , where an underscore is to be read as or ',Marcus,'nsubj_example_Below cop_example_is det_example_an prep_example_of det_initial-state_the measure_tagging_initial-state amod_read_tagging prep_tagging_of det_sentence_a pobj_of_sentence prep_sentence_from det__the nn__Penn nn__Treebank pobj_from_ advmod_is_where det_underscore_an nsubj_is_underscore rcmod__is aux_be_to xcomp_is_be pobj_of_read dep_read_as cc_read_or ' +o,6257,'The performance of cross-language information retrieval with a uniform T is likely to be limited in the same way as the performance of conventional information retrieval without term-frequency information , ie , where the system knows which terms occur in which documents , but not how often ',Dunning,'det_performance_The nsubj_likely_performance prep_performance_of amod_retrieval_cross-language nn_retrieval_information pobj_of_retrieval prep_retrieval_with det_T_a amod_T_uniform pobj_with_T cop_likely_is aux_limited_to auxpass_limited_be xcomp_likely_limited prep_limited_in det_way_the amod_way_same pobj_in_way prep_way_as det_performance_the pobj_as_performance prep_performance_of amod_retrieval_conventional nn_retrieval_information pobj_of_retrieval prep_limited_without amod_information_term-frequency pobj_without_information appos_information_ie advmod_knows_where det_system_the nsubj_knows_system rcmod_information_knows dobj_occur_which nsubj_occur_terms ccomp_knows_occur prt_occur_in dep_documents_which dobj_occur_documents cc_documents_but dep_how_not advmod__how advmod_how_often conj_documents_ ' +o,6258,'Although LDD annotation is actually provided in Treebanks such as the Penn Treebank over which they are typically trained , most probabilistic parsers largely or fully ignore this information ',Marcus,'mark_provided_Although nn_annotation_LDD nsubjpass_provided_annotation auxpass_provided_is advmod_provided_actually advcl_ignore_provided prep_provided_in pobj_in_Treebanks dep_as_such prep_Treebanks_as det__the nn__Penn nn__Treebank pobj_as_ prep_provided_over dobj_trained_which nsubj_trained_they cop_trained_are advmod_trained_typically pcomp_over_trained advmod_parsers_most amod_parsers_probabilistic nsubj_ignore_parsers advmod_ignore_largely cc_largely_or conj_largely_fully det_information_this dobj_ignore_information ' +o,6259,'Log-likelihood ratio -LRB- G2 -RRB- with respect to a large reference corpus , Web 1T 5-gram Corpus , is used to capture the contextually relevant nouns ',Dunning,'amod_ratio_Log-likelihood nsubjpass_used_ratio appos_ratio_G2 partmod_ratio_ prep__with pobj_with_respect prep__to det_corpus_a amod_corpus_large nn_corpus_reference pobj_to_corpus nn__Web num__1T num__5-gram nn__Corpus appos_corpus_ auxpass_used_is aux_capture_to xcomp_used_capture det_nouns_the advmod_relevant_contextually amod_nouns_relevant dobj_capture_nouns ' +o,6260,'In , target trees were employed to improve the scoring of translation theories ',Shen,'prep_employed_In pobj_In_ nn_trees_target nsubjpass_employed_trees auxpass_employed_were aux_improve_to xcomp_employed_improve det_scoring_the dobj_improve_scoring prep_scoring_of nn_theories_translation pobj_of_theories ' +o,6261,'As with other randomised models we construct queries with the appropriate sanity checks to lower the error rate efficiently ',Talbot,'prep_construct_As dep_As_with amod_models_other amod_models_randomised pobj_with_models nsubj_construct_we dobj_construct_queries prep_construct_with det_checks_the amod_checks_appropriate amod_checks_sanity pobj_with_checks aux_lower_to xcomp_construct_lower det_rate_the nn_rate_error dobj_lower_rate advmod_lower_efficiently ' +o,6262,'3.1 Part-of-Speech (POS) of Neighboring Words We use 7 features to encode this knowledge source: a0a2a1a4a3a6a5a7a0a8a1a10a9a11a5a7a0a8a1a13a12a14a5a15a0a17a16a6a5a15a0a2a12a18a5a7a0a19a9a20a5a15a0a17a3, where a0a8a1 a21 (a0 a21 ) is the POS of thea6 th token to the left (right) ofa0, and a0a17a16 is the POS of a0 . A token can be a word or a punctuation symbol, and each of these neighboring tokens must be in the same sentence asa0 . We use a sentence segmentation program (Reynar and Ratnaparkhi, 1997) and a POS tagger (Ratnaparkhi, 1996) to segment the tokens surroundinga0 into sentences and assign POS tags to these tokens.',Ratnaparkhi,'' +o,6263,'35 Adding Context to the Model Next , we added of a stochastic POS tagger to provide a model of context ',Marcus,'dep_added_Adding dobj_Adding_Context prep_Adding_to det_Next_the nn_Next_Model pobj_to_Next nsubj_added_we dep_35_added prep_added_of det__a amod__stochastic dep__POS nn__tagger pobj_of_ aux_provide_to xcomp_added_provide det_model_a dobj_provide_model prep_model_of pobj_of_context ' +o,6264,'In examining the combination of the two types of parsing , utilized similar approaches to our empirical analysis ',McDonald,'prep_utilized_In pcomp_In_examining det_combination_the dobj_examining_combination prep_combination_of det_types_the num_types_two pobj_of_types prep_types_of pobj_of_parsing nsubj_utilized_ amod_approaches_similar dobj_utilized_approaches prep_utilized_to poss_analysis_our amod_analysis_empirical pobj_to_analysis ' +o,6265,'We have achieved average results in the CoNLL domain adaptation track open submission ',Marcus,'nsubj_achieved_We aux_achieved_have amod_results_average dobj_achieved_results prep_achieved_in det_submission_the nn_submission_CoNLL nn_submission_domain nn_submission_adaptation nn_submission_track nn_submission_open pobj_in_submission ' +o,6266,' also worked on one of our data sets ',Haghighi,'nsubj_worked_ advmod_worked_also prep_worked_on pobj_on_one prep_one_of poss_sets_our nn_sets_data pobj_of_sets ' +o,6267,'To make the model more practical in parameter estimation , we assume the features in feature set FS are independent from each other , thus : = FSFi AFiPAFSP -RRB- , -LRB- -RRB- , -LRB- -LRB- 5 -RRB- Under this PCFG+PF model , the goal of a parser is to choose a parse that maximizes the following score : -RRB- , -LRB- maxarg -RRB- -LRB- 1 AFS i i i n i T PSTScore = = -LRB- 6 -RRB- Our model is thus a simplification of more sophisticated models which integrate PCFGs with features , such as those in Magerman , and Goodman -LRB- 1997 -RRB- ',Collins,'aux_make_To dep_assume_make det_model_the nsubj_practical_model advmod_practical_more xcomp_make_practical prep_practical_in nn_estimation_parameter pobj_in_estimation nsubj_assume_we det_features_the nsubj_set_features prep_features_in pobj_in_feature ccomp_assume_set nsubj_independent_FS cop_independent_are ccomp_set_independent prep_independent_from det_other_each pobj_from_other dep_AFiPAFSP_thus nn_AFiPAFSP_= nn_AFiPAFSP_FSFi dep_other_AFiPAFSP dep_model_5 dep_5_Under pobj_Under_this amod_model_PCFG+PF nsubj_is_model det_goal_the appos_model_goal prep_goal_of det_parser_a pobj_of_parser dep_other_is aux_choose_to xcomp_is_choose det_parse_a dobj_choose_parse nsubj_maximizes_that rcmod_parse_maximizes det_score_the amod_score_following dobj_maximizes_score appos_other_maxarg num_AFS_1 dep_maxarg_AFS nn_PSTScore_i nn_PSTScore_i nn_PSTScore_i nn_PSTScore_n nn_PSTScore_i nn_PSTScore_T nsubj_=_PSTScore dep_PSTScore_= rcmod_AFS_= dep_model_6 poss_model_Our dep_=_model cop_simplification_is advmod_simplification_thus det_simplification_a rcmod_model_simplification prep_simplification_of advmod_sophisticated_more amod_models_sophisticated pobj_of_models nsubj_integrate_which rcmod_simplification_integrate dobj_integrate_PCFGs prep_PCFGs_with pobj_with_features dep_as_such prep_maxarg_as pobj_as_those prep_those_in nn__Magerman pobj_in_ conj__ cc__and conj__Goodman appos_maxarg_1997 ' +o,6268,'The relatedness between two word senses is computed using a measure of semantic relatedness defined in the WordNet : : Similarity software package , which is a suite of Perl modules implementing a number WordNet-based measures of semantic relatedness ',Pedersen,'det_relatedness_The nsubjpass_computed_relatedness prep_relatedness_between num_senses_two nn_senses_word pobj_between_senses auxpass_computed_is xcomp_computed_using det_measure_a dobj_using_measure prep_measure_of amod_relatedness_semantic pobj_of_relatedness partmod_relatedness_defined prep_defined_in det_WordNet_the pobj_in_WordNet nn__Similarity nn__software nn__package nsubj_modules_ nsubj_suite_which cop_suite_is det_suite_a rcmod__suite prep_suite_of pobj_of_Perl parataxis_computed_modules xcomp_modules_implementing det_measures_a nn_measures_number amod_measures_WordNet-based dobj_implementing_measures prep_measures_of amod_relatedness_semantic pobj_of_relatedness ' +o,6269,'24 Comparison with Hybrid Model SSL based on a hybrid generative\\/discriminative approach proposed in has been defined as a log-linear model that discriminatively combines several discriminative models , pDi , and generative models , pGj , such that : R -LRB- y x ; , , -RRB- = producttext i p Di -LRB- y x ; i -RRB- i producttext j p Gj -LRB- xj , y ; j -RRB- j summationtext y producttext i p Di -LRB- y x ; i -RRB- i producttext j p Gj -LRB- xj , y ; j -RRB- j , where = -LCB- i -RCB- Ii = 1 , and = -LCB- -LCB- i -RCB- Ii = 1 , -LCB- j -RCB- I+J j = I +1 -RCB- ',Suzuki,'num_Comparison_24 nsubjpass_defined_Comparison prep_Comparison_with nn_SSL_Hybrid nn_SSL_Model pobj_with_SSL partmod_SSL_based prep_based_on det_approach_a amod_approach_hybrid nn_approach_generative\\/discriminative pobj_on_approach partmod_approach_proposed prep_proposed_in pobj_in_ aux_defined_has auxpass_defined_been mark_producttext_as det_model_a amod_model_log-linear nsubj_producttext_model nsubj_combines_that advmod_combines_discriminatively rcmod_model_combines amod_models_several nn_models_discriminative dobj_combines_models appos_models_pDi dep_,_and amod_models_generative dep_and_models appos_models_pGj predet_that_such appos_models_that dep_that_R dep_R_y dep_y_x dep_producttext_= advcl_defined_producttext dep_Di_i nsubj_Di_p dep_defined_Di dep_Di_y dep_y_x dep_y_i dep_y_i amod_j_producttext nn_j_j nn_j_p nn_j_Gj appos_j_xj nn_j_y dep_xj_j dep_y_j nn_Di_summationtext nn_Di_y nn_Di_producttext nn_Di_i nn_Di_p dep_j_Di parataxis_Di_y dobj_y_x dep_producttext_i nsubj_producttext_i dep_x_producttext nn_j_j nn_j_p nn_j_Gj appos_j_xj nn_j_y dep_xj_j dobj_producttext_j advmod_=_where advcl_producttext_= nn_Ii_i dobj_=_Ii dep_1_= dep_Ii_1 cc_producttext_and nsubj_I+J_= dep_=_i partmod_i_Ii dep_1_= ccomp_Ii_1 appos_=_j conj_producttext_I+J nsubj_=_j ccomp_I+J_= iobj_=_I dobj_=_+1 ' +o,6270,'We present results in the form of search error analysis and translation quality as measured by the BLEU score on the IWSLT 06 text translation task 1 , comparing Cube Pruning with our two-pass approach ',Papineni,'nsubj_present_We dobj_present_results prep_results_in det_form_the pobj_in_form prep_form_of nn_analysis_search nn_analysis_error pobj_of_analysis cc_analysis_and nn_quality_translation conj_analysis_quality mark__as csubj__measured prep_measured_by det__the amod__BLEU nn__score pobj_by_ prep__on det_task_the nn_task_IWSLT num_task_06 nn_task_text nn_task_translation pobj_on_task dep_analysis_ dobj__1 xcomp__comparing nn_Pruning_Cube dobj_comparing_Pruning prep_Pruning_with poss_approach_our amod_approach_two-pass pobj_with_approach ' +p,6271,'Recently , an elegant approach to inference in discourse interpretation has been developed at a number of sites , all based on tim notion of abduction , and we have begun to explore its potential application to machine translation ',Hobbs,'advmod_developed_Recently det_approach_an amod_approach_elegant nsubjpass_developed_approach aux_inference_to infmod_approach_inference prep_inference_in nn_interpretation_discourse pobj_in_interpretation aux_developed_has auxpass_developed_been prep_developed_at det_number_a pobj_at_number prep_number_of nn__sites pobj_of_ nsubj_based_all ccomp_developed_based prep_based_on nn_notion_tim pobj_on_notion prep_notion_of pobj_of_abduction cc_developed_and nsubj_begun_we aux_begun_have conj_developed_begun aux_explore_to xcomp_begun_explore poss_application_its amod_application_potential dobj_explore_application prep_explore_to nn_translation_machine pobj_to_translation ' +o,6272,'The features are the same as those in ',Ratnaparkhi,'det_features_The nsubj_same_features cop_same_are det_same_the prep_same_as pobj_as_those advmod_same_in ' +o,6273,'In the usual case considered by and discussed by Manning and Sch utze -LRB- 1999 -RRB- , the right-hand side of the equation is larger than the left-hand side ',Dunning,'mark_utze_In det_case_the amod_case_usual nsubj_utze_case dep_case_considered prep_considered_by pobj_by_ cc_considered_and conj_considered_discussed prep_discussed_by pobj_by_Manning cc_Manning_and conj_Manning_Sch advcl_larger_utze dep_utze_1999 det_side_the amod_side_right-hand nsubj_larger_side prep_side_of det_equation_the pobj_of_equation cop_larger_is prep_larger_than det_side_the amod_side_left-hand pobj_than_side ' +o,6274,'We examine Structural Correspondence Learning -LRB- SCL -RRB- for this task , and compare it to several variants of Self-training ',Blitzer,'nsubj_examine_We nn__Structural nn__Correspondence nn__Learning abbrev__SCL dobj_examine_ prep__for det_task_this pobj_for_task cc_examine_and conj_examine_compare dobj_compare_it prep_compare_to amod_variants_several pobj_to_variants prep_variants_of nn__Self-training pobj_of_ ' +o,6275,' and use syntactic markers to increase the significance of the data ',Hindle,'nsubj_use_ cc__and conj__ amod_markers_syntactic nsubj_increase_markers aux_increase_to xcomp_use_increase det_significance_the dobj_increase_significance prep_significance_of det_data_the pobj_of_data ' +o,6276,'Such studies follow the empiricist approach to word meaning summarized best in the famous dictum of the British 3 linguist J.R. Firth: You shall know a word by the company it keeps. (Firth, 1957, p. 11) Context similarity has been used as a means of extracting collocations from corpora, e.g. by Church & Hanks (1990) and by Dunning (1993), of identifying word senses, e.g. by Yarowski (1995) and by Schutze (1998), of clustering verb classes, e.g. by Schulte im Walde (2003), and of inducing selectional restrictions of verbs, e.g. by Resnik (1993), by Abe & Li (1996), by Rooth et al.',Dunning,'' +n,6277,'1 Introduction Currently , most of the phrase-based statistical machine translation -LRB- PBSMT -RRB- models adopt full matching strategy for phrase translation , which means that a phrase pair -LRB- tildewidef , tildewidee -RRB- can be used for translating a source phrase f , only if tildewidef = f Due to lack of generalization ability , the full matching strategy has some limitations ',Koehn,'num_Introduction_1 ccomp_adopt_Introduction advmod_Introduction_Currently nsubj_adopt_most prep_most_of det__the amod__phrase-based dep_phrase-based_statistical nn__machine nn__translation abbrev__PBSMT nn__models pobj_of_ amod_strategy_full amod_strategy_matching dobj_adopt_strategy prep_adopt_for nn_translation_phrase pobj_for_translation nsubj_means_which rcmod_translation_means complm_used_that det_pair_a nn_pair_phrase nsubjpass_used_pair appos_pair_tildewidef dep_tildewidef_tildewidee aux_used_can auxpass_used_be ccomp_means_used prep_used_for pcomp_for_translating det_phrase_a nn_phrase_source dobj_translating_phrase dep_translating_f advmod_=_only dep_=_if nsubj_=_tildewidef dep_adopt_= dep_has_f dep_to_Due prep_has_to pobj_to_lack prep_lack_of nn_ability_generalization pobj_of_ability det_strategy_the amod_strategy_full nn_strategy_matching nsubj_has_strategy dep_adopt_has det_limitations_some dobj_has_limitations ' +o,6278,'2 Related Work Question Answering has attracted much attention from the areas of Natural Language Processing , Information Retrieval and Data Mining ',Fleischman,'num_Answering_2 amod_Answering_Related nn_Answering_Work nn_Answering_Question nsubj_attracted_Answering aux_attracted_has amod_attention_much dobj_attracted_attention prep_attracted_from det_areas_the pobj_from_areas prep_areas_of nn_Processing_Natural nn_Processing_Language pobj_of_Processing nn_Retrieval_Information conj_Processing_Retrieval cc_Processing_and nn_Mining_Data conj_Processing_Mining ' +o,6279,'3 Algorithm As in previous work , our computations are based on a partially lemmatized version of the British National Corpus -LRB- BNC -RRB- which has the function words removed ',Rapp,'num_Algorithm_3 dep_As_Algorithm dep_in_As prep_based_in amod__previous nn__work pobj_in_ poss_computations_our nsubjpass_based_computations auxpass_based_are prep_based_on det_version_a advmod_lemmatized_partially amod_version_lemmatized pobj_on_version prep_version_of det_Corpus_the nn_Corpus_British nn_Corpus_National pobj_of_Corpus abbrev_Corpus_BNC nsubj_has_which rcmod_Corpus_has det_words_the nn_words_function dobj_has_words partmod_words_removed ' +o,6280,'toilet/bathroom Since the word \'facility\' is the subject of \'employ\' and is modified by \'new\' in (3), we retrieve other words that appeared in the same contexts and obtain the following two groups of selectors (the log A column shows the likelihood ratios (Dunning, 1993) of these words in the local contexts): Subjects of \'employ\' with top-20 highest likelihood ratios: word freq, Iog,k word freq ORG\' 64 50.4 plant 14 31.0 company 27 28.6 operation 8 23.0 industry 9 14.6 firm 8 13.5 pirate 2 12.1 unit 9 9.32 shift 3 8.48 postal service 2 7.73 machine 3 6.56 corporation 3 6.47 manufacturer 3 6.21 insurance company 2 6.06 aerospace 2 5.81 memory device 1 5.79 department 3 5.55 foreign office 1 5.41 enterprise 2 5.39 pilot 2 537 *ORG includes all proper names recognized as organizations 18 Modifiees of \'new\' with top-20 highest likelihood ratios: word freq log,k post 432 952.9 issue 805 902.8 product 675 888.6 rule 459 875.8 law 356 541.5 technology 237 382.7 generation 150 323.2 model 207 319.3 job 260 269.2 system 318 251.8 word freq log )~ bonds 223 245.4 capital 178 241.8 order 228 236.5 version 158 223.7 position 236 207.3 high 152 201.2 contract 279 198.1 bill 208 194.9 venture 123 193.7 program 283 183.8 Since the similarity between Sense 1 of \'facility\' and the selectors is greater than that of other senses, the word \'facility\' in (3) is tagged \'Sense The key innovation of our algorithm is that a polysemous word is disambiguated with past usages of other words.',Dunning,'' +o,6281,'In recent years , many researchers have tried to automatically evaluate the quality of MT and improve the performance of automatic MT evaluations because improving the performance of automatic MT evaluation is expected to enable us to use and improve MT systems efficiently ',Banerjee,'prep_tried_In amod_years_recent pobj_In_years amod_researchers_many nsubj_tried_researchers aux_tried_have aux_evaluate_to advmod_evaluate_automatically xcomp_tried_evaluate det_quality_the dobj_evaluate_quality prep_quality_of pobj_of_MT cc_evaluate_and conj_evaluate_improve det_performance_the dobj_improve_performance prep_performance_of amod_evaluations_automatic nn_evaluations_MT pobj_of_evaluations xcomp_improve_ mark_expected_because csubjpass_expected_improving det_performance_the dobj_improving_performance prep_performance_of amod_evaluation_automatic nn_evaluation_MT pobj_of_evaluation auxpass_expected_is advcl__expected aux_enable_to xcomp_expected_enable nsubj_use_us aux_use_to xcomp_enable_use cc_use_and conj_use_improve nn_systems_MT dobj_use_systems advmod_use_efficiently ' +o,6282,'The model we use is similar to that of ',Ratnaparkhi,'det_model_The nsubj_similar_model nsubj_use_we rcmod_model_use cop_similar_is prep_similar_to pobj_to_that prep_that_of ' +o,6283,'We evaluate the system generated summaries using the automatic evaluation toolkit ROUGE ',Lin,'nsubj_evaluate_We det_system_the dobj_evaluate_system partmod_system_generated dobj_generated_summaries xcomp_generated_using det_toolkit_the amod_toolkit_automatic nn_toolkit_evaluation dobj_using_toolkit advmod_using_ROUGE ' +o,6284,' has presented an unsupervised opinion classification algorithm called SO-PMI -LRB- Semantic Orientation Using Pointwise Mutual Information -RRB- ',Turney,'nsubj_presented_ aux_presented_has det_SO-PMI_an amod_SO-PMI_unsupervised nn_SO-PMI_opinion nn_SO-PMI_classification nn_SO-PMI_algorithm amod_SO-PMI_called dobj_presented_SO-PMI nn_Orientation_Semantic dep_SO-PMI_Orientation dep_Orientation_Using nn_Information_Pointwise nn_Information_Mutual dobj_Using_Information ' +n,6285,'1 Introduction For statistical machine translation -LRB- SMT -RRB- , phrasebased methods and syntax-based methods outperform word-based methods ',Brown,'num_Introduction_1 dep_Introduction_For amod_translation_statistical nn_translation_machine pobj_For_translation abbrev_translation_SMT amod__phrasebased nn__methods conj_translation_ cc_translation_and amod_methods_syntax-based conj_translation_methods partmod_methods_ ccomp__outperform amod_methods_word-based dobj_outperform_methods ' +n,6286,'To our knowledge no systems directly address Problem 1 , instead choosing to ignore the problem by using one or a small handful of reference derivations in an n-best list , or else making local independence assumptions which side-step the issue ',Liang,'poss_knowledge_our pobj_To_knowledge det_systems_no nsubj_address_systems advmod_address_directly rcmod_knowledge_address dobj_address_Problem num_Problem_1 advmod_choosing_instead dep_To_choosing aux_ignore_to xcomp_choosing_ignore det_problem_the dobj_ignore_problem prep_ignore_by pcomp_by_using dobj_using_one cc_one_or det_handful_a amod_handful_small conj_one_handful prep_handful_of nn_derivations_reference pobj_of_derivations prep_using_in det__an amod__n-best nn__list pobj_in_ cc_choosing_or advmod_making_else conj_choosing_making amod_assumptions_local nn_assumptions_independence dobj_making_assumptions nsubj_side-step_which rcmod_assumptions_side-step det__the nn__issue dobj_side-step_ ' +o,6287,'Standard sequence prediction models are highly effective for supertagging , including Hidden Markov Models , Maximum Entropy Markov Models , and Conditional Random Fields ',Johnson,'nn_models_Standard nn_models_sequence nn_models_prediction nsubj_effective_models cop_effective_are advmod_effective_highly prep_effective_for pobj_for_supertagging prep_supertagging_including nn__Hidden nn__Markov nn__Models pobj_including_ nn__Maximum nn__Entropy nn__Markov nn__Models conj__ cc__and nn__Conditional nn__Random nn__Fields conj__ ' +o,6288,'11 This low agreement ratio is also re ected in a measure called the statistic ',Carletta,'det_ratio_This amod_ratio_low nn_ratio_agreement nsubj_re_ratio cop_re_is advmod_re_also dep_11_re partmod_re_ected prep_ected_in det_measure_a pobj_in_measure partmod_measure_called det_statistic_the dobj_called_statistic ' +o,6289,'Early work by falls within this framework ',Yarowsky,'advmod_work_Early nsubj_falls_work prep_work_by pobj_by_ prep_falls_within det_framework_this pobj_within_framework ' +o,6290,'In tabh ; 2 , the accuracy rate of the Net-Tagger is cOrolLated to that of a trigram l -RRB- msed tagger and a lIidden Markov Model tagger which were ',Cutting,'prep_msed_In pobj_In_tabh dep_rate_2 det_rate_the nn_rate_accuracy dep_tabh_rate prep_rate_of det_Net-Tagger_the pobj_of_Net-Tagger auxpass_cOrolLated_is rcmod_Net-Tagger_cOrolLated prep_cOrolLated_to pobj_to_that prep_that_of det_trigram_a pobj_of_trigram tmod_cOrolLated_l nn__tagger nsubj_msed_ cc__and det__a amod__lIidden nn__Markov nn__Model nn__tagger conj__ nsubj_were_which rcmod__were ' +o,6291,'In the sequel , we use Collinss statistical parser as our canonical automated approximation of the Treebank ',Collins,'prep_use_In det_sequel_the pobj_In_sequel nsubj_use_we nn__Collinss amod__statistical nn__parser dobj_use_ prep__as poss_approximation_our amod_approximation_canonical amod_approximation_automated pobj_as_approximation prep_approximation_of det_Treebank_the pobj_of_Treebank ' +o,6292,'For details on these feature functions , please refer to ',Koehn,'prep_please_For pobj_For_details prep_details_on det_functions_these nn_functions_feature pobj_on_functions dep_please_refer prep_refer_to ' +o,6293,'In , shallow syntactic analysis such as POS tagging and morphological analysis were incorporated in a phrasal decoder ',Koehn,'prep_incorporated_In pobj_In_ nn_analysis_shallow amod_analysis_syntactic nsubjpass_incorporated_analysis dep_as_such prep_analysis_as dep_tagging_POS pobj_as_tagging cc_tagging_and amod_analysis_morphological conj_tagging_analysis auxpass_incorporated_were prep_incorporated_in det_decoder_a amod_decoder_phrasal pobj_in_decoder ' +o,6294,' , a robust risk minimization classifier , based on a regularized winnow method -LRB- henceforth RRM -RRB- and a maximum entropy classifier -LRB- henceforth MaxEnt -RRB- ',Berger,'det_classifier_a amod_classifier_robust nn_classifier_risk nn_classifier_minimization appos__classifier prep__based dep_based_on det__a amod__regularized nn__winnow nn__method pobj_on_ nn_RRM_henceforth abbrev__RRM cc__and det__a amod__maximum amod__entropy nn__classifier conj__ nn_MaxEnt_henceforth appos__MaxEnt ' +o,6295,'Various machine learning strategies have been proposed to address this problem , including semi-supervised learning , domain adaptation , multi-task learning , self-taught learning , etc A commonality among these methods is that they all require the training data and test data to be in the same feature space ',Blitzer,'amod_strategies_Various nn_strategies_machine nn_strategies_learning nsubjpass_proposed_strategies aux_proposed_have auxpass_proposed_been aux_address_to xcomp_proposed_address det_problem_this dobj_address_problem prep_proposed_including amod__semi-supervised nn__learning pobj_including_ nn__domain nn__adaptation appos__ amod__multi-task nn__learning appos__ amod__self-taught nn__learning appos__ appos__etc det_commonality_A nsubj_is_commonality prep_commonality_among det_methods_these pobj_among_methods dep_proposed_is complm_require_that nsubj_require_they advmod_require_all ccomp_is_require det_data_the nn_data_training dobj_require_data cc_data_and nn_data_test conj_data_data aux_be_to xcomp_require_be prep_be_in det_space_the amod_space_same nn_space_feature pobj_in_space ' +o,6296,'Since we need knowledge-poor Daille , 1996 -RRB- induction , we can not use human-suggested filtering Chi-squared -LRB- G24 -RRB- 2 Z-Score Students t-Score n-gram list in accordance to each probabilistic algorithm ',Church,'mark_need_Since nsubj_need_we advcl_use_need amod_induction_knowledge-poor dep_knowledge-poor_Daille appos_Daille_1996 dobj_need_induction nsubj_use_we aux_use_can neg_use_not amod_Chi-squared_human-suggested amod_Chi-squared_filtering dobj_use_Chi-squared appos_Chi-squared_G24 number__2 num_Students_ amod_Students_Z-Score nn_Students_ nsubj__Students advmod__t-Score rcmod_Chi-squared_ amod_list_n-gram dobj__list prep__in pobj_in_accordance prep__to det_algorithm_each amod_algorithm_probabilistic pobj_to_algorithm ' +o,6297,'That is , phrases are heuristically extracted from word-level alignments produced by doing GIZA + + training on the corresponding parallel corpora ',Koehn,'nsubj_is_That nsubjpass_extracted_phrases auxpass_extracted_are advmod_extracted_heuristically ccomp_is_extracted prep_extracted_from amod_alignments_word-level pobj_from_alignments partmod_alignments_produced prep_produced_by pcomp_by_doing amod_training_GIZA cc_GIZA_+ conj_GIZA_+ dobj_doing_training prep_doing_on det_corpora_the amod_corpora_corresponding amod_corpora_parallel pobj_on_corpora ' +o,6298,'The system used for baseline experiments is two runs of IBM Model 4 in the GIZA + + implementation , which includes smoothing extensions to Model 4 ',Brown,'det_system_The nsubj_runs_system partmod_system_used prep_used_for nn_experiments_baseline pobj_for_experiments cop_runs_is num_runs_two prep_runs_of nn__IBM nn__Model num__4 pobj_of_ prep__in det_implementation_the nn_implementation_GIZA amod_implementation_+ cc_+_+ conj_+_ pobj_in_implementation nsubj_includes_which rcmod_implementation_includes xcomp_includes_smoothing dobj_smoothing_extensions prep_smoothing_to pobj_to_Model num_Model_4 ' +o,6299,'The problem itself has started to get attention only recently ',McClosky,'det_problem_The nsubj_started_itself aux_started_has dep_problem_started aux_get_to xcomp_started_get dobj_get_attention advmod_recently_only advmod_get_recently ' +o,6300,'They are based on the sourcechannel approach to statistical machine translation ',Brown,'nsubjpass_based_They auxpass_based_are prep_based_on det_approach_the amod_approach_sourcechannel pobj_on_approach prep_based_to amod_translation_statistical nn_translation_machine pobj_to_translation ' +o,6301,'1 Introduction Sentiment classification is a special task of text categorization that aims to classify documents according to their opinion of , or sentiment toward a given subject -LRB- eg , if an opinion is supported or not -RRB- ',Pang,'num_classification_1 nn_classification_Introduction nn_classification_Sentiment nsubj_task_classification cop_task_is det_task_a amod_task_special prep_task_of nn_categorization_text pobj_of_categorization nsubj_aims_that rcmod_task_aims aux_classify_to xcomp_aims_classify dobj_classify_documents prep_classify_according dep_according_to poss_opinion_their pobj_to_opinion prep_opinion_of cc_task_or conj_task_sentiment prep_sentiment_toward det_subject_a amod_subject_given pobj_toward_subject dep_subject_eg mark_supported_if det_opinion_an nsubjpass_supported_opinion auxpass_supported_is advcl_eg_supported cc_supported_or conj_supported_not ' +o,6302,'1 Introduction Deep and accurate text analysis based on discriminative models is not yet efficient enough as a component of real-time applications , and it is inadequate to process Web-scale corpora for knowledge acquisition or semi-supervised learning ',McClosky,'dep_efficient_1 nn_Deep_Introduction nsubj_efficient_Deep cc_Deep_and amod_analysis_accurate nn_analysis_text conj_Deep_analysis partmod_analysis_based prep_based_on amod_models_discriminative pobj_on_models cop_efficient_is neg_efficient_not advmod_efficient_yet advmod_efficient_enough prep_efficient_as det_component_a pobj_as_component prep_component_of amod_applications_real-time pobj_of_applications cc_efficient_and nsubj_inadequate_it cop_inadequate_is conj_efficient_inadequate aux_process_to xcomp_inadequate_process amod_corpora_Web-scale dobj_process_corpora prep_process_for nn__knowledge nn__acquisition pobj_for_ cc__or amod_learning_semi-supervised conj__learning ' +p,6303,'The state of the art technology for relation extraction primarily relies on pattern-based approaches ',Snow,'det_state_The nsubj_relies_state prep_state_of det_technology_the nn_technology_art pobj_of_technology prep_technology_for nn_extraction_relation pobj_for_extraction advmod_relies_primarily prep_relies_on amod_approaches_pattern-based pobj_on_approaches ' +o,6304,'3 Length Model : Dynamic Programming Given the word fertility de nitions in IBM Models , we can compute a probability to predict phrase length : given the candidate target phrase -LRB- English -RRB- eI1 , and a source phrase -LRB- French -RRB- of length J , the model gives the estimation of P -LRB- J eI1 -RRB- via a dynamic programming algorithm using the source word fertilities ',Brown,'num_Model_3 nn_Model_Length nn_Programming_Dynamic dep_Model_Programming dep_compute_Given det_fertility_the nn_fertility_word dep_Given_fertility dep_fertility_de dep_fertility_nitions prep_nitions_in nn__IBM nn__Models pobj_in_ nsubj_compute_we aux_compute_can rcmod_Programming_compute det_probability_a dobj_compute_probability aux_predict_to infmod_probability_predict nn_length_phrase dobj_predict_length prep_gives_given det_eI1_the nn_eI1_candidate nn_eI1_target nn_eI1_phrase appos_eI1_English dep_given_eI1 cc_eI1_and det_phrase_a nn_phrase_source conj_eI1_phrase dep_phrase_French prep_phrase_of nn_J_length pobj_of_J det_model_the nsubj_gives_model parataxis_compute_gives det_estimation_the nsubj_fertilities_estimation prep_estimation_of pobj_of_P nn_eI1_J appos_estimation_eI1 prep_estimation_via det_algorithm_a amod_algorithm_dynamic nn_algorithm_programming pobj_via_algorithm partmod_algorithm_using det_word_the nn_word_source dobj_using_word ccomp_gives_fertilities ' +o,6305,'More recently , the problem has been tackled using statistics-based and learning-based methods ',Bean,'advmod_recently_More advmod_tackled_recently det_problem_the nsubjpass_tackled_problem aux_tackled_has auxpass_tackled_been xcomp_tackled_using dobj_using_statistics-based dep_statistics-based_ cc_statistics-based_and amod_methods_learning-based nn_methods_ conj_statistics-based_methods ' +o,6306,'These were : BLEU , NIST , WER -LRB- Word Error Rate -RRB- , PER -LRB- Position-independent WER -RRB- , GTM -LRB- General Text Matcher -RRB- , and METEOR ',Banerjee,'dep_These_were nn__BLEU dep_These_ nn__NIST appos__ appos__WER nn_Rate_Word nn_Rate_Error appos_WER_Rate conj_WER_PER nn_WER_Position-independent abbrev_PER_WER conj_WER_GTM nn_Matcher_General nn_Matcher_Text appos_GTM_Matcher cc_WER_and conj_WER_METEOR ' +o,6307,'Accordingly , in we used a machine learning based coreference resolution system to provide an extrinsic evaluation of the utility of WordNet and Wikipedia relatedness measures for NLP applications ',Ponzetto,'advmod_used_Accordingly prep_used_in pobj_in_ nsubj_used_we det_machine_a dobj_used_machine partmod_machine_learning prep_learning_based amod_system_coreference nn_system_resolution dep_based_system aux_provide_to infmod_system_provide det_evaluation_an amod_evaluation_extrinsic dobj_provide_evaluation prep_evaluation_of det_utility_the pobj_of_utility prep_utility_of pobj_of_WordNet cc_used_and nsubj_relatedness_Wikipedia conj_used_relatedness dobj_relatedness_measures prep_measures_for nn_applications_NLP pobj_for_applications ' +o,6308,'The maximum entropy classier used is Le Zhang \'s Maximum Entropy Modeling Toolkit and the L-BFGS parameter estimation algorithm with gaussian prior smoothing ',Berger,'det_classier_The amod_classier_maximum nn_classier_entropy nsubj_Toolkit_classier advmod_used_ amod_classier_used cop_Toolkit_is nn_Zhang_Le poss_Toolkit_Zhang possessive_Zhang_\'s nn_Toolkit_Maximum nn_Toolkit_Entropy nn_Toolkit_Modeling cc_Toolkit_and det_algorithm_the nn_algorithm_L-BFGS nn_algorithm_parameter nn_algorithm_estimation conj_Toolkit_algorithm prep_algorithm_with amod__gaussian amod__prior nn__smoothing pobj_with_ ' +p,6309,'Many mainstream systems and formalisms would satisfy these criteria , including ones such as the University of Pennsylvania Treebank which are purely syntactic -LRB- though of course , only syntactic properties could then be extracted -RRB- ',Marcus,'amod_systems_Many nn_systems_mainstream nsubj_satisfy_systems cc_systems_and conj_systems_formalisms aux_satisfy_would det_criteria_these dobj_satisfy_criteria prep_criteria_including pobj_including_ones dep_as_such prep_ones_as det_University_the pobj_as_University prep_University_of nn__Pennsylvania nn__Treebank pobj_of_ nsubj_syntactic_which cop_syntactic_are advmod_syntactic_purely rcmod__syntactic mark_extracted_though prep_extracted_of pobj_of_course advmod_properties_only amod_properties_syntactic nsubjpass_extracted_properties aux_extracted_could advmod_extracted_then auxpass_extracted_be dep__extracted ' +n,6310,'For example , the statistical word alignment in IBM translation models can only handle word to word and multi-word to word alignments ',Brown,'prep_handle_For pobj_For_example det_alignment_the amod_alignment_statistical nn_alignment_word nsubj_handle_alignment prep_alignment_in nn_models_IBM nn_models_translation pobj_in_models advmod_handle_ aux_handle_can advmod_handle_only dobj_handle_word prep_handle_to pobj_to_word cc_handle_and conj_handle_multi-word prep_multi-word_to nn_alignments_word pobj_to_alignments ' +o,6311,'One option would be to leverage unannotated text ',Smith,'num_option_One nsubj_be_option aux_be_would prep_be_to nn_text_leverage nn_text_unannotated pobj_to_text ' +o,6312,'Model Bits \\/ Character ASCII Huffman code each char Lempel-Ziv -LRB- Unix TM compress -RRB- Unigram -LRB- Huffman code each word -RRB- Trigram Human Performance 8 5 443 21 -LRB- Brown , personal communication -RRB- 176 125 The cross entropy , H , of a code and a source is given by : H -LRB- source , code -RRB- = ~ ~ Pr -LRB- s , h I source -RRB- log 2 Pr -LRB- s I h , code -RRB- s h where Pr -LRB- s , h I source -RRB- is the joint probability of a symbol s following a history h given the source ',Brown,'nn_Huffman_Model nn_Huffman_Bits nn_Huffman_\\/ nn_Huffman_Character nn_Huffman_ASCII nsubj_code_Huffman det_char_each iobj_code_char nn_Unigram_Lempel-Ziv num_compress_Unix nn_compress_TM appos_Unigram_compress dobj_code_Unigram amod_code_Huffman dep_Unigram_code det_word_each dep_code_word nn_Performance_Trigram nn_Performance_Human dobj_code_Performance amod_Performance_8 dep_8_5 number_21_443 dobj_code_21 appos_21_Brown amod_communication_personal dep_Brown_communication num__176 dobj_code_ number__125 num__ det_entropy_The amod_entropy_cross dobj_code_entropy appos_entropy_H prep_code_of det_code_a pobj_of_code cc_code_and det_source_a nsubjpass_given_source auxpass_given_is conj_code_given prep_given_by nsubj_=_H appos_H_source dep_source_code parataxis_code_= nn_Pr_~ nn_Pr_~ nsubj_log_Pr nn_source_s dep_source_h nn_source_I appos_Pr_source ccomp_=_log num_Pr_2 nsubj_h_Pr num_h_s nn_h_I appos_Pr_h dep_h_code dep_h_s ccomp_log_h advmod_probability_where nsubj_probability_Pr nn_source_s dep_source_h nn_source_I appos_Pr_source cop_probability_is det_probability_the amod_probability_joint rcmod_h_probability prep_probability_of det_s_a nn_s_symbol pobj_of_s prep_s_following det_h_a nn_h_history pobj_following_h partmod_h_given det_source_the dobj_given_source ' +o,6313,'These later inductive phases may rely on some level of a priori knowledge , like for example the naive case relations used in the ARIOSTO_LEX system ',Marcus,'det_inductive_These amod_inductive_later nsubj_rely_phases aux_rely_may rcmod_inductive_rely prep_rely_on det_level_some pobj_on_level prep_level_of det_knowledge_a amod_knowledge_priori pobj_of_knowledge dep_for_like prep_level_for pobj_for_example dep_naive_the dep_relations_naive nn_relations_case dep_inductive_relations partmod_relations_used prep_used_in det__the nn__ARIOSTO_LEX nn__system pobj_in_ ' +o,6314,' says that 067 a10a14a11a15a10 08 allows just tentative conclusions to be drawn ',Carletta,'advmod_says_ dep_allows_that num_08_067 amod_08_a10a14a11a15a10 nsubj_allows_08 dep_says_allows advmod_conclusions_just amod_conclusions_tentative dobj_allows_conclusions aux_drawn_to auxpass_drawn_be xcomp_allows_drawn ' +o,6315,'Rmnshaw and Marcus introdu -LRB- : e -LRB- l a 1 -RRB- aseNl \' whi -LRB- : h is a non-re -LRB- : ursive NIL They used trmlsfornmtion-1 -RRB- ase -LRB- l learning to i -LRB- lentif ~ y n -LRB- \\/ nrecto ` sire l -RRB- aseNPs in a s -LRB- mtence ',Ramshaw,'nsubj_mtence_Rmnshaw cc_Rmnshaw_and conj_Rmnshaw_Marcus advmod_introdu_ partmod_Rmnshaw_introdu dobj_introdu_e appos_aseNl_l quantmod_1_a num_l_1 prep_e_aseNl dep_aseNl_whi dep_whi_h prep_h_is det_non-re_a dep_is_non-re dep_-LRB-_ursive dep_ursive_NIL dep_NIL_They amod_They_used dep_used_trmlsfornmtion-1 nn_l_ase dep_h_l partmod_l_learning prep_learning_to nn_l_i nn_l_lentif nn_l_~ nn_l_y nn_l_n nn_l_\\/ nn_l_nrecto nn_l_sire pobj_to_l dep_whi_aseNPs prep_e_in dep_in_a aux_mtence_s ' +o,6316,'We are encoding the knowledge as axioms in what is for the most part first-order logic , described in , although quantification over predicates is sometimes convenient ',Hobbs,'nsubj_encoding_We aux_encoding_are det_knowledge_the dobj_encoding_knowledge prep_encoding_as pobj_as_axioms prep_axioms_in nsubj_is_what pcomp_in_is prep_is_for det_logic_the amod_logic_most nn_logic_part nn_logic_first-order pobj_for_logic partmod_logic_described prep_described_in pobj_in_ mark_convenient_although nsubj_convenient_quantification prep_quantification_over pobj_over_predicates cop_convenient_is advmod_convenient_sometimes advcl_is_convenient ' +o,6317,'3 and Kleins Coreference Model To gauge the performance of our model , we compare it with a Bayesian model for unsupervised coreference resolution that was recently proposed by and Klein ',Haghighi,'num__3 cc__and nn_Model_Kleins nn_Model_Coreference conj__Model aux_gauge_To dep_compare_gauge det_performance_the dobj_gauge_performance prep_performance_of poss_model_our pobj_of_model nsubj_compare_we dep__compare dobj_compare_it prep_compare_with det_model_a amod_model_Bayesian pobj_with_model prep_model_for amod_resolution_unsupervised nn_resolution_coreference pobj_for_resolution nsubjpass_proposed_that auxpass_proposed_was advmod_proposed_recently rcmod_resolution_proposed prep_proposed_by nn__ cc__and conj__Klein pobj_by_ ' +o,6318,'The association relationship between two words can be indicated by their mutual information , which can be further used to discover phrases -LRB- -RRB- ',Church,'det_relationship_The nn_relationship_association nsubjpass_indicated_relationship prep_relationship_between num_words_two pobj_between_words aux_indicated_can auxpass_indicated_be prep_indicated_by poss_information_their amod_information_mutual pobj_by_information nsubjpass_used_which aux_used_can auxpass_used_be advmod_used_further rcmod_information_used aux_discover_to purpcl_used_discover dobj_discover_phrases appos_phrases_ ' +o,6319,' used a Bloomier filter to encode a LM ',Talbot,'nsubj_used_ det_filter_a nn_filter_Bloomier dobj_used_filter aux_encode_to xcomp_used_encode det_LM_a dobj_encode_LM ' +o,6320,' from the Penn Treebank WSJ corpus ',Marcus,'prep__from det_corpus_the nn_corpus_Penn nn_corpus_Treebank nn_corpus_ nn_corpus_WSJ pobj_from_corpus ' +o,6321,'Several non-linear objective functions , such as F-score for text classification , and BLEU-score and some other evaluation measures for statistical machine translation , have been introduced with reference to the framework of MCE criterion training ',Och,'amod_functions_Several amod_functions_non-linear amod_functions_objective nsubjpass_introduced_functions dep_as_such prep_functions_as pobj_as_F-score prep_F-score_for nn__text nn__classification pobj_for_ cc_F-score_and conj_F-score_BLEU-score cc_F-score_and det_measures_some amod_measures_other nn_measures_evaluation conj_F-score_measures prep_measures_for amod__statistical nn__machine nn__translation pobj_for_ aux_introduced_have auxpass_introduced_been prep_introduced_with pobj_with_reference prep_introduced_to det_framework_the pobj_to_framework prep_framework_of nn_training_MCE nn_training_criterion pobj_of_training ' +o,6322,'Firstly , there is also H -LRB- RB -RRB- A -LRB- ADVP -RRB- declined H -LRB- VBD -RRB- H -LRB- VP -RRB- the dollar A -LRB- DT -RRB- H -LRB- NN -RRB- C -LRB- NP-SBJ -RRB- H -LRB- VP -RRB- H -LRB- S -RRB- Figure 2 : A tree with constituents marked the top-down method , which is a version of the algorithm described by Hockenmaier et al , but used for translating into simple -LRB- AB -RRB- CG rather than the Steedmans Combinatory Categorial Grammar -LRB- CCG -RRB- ',Marcus,'advmod_is_Firstly expl_is_there advmod_is_also nsubj_is_H appos_H_RB det_ADVP_A nsubj_declined_ADVP rcmod_H_declined nn_H_H abbrev_H_VBD dobj_declined_H abbrev_H_VP det_A_the nn_A_dollar dep_H_A abbrev_A_DT nn_C_H abbrev_C_NN dep_H_C abbrev_C_NP-SBJ nn_H_H abbrev_H_VP prep_C_H appos_H_S dep_H_Figure num_Figure_2 det_tree_A dep_Figure_tree prep_tree_with pobj_with_constituents partmod_constituents_marked det_method_the amod_method_top-down dobj_marked_method nsubj_version_which cop_version_is det_version_a rcmod_H_version prep_version_of det_algorithm_the pobj_of_algorithm dep_algorithm_described prep_described_by nn__Hockenmaier cc_Hockenmaier_et conj_Hockenmaier_al pobj_by_ cc_described_but conj_described_used prep_used_for pcomp_for_translating prep_translating_into amod_CG_simple abbrev_CG_AB nn__CG cc_CG_rather dep_rather_than det_Grammar_the nn_Grammar_Steedmans nn_Grammar_Combinatory nn_Grammar_Categorial conj_CG_Grammar abbrev_Grammar_CCG pobj_into_ ' +o,6323,'(Turney, 2002; Pang et al., 2002; Dave at al., 2003).',Pang,'' +o,6324,'Ramshaw and Marcus first represented base noun phrase recognition as a machine learning problem ',Ramshaw,'nn__Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus nsubj_represented_ advmod_represented_first nn_recognition_base nn_recognition_noun nn_recognition_phrase dobj_represented_recognition prep_represented_as det_machine_a pobj_as_machine partmod_machine_learning dobj_learning_problem ' +o,6325,'Approaches include word substitution systems , phrase substitution systems , and synchronous context-free grammar systems , all of which train on string pairs and seek to establish connections between source and target strings ',Brown,'nsubj_include_Approaches nn_systems_word nn_systems_substitution nsubj__systems xcomp_include_ nn_systems_phrase nn_systems_substitution nsubj__systems ccomp_include_ cc_include_and amod_systems_synchronous amod_systems_context-free nn_systems_grammar nsubj__systems conj_include_ dep_of_all nsubj_train_of pobj_of_which ccomp__train prep_train_on nn_pairs_string pobj_on_pairs cc_train_and conj_train_seek aux_establish_to xcomp_seek_establish dobj_establish_connections prep_connections_between nn_strings_source cc_source_and conj_source_target pobj_between_strings ' +o,6326,'A variety of unsupervised WSD methods , which use a machinereadable dictionary or thesaurus in addition to a corpus , have also been proposed ',Yarowsky,'det_variety_A nsubjpass_proposed_variety prep_variety_of amod_methods_unsupervised nn_methods_WSD pobj_of_methods nsubj_use_which rcmod_variety_use det_dictionary_a amod_dictionary_machinereadable dobj_use_dictionary cc_dictionary_or conj_dictionary_thesaurus prep_dictionary_in dep_in_addition dep_in_to det_corpus_a pobj_in_corpus aux_proposed_have advmod_proposed_also auxpass_proposed_been ' +p,6327,'The Penn Treebank has until recently been the only such corpus , covering 45M words in a single genre of financial reporting ',Marcus,'det__The nn__Penn nn__Treebank nsubj_been_ aux_been_has prep_been_until dep_until_recently det_corpus_the advmod_corpus_only amod_corpus_such dobj_been_corpus xcomp_been_covering num_words_45M dobj_covering_words prep_covering_in det_genre_a amod_genre_single pobj_in_genre prep_genre_of amod_reporting_financial pobj_of_reporting ' +o,6328,'In his analysis of Yarowsky , formulates several variants of bootstrapping ',Abney,'prep_formulates_In poss_analysis_his pobj_In_analysis prep_analysis_of nn__Yarowsky pobj_of_ nsubj_formulates_ amod_variants_several dobj_formulates_variants prep_variants_of pobj_of_bootstrapping ' +o,6329,'To make this paper comparable to , we use English-French notation in this section ',Brown,'aux_make_To dep_use_make det_paper_this nsubj_comparable_paper xcomp_make_comparable prep_comparable_to pobj_to_ nsubj_use_we amod_notation_English-French dobj_use_notation prep_notation_in det_section_this pobj_in_section ' +n,6330,'Our graphical representation has two advantages over previous work : unifying sentence relations and incorporating question interactions ',Ding,'poss_representation_Our amod_representation_graphical dep_representation_has num_advantages_two dobj_has_advantages prep_advantages_over amod__previous nn__work pobj_over_ amod_relations_unifying nn_relations_sentence dep_representation_relations cc_relations_and amod_interactions_incorporating nn_interactions_question conj_relations_interactions ' +o,6331,' reports that we should not rely on the assumption of a normal distribution when performing statistical text analysis and suggests that parametric analysis based on the binomial or multinomial distributions is a better alternative for smaller texts ',Dunning,'advmod_reports_ dep_rely_that nsubj_rely_we aux_rely_should neg_rely_not dep_reports_rely prep_rely_on det_assumption_the pobj_on_assumption prep_assumption_of det_distribution_a amod_distribution_normal pobj_of_distribution advmod_performing_when advcl_rely_performing amod_analysis_statistical nn_analysis_text dobj_performing_analysis cc_performing_and conj_performing_suggests complm_alternative_that amod_analysis_parametric nsubj_alternative_analysis prep_analysis_based dep_based_on det_distributions_the amod_distributions_binomial cc_binomial_or conj_binomial_multinomial pobj_on_distributions cop_alternative_is det_alternative_a amod_alternative_better ccomp_suggests_alternative prep_alternative_for amod_texts_smaller pobj_for_texts ' +p,6332,'Some methods which can offer powerful reordering policies have been proposed like syntax based machine translation and Inversion Transduction Grammar ',Wu,'det_methods_Some nsubjpass_proposed_methods nsubj_offer_which aux_offer_can rcmod_methods_offer amod_policies_powerful nn_policies_reordering dobj_offer_policies aux_proposed_have auxpass_proposed_been prep_proposed_like pobj_like_syntax partmod_syntax_based nn__machine nn__translation dobj_based_ cc__and nn_Grammar_Inversion nn_Grammar_Transduction conj__Grammar ' +o,6333,' used transformation-based learning , an error-driven learning technique introduced by Eric Bn11 , to locate chunks in the tagged corpus ',Ramshaw,'advmod_used_ amod_learning_transformation-based nsubj_used_learning det_technique_an amod_technique_error-driven nn_technique_learning appos_learning_technique partmod_technique_introduced prep_introduced_by nn__Eric nn__Bn11 pobj_by_ aux_locate_to xcomp_introduced_locate dobj_locate_chunks prep_locate_in det_corpus_the amod_corpus_tagged pobj_in_corpus ' +o,6334,'7 For a more detailed discussion , see Berger , Della Pietra , and Della Pietra and ',Ratnaparkhi,'dep_For_7 prep_see_For det_discussion_a advmod_detailed_more amod_discussion_detailed pobj_For_discussion dobj_see_Berger nn_Pietra_Della conj_Berger_Pietra cc_Berger_and nn__Della nn__Pietra conj_Berger_ cc__and nn__ conj__ ' +o,6335,'146 23 Approximating ISBNs proposes two approximations for inference in ISBNs , both based on variational methods ',Titov,'number_23_146 num__23 nn__Approximating nn__ISBNs nsubj_proposes_ num_approximations_two dobj_proposes_approximations prep_approximations_for pobj_for_inference prep_proposes_in pobj_in_ISBNs dep_proposes_both conj_proposes_based prep_based_on amod_methods_variational pobj_on_methods ' +o,6336,'Because of its central role in building machine translation systems and because of the complexity of the task , sub-sentential alignment of parallel corpora continues to be an active area of research , and this implies a continuing demand for manually created or human-verified gold standard alignments for development and evaluation purposes ',Fraser,'dep_Because_of poss_role_its amod_role_central pobj_of_role prep_role_in pcomp_in_building nn_systems_machine nn_systems_translation dobj_building_systems cc_Because_and mark_continues_because prep_continues_of det_complexity_the pobj_of_complexity prep_complexity_of det_task_the pobj_of_task amod_alignment_sub-sentential nsubj_continues_alignment prep_alignment_of amod_corpora_parallel pobj_of_corpora conj_Because_continues aux_area_to cop_area_be det_area_an amod_area_active xcomp_continues_area prep_area_of nn__research pobj_of_ cc_continues_and nsubj_implies_this conj_continues_implies det_demand_a amod_demand_continuing dobj_implies_demand prep_demand_for advmod_created_manually amod_alignments_created cc_created_or conj_created_human-verified nn_alignments_gold nn_alignments_standard pobj_for_alignments prep_alignments_for nn_purposes_development cc_development_and conj_development_evaluation pobj_for_purposes ' +o,6337,'In future work we plan to experiment with richer representations , eg including long-range n-grams , class n-grams , grammatical features , etc \' ',Brown,'prep__In amod_work_future pobj_In_work nsubj_plan_we parataxis__plan aux_experiment_to xcomp_plan_experiment prep_experiment_with amod_representations_richer pobj_with_representations appos_representations_eg prep_eg_including amod__long-range amod__n-grams pobj_including_ nn_n-grams_class nsubj__n-grams amod_features_grammatical nsubj__features parataxis__ dobj__etc ' +p,6338,'Recent work emphasizes corpus-based unsupervised approach that avoids the need for costly truthed training data ',Yarowsky,'amod_work_Recent nsubj_emphasizes_work amod__corpus-based amod__unsupervised nn__approach dobj_emphasizes_ nsubj_avoids_that rcmod__avoids det_need_the dobj_avoids_need prep_need_for amod_data_costly amod_data_truthed nn_data_training pobj_for_data ' +o,6339,'The approach is in the spirit of on retrieving collocations from text corpora , but is more integrated with parsing ',Smadja,'det_approach_The nsubj_is_approach prep_is_in det_spirit_the pobj_in_spirit prep_spirit_of pobj_of_ prep_is_on pcomp_on_retrieving dobj_retrieving_collocations prep_retrieving_from nn_corpora_text pobj_from_corpora cc_is_but cop_integrated_is advmod_integrated_more conj_is_integrated prep_integrated_with pcomp_with_parsing ' +o,6340,'Experimentation The corpus used in shallow parsing is extracted from the PENN TreeBank of 1 million words -LRB- 25 sections -RRB- by a program provided by Sabine Buchholz from Tilburg University ',Marcus,'det_corpus_The nsubjpass_extracted_corpus partmod_corpus_used prep_used_in nn_parsing_shallow pobj_in_parsing auxpass_extracted_is dep_Experimentation_extracted prep_extracted_from det__the nn__PENN nn__TreeBank pobj_from_ prep__of number_million_1 num_words_million pobj_of_words num_sections_25 appos_words_sections prep_extracted_by det_program_a pobj_by_program partmod_program_provided prep_provided_by nn_Buchholz_Sabine pobj_by_Buchholz prep_Buchholz_from nn_University_Tilburg pobj_from_University ' +o,6341,'We collected training samples from the Brown Corpus distributed with the Penn Treebank ',Marcus,'nsubj_collected_We nn_samples_training dobj_collected_samples prep_collected_from det_Corpus_the nn_Corpus_Brown pobj_from_Corpus partmod_Corpus_distributed prep_distributed_with det_Treebank_the nn_Treebank_Penn pobj_with_Treebank ' +o,6342,'Recent research in open information extraction has shown that we can extract large amounts of relational data from open-domain text with high accuracy ',Banko,'amod_research_Recent nsubj_shown_research prep_research_in amod__open nn__information nn__extraction pobj_in_ aux_shown_has complm_extract_that nsubj_extract_we aux_extract_can ccomp_shown_extract amod_amounts_large dobj_extract_amounts prep_amounts_of amod_data_relational pobj_of_data prep_extract_from amod_text_open-domain pobj_from_text prep_extract_with amod_accuracy_high pobj_with_accuracy ' +o,6343,'Rather than learning how strings in one language map to strings in another, however, translation now involves learning how systematic patterns of errors in ESL learners English map to corresponding patterns in native English 2.2 A Noisy Channel Model of ESL Errors If ESL error correction is seen as a translation task, the task can be treated as an SMT problem using the noisy channel model of (Brown et al. , 1993): here the L2 sentence produced by the learner can be regarded as having been corrupted by noise in the form of interference from his or her L1 model and incomplete language models internalized during language learning.',Brown,'' +o,6344,'Results on the provided 2000sentence development set are reported using the BLEU metric ',Papineni,'nsubjpass_reported_Results prep_Results_on det_set_the amod_set_provided amod_set_2000sentence nn_set_development pobj_on_set auxpass_reported_are xcomp_reported_using det_metric_the amod_metric_BLEU dobj_using_metric ' +o,6345,'3 Evaluation of Algorithms All four algorithms were run on a 3900 utterance subset of the Penn Treebank annotated corpus provided by Charniak and Ge -LRB- 1998 -RRB- ',Marcus,'num_Evaluation_3 prep_Evaluation_of pobj_of_Algorithms det_algorithms_All num_algorithms_four nsubjpass_run_algorithms auxpass_run_were rcmod_Evaluation_run prep_run_on det_subset_a num_subset_3900 nn_subset_utterance pobj_on_subset prep_Evaluation_of det__the nn__Penn nn__Treebank amod__annotated nn__corpus pobj_of_ partmod__provided prep_provided_by pobj_by_Charniak cc_Charniak_and conj_Charniak_Ge dep_provided_1998 ' +o,6346,'Like , we give our model information about the basic types of pronouns in English ',Haghighi,'prep_give_Like pobj_Like_ nsubj_give_we poss_information_our nn_information_model dep_give_information advmod_types_about det_types_the amod_types_basic dobj_give_types prep_types_of pobj_of_pronouns prep_pronouns_in pobj_in_English ' +o,6347,'Averaging has been shown to help reduce overfitting ',Collins,'nsubjpass_shown_Averaging aux_shown_has auxpass_shown_been aux_help_to xcomp_shown_help xcomp_help_reduce dobj_reduce_overfitting ' +p,6348,'We examine the effectiveness of Structural Correspondence Learning -LRB- SCL -RRB- for this task , a recently proposed adaptation technique shown to be effective for PoS tagging and Sentiment Analysis ',Blitzer,'nsubj_examine_We det_effectiveness_the dobj_examine_effectiveness prep_effectiveness_of pcomp_of_Structural nn__Correspondence nn__Learning abbrev__SCL dobj_Structural_ prep__for det_task_this pobj_for_task det_technique_a advmod_technique_recently amod_technique_proposed nn_technique_adaptation appos_task_technique partmod_technique_shown aux_effective_to cop_effective_be xcomp_shown_effective prep_effective_for amod_tagging_PoS pobj_for_tagging cc_tagging_and nn_Analysis_Sentiment conj_tagging_Analysis ' +o,6349,'Though several algorithms have been proposed 100 -LRB- 9o -LRB- 80 -LRB- 4O -LRB- 20 -LRB- 1000 goo 80 ~ 41111 2 @ 5 10 15 20 25 30 5 10 15 20 25 30 iteration of EM iteration of EM -LRB- a -RRB- -LRB- b -RRB- Figure 1 : Plots of -LRB- a -RRB- training and -LRB- b -RRB- test perplexity versus number of iterations of the EM algorithm , for the aggregate Markov model with C = 32 classes ',Brown,'mark_proposed_Though amod_algorithms_several nsubjpass_proposed_algorithms advmod_proposed_ aux_proposed_have auxpass_proposed_been dobj_proposed_100 dep_100_9o dep_9o_80 dep_80_4O dep_4O_20 num_iteration_1000 number_80_goo num_iteration_80 amod_iteration_~ number_30_41111 punct_30_2 number_30_@ number_30_5 number_30_10 number_30_15 number_30_20 number_30_25 number_30_30 number_30_5 number_30_10 number_30_15 number_30_20 number_30_25 num_iteration_30 dep_20_iteration prep_iteration_of nn_iteration_EM pobj_of_iteration prep_iteration_of pobj_of_EM dep_-LRB-_a dep_-LRB-_b dep_iteration_Figure num_Figure_1 dep_iteration_Plots prep_Plots_of dep_-LRB-_a pobj_of_training cc_100_and dep_-RRB-_b nn_perplexity_test conj_100_perplexity prep_perplexity_versus pobj_versus_number prep_perplexity_of pobj_of_iterations prep_iterations_of det_algorithm_the nn_algorithm_EM pobj_of_algorithm mark_classes_for det_model_the amod_model_aggregate nn_model_Markov nsubj_classes_model prep_model_with pobj_with_C dep_classes_= num_classes_32 advcl_proposed_classes ' +o,6350,'These weights or scaling factors can be optimized with respect to some evaluation criterion ',Och,'det_weights_These nsubjpass_optimized_weights cc_weights_or nn_factors_scaling conj_weights_factors aux_optimized_can auxpass_optimized_be prep_optimized_with pobj_with_respect prep_optimized_to det_criterion_some nn_criterion_evaluation pobj_to_criterion ' +p,6351,'Online votedperceptrons have been reported to work well in a number of NLP tasks ',Liang,'amod_votedperceptrons_Online nsubjpass_reported_votedperceptrons aux_reported_have auxpass_reported_been aux_work_to xcomp_reported_work advmod_work_well prep_work_in det_number_a pobj_in_number prep_number_of nn_tasks_NLP pobj_of_tasks ' +o,6352,' , various classification models and linguistic features have been proposed to improve the classification performance ',Pang,'amod_models_ amod_models_various nn_models_classification nsubjpass_proposed_models cc_models_and amod_features_linguistic conj_models_features aux_proposed_have auxpass_proposed_been aux_improve_to xcomp_proposed_improve det_performance_the nn_performance_classification dobj_improve_performance ' +o,6353,'Several models were introduced for these problems , for example , the Hidden Markov Model -LRB- HMM -RRB- , Maximum Entropy Model -LRB- ME -RRB- , and Conditional Random Fields -LRB- CRFs -RRB- ',Ratnaparkhi,'amod_models_Several nsubjpass_introduced_models auxpass_introduced_were prep_introduced_for det_problems_these pobj_for_problems prep_introduced_for pobj_for_example det__the amod__Hidden nn__Markov nn__Model abbrev__HMM appos_example_ nn__Maximum nn__Entropy nn__Model abbrev__ME conj__ cc__and nn__Conditional nn__Random nn__Fields abbrev__CRFs conj__ ' +o,6354,'Our framework makes use of the log-frequency Bloom filter presented in , and described briefly below , to compute smoothed conditional n-gram probabilities on the fly ',Talbot,'poss_framework_Our nsubj_makes_framework dobj_makes_use prep_use_of det_filter_the amod_filter_log-frequency nn_filter_Bloom pobj_of_filter dep_filter_presented prep_presented_in pobj_in_ cc_presented_and conj_presented_described dep_below_briefly advmod_described_below aux_compute_to xcomp_described_compute amod_probabilities_smoothed amod_probabilities_conditional amod_probabilities_n-gram dobj_compute_probabilities prep_probabilities_on det_fly_the pobj_on_fly ' +o,6355,'5 Related Work As discussed in footnote 3 , and McDonald et al ',Collins,'num_Work_5 amod_Work_Related dep_As_Work advmod_discussed_As prep_discussed_in pobj_in_footnote nsubj_discussed_3 appos_3_ cc__and conj__McDonald cc_McDonald_et conj_McDonald_al ' +o,6356,'214 Model Features Our MST models are based on the features described in ; specifically , we use features based on a dependency nodes form , lemma , coarse and fine part-of-speech tag , and morphologicalstring attributes ',Hall,'num_models_214 nn_models_Model nn_models_Features nn_models_Our nn_models_MST nsubjpass_based_models auxpass_based_are prep_based_on det_features_the pobj_on_features partmod_features_described prep_described_in pobj_in_ advmod_use_specifically nsubj_use_we parataxis_based_use dobj_use_features prep_use_based dep_based_on det_form_a amod_form_dependency nn_form_nodes pobj_on_form conj_form_lemma conj_form_coarse cc_form_and amod_tag_fine amod_tag_part-of-speech conj_form_tag cc_based_and nsubj_attributes_morphologicalstring conj_based_attributes ' +o,6357,'We compare an ordinary PCFG estimated with maximum likelihood and the HDP-PCFG estimated using the variational inference algorithm described in Section 26 ',Matsuzaki,'nsubj_compare_We det_PCFG_an amod_PCFG_ordinary dobj_compare_PCFG partmod_PCFG_estimated prep_estimated_with amod__maximum nn__likelihood pobj_with_ cc_PCFG_and det_HDP-PCFG_the conj_PCFG_HDP-PCFG partmod_HDP-PCFG_estimated xcomp_estimated_using det_algorithm_the amod_algorithm_variational nn_algorithm_inference dobj_using_algorithm partmod_algorithm_described prep_described_in pobj_in_Section num_Section_26 ' +o,6358,'See for additional work using perceptron algorithms to train tagging models , and a more thorough description of the theory underlying the perceptron algorithm applied to ranking problems ',Collins,'dobj_See_ prep__for amod_work_additional pobj_for_work partmod_work_using nn_algorithms_perceptron dobj_using_algorithms aux_train_to xcomp_using_train amod_models_tagging dobj_train_models cc__and det_description_a amod_description_more amod_description_thorough conj__description prep_description_of det_theory_the pobj_of_theory partmod_theory_underlying det_algorithm_the nn_algorithm_perceptron dobj_underlying_algorithm partmod_algorithm_applied prep_applied_to amod_problems_ranking pobj_to_problems ' +p,6359,'1 Introduction Large scale annotated corpora , eg , the Penn TreeBank -LRB- PTB -RRB- project , have played an important role in text-mining ',Marcus,'num_corpora_1 amod_corpora_Introduction amod_corpora_Large nn_corpora_scale nn_corpora_annotated nsubj_played_corpora appos_corpora_eg det__the nn_TreeBank_Penn nn__TreeBank abbrev_TreeBank_PTB nn__project appos_eg_ aux_played_have det_role_an amod_role_important dobj_played_role prep_role_in pobj_in_text-mining ' +o,6360,'2 The ME Tagger The ME tagger is based on s POS tagger and is described in Curran and Clark -LRB- 2003 -RRB- ',Ratnaparkhi,'nsubj_Tagger_2 det_Tagger_The nn_Tagger_ME det_tagger_The nn_tagger_ME nsubjpass_based_tagger auxpass_based_is rcmod_Tagger_based prep_based_on amod_tagger_ nn_tagger_s dep_tagger_POS pobj_on_tagger cc_based_and auxpass_described_is conj_based_described prep_described_in pobj_in_Curran cc_Curran_and conj_Curran_Clark dep_Tagger_2003 ' +o,6361,'This could , for example , aid machine-translation evaluation , where it has become common to evaluate systems by comparing their output against a bank of several reference translations for the same sentences ',Papineni,'nsubj_aid_This aux_aid_could prep_aid_for pobj_for_example amod_evaluation_machine-translation dobj_aid_evaluation advmod_common_where nsubj_common_it aux_common_has cop_common_become rcmod_evaluation_common aux_evaluate_to xcomp_common_evaluate dobj_evaluate_systems prep_evaluate_by pcomp_by_comparing poss_output_their dobj_comparing_output prep_comparing_against det_bank_a pobj_against_bank prep_bank_of amod_translations_several nn_translations_reference pobj_of_translations prep_translations_for det_sentences_the amod_sentences_same pobj_for_sentences ' +o,6362,'methods for syntactic SMT held to this assumption in its entirety ',Wu,'dep_held_methods prep_methods_for amod_SMT_syntactic pobj_for_SMT prep_held_to det_assumption_this pobj_to_assumption prep_assumption_in poss_entirety_its pobj_in_entirety ' +o,6363,'This corpus contains annotations of semantic PASs superimposed on the Penn Treebank -LRB- PTB -RRB- ',Marcus,'det_corpus_This nsubj_contains_corpus dobj_contains_annotations prep_annotations_of amod_PASs_semantic pobj_of_PASs partmod_PASs_superimposed prep_superimposed_on det_Treebank_the nn_Treebank_Penn pobj_on_Treebank abbrev_Treebank_PTB ' +p,6364,'The fluency models hold promise for actual improvements in machine translation output quality ',Zwarts,'det_models_The amod_models_fluency nsubj_hold_models dobj_hold_promise prep_promise_for amod_improvements_actual pobj_for_improvements prep_improvements_in nn_quality_machine nn_quality_translation nn_quality_output pobj_in_quality ' +o,6365,' , , Dave et al ',Turney,'nn__ appos__ conj__Dave cc__et conj__al ' +n,6366,' presented a history-based generation model to overcome some of the inappropriate independence assumptions in the basic generation model of ',Cahill,'partmod__presented det_model_a amod_model_history-based nn_model_generation dobj_presented_model aux_overcome_to xcomp_presented_overcome dobj_overcome_some prep_some_of det_assumptions_the amod_assumptions_inappropriate nn_assumptions_independence pobj_of_assumptions prep_assumptions_in det_model_the amod_model_basic nn_model_generation pobj_in_model dep__of ' +o,6367,'Significant neighbor-based co-occurrence : As discussed in , it is possible to measure the amount of surprise to see two neighboring words in a corpus at a certain frequency under the assumption of independence ',Dunning,'amod_co-occurrence_Significant amod_co-occurrence_neighbor-based dep_possible_co-occurrence mark_discussed_As advcl_possible_discussed prep_discussed_in pobj_in_ nsubj_possible_it cop_possible_is aux_measure_to xcomp_possible_measure det_amount_the dobj_measure_amount prep_amount_of pobj_of_surprise aux_see_to xcomp_measure_see num_words_two amod_words_neighboring dobj_see_words prep_words_in det_corpus_a pobj_in_corpus prep_corpus_at det_frequency_a amod_frequency_certain pobj_at_frequency prep_see_under det_assumption_the pobj_under_assumption prep_assumption_of pobj_of_independence ' +o,6368,'2 Block Orientation Bigrams This section describes a phrase-based model for SMT similar to the models presented in ',Koehn,'num_Bigrams_2 nn_Bigrams_Block nn_Bigrams_Orientation det_section_This nsubj_describes_section dep_Bigrams_describes det_model_a amod_model_phrase-based nsubj_similar_model prep_model_for pobj_for_SMT xcomp_describes_similar prep_similar_to det_models_the pobj_to_models partmod_models_presented prep_presented_in ' +o,6369,'For the evaluation of translation quality , we used the BLEU metric , which measures the n-gram overlap between the translated output and one or more reference translations ',Papineni,'prep_used_For det_evaluation_the pobj_For_evaluation prep_evaluation_of nn_quality_translation pobj_of_quality nsubj_used_we det__the amod__BLEU amod__metric dobj_used_ nsubj_measures_which rcmod__measures det_overlap_the amod_overlap_n-gram dobj_measures_overlap prep_overlap_between det_output_the amod_output_translated pobj_between_output cc_output_and num_translations_one cc_one_or conj_one_more nn_translations_reference conj_output_translations ' +o,6370,'The decoding process is very similar to those described in : It starts from an initial empty hypothesis ',Koehn,'det_process_The amod_process_decoding nsubj_similar_process cop_similar_is advmod_similar_very prep_similar_to pobj_to_those partmod_those_described prep_described_in pobj_in_ nsubj_starts_It parataxis_similar_starts prep_starts_from det_hypothesis_an amod_hypothesis_initial amod_hypothesis_empty pobj_from_hypothesis ' +o,6371,'Moreover , the inference procedure for each sentence pair is non-trivial , proving NP-complete for learning phrase based models or a high order polynomial -LRB- O -LRB- f 3 e 3 -RRB- -RRB- 1 for a sub-class of weighted synchronous context free grammars ',Wu,'advmod_non-trivial_Moreover det_procedure_the amod_procedure_inference nsubj_non-trivial_procedure prep_procedure_for det_pair_each nn_pair_sentence pobj_for_pair aux_non-trivial_is dep_non-trivial_proving amod__NP-complete prep_NP-complete_for pcomp_for_learning dobj_learning_phrase partmod_phrase_based dobj_based_models dobj_proving_ cc__or det_polynomial_a amod_polynomial_high nn_polynomial_order conj__polynomial dep_polynomial_O dep_3_f dep_-LRB-_3 num_3_e dep_3_3 dobj_non-trivial_1 prep_non-trivial_for det_sub-class_a pobj_for_sub-class prep_sub-class_of amod_grammars_weighted amod_grammars_synchronous nn_grammars_context amod_grammars_free pobj_of_grammars ' +o,6372,' 866 867 119 611 Collins 887 885 092 667 Charniak and Johnson 901 901 074 701 This Paper 903 900 078 685 all sentences LP LR CB 0CB Klein and Manning 863 851 131 572 et al ',Matsuzaki,'number_667_ cc_667_866 conj_667_867 number_667_119 number_667_611 cc_667_Collins conj_667_ number_667_887 number_667_885 number_667_092 dep_Klein_667 dep_667_Charniak cc_667_and nn_Paper_Johnson num_Paper_ num_Paper_901 num_Paper_901 num_Paper_074 num_Paper_701 nn_Paper_This conj_667_Paper number_900_903 num_685_900 number_685_078 dep_Paper_685 advmod_sentences_all dep_Klein_sentences nn_Klein_LP nn_Klein_LR nn_Klein_CB num_Klein_0CB cc_Klein_and nn__Manning num__ num__863 num__851 num__131 nn__572 conj_Klein_ cc__et conj__al ' +o,6373,'With this constraint , each of these binary trees is unique and equivalent to a parse tree of the canonical-form grammar in ',Wu,'prep_unique_With det_constraint_this pobj_With_constraint nsubj_unique_each prep_each_of det_trees_these amod_trees_binary pobj_of_trees cop_unique_is cc_unique_and conj_unique_equivalent prep_unique_to det_tree_a amod_tree_parse pobj_to_tree prep_tree_of det_grammar_the amod_grammar_canonical-form pobj_of_grammar prep_unique_in ' +o,6374,'73 122 Baseline System and Experimental Setup We take BBNs HierDec , a string-to-dependency decoder as described in , as our baseline for the following two reasons : It provides a strong baseline , which ensures the validity of the improvement we would obtain ',He,'num_System_73 num_System_122 nn_System_Baseline cc_System_and nn_Setup_Experimental conj_System_Setup nsubj_take_We dep_System_take nn_HierDec_BBNs dobj_take_HierDec det_decoder_a amod_decoder_string-to-dependency appos_HierDec_decoder advmod_described_as amod_decoder_described prep_take_in pobj_in_ prep_take_as poss_baseline_our pobj_as_baseline prep_baseline_for det_reasons_the amod_reasons_following num_reasons_two pobj_for_reasons nsubj_provides_It parataxis_take_provides det_baseline_a amod_baseline_strong dobj_provides_baseline nsubj_ensures_which rcmod_baseline_ensures det_validity_the dobj_ensures_validity prep_validity_of det_improvement_the pobj_of_improvement nsubj_obtain_we aux_obtain_would dep_System_obtain ' +o,6375,'This tagging scheme is the IOB scheme originally put forward by Ramshaw and Marcus ',Ramshaw,'det_scheme_This amod_scheme_tagging nsubj_scheme_scheme cop_scheme_is det_scheme_the nn_scheme_IOB advmod_put_originally partmod_scheme_put advmod_put_forward prep_put_by pobj_by_Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus ' +o,6376,' and et al ',Cahill,'cc__and conj__ cc__et conj__al ' +o,6377,'The implementation of the algorithm is one that has a core of code that can run on either the Penn Treebank or on the Chinese Treebank ',Marcus,'det_implementation_The nsubj_one_implementation prep_implementation_of det_algorithm_the pobj_of_algorithm cop_one_is nsubj_has_that rcmod_one_has det_core_a dobj_has_core prep_core_of pobj_of_code nsubj_run_that aux_run_can rcmod_core_run prep_run_on dep_on_either det__the nn__Penn nn__Treebank conj_on_ cc_on_or conj_on_on det_Treebank_the amod_Treebank_Chinese pobj_on_Treebank ' +o,6378,'One approach to translate terms consists in using a domain-specific parallel corpus with standard alignment techniques to mine new translations ',Brown,'num_approach_One aux_translate_to infmod_approach_translate nsubj_consists_terms ccomp_translate_consists prep_consists_in pcomp_in_using det_corpus_a amod_corpus_domain-specific amod_corpus_parallel dobj_using_corpus prep_using_with amod_techniques_standard nn_techniques_alignment pobj_with_techniques amod_translations_ prep__to pobj_to_mine amod_translations_new dep_approach_translations ' +o,6379,'We use SUMMA to generate generic and query-based multi-document summaries and evaluate them using ROUGE evaluation metrics relative to human generated summaries ',Lin,'nsubj_use_We nn__SUMMA nsubj_generate_ aux_generate_to xcomp_use_generate amod_summaries_generic cc_generic_and conj_generic_query-based amod_summaries_multi-document dobj_generate_summaries cc_generate_and conj_generate_evaluate nsubj_using_them dep_evaluate_using nn_metrics_ROUGE nn_metrics_evaluation nsubj_relative_metrics advmod_relative_ xcomp_using_relative prep_relative_to amod_summaries_human amod_summaries_generated pobj_to_summaries ' +p,6380,'The notion that nouns have only one sense per discourse\\/collocation was also exploited by in his seminal work on bootstrapping for word sense disambiguation ',Yarowsky,'det_notion_The nsubjpass_exploited_notion complm_have_that nsubj_have_nouns dep_notion_have advmod_sense_only num_sense_one dobj_have_sense prep_sense_per pobj_per_discourse\\/collocation auxpass_exploited_was advmod_exploited_also prep_exploited_by pobj_by_ prep__in poss_work_his amod_work_seminal pobj_in_work prep_exploited_on pcomp_on_bootstrapping prep_bootstrapping_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation ' +o,6381,'The initial phase relies on a parser that draws on the SPECIALIST Lexicon and the Xerox Part-of-Speech Tagger to produce an underspecified categorial analysis ',Cutting,'det_phase_The amod_phase_initial nsubj_relies_phase prep_relies_on det_parser_a pobj_on_parser nsubj_draws_that rcmod_parser_draws prep_draws_on det__the nn__SPECIALIST nn__Lexicon pobj_on_ cc__and det__the nn__Xerox nn__Part-of-Speech nn__Tagger conj__ aux_produce_to xcomp_draws_produce det_analysis_an amod_analysis_underspecified amod_analysis_categorial dobj_produce_analysis ' +o,6382,'In marked contrast to annotated training material for partof-speech tagging, (a) there is no coarse-level set of sense distinctions widely agreed upon (whereas part-of-speech tag sets tend to differ in the details); (b) sense annotation has a comparatively high error rate (Miller, personal communication, reports an upper bound for human annotators of around 90\% for ambiguous cases, using a non-blind evaluation method that may make even this estimate overly optimistic); and (c) no fully automatic method provides high enough quality output to support the \'annotate automatically, correct manually\' methodology used to provide high volume annotation by data providers like the Penn Treebank project (Marcus et al. , 1993).',Marcus,'' +o,6383,'This negation handling is similar to that used in ',Pang,'det_handling_This nn_handling_negation nsubj_similar_handling cop_similar_is prep_similar_to pobj_to_that partmod_that_used prep_used_in ' +o,6384,'To estimate combination weights , we extend the F 1 - score maximization training algorithm for LRM described in ',Jansche,'aux_estimate_To dep_extend_estimate nn_weights_combination dobj_estimate_weights nsubj_extend_we det_F_the dobj_extend_F num_F_1 nn_algorithm_score nn_algorithm_maximization nn_algorithm_training nsubj_described_algorithm prep_algorithm_for pobj_for_LRM parataxis_extend_described prt_described_in ' +o,6385,' grouped nouns into thesaurus-like lists based on the similarity of their syntactic contexts ',Hindle,'nn_grouped_ nsubj_nouns_grouped prep_nouns_into amod_lists_thesaurus-like pobj_into_lists partmod_lists_based prep_based_on det_similarity_the pobj_on_similarity prep_similarity_of poss_contexts_their amod_contexts_syntactic pobj_of_contexts ' +o,6386,'Our method was applied to 23 million words of the WSJ that were automatically tagged with Ratnaparkhi \'s maximum entropy tagger and chunked with the partial parser CASS ',Ratnaparkhi,'poss_method_Our nsubjpass_applied_method auxpass_applied_was prep_applied_to number_million_23 num_words_million pobj_to_words prep_words_of det_WSJ_the pobj_of_WSJ nsubjpass_tagged_that auxpass_tagged_were advmod_tagged_automatically rcmod_words_tagged prep_tagged_with poss__Ratnaparkhi possessive_Ratnaparkhi_\'s amod__maximum amod__entropy nn__tagger pobj_with_ cc_tagged_and conj_tagged_chunked prep_chunked_with det__the amod__partial nn__parser nn__CASS pobj_with_ ' +o,6387,'task , and reported errors in the range of 26 \% are common ',Marcus,'nsubj__task cc__and conj__reported nsubj_common_errors prep_errors_in det_range_the pobj_in_range prep_range_of num_\%_26 pobj_of_\% cop_common_are ccomp_reported_common ' +n,6388,'Unfortunately , longer sentences -LRB- up to 100 tokens , rather than 40 -RRB- , longer phrases -LRB- up to 10 tokens , rather than 7 -RRB- , two LMs -LRB- rather than just one -RRB- , higher-order LMs -LRB- order 7 , rather than 3 -RRB- , multiple higher-order lexicalized re-ordering models -LRB- up to 3 -RRB- , etc all contributed to increased system ? s complexity , and , as a result , time limitations prevented us from performing minimum-error-rate training -LRB- MERT -RRB- for ucb3 , ucb4 and ucb5 ',Och,'advmod_complexity_Unfortunately amod_sentences_longer nsubj_complexity_sentences dep_sentences_up dep_up_to num_tokens_100 pobj_to_tokens cc_tokens_rather dep_rather_than conj_tokens_40 amod_phrases_longer appos_sentences_phrases dep_phrases_up dep_up_to num_tokens_10 pobj_to_tokens cc_tokens_rather dep_rather_than conj_tokens_7 num_LMs_two appos_sentences_LMs dep_than_rather dep_LMs_than advmod_one_just pobj_than_one amod_LMs_higher-order appos_sentences_LMs dep_LMs_order num_order_7 dep_than_rather prep_order_than pobj_than_3 amod_models_multiple amod_models_higher-order amod_models_lexicalized amod_models_re-ordering appos_sentences_models quantmod_3_up dep_3_to dep_models_3 advmod_all_etc amod_sentences_all partmod_sentences_contributed prep_contributed_to amod_system_increased pobj_to_system dep_complexity_s cc_complexity_and prep_prevented_as det_result_a pobj_as_result nn_limitations_time nsubj_prevented_limitations conj_complexity_prevented dobj_prevented_us prep_prevented_from pcomp_from_performing amod__minimum-error-rate nn__training abbrev__MERT dobj_performing_ prep__for pobj_for_ucb3 conj_ucb3_ucb4 cc_ucb3_and conj_ucb3_ucb5 ' +o,6389,'A pipage approach has been proposed for MCKP , but we do not use this algorithm , since it requires costly partial enumeration and solutions to many linear relaxation problems ',Lin,'det_approach_A amod_approach_pipage nsubjpass_proposed_ aux_proposed_has auxpass_proposed_been dep_approach_proposed prep_proposed_for pobj_for_MCKP cc_proposed_but nsubj_use_we aux_use_do neg_use_not conj_proposed_use det_algorithm_this dobj_use_algorithm mark_requires_since nsubj_requires_it advcl_use_requires amod_enumeration_costly amod_enumeration_partial dobj_requires_enumeration cc_enumeration_and conj_enumeration_solutions prep_requires_to amod_problems_many amod_problems_linear nn_problems_relaxation pobj_to_problems ' +o,6390,'The acquisition of clues is a key technology in these research efforts , as seen in learning methods for document-level SA and for phraselevel SA ',Turney,'det_acquisition_The nsubj_technology_acquisition prep_acquisition_of pobj_of_clues cop_technology_is det_technology_a amod_technology_key prep_technology_in det_efforts_these nn_efforts_research pobj_in_efforts mark_seen_as advcl_technology_seen prep_seen_in pcomp_in_learning dobj_learning_methods prep_learning_for amod__document-level nn__SA pobj_for_ cc_for_and conj_for_for amod__phraselevel nn__SA pobj_for_ ' +o,6391,'The feature weights i are trained in concert with the LM weight via minimum error rate -LRB- MER -RRB- training ',Och,'det_i_The nn_i_feature nn_i_weights nsubjpass_trained_i auxpass_trained_are prep_trained_in pobj_in_concert prep_trained_with det_weight_the nn_weight_LM pobj_with_weight prep_weight_via amod_training_minimum nn_training_error nn_training_rate nn_training_MER pobj_via_training ' +o,6392,'Our next steps will be to take a closer look at the following work : clustering of similar words , topic signatures and Kilgariffs sketch engine ',Lin,'poss_steps_Our amod_steps_next aux_be_will dep_steps_be aux_take_to xcomp_be_take det_look_a amod_look_closer dobj_take_look prep_take_at det_work_the amod_work_following pobj_at_work dep_steps_clustering prep_clustering_of amod__similar nn__words nn__ dep__topic nn__signatures pobj_of_ cc__and nn__Kilgariffs nn__sketch nn__engine conj__ ' +o,6393,'31 NP Our NP chunks are very similar to the ones of ',Ramshaw,'num_chunks_31 nn_chunks_NP nn_chunks_Our nn_chunks_NP nsubj_similar_chunks cop_similar_are advmod_similar_very prep_similar_to det_ones_the pobj_to_ones prep_ones_of ' +o,6394,'Most of the early work in this area was based on postulating generative probability models of language that included parse structure ',Collins,'nsubjpass_based_Most prep_Most_of det_work_the amod_work_early pobj_of_work prep_work_in det_area_this pobj_in_area auxpass_based_was prep_based_on pcomp_on_postulating amod_models_generative nn_models_probability dobj_postulating_models prep_models_of pobj_of_language nsubj_included_that rcmod_models_included amod_structure_parse dobj_included_structure ' +o,6395,'The model weights are trained using the improved iterative scaling algorithm ',Berger,'det_weights_The nn_weights_model nsubjpass_trained_weights auxpass_trained_are xcomp_trained_using det_algorithm_the amod_algorithm_improved nn_algorithm_iterative amod_algorithm_scaling dobj_using_algorithm ' +o,6396,'We experimented with two independent , arguably complementary techniques for clustering and aligning a predicate argument based approach that extracts more general templates containing one predicate and a ROUGE based 265 approach that can extract templates containing multiple verbs ',Lin,'nsubj_experimented_We prep_experimented_with pobj_with_two amod_techniques_independent advmod_techniques_arguably amod_techniques_complementary dobj_experimented_techniques prep_techniques_for pcomp_for_clustering cc_clustering_and conj_clustering_aligning det_argument_a amod_argument_predicate dobj_clustering_argument partmod_argument_based dobj_based_approach complm_templates_that nsubj_templates_extracts advmod_general_more amod_extracts_general ccomp_based_templates xcomp_templates_containing num_predicate_one dobj_containing_predicate cc_predicate_and det__a amod__ROUGE conj_predicate_ prep_containing_based num_approach_265 dep_based_approach nsubj_extract_that aux_extract_can rcmod_approach_extract dobj_extract_templates partmod_templates_containing amod_verbs_multiple dobj_containing_verbs ' +o,6397,'We show that the method of , which was presented as a simple preprocessing step , is actually equivalent , except our representation explicitly separates hyperparameters which were tied in his work ','Daume III','nsubj_show_We complm_equivalent_that det_method_the nsubj_equivalent_method prep_method_of pobj_of_ nsubjpass_presented_which auxpass_presented_was rcmod_method_presented prep_presented_as det_step_a amod_step_simple nn_step_preprocessing pobj_as_step cop_equivalent_is advmod_equivalent_actually ccomp_show_equivalent mark_separates_except poss_representation_our nsubj_separates_representation advmod_separates_explicitly advcl_equivalent_separates dobj_separates_hyperparameters nsubjpass_tied_which auxpass_tied_were rcmod_hyperparameters_tied prep_tied_in poss_work_his pobj_in_work ' +o,6398,'The loglinear model feature weights were learned using minimum error rate training -LRB- MERT -RRB- with BLEU score as the objective function ',Och,'det_model_The nn_model_loglinear nsubj_feature_model nsubjpass_learned_weights auxpass_learned_were ccomp_feature_learned xcomp_learned_using amod__minimum nn__error nn__rate nn__training abbrev__MERT dobj_using_ prep__with amod__BLEU nn__score pobj_with_ prep_using_as det_function_the amod_function_objective pobj_as_function ' +o,6399,'To reduce it we exploit the one sense per collocation property ',Yarowsky,'aux_reduce_To nsubj_exploit_it dep_exploit_we xcomp_reduce_exploit det_sense_the num_sense_one dobj_exploit_sense prep_sense_per nn_property_collocation pobj_per_property ' +o,6400,'WordNet has been criticized for being overly finegrained , we are using it here because it is the sense inventory used by Erk et al ',Navigli,'nsubjpass_criticized_WordNet aux_criticized_has auxpass_criticized_been prep_criticized_for cop_finegrained_being advmod_finegrained_overly pcomp_for_finegrained advmod_it_ nsubj_it_we aux_it_are dep_it_using dep_criticized_it advmod_inventory_here dep_inventory_because nsubj_inventory_it cop_inventory_is det_inventory_the nn_inventory_sense dep_it_inventory partmod_inventory_used prep_used_by pobj_by_Erk cc_Erk_et conj_Erk_al ' +o,6401,'2 Related Work Two different approaches have been proposed for Sentence Compression : purely statistical methodologies and hybrid linguistic\\/statistic methodologies ',Barzilay,'dep_proposed_2 nn_Work_Related nsubjpass_proposed_Work num_approaches_Two amod_approaches_different dep_Work_approaches aux_proposed_have auxpass_proposed_been prep_proposed_for nn_Compression_Sentence pobj_for_Compression advmod_methodologies_purely amod_methodologies_statistical nsubj__methodologies amod_methodologies_ cc_methodologies_and amod_methodologies_hybrid amod_methodologies_linguistic\\/statistic conj_methodologies_methodologies parataxis_proposed_ ' +o,6402,'This is concordant with the usage in the maximum entropy literature ',Berger,'nsubj_concordant_This cop_concordant_is prep_concordant_with det_usage_the pobj_with_usage prep_usage_in det_literature_the amod_literature_maximum nn_literature_entropy pobj_in_literature ' +o,6403,'43 Baselines 431 Word Alignment We used the GIZA + + implementation of IBM word alignment model 4 for word alignment , and the heuristics described in to derive the intersection and refined alignment ',Och,'num_Baselines_43 measure_431_Baselines amod_Alignment_431 nn_Alignment_Word nsubj_used_We dep_Alignment_used det_implementation_the amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dobj_used_implementation prep_implementation_of nn_model_IBM nn_model_word nn_model_alignment pobj_of_model dep__4 amod_model_ prep_used_for nn_alignment_word pobj_for_alignment cc_used_and det_heuristics_the nsubj_described_heuristics conj_used_described prep_described_in pobj_in_ aux_derive_to xcomp_described_derive det_intersection_the dobj_derive_intersection cc_intersection_and amod_alignment_refined conj_intersection_alignment ' +o,6404,'Much previous work has been done on this problem and many different methods have been used: Church\'s PARTS (1988) program uses a Markov model; Bourigault (1992) uses heuristics along with a grammar; Voutilainen\'s NPTool (1993) uses a lexicon combined with a constraint grammar; Juteson and Katz (1995) use repeated phrases; Veenstra (1998), Argamon, Dagan & Krymolowski(1998) and Daelemaus, van den Bosch & Zavrel (1999) use memory-based systems; Ramshaw & Marcus (In Press) and Cardie & Pierce (1998) use rule-based systems.',Ramshaw,'' +o,6405,'There are similarities with dependency grammars here because such constraint graphs are also produced by dependency grammars ',Kashket,'expl_are_There dep_are_similarities prep_similarities_with amod_grammars_dependency pobj_with_grammars advmod_produced_here dep_produced_because amod_graphs_such nn_graphs_constraint nsubjpass_produced_graphs auxpass_produced_are advmod_produced_also ccomp_similarities_produced prep_produced_by amod__dependency nn__grammars pobj_by_ ' +o,6406,'Learned vowels include -LRB- in order of generation probability -RRB- : e , a , o , u , i , y Learned sonorous consonants include : n , s , r , l , m Learned non-sonorous consonants include : d , c , t , l , b , m , p , q The model bootstrapping is good for dealing with too many parameters ; we see a similar approach in march from Model 1 to Model 5 ',Brown,'amod_vowels_Learned nsubj_include_vowels dep_include_in pobj_in_order prep_order_of nn_probability_generation pobj_of_probability dep_q_e nsubj_q_a appos_a_o appos_o_u dep_Learned_i nn_Learned_y appos_o_Learned amod_consonants_sonorous nsubj_include_consonants rcmod_Learned_include dobj_include_n dep_r_s dobj_include_r tmod_include_l nn_consonants_m nn_consonants_Learned nn_consonants_non-sonorous nsubj_include_consonants rcmod_n_include dep_o_d appos_d_c appos_d_t appos_d_l appos_d_b appos_d_m appos_d_p parataxis_include_q det_bootstrapping_The nn_bootstrapping_model nsubj_good_bootstrapping cop_good_is ccomp_q_good prep_good_for pcomp_for_dealing prep_dealing_with advmod_many_too amod_parameters_many pobj_with_parameters nsubj_see_we parataxis_q_see det_approach_a amod_approach_similar dobj_see_approach prep_approach_in amod_march_ pobj_in_march prep_march_from pobj_from_Model num_Model_1 prep_see_to pobj_to_Model num_Model_5 ' +o,6407,' managed to extract LFG subcategorisation frames and paths linking long distance dependencies reentrancies from f-structures generated automatically for the PennII treebank trees and used them in an long distance dependency resolution algorithm to parse new text ',Cahill,'advmod_managed_ amod_reentrancies_managed aux_extract_to xcomp_managed_extract nn_frames_LFG nn_frames_subcategorisation nsubj_linking_frames cc_frames_and conj_frames_paths dep_extract_linking amod_dependencies_long nn_dependencies_distance dobj_linking_dependencies prep_reentrancies_from pobj_from_f-structures dep_f-structures_generated advmod_generated_automatically prep_generated_for det_trees_the nn_trees_PennII nn_trees_treebank pobj_for_trees cc_generated_and conj_generated_used dobj_used_them prep_used_in det_algorithm_an amod_algorithm_long nn_algorithm_distance nn_algorithm_dependency nn_algorithm_resolution pobj_in_algorithm aux_parse_to xcomp_used_parse amod_text_new dobj_parse_text ' +o,6408,'Second , phrase translation pairs are extracted from the word alignment corpus ',Koehn,'advmod_extracted_Second nn_pairs_phrase nn_pairs_translation nsubjpass_extracted_pairs auxpass_extracted_are prep_extracted_from det_corpus_the nn_corpus_word nn_corpus_alignment pobj_from_corpus ' +o,6409,'The scores were then weighted by the inverse of their height in the tree and then summed together , similarly to the procedure in ',Dunning,'det_scores_The nsubjpass_weighted_scores auxpass_weighted_were advmod_weighted_then prep_weighted_by det_inverse_the pobj_by_inverse prep_inverse_of poss_height_their pobj_of_height prep_height_in det_tree_the pobj_in_tree cc_weighted_and advmod_weighted_then conj_weighted_summed advmod_summed_together advmod_summed_similarly prep_summed_to det_procedure_the pobj_to_procedure prep_procedure_in pobj_in_ ' +o,6410,'Then , we run GIZA + + on the corpus to obtain word alignments in both directions ',Och,'advmod_run_Then nsubj_run_we dobj_run_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_run_on det_corpus_the pobj_on_corpus aux_obtain_to infmod_corpus_obtain nn_alignments_word dobj_obtain_alignments prep_obtain_in det_directions_both pobj_in_directions ' +o,6411,'Berry et al (1993)) to yield W W = U S V T as Figure 3 shows, where, for some order R lessmuch min(M,N) of the decomposition, U is a MR left singular matrix with rows ui, i = 1,,M, S is a RR diagonal matrix of singular values s1 s2 sR greatermuch 0, and V is NR a right singular matrix with rows vj, j = 1,,N. For each i, the scaled R-vector uiS may be viewed as representing wi, thei-th word in the vocabulary, and similarly the scaled R-vector vjS as representing dj, j-th document in the corpus.',Brown,'' +o,6412,'This cost can often be substantial , as with the Penn Treebank ',Brown,'det_cost_This nsubj_substantial_cost aux_substantial_can advmod_substantial_often cop_substantial_be advmod_substantial_as dep_as_with det__the nn__Penn nn__Treebank pobj_with_ ' +o,6413,'This characteristic of our corpus is similar to problems with noisy and comparable corpora , and it prevents us from using methods developed in the MT community based on clean parallel corpora , such as ',Brown,'det_characteristic_This nsubj_similar_characteristic prep_characteristic_of poss_corpus_our pobj_of_corpus cop_similar_is prep_similar_to pobj_to_problems prep_problems_with amod__noisy cc_noisy_and conj_noisy_comparable nn__corpora pobj_with_ cc_similar_and nsubj_prevents_it conj_similar_prevents dobj_prevents_us prep_prevents_from pcomp_from_using dobj_using_methods partmod_methods_developed prep_developed_in det_community_the nn_community_MT pobj_in_community prep_developed_based dep_based_on amod_corpora_clean amod_corpora_parallel pobj_on_corpora dep_as_such prep_prevents_as ' +o,6414,'It forms a baseline for performance evaluations , but is prone to sparse data problems ',Dunning,'nsubj_forms_It det_baseline_a dobj_forms_baseline prep_baseline_for nn_evaluations_performance pobj_for_evaluations cc_forms_but cop_prone_is conj_forms_prone aux_sparse_to xcomp_prone_sparse nn_problems_data dobj_sparse_problems ' +o,6415,'1 Introduction Word alignmentdetection of corresponding words between two sentences that are translations of each otheris usually an intermediate step of statistical machine translation -LRB- MT -RRB- , but also has been shown useful for other applications such as construction of bilingual lexicons , word-sense disambiguation , projection of resources , and crosslanguage information retrieval ',Och,'num_alignmentdetection_1 nn_alignmentdetection_Introduction nn_alignmentdetection_Word prep_alignmentdetection_of amod_words_corresponding pobj_of_words prep_words_between num_sentences_two pobj_between_sentences nsubj_translations_that cop_translations_are rcmod_sentences_translations prep_translations_of det_otheris_each pobj_of_otheris advmod_translations_usually det_step_an amod_step_intermediate nsubjpass__step prep_step_of amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_MT dep_alignmentdetection_ cc__but advmod_shown_also aux_shown_has auxpass_shown_been conj__shown acomp_shown_useful prep_shown_for amod_applications_other pobj_for_applications dep_as_such prep_applications_as pobj_as_construction prep_construction_of amod_lexicons_bilingual pobj_of_lexicons amod_disambiguation_word-sense conj_lexicons_disambiguation conj_lexicons_projection prep_projection_of pobj_of_resources cc_lexicons_and amod_retrieval_crosslanguage nn_retrieval_information conj_lexicons_retrieval ' +o,6416,'The basic phrase reordering model is a simple unlexicalized , context-insensitive distortion penalty model ',Koehn,'det_model_The amod_model_basic nn_model_phrase nn_model_reordering nsubj_model_model cop_model_is det_model_a amod_model_simple amod_model_unlexicalized amod_model_context-insensitive nn_model_distortion nn_model_penalty ' +p,6417,'Using the components of the row-vector bm as feature function values for the candidate translation em -LRB- m a16 1 , , M -RRB- , the system prior weights can easily be trained using the Minimum Error Rate Training described in ',Och,'det_components_the dobj_Using_components prep_components_of det_bm_the amod_bm_row-vector pobj_of_bm prep_bm_as nn_function_feature pobj_as_function dep_Using_values prep_values_for det_translation_the nn_translation_candidate pobj_for_translation nsubj_Using_em nn_a16_m dep_em_a16 dep_a16_1 appos_a16_M det_system_the appos_em_system amod_weights_prior nsubjpass_trained_weights aux_trained_can advmod_trained_easily auxpass_trained_be rcmod_system_trained purpcl_trained_using det_Training_the nn_Training_Minimum nn_Training_Error nn_Training_Rate dobj_using_Training partmod_Training_described prep_described_in ' +p,6418,'Bootstrapping a PMTG from a lower-dimensional PMTG and a word-to-word translation model is similar in spirit to the way that regular grammars can help to estimate CFGs , and the way that simple translation models can help to bootstrap more sophisticated ones ',Brown,'det_PMTG_a dobj_Bootstrapping_PMTG prep_Bootstrapping_from det_PMTG_a amod_PMTG_lower-dimensional pobj_from_PMTG cc_Bootstrapping_and det_model_a amod_model_word-to-word nn_model_translation nsubj_similar_model cop_similar_is conj_Bootstrapping_similar prep_similar_in pobj_in_spirit prep_spirit_to det_way_the pobj_to_way dep_help_that amod_grammars_regular nsubj_help_grammars aux_help_can dep_similar_help aux_estimate_to xcomp_help_estimate nn__CFGs dobj_estimate_ cc_help_and det_way_the dep_help_way dep_help_that amod_models_simple nn_models_translation nsubj_help_models aux_help_can conj_help_help aux_bootstrap_to xcomp_help_bootstrap advmod_sophisticated_more amod_ones_sophisticated dobj_bootstrap_ones ' +n,6419,'Unsupervised methods have been developed for WSD , but despite modest success have not always been well understood statistically ',Abney,'amod_methods_Unsupervised nsubjpass_developed_methods aux_developed_have auxpass_developed_been dep_understood_developed prep_developed_for pobj_for_WSD cc_for_but conj_for_despite amod_success_modest pobj_despite_success aux_understood_have neg_understood_not advmod_understood_always auxpass_understood_been advmod_understood_well nsubjpass_understood_statistically ' +n,6420,'In addition , uniform conditioning on mother grammatical function is more general than the case-phenomena specific generation grammar transform of , in that it applies to each and every sub-part of a recursive input f-structure driving generation , making available relevant generation history -LRB- context -RRB- to guide local generation decisions ',Cahill,'prep_general_In pobj_In_addition amod_conditioning_uniform nsubj_general_conditioning prep_conditioning_on nn_function_mother amod_function_grammatical pobj_on_function cop_general_is advmod_general_more mark_transform_than det_grammar_the amod_grammar_case-phenomena amod_grammar_specific nn_grammar_generation nsubj_transform_grammar advcl_general_transform prep_transform_of pobj_of_ rel_applies_in pobj_in_that nsubj_applies_it rcmod__applies prep_applies_to pobj_to_each cc_each_and det_sub-part_every conj_each_sub-part prep_sub-part_of det_generation_a amod_generation_recursive nn_generation_input amod_generation_f-structure amod_generation_driving pobj_of_generation partmod_generation_making amod_history_available dep_available_relevant nn_history_generation dobj_making_history appos_history_context aux_guide_to xcomp_making_guide amod_decisions_local nn_decisions_generation dobj_guide_decisions ' +o,6421,'Corpus Time Period Size Articles Words New Indian Express -LRB- English -RRB- 20070101 to 20070831 2,359 347,050 Dinamani -LRB- Tamil -RRB- 20070101 to 20070831 2,359 256,456 Table 1 : Statistics on Comparable Corpora From the above corpora , we first extracted all the NEs from the English side , using the Stanford NER tool ',Finkel,'nn_Time_Corpus nsubj_Period_Time iobj_Period_Size nn_Express_Articles nn_Express_Words nn_Express_New nn_Express_Indian dobj_Period_Express appos_Express_English number_2,359_20070101 dep_2,359_to number_2,359_20070831 num_347,050_2,359 dobj_Period_347,050 dobj_Period_Dinamani appos_Dinamani_Tamil number_2,359_20070101 dep_2,359_to number_2,359_20070831 num_256,456_2,359 dobj_Period_256,456 prep_Period_Table pobj_Table_1 dobj_Period_Statistics prep_Statistics_on amod_Corpora_Comparable pobj_on_Corpora prep_extracted_From det_corpora_the amod_corpora_above pobj_From_corpora nsubj_extracted_we advmod_extracted_first dep_Period_extracted predet_NEs_all det_NEs_the dobj_extracted_NEs prep_extracted_from det_side_the amod_side_English pobj_from_side xcomp_extracted_using det_tool_the nn_tool_Stanford nn_tool_NER dobj_using_tool ' +o,6422,'In general , they can be divided into two major categories , namely lexicalized models and un-lexicalized models ',Collins,'prep_divided_In pobj_In_general nsubjpass_divided_they aux_divided_can auxpass_divided_be prep_divided_into num_categories_two amod_categories_major pobj_into_categories advmod_models_namely amod_models_lexicalized conj_categories_models amod_models_ cc_categories_and amod__un-lexicalized nn__models conj_categories_ ' +o,6423,'The candidates were then ranked according to the scores assigned by four association measures : the log-likelihood ratio G2 , Pearsons chi-squared statistic X2 , the t-score statistic t , and mere cooccurrence frequency f4 TPs were identified according to the definition of Krenn ',Dunning,'det_candidates_The nsubjpass_ranked_candidates auxpass_ranked_were advmod_ranked_then prep_ranked_according dep_according_to det_scores_the pobj_to_scores partmod_scores_assigned prep_assigned_by num_measures_four nn_measures_association pobj_by_measures det__the amod__log-likelihood nn__ratio nn__G2 nsubjpass_identified_ nn__Pearsons amod__chi-squared nn__statistic nn__X2 conj__ det__the amod__t-score nn__statistic nn__t conj__ cc__and amod_TPs_mere nn_TPs_cooccurrence nn_TPs_frequency nn_TPs_f4 conj__TPs auxpass_identified_were parataxis_ranked_identified prep_identified_according dep_according_to det_definition_the pobj_to_definition prep_definition_of nn__Krenn pobj_of_ ' +o,6424,'5 Conclusions and Future Work The paper compares Structural Correspondence Learning with -LRB- various instances of -RRB- self-training for the adaptation of a parse selection model to Wikipedia domains ',Blitzer,'num_Conclusions_5 nsubj_compares_Conclusions cc_Conclusions_and nn_Work_Future conj_Conclusions_Work det_paper_The dep_Work_paper amod__Structural nn__Correspondence nn__Learning dobj_compares_ prep__with amod_instances_various dep__instances prep_instances_of amod__self-training pobj_with_ prep__for det_adaptation_the pobj_for_adaptation prep_adaptation_of det_model_a amod_model_parse nn_model_selection pobj_of_model prep_compares_to nn_domains_Wikipedia pobj_to_domains ' +o,6425,'Second , we discuss the work done by who use clustering of paraphrases to induce rewriting rules ',Barzilay,'advmod_discuss_Second nsubj_discuss_we det_work_the dobj_discuss_work partmod_work_done prep_done_by pobj_by_ nsubj_use_who rcmod_work_use nsubj_induce_clustering prep_clustering_of pobj_of_paraphrases aux_induce_to xcomp_use_induce amod_rules_rewriting dobj_induce_rules ' +o,6426,'Word alignment is newer , found only in a few places ',Brown,'amod_alignment_Word nsubj_newer_alignment cop_newer_is partmod_newer_found advmod_found_only prep_found_in det_places_a amod_places_few pobj_in_places ' +o,6427,'Uses for k-best lists include minimum Bayes risk decoding , discriminative reranking , and discriminative training ',Och,'nsubj_include_Uses prep_Uses_for amod_lists_k-best pobj_for_lists amod__minimum nn__Bayes nn__risk amod__decoding dobj_include_ amod__discriminative dep_discriminative_reranking conj__ cc__and amod_training_discriminative conj__training ' +n,6428,'Section 5 presents an error analysis for lexicalized model , which shows that the head-head dependencies used in this model fail to cope well with the flat structures in Negra ',Collins,'nsubj_presents_Section num_Section_5 det_analysis_an nn_analysis_error dobj_presents_analysis prep_presents_for amod_model_ amod_model_lexicalized pobj_for_model nsubj_shows_which rcmod_model_shows complm_fail_that det_dependencies_the amod_dependencies_head-head nsubj_fail_dependencies partmod_dependencies_used prep_used_in det_model_this pobj_in_model ccomp_shows_fail aux_cope_to xcomp_fail_cope advmod_cope_well prep_cope_with det_structures_the amod_structures_flat pobj_with_structures prep_structures_in pobj_in_Negra ' +o,6429,' , -RRB- ',Pang,'appos__ ' +o,6430,'The disambiguation model of this parser is based on a maximum entropy model ',Berger,'det_model_The nn_model_disambiguation nsubjpass_based_model prep_model_of det_parser_this pobj_of_parser auxpass_based_is prep_based_on det_model_a amod_model_maximum nn_model_entropy pobj_on_model ' +o,6431,'The reliability for the two annotation tasks -LRB- - statistics -RRB- was of 094 and 090 respectively ',Carletta,'det_reliability_The nsubj_was_reliability prep_reliability_for det_tasks_the num_tasks_two nn_tasks_annotation pobj_for_tasks measure__statistics dep_reliability_ prep_was_of pobj_of_094 cc_094_and conj_094_090 dep_of_respectively ' +o,6432,'2 Previous Work We briefly outline the most important existing methods and cite error rates on a standard English data set , sections 03-06 of the Wall Street Journal -LRB- WSJ -RRB- corpus , containing nearly 27,000 examples ',Marcus,'num_Work_2 amod_Work_Previous nsubj_outline_We advmod_outline_briefly dep_Work_outline det_methods_the advmod_important_most amod_methods_important amod_methods_existing dobj_outline_methods cc_outline_and conj_outline_cite nn_rates_error dobj_cite_rates prep_cite_on det_data_a amod_data_standard amod_data_English pobj_on_data nsubj_containing_set appos_set_sections num_sections_03-06 prep_set_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal abbrev_set_WSJ nn__corpus dep_set_ dep_Work_containing quantmod_27,000_nearly num_examples_27,000 dobj_containing_examples ' +o,6433,' -RRB- , and emotion studies -LRB- eg ',Pang,'dep_and_ partmod_and_emotion nn_eg_studies dep_and_eg ' +o,6434,' , bilingual sentences are trained by GIZA + + in two directions -LRB- from source to target and target to source -RRB- ',Och,'advmod_trained_ amod_sentences_bilingual nsubj_trained_sentences cop_trained_are prep_trained_by nn_+_GIZA pobj_by_+ cc_trained_+ conj_trained_ prep__in num_directions_two pobj_in_directions dep_trained_from pobj_from_source dep_from_to pobj_to_target cc_target_and conj_target_target dep_from_to pobj_to_source ' +o,6435,'We have begun experimenting with log likelihood ratio as a thresholding technique ',Dunning,'nsubj_begun_We aux_begun_have xcomp_begun_experimenting prep_experimenting_with nn__log nn__likelihood nn__ratio pobj_with_ prep_experimenting_as det_technique_a amod_technique_thresholding pobj_as_technique ' +o,6436,'33 Language Model -LRB- LM -RRB- As a second baseline we use the classification based on the language model using overlapping ngram sequences -LRB- n was set to 8 -RRB- as suggested by for the English language ',Pang,'nn_Model_Language dep_33_Model abbrev_Model_LM dep_33_As det_baseline_a amod_baseline_second pobj_As_baseline nsubj_use_we rcmod_baseline_use det_classification_the dobj_use_classification prep_use_based dep_based_on det_model_the nn_model_language pobj_on_model partmod_model_using amod_sequences_overlapping nn_sequences_ngram dobj_using_sequences nsubjpass_set_n auxpass_set_was dep_sequences_set prep_set_to pobj_to_8 mark_suggested_as advcl_using_suggested prep_suggested_by pobj_by_ prep_suggested_for det_language_the amod_language_English pobj_for_language ' +o,6437,'Method Source Spearman Wikipedia 019048 WordNet 033035 Rogets 055 WordNet 055 Web corpus , WN 056 ODP 065 Wikipedia 075 SVM Web corpus , WN 078 Table 9 : Comparison with previous work for WordSim353 ',Hughes,'nn_Wikipedia_Method nn_Wikipedia_Source nn_Wikipedia_Spearman nn_Wikipedia_ number__019048 num_WN_ nn_WN_WordNet num_WN_033035 nn_corpus_ nn_corpus_Rogets nn_corpus_055 nn_corpus_ nn_corpus_WordNet nn_corpus_055 nn_corpus_ nn_corpus_Web dep_WN_corpus nn_corpus_WN nn_corpus_056 nn_corpus_ nn_corpus_ODP nn_corpus_065 nn_corpus_ nn_corpus_Wikipedia nn_corpus_075 nn_corpus_SVM nn_corpus_Web dep_corpus_corpus dep_Wikipedia_WN num_Table_078 dep_WN_Table num_Table_9 dep_WN_Comparison prep_Comparison_with amod_work_previous pobj_with_work prep_work_for pobj_for_WordSim353 ' +o,6438,'Following the suggestions in , Core et al consider kappa scores above 067 to indicate significant agreement and scores above 08 reliable agreement ',Carletta,'prep_consider_Following det_suggestions_the pobj_Following_suggestions prep_suggestions_in pobj_in_ nsubj_consider_Core cc_Core_et conj_Core_al nn_scores_kappa dobj_consider_scores prep_consider_above pobj_above_067 aux_indicate_to xcomp_consider_indicate amod_agreement_significant dobj_indicate_agreement cc_agreement_and conj_agreement_scores prep_indicate_above num_agreement_08 amod_agreement_reliable pobj_above_agreement ' +o,6439,'These rules are learned using a word alignment model , which finds an optimal mapping from words to MR predicates given a set of training sentences and their correct MRs Word alignment models have been widely used for lexical acquisition in SMT ',Brown,'det_rules_These nsubjpass_learned_rules auxpass_learned_are xcomp_learned_using det_model_a nn_model_word nn_model_alignment dobj_using_model nsubj_finds_which rcmod_model_finds det_mapping_an amod_mapping_optimal dobj_finds_mapping prep_finds_from pobj_from_words prep_words_to nn_predicates_MR pobj_to_predicates partmod_predicates_given det_set_a dobj_given_set prep_set_of nn_sentences_training pobj_of_sentences cc_sentences_and poss_alignment_their amod_alignment_correct nn_alignment_MRs nn_alignment_Word conj_sentences_alignment nsubjpass_used_models aux_used_have auxpass_used_been advmod_used_widely dep_learned_used prep_used_for amod_acquisition_lexical pobj_for_acquisition prep_acquisition_in pobj_in_SMT ' +o,6440,'While work on subjectivity analysis in other languages is growing -LRB- eg , Japanese data are used in , Chinese data are used in , and German data are used in -RRB- , much of the work in subjectivity analysis has been applied to English data ',Kanayama,'mark_growing_While nsubj_growing_work prep_work_on amod_analysis_subjectivity pobj_on_analysis prep_analysis_in amod_languages_other pobj_in_languages cop_growing_is advcl_applied_growing nn_data_eg amod_data_Japanese nsubjpass_used_data auxpass_used_are parataxis_growing_used prep_used_in pobj_in_ amod_data_Chinese nsubjpass_used_data auxpass_used_are ccomp_used_used prep_used_in pobj_in_ cc_used_and amod_data_German nsubjpass_used_data auxpass_used_are conj_used_used prep_used_in pobj_in_ nsubjpass_applied_much prep_much_of det_work_the pobj_of_work prep_work_in amod_analysis_subjectivity pobj_in_analysis aux_applied_has auxpass_applied_been prep_applied_to amod_data_English pobj_to_data ' +o,6441,'While this technique has been sttccessfully applied to parsing lhe ATIS portion in the Penn Treebank , it is extremely time consuming ',Marcus,'mark_applied_While det_technique_this nsubjpass_applied_technique aux_applied_has auxpass_applied_been advmod_applied_sttccessfully advcl_consuming_applied prep_applied_to pcomp_to_parsing nn_portion_lhe nn_portion_ATIS dobj_parsing_portion prep_parsing_in det__the nn__Penn nn__Treebank pobj_in_ nsubj_consuming_it cop_consuming_is advmod_consuming_extremely nn_consuming_time ' +o,6442,' and Lin and Och -LRB- 2004 -RRB- proposed an LCS-based automatic evaluation measure called ROUGE-L ',Lin,'nsubj_proposed_ cc__and conj__Lin cc_Lin_and conj_Lin_Och appos_Lin_2004 det_measure_an amod_measure_LCS-based amod_measure_automatic nn_measure_evaluation nsubj_called_measure ccomp_proposed_called acomp_called_ROUGE-L ' +o,6443,'In , the definition words were used as initial sense indicators , automatically tagging the target word examples containing them ',Yarowsky,'prep_used_In pobj_In_ det_words_the nn_words_definition nsubjpass_used_words auxpass_used_were prep_used_as amod_indicators_initial nn_indicators_sense pobj_as_indicators advmod_tagging_automatically xcomp_used_tagging det_examples_the nn_examples_target nn_examples_word nsubj_containing_examples dep_tagging_containing dobj_containing_them ' +o,6444,'We follow IBM Model 1 and assume that each word in an utterance is generated by exactly one role in the parallel frame Using standard EM to learn the role to word mapping is only sufficient if one knows to which level in the tree the utterance should be mapped ',Brown,'nsubj_follow_We nn__IBM nn__Model num__1 dobj_follow_ cc_follow_and conj_follow_assume complm_generated_that det_word_each nsubjpass_generated_word prep_word_in det_utterance_an pobj_in_utterance auxpass_generated_is ccomp_assume_generated prep_generated_by advmod_role_exactly num_role_one pobj_by_role prep_role_in det_frame_the amod_frame_parallel pobj_in_frame csubj_sufficient_Using amod_EM_standard dobj_Using_EM aux_learn_to xcomp_Using_learn det_role_the dobj_learn_role prep_learn_to nn_mapping_word pobj_to_mapping cop_sufficient_is advmod_sufficient_only dep_follow_sufficient mark_knows_if nsubj_knows_one advcl_sufficient_knows prep_knows_to dobj_mapped_which nsubjpass_mapped_level prep_level_in det_tree_the pobj_in_tree det_utterance_the dep_tree_utterance aux_mapped_should auxpass_mapped_be pcomp_to_mapped ' +o,6445,'We compared this nonprobabilistic DOP model against tile probabilistic DOP model -LRB- which estimales the most probable parse for each sentence -RRB- on three different domains : tbe Penn ATIS treebank , the Dutch OVIS treebank and tile Penn Wall Street Journal -LRB- WSJ -RRB- treebank ',Marcus,'nsubj_compared_We det_model_this amod_model_nonprobabilistic nn_model_DOP dobj_compared_model prep_compared_against num_model_tile amod_model_probabilistic nn_model_DOP pobj_against_model nsubj_estimales_which parataxis_compared_estimales det_parse_the advmod_probable_most amod_parse_probable dobj_estimales_parse prep_parse_for det_sentence_each pobj_for_sentence prep_compared_on num_domains_three amod_domains_different pobj_on_domains nn__tbe nn__Penn nn__ATIS nn__treebank nsubj_treebank_ det__the amod__Dutch nn__OVIS nn__treebank conj__ cc__and nn_Journal_tile nn_Journal_Penn nn_Journal_Wall nn_Journal_Street conj__Journal abbrev__WSJ parataxis_compared_treebank ' +o,6446,' used the averaged perceptron ',Collins,'nsubj_used_ det_perceptron_the amod_perceptron_averaged dobj_used_perceptron ' +o,6447,'However , with the algorithms proposed in , it is possible to develop a general-purpose decoder that can be used by all the parsing-based systems ',Huang,'advmod_possible_However prep_possible_with det_algorithms_the pobj_with_algorithms partmod_algorithms_proposed prep_proposed_in pobj_in_ nsubj_possible_it cop_possible_is aux_develop_to xcomp_possible_develop det_decoder_a amod_decoder_general-purpose dobj_develop_decoder nsubjpass_used_that aux_used_can auxpass_used_be rcmod_decoder_used prep_used_by predet_systems_all det_systems_the amod_systems_parsing-based pobj_by_systems ' +n,6448,'Even the creators of BLEU point out that it may not correlate particularly well with human judgment at the sentence level , a problem also noted by and ',Papineni,'advmod_point_Even det_creators_the nsubj_point_creators prep_creators_of pobj_of_BLEU dep_noted_point prt_point_out complm_correlate_that nsubj_correlate_it aux_correlate_may neg_correlate_not ccomp_point_correlate advmod_well_particularly advmod_correlate_well prep_correlate_with amod_judgment_human pobj_with_judgment prep_judgment_at det__the nn__sentence nn__level pobj_at_ det_problem_a nsubj_noted_problem advmod_noted_also prep_noted_by pobj_by_ cc__and conj__ ' +o,6449,'We also show that integrating our case prediction model improves the quality of translation according to BLEU g2 and human evaluation ',Papineni,'nsubj_show_We advmod_show_also complm_improves_that csubj_improves_integrating poss_model_our nn_model_case nn_model_prediction dobj_integrating_model ccomp_show_improves det_quality_the dobj_improves_quality prep_quality_of pobj_of_translation prep_improves_according dep_according_to amod_g2_BLEU amod_g2_ pobj_to_g2 cc_g2_and amod_evaluation_human conj_g2_evaluation ' +o,6450,'A CYK-style decoder has to rely on binarization to preprocess the grammar as did in to handle multi-nonterminal rules ',Zhang,'det_decoder_A amod_decoder_CYK-style nsubj_has_decoder aux_rely_to xcomp_has_rely prep_rely_on pobj_on_binarization aux_preprocess_to xcomp_has_preprocess det_grammar_the dobj_preprocess_grammar mark_did_as advcl_preprocess_did prep_did_in pobj_in_ aux_handle_to xcomp_did_handle amod_rules_multi-nonterminal dobj_handle_rules ' +o,6451,'The results of these studies have important applications in lexicography , to detect lexicosyntactic regularities -LRB- and Hanks , 19901 , such as , for example ~ support verbs -LRB- eg ` make-decision \' -RRB- prepositional verbs -LRB- eg ` rely-upon \' -RRB- idioms , semantic relations -LRB- eg ` part_of \' -RRB- and fixed expressions -LRB- eg ` kick the bucket \' -RRB- ',Church,'det_results_The nsubj_have_results prep_results_of det_studies_these pobj_of_studies amod_applications_important dobj_have_applications prep_applications_in pobj_in_lexicography aux_detect_to xcomp_have_detect amod_regularities_lexicosyntactic dobj_detect_regularities dep_regularities_ cc__and conj__Hanks num__19901 dep__ dep_as_such dep_regularities_as prep_detect_for nn_verbs_example nn_verbs_~ nn_verbs_support pobj_for_verbs dep_verbs_eg dep_eg_make-decision amod_verbs_prepositional dep_verbs_verbs nn_rely-upon_eg dep_verbs_rely-upon amod_relations_idioms amod_relations_semantic dep_verbs_relations nn_part_of_eg dep_relations_part_of cc_verbs_and amod_expressions_fixed conj_verbs_expressions dep_expressions_eg xcomp_eg_kick det_bucket_the dobj_kick_bucket ' +o,6452,'Following the perspective of , a minimal set of phrase blocks with lengths -LRB- m , n -RRB- where either m or n must be greater than zero results in the following types of blocks : 1 ',Brown,'det_perspective_the pobj_Following_perspective prep_perspective_of pobj_of_ det_set_a amod_set_minimal appos__set prep_set_of nn_blocks_phrase pobj_of_blocks prep_blocks_with pobj_with_lengths appos_lengths_m dep_m_n advmod_greater_where preconj_m_either nsubj_greater_m cc_m_or conj_m_n aux_greater_must cop_greater_be rcmod_lengths_greater prep_greater_than num_results_zero pobj_than_results prep_results_in det_types_the amod_types_following pobj_in_types prep_types_of pobj_of_blocks dep_blocks_1 ' +o,6453,'For a class bigram model , find : V -- + C to maximize ~ -LRB- T -RRB- = ~ I\\/L = I p -LRB- wi I -LRB- wl -RRB- -RRB- p -LRB- -LRB- wi -RRB- l -LRB- wi-1 -RRB- -RRB- -RRB- -RRB- Alternatively , perplexity or average mutual information can be used as the characteristic value for optimization ',Brown,'prep_find_For det_model_a nn_model_class nn_model_bigram pobj_For_model nsubj_~_V amod_C_+ dep_V_C aux_maximize_to infmod_C_maximize dep_find_~ nsubj_=_T ccomp_~_= nn_I\\/L_~ dobj_=_I\\/L dep_I\\/L_= iobj_=_I dobj_=_p dep_p_wi dep_wi_I appos_I_wl dep_I_p nn_l_wi dep_p_l appos_p_wi-1 advmod_p_Alternatively advmod_~_perplexity nn__ cc__or conj__average amod__mutual nn__information nsubjpass_used_ aux_used_can auxpass_used_be dep_~_used prep_used_as det_value_the amod_value_characteristic pobj_as_value prep_value_for pobj_for_optimization ' +o,6454,'O ` Hara and Wiebe make use of Penn Treebank and FrameNet to classify prepositions ',Marcus,'nn__Hara cc_Hara_and conj_Hara_Wiebe nsubj_make_ dep_O_make nsubj_classify_use prep_use_of nn__Penn nn__Treebank pobj_of_ cc__and nn__FrameNet conj__ aux_classify_to xcomp_make_classify dobj_classify_prepositions ' +n,6455,'Methods like McDonalds , including the wellknown Maximal Marginal Relevance -LRB- MMR -RRB- algorithm , are subject to another problem : Summary-level redundancy is not always well modeled by pairwise sentence-level redundancy ',Goldstein,'nsubj_subject_Methods prep_Methods_like pobj_like_McDonalds prep_Methods_including det_algorithm_the amod_algorithm_wellknown amod_algorithm_Maximal amod_algorithm_Marginal nn_algorithm_Relevance abbrev_algorithm_MMR pobj_including_algorithm partmod_algorithm_ cop_subject_are prep_subject_to det_problem_another pobj_to_problem amod_redundancy_Summary-level nsubjpass_modeled_redundancy auxpass_modeled_is neg_modeled_not advmod_modeled_always advmod_modeled_well parataxis_subject_modeled prep_modeled_by amod_redundancy_pairwise amod_redundancy_sentence-level pobj_by_redundancy ' +o,6456,'6 The Experiments We used the Penn Treebank to perform empirical experiments on the proposed parsing models ',Marcus,'num_Experiments_6 nn_Experiments_The nsubj_used_We dep_Experiments_used det__the nn__Penn nn__Treebank nsubj_perform_ aux_perform_to xcomp_used_perform amod_experiments_empirical dobj_perform_experiments prep_perform_on det_models_the amod_models_proposed nn_models_parsing pobj_on_models ' +p,6457,'Semantic collocations are harder to extract than cooccurrence patterns--the state of the art does not enable us to find semantic collocations automatically t. This paper however argues that if we take advantage of lexicai paradigmatic behavior underlying the lexicon, we can at least achieve semi-automatic extraction of semantic collocations (see also Calzolari and Bindi (1990) I But note the important work by Hindle [HindlegO] on extracting semantically similar nouns based on their substitutability in certain verb contexts.',Hindle,'' +o,6458,'For example , developed a system to identify inflammatory texts and developed methods for classifying reviews as positive or negative ',Turney,'prep_developed_For pobj_For_example nsubj_developed_ det_system_a dobj_developed_system aux_identify_to infmod_system_identify amod_texts_inflammatory dobj_identify_texts cc_identify_and conj_identify_ amod_methods_developed dobj__methods prep__for pcomp_for_classifying nsubj_positive_reviews advmod_positive_as xcomp_classifying_positive cc_positive_or conj_positive_negative ' +o,6459,'Parameters were tuned with minimum error-rate training on the NIST evaluation set of 2006 -LRB- MT06 -RRB- for both C-E and A-E ',Och,'nsubjpass_tuned_Parameters auxpass_tuned_were prep_tuned_with amod__minimum amod__error-rate nn__training pobj_with_ prep__on det_evaluation_the amod_evaluation_NIST pobj_on_evaluation partmod_evaluation_set prep_set_of num_MT06_2006 pobj_of_MT06 prep_MT06_for preconj_C-E_both pobj_for_C-E cc_C-E_and conj_C-E_A-E ' +p,6460,'Extracting semantic information from word co-occurrence statistics has been effective , particularly for sense disambiguation ',Yarowsky,'amod_information_Extracting amod_information_semantic nsubj_effective_information prep_information_from nn_statistics_word nn_statistics_co-occurrence pobj_from_statistics aux_effective_has dep_effective_been dep_for_particularly prep_effective_for nn_disambiguation_sense pobj_for_disambiguation ' +p,6461,'High correlation is reported between the BLEU score and human evaluations for translations from Arabic , Chinese , French , and Spanish to English ',Papineni,'nn_correlation_High nsubj_is_correlation dep_is_reported prep_reported_between det_score_the amod_score_BLEU pobj_between_score cc_score_and amod_evaluations_human conj_score_evaluations prep_reported_for pobj_for_translations prep_translations_from nn_French_Arabic appos_French_Chinese pobj_from_French cc_reported_and conj_reported_Spanish prep_Spanish_to pobj_to_English ' +o,6462,'We used a maximummatching algorithm and a dictionary compiled from the CTB to do segmentation , and trained a maximum entropy part-ofspeech tagger and TAG-based parser on the CTB to do tagging and parsing4 Then the same feature extraction and model-training was done for the PDN corpus as for the CTB ',Berger,'nsubj_used_We det_algorithm_a amod_algorithm_maximummatching dobj_used_algorithm cc_algorithm_and det_dictionary_a conj_algorithm_dictionary partmod_dictionary_compiled prep_compiled_from det__the nn__CTB pobj_from_ aux_do_to xcomp_compiled_do dobj_do_segmentation cc_do_and conj_do_trained det__a amod__maximum advmod_part-ofspeech_entropy amod__part-ofspeech nn__tagger dobj_trained_ cc__and amod__TAG-based nn__parser conj__ prep__on det_CTB_the pobj_on_CTB aux_do_to infmod_CTB_do dobj_do_tagging cc_tagging_and conj_tagging_parsing4 advmod_do_Then det_extraction_the amod_extraction_same nn_extraction_feature nsubjpass_done_extraction cc_extraction_and conj_extraction_model-training auxpass_done_was dep_used_done prep_done_for det_corpus_the nn_corpus_PDN pobj_for_corpus prep_done_as dep_as_for det_CTB_the pobj_for_CTB ' +o,6463,' proposed a coreference resolution approach which also explores the information from the syntactic parse trees ',Luo,'nsubj_proposed_ det_approach_a amod_approach_coreference nn_approach_resolution dobj_proposed_approach nsubj_explores_which advmod_explores_also rcmod_approach_explores det_information_the dobj_explores_information prep_explores_from det_trees_the amod_trees_syntactic amod_trees_parse pobj_from_trees ' +o,6464,'Therefore , we determine the maximal translation probability of the target word e over the source sentence words : p ibm1 -LRB- e f J 1 -RRB- = max j = 0 , , J p -LRB- e f j -RRB- -LRB- 18 -RRB- where f 0 is the empty source word ',Brown,'advmod_determine_Therefore nsubj_determine_we det_probability_the amod_probability_maximal nn_probability_translation nsubj_=_probability prep_probability_of det_word_the nn_word_target pobj_of_word dep_probability_e prep_probability_over det_words_the nn_words_source nn_words_sentence pobj_over_words nn_ibm1_p dep_words_ibm1 dep_f_e dep_-LRB-_f num_1_J dep_f_1 ccomp_determine_= dobj_=_max partmod_max_j dep_0_= ccomp_j_0 nn_p_J appos_max_p dep_f_e dep_-LRB-_f dep_f_j appos_max_18 advmod_word_where dep_where_f nsubj_word_0 cop_word_is det_word_the amod_word_empty nn_word_source rcmod_max_word ' +o,6465,'-LRB- KD1 , 2371 -RRB- 23 Reliability To evaluate the reliability of the annotation , we use the kappa coe cient -LRB- K -RRB- , which measures pairwise agreement between a set of coders making category judgements , correcting for expected chance agreement ',Carletta,'dep_evaluate_KD1 dep_KD1_2371 num_Reliability_23 nsubj_evaluate_Reliability aux_evaluate_To dep_use_evaluate det_reliability_the dobj_evaluate_reliability prep_reliability_of det_annotation_the pobj_of_annotation nsubj_use_we det__the nn__kappa nn__coe nn__cient appos__K dobj_use_ nsubj_measures_which rcmod__measures amod_agreement_pairwise dobj_measures_agreement prep_agreement_between det_set_a pobj_between_set prep_set_of pobj_of_coders partmod_coders_making nn_judgements_category dobj_making_judgements partmod_agreement_correcting prep_correcting_for amod_agreement_expected nn_agreement_chance pobj_for_agreement ' +o,6466,'Occasionally , in 59 sentences out of 2416 on section 23 of the Wall Street Journal Penn Treebank , the shift-reduce parser fails to attach a node to a head , producing a disconnected graph ',Marcus,'advmod_fails_Occasionally prep_fails_in num_sentences_59 pobj_in_sentences dep_of_out prep_sentences_of pobj_of_2416 prep_2416_on pobj_on_section num_section_23 prep_section_of det__the nn__Wall nn__Street nn__Journal nn__Penn nn__Treebank pobj_of_ det_parser_the amod_parser_shift-reduce nsubj_fails_parser aux_attach_to xcomp_fails_attach det_node_a dobj_attach_node prep_attach_to det_head_a pobj_to_head xcomp_fails_producing det_graph_a amod_graph_disconnected dobj_producing_graph ' +o,6467,'We build sentencespecific zero-cutoff stupid-backoff 5-gram language models , estimated using 47B words of English newswire text , and apply them to rescore each 10000-best list ',Brants,'nsubj_build_We amod_models_sentencespecific amod_models_zero-cutoff amod_models_stupid-backoff amod_models_ amod_models_5-gram nn_models_language dobj_build_models partmod_models_estimated xcomp_estimated_using num_words_47B dobj_using_words prep_words_of amod_text_English nn_text_newswire pobj_of_text cc_build_and conj_build_apply nsubj_rescore_them aux_rescore_to xcomp_apply_rescore det_list_each amod_list_10000-best dobj_rescore_list ' +o,6468,'The techniques examined are Structural Correspondence Learning -LRB- SCL -RRB- and Self-training ',Blitzer,'det_techniques_The nsubj_Structural_techniques partmod_techniques_examined aux_Structural_are nn__Correspondence nn__Learning abbrev__SCL dobj_Structural_ cc__and nn__Self-training conj__ ' +o,6469,'52 Evaluation Criteria For the automatic evaluation , we used the criteria from the IWSLT evaluation campaign , namely word error rate -LRB- WER -RRB- , positionindependent word error rate -LRB- PER -RRB- , and the BLEU and NIST scores ',Papineni,'num_Criteria_52 nn_Criteria_Evaluation prep_used_For det_evaluation_the amod_evaluation_automatic pobj_For_evaluation nsubj_used_we dep_Criteria_used det_criteria_the dobj_used_criteria prep_used_from det__the amod__IWSLT nn__evaluation nn__campaign pobj_from_ advmod_rate_namely nn_rate_word nn_rate_error appos__rate abbrev_rate_WER amod_rate_positionindependent nn_rate_word nn_rate_error conj_rate_rate abbrev_rate_PER cc_rate_and det_BLEU_the conj_rate_BLEU cc_BLEU_and amod_scores_NIST conj_BLEU_scores ' +o,6470,'Our process of extraction of rules as synchronous trees and then converting them to synchronous CFG rules is most similar to that of ',Galley,'poss_process_Our nsubj_similar_process prep_process_of pobj_of_extraction prep_extraction_of pobj_of_rules prep_rules_as amod_trees_synchronous pobj_as_trees cc_trees_and advmod_converting_then conj_trees_converting dobj_converting_them prep_converting_to amod_rules_synchronous nn_rules_CFG pobj_to_rules cop_similar_is advmod_similar_most prep_similar_to pobj_to_that prep_that_of ' +o,6471,'SGD was recently used for NLP tasks including machine translation and syntactic parsing ',Tillmann,'nsubjpass_used_SGD auxpass_used_was advmod_used_recently prep_used_for nn_tasks_NLP pobj_for_tasks prep_tasks_including nn__machine nn__translation pobj_including_ cc__and amod__syntactic amod__parsing conj__ ' +o,6472,'22 STT : A Statistical Tree-based Tagger The aim of statistical or probabilistic tagging is to assign the most likely sequence of tags given the observed sequence of words ',Cutting,'num_STT_22 dep_is_STT det_Tagger_A nn_Tagger_Statistical nn_Tagger_Tree-based nsubj_is_Tagger det_aim_The dep_Tagger_aim prep_Tagger_of amod__statistical cc_statistical_or conj_statistical_probabilistic nn__tagging pobj_of_ aux_assign_to xcomp_is_assign det_sequence_the advmod_likely_most amod_sequence_likely dobj_assign_sequence prep_sequence_of pobj_of_tags prep_assign_given det_sequence_the amod_sequence_observed dep_given_sequence prep_sequence_of pobj_of_words ' +n,6473,'While other systems , such as , have addressed these tasks to some degree , OPINE is the first to report results ',Turney,'mark_addressed_While amod_systems_other nsubj_addressed_systems dep_as_such prep_systems_as pobj_as_ aux_addressed_have advcl_first_addressed det_tasks_these dobj_addressed_tasks prep_addressed_to det_degree_some pobj_to_degree nsubj_first_OPINE cop_first_is det_first_the aux_report_to dep_first_report dobj_report_results ' +o,6474,'The NIST MT03 test set is used for development , particularly for optimizing the interpolation weights using Minimum Error Rate training ',Och,'det_set_The nn_set_NIST nn_set_MT03 nn_set_test nsubjpass_used_set auxpass_used_is prep_used_for pobj_for_development dep_for_particularly prep_used_for pcomp_for_optimizing det_weights_the nn_weights_interpolation dobj_optimizing_weights xcomp_optimizing_using nn_training_Minimum nn_training_Error nn_training_Rate dobj_using_training ' +o,6475,'For instance , all automatically acquire large TAGs for English from the Penn Treebank ',Marcus,'prep__For nn__instance pobj_For_ nsubj__ advmod__all advmod__automatically dep__acquire amod_TAGs_large dobj_acquire_TAGs prep_TAGs_for pobj_for_English prep_acquire_from det_Treebank_the nn_Treebank_Penn pobj_from_Treebank ' +o,6476,'Toward a Task-based Gold Standard for Evaluation of NP Chunks and Technical Terms Nina Wacholder Rutgers University nina @ scilsrutgersedu Peng Song Rutgers University psong @ paulrutgersedu Abstract We propose a gold standard for evaluating two types of information extraction output - noun phrase -LRB- NP -RRB- chunks and technical terms ',Ramshaw,'dep_Standard_Toward det_Standard_a amod_Standard_Task-based nn_Standard_Gold prep_Standard_for pobj_for_Evaluation prep_Evaluation_of nn_Chunks_NP nn_nina_Chunks cc_Chunks_and nn_Terms_Technical conj_Chunks_Terms nn_nina_Nina nn_nina_Wacholder nn_nina_Rutgers nn_nina_University pobj_of_nina dep_University_@ nn_University_scilsrutgersedu nn_University_Peng nn_University_Song nn_University_Rutgers dep_Standard_University partmod_Standard_psong dep_Abstract_@ nn_Abstract_paulrutgersedu dep_psong_Abstract nsubj_propose_We rcmod_Abstract_propose det_standard_a amod_standard_gold dobj_propose_standard prep_propose_for pcomp_for_evaluating num_types_two dobj_evaluating_types prep_types_of nn_output_information nn_output_extraction pobj_of_output amod_chunks_noun nn_chunks_phrase abbrev_chunks_NP dep_Abstract_chunks amod__ cc__and conj__technical nn__terms dep_chunks_ ' +o,6477,'The relationship between the translation model and the alignment model is given by : Pr -LRB- fJ1 jeI1 -RRB- = X aJ1 Pr -LRB- fJ1 ; aJ1jeI1 -RRB- -LRB- 3 -RRB- In this paper , we use the models IBM-1 , IBM4 from and the HiddenMarkovalignmentmodel -LRB- HMM -RRB- from ',Brown,'det_relationship_The nsubjpass_given_relationship prep_relationship_between det_model_the nn_model_translation pobj_between_model cc_model_and det_model_the amod_model_alignment conj_model_model auxpass_given_is prep_given_by nsubj_=_Pr amod_jeI1_fJ1 appos_Pr_jeI1 dep_by_= nn_Pr_X nn_Pr_aJ1 dobj_=_Pr nn_aJ1jeI1_fJ1 appos_Pr_aJ1jeI1 appos_Pr_3 prep_use_In det_paper_this pobj_In_paper nsubj_use_we dep_given_use det_IBM-1_the nn_IBM-1_models dobj_use_IBM-1 conj_IBM-1_IBM4 prep_IBM4_from pobj_from_ cc_IBM-1_and det_HiddenMarkovalignmentmodel_the conj_IBM-1_HiddenMarkovalignmentmodel abbrev_HiddenMarkovalignmentmodel_HMM prep_HiddenMarkovalignmentmodel_from pobj_from_ ' +o,6478,'Based on this theoretical cornerstone , presented a PCFG-based chart generator using wide-coverage LFG approximations automatically extracted from the Penn-II treebank ',Cahill,'prep_presented_Based dep_Based_on det_cornerstone_this amod_cornerstone_theoretical pobj_on_cornerstone nsubj_presented_ det_generator_a amod_generator_PCFG-based nn_generator_chart dobj_presented_generator xcomp_presented_using amod_approximations_wide-coverage nn_approximations_LFG dobj_using_approximations advmod_extracted_automatically partmod_approximations_extracted prep_extracted_from det_treebank_the amod_treebank_Penn-II pobj_from_treebank ' +n,6479,'Thirdly , deploys the dependency language model to augment the lexical language model probability be1183 tween two head words but never seek a full dependency graph ',Shen,'advmod_deploys_Thirdly nsubj_deploys_ det_model_the amod_model_dependency nn_model_language dobj_deploys_model aux_augment_to xcomp_deploys_augment det_be1183_the amod_be1183_lexical nn_be1183_language nn_be1183_model nn_be1183_probability nsubj_tween_be1183 ccomp_augment_tween num_words_two nn_words_head dobj_tween_words cc_tween_but neg_seek_never conj_tween_seek det_graph_a amod_graph_full amod_graph_dependency dobj_seek_graph ' +o,6480,'Many grammars , such as finite-state grammars -LRB- FSG -RRB- , bracket\\/inversion transduction grammars -LRB- BTG\\/ITG -RRB- , context-free grammar -LRB- CFG -RRB- , tree substitution grammar -LRB- TSG -RRB- and their synchronous versions , have been explored in SMT ',Wu,'amod_grammars_Many nsubjpass_explored_grammars dep_as_such prep_grammars_as amod_grammars_finite-state pobj_as_grammars abbrev_grammars_FSG amod__bracket\\/inversion nn__transduction nn__grammars abbrev__BTG\\/ITG appos_grammars_ amod_grammar_context-free appos_grammars_grammar abbrev_grammar_CFG nn_grammar_tree nn_grammar_substitution measure__grammar abbrev_grammar_TSG amod_grammars_ cc_grammars_and poss_versions_their amod_versions_synchronous conj_grammars_versions aux_explored_have auxpass_explored_been prep_explored_in pobj_in_SMT ' +o,6481,'The problem is that with such a definition of collocations , even when improved , one identifies not only collocations but freecombining pairs frequently appearing together such as lawyer-client ; doctor-hospital , as pointed out by ',Smadja,'det_problem_The nsubj_is_problem complm_identifies_that prep_identifies_with predet_definition_such det_definition_a pobj_with_definition prep_definition_of pobj_of_collocations advmod_when_even advmod_improved_when dep_collocations_improved nsubj_identifies_one ccomp_is_identifies cc_collocations_not dep_not_only dobj_identifies_collocations cc_collocations_but amod_pairs_freecombining conj_collocations_pairs advmod_appearing_frequently partmod_pairs_appearing prt_appearing_together dep_as_such prep_appearing_as pobj_as_lawyer-client advmod_lawyer-client_doctor-hospital mark_pointed_as dep_lawyer-client_pointed prt_pointed_out prep_pointed_by ' +o,6482,'The parsing algorithm was CKY-style parsing with beam thresholding , which was similar to ones used in ',Berger,'det_algorithm_The amod_algorithm_parsing nsubj_parsing_algorithm cop_parsing_was amod_parsing_CKY-style prep_parsing_with nn_thresholding_beam pobj_with_thresholding nsubj_similar_which cop_similar_was rcmod_thresholding_similar prep_similar_to pobj_to_ones partmod_ones_used prep_used_in pobj_in_ ' +o,6483,'The experiments were performed using the Wall Street Journal -LRB- WSJ -RRB- corpus of the University of Pennsylvania modified as described in and ',Marcus,'det_experiments_The nsubjpass_performed_experiments auxpass_performed_were xcomp_performed_using det_corpus_the nn_Journal_Wall nn_Journal_Street nn_corpus_Journal abbrev_Journal_WSJ dobj_using_corpus prep_corpus_of det_University_the pobj_of_University prep_corpus_of nn__Pennsylvania pobj_of_ partmod__modified mark_described_as advcl_modified_described prep_described_in pobj_in_ cc__and conj__ ' +o,6484,'The approach is related , but not identical , to distributional similarity -LRB- for details , see and -RRB- ',Brown,'det_approach_The nsubj_related_approach cop_related_is dep_not_but cc_related_not conj_related_identical prep_identical_to amod_similarity_distributional pobj_to_similarity dep_related_for pobj_for_details dep_details_see acomp_see_ cc__and conj__ ' +o,6485,'While the BBN model does not perform at the level of Model 2 of on Wall Street Journal text , it is also less language-dependent , eschewing the distance metric -LRB- which relied on specific features of the English Treebank -RRB- in favor of the ` bigrams on nonterminals \' model ',Collins,'mark_perform_While det_model_the nn_model_BBN nsubj_perform_model aux_perform_does neg_perform_not advcl_language-dependent_perform prep_perform_at det_level_the pobj_at_level prep_level_of pobj_of_Model num_Model_2 prep_level_of pobj_of_ prep_perform_on nn_text_Wall nn_text_Street nn_text_Journal pobj_on_text nsubj_language-dependent_it cop_language-dependent_is advmod_language-dependent_also advmod_language-dependent_less xcomp_language-dependent_eschewing det_metric_the nn_metric_distance dobj_eschewing_metric nsubj_relied_which dep_metric_relied prep_relied_on amod_features_specific pobj_on_features prep_features_of det_Treebank_the nn_Treebank_English pobj_of_Treebank prep_metric_in pobj_in_favor prep_favor_of det_bigrams_the pobj_of_bigrams prep_bigrams_on poss_model_nonterminals pobj_on_model ' +o,6486,'Maximum Entropy Modeling -LRB- MaxEnt -RRB- and Support Vector Machine -LRB- SVM -RRB- were used to build the classifiers in our solution ',Berger,'nn_Modeling_Maximum nn_Modeling_Entropy appos_Modeling_MaxEnt amod__ cc__and conj__Support nn__Vector nn__Machine abbrev__SVM nsubjpass_used_ auxpass_used_were dep_Modeling_used aux_build_to xcomp_used_build det_classifiers_the dobj_build_classifiers prep_build_in poss_solution_our pobj_in_solution ' +o,6487,'They can be seen as extensions of the simpler IBM models 1 and 2 ',Brown,'nsubjpass_seen_They aux_seen_can auxpass_seen_be prep_seen_as pobj_as_extensions prep_extensions_of det_models_the amod_models_simpler nn_models_IBM pobj_of_models dobj_seen_1 cc_1_and conj_1_2 ' +o,6488,'Previous authors have used numerous HMM-based models and other types of networks including maximum entropy models , conditional Markov models , conditional random elds -LRB- CRF -RRB- , and cyclic dependency networks ',Ratnaparkhi,'amod_authors_Previous nsubj_used_authors aux_used_have amod_models_numerous amod_models_HMM-based dobj_used_models amod_types_ cc__and conj__other nsubj__types prep_types_of pobj_of_networks prep_networks_including amod_models_maximum nn_models_entropy pobj_including_models rcmod_models_ amod_models_conditional nn_models_Markov dep__models ccomp_used_ amod_elds_conditional amod_elds_random nsubj__elds abbrev_elds_CRF ccomp_used_ cc_used_and amod_networks_cyclic nn_networks_dependency nsubj__networks conj_used_ ' +o,6489,'Unlike , in the shared task we used only the simplest feed-forward approximation , which replicates the computation of a neural network of the type proposed in ',Titov,'prep_used_Unlike pobj_Unlike_ prep__in det_task_the amod_task_shared pobj_in_task nsubj_used_we advmod_approximation_only det_approximation_the amod_approximation_simplest amod_approximation_feed-forward dobj_used_approximation nsubj_replicates_which rcmod_approximation_replicates det_computation_the dobj_replicates_computation prep_computation_of det_network_a amod_network_neural pobj_of_network prep_network_of det_type_the pobj_of_type partmod_type_proposed prep_proposed_in pobj_in_ ' +o,6490,' used a corpus-based algorithm ',Turney,'nsubj_used_ det_algorithm_a amod_algorithm_corpus-based dobj_used_algorithm ' +o,6491,'More recently , EM has been used to learn hidden variables in parse trees ; these can be head-childannotations , latent head features , or hierarchicallysplit nonterminal states ',Matsuzaki,'advmod_recently_More advmod_used_recently nsubjpass_used_EM aux_used_has auxpass_used_been aux_learn_to xcomp_used_learn amod_variables_hidden dobj_learn_variables prep_learn_in amod_trees_parse pobj_in_trees nsubj__these aux__can cop__be amod__head-childannotations parataxis_used_ amod_features_latent nn_features_head nsubj__features ccomp_used_ cc_used_or amod_states_hierarchicallysplit amod_states_nonterminal nsubj__states conj_used_ ' +o,6492,'33 CRFs and Perceptron Learning Perceptron training for conditional models is an approximation to the SGD algorithm , using feature counts from the Viterbi label sequence in lieu of expected feature counts ',Collins,'num_CRFs_33 nsubj_training_CRFs cc_CRFs_and nn_Perceptron_Perceptron nn_Perceptron_Learning conj_CRFs_Perceptron prep_training_for amod_models_conditional pobj_for_models nsubj_approximation_ cop_approximation_is det_approximation_an dep_training_approximation prep_approximation_to det_algorithm_the nn_algorithm_SGD pobj_to_algorithm partmod_approximation_using nn_counts_feature dobj_using_counts prep_using_from det_sequence_the nn_sequence_Viterbi nn_sequence_label pobj_from_sequence prep_sequence_in pobj_in_lieu prep_lieu_of amod_counts_expected nn_counts_feature pobj_of_counts ' +p,6493,'The maximum entropy approach is known to be well suited to solve the classification problem ',Berger,'det__The amod__maximum amod__entropy nn__approach nsubjpass_known_ auxpass_known_is aux_suited_to auxpass_suited_be advmod_suited_well xcomp_known_suited aux_solve_to purpcl_suited_solve det_problem_the nn_problem_classification dobj_solve_problem ' +p,6494,'We compared a baseline system , the state-of-the-art phrase-based system Pharaoh , against our system ',Koehn,'dep_We_compared det_system_a amod_system_baseline dep_compared_system det_system_the amod_system_state-of-the-art amod_system_phrase-based appos_system_system nn__Pharaoh dep_system_ prep_system_against poss_system_our pobj_against_system ' +o,6495,'In none of these cases did we repeat minimum-error-rate training ; all these systems were trained using max-B The metrics we tested were : METEOR , version 06,usingtheexact,Porter-stemmer , andWordNet synonmy stages , and the optimized parameters = 081 , = 083 , = 028 as reported in ',Banerjee,'prep_did_In pobj_In_none prep_none_of det_cases_these pobj_of_cases nsubj_repeat_we ccomp_did_repeat amod_training_minimum-error-rate dobj_repeat_training predet_systems_all det_systems_these nsubjpass_trained_systems auxpass_trained_were dep_training_trained xcomp_trained_using dobj_using_max-B det_metrics_The nsubj_did_metrics nsubj_tested_we rcmod_metrics_tested ccomp_tested_were nn__METEOR dobj_were_ nn_06,usingtheexact,Porter-stemmer_version conj__06,usingtheexact,Porter-stemmer amod_stages_andWordNet nn_stages_synonmy conj__stages cc__and det_parameters_the amod_parameters_optimized conj__parameters dep_081_= rcmod_parameters_081 dep_were_= dobj_=_083 dep_were_= dobj_=_028 mark_reported_as advcl_were_reported prep_reported_in pobj_in_ ' +o,6496,'Each item is associated with a stack whose signa12Specifically a B-hypergraph , equivalent to an and-or graph or context-free grammar ',Brown,'det_item_Each auxpass_associated_is rcmod_item_associated prep_associated_with det_stack_a pobj_with_stack poss_signa12Specifically_whose dep_stack_signa12Specifically dep_B-hypergraph_a dep_item_B-hypergraph dep_B-hypergraph_equivalent prep_equivalent_to det_graph_an amod_graph_and-or pobj_to_graph dep_B-hypergraph_ cc_B-hypergraph_or amod__context-free nn__grammar conj_B-hypergraph_ ' +o,6497,'Under the maximum entropy framework , evidence from different features can be combined with no assumptions of feature independence ',Berger,'prep_combined_Under det__the amod__maximum amod__entropy nn__framework pobj_Under_ nsubjpass_combined_evidence prep_evidence_from amod_features_different pobj_from_features aux_combined_can auxpass_combined_be prep_combined_with det_assumptions_no pobj_with_assumptions prep_assumptions_of nn_independence_feature pobj_of_independence ' +o,6498,'Paraphrases can also be automatically acquired using statistical methods as shown by ',Barzilay,'nsubjpass_acquired_Paraphrases aux_acquired_can advmod_acquired_also auxpass_acquired_be advmod_acquired_automatically purpcl_acquired_using amod_methods_statistical dobj_using_methods mark_shown_as advcl_using_shown prep_shown_by ' +o,6499,'The formally syntax-based model for SMT was first advocated by ',Wu,'det_model_The advmod_syntax-based_formally amod_model_syntax-based nsubjpass_advocated_model prep_model_for pobj_for_SMT auxpass_advocated_was advmod_advocated_first prep_advocated_by ' +o,6500,'Therefore , the results are more informative than a simple agreement average ',Carletta,'advmod_informative_Therefore det_results_the nsubj_informative_results cop_informative_are advmod_informative_more prep_informative_than det_average_a amod_average_simple nn_average_agreement pobj_than_average ' +o,6501,'in that order ',Banerjee,'det_order_that pobj_in_order ' +o,6502,'Aggregate models based on higher-order n-grams might be able to capture multi-word structures such as noun phrases ',Brown,'amod_models_Aggregate nsubj_able_models prep_models_based dep_based_on amod__higher-order amod__n-grams pobj_on_ aux_able_might cop_able_be aux_capture_to xcomp_able_capture amod_structures_multi-word dobj_capture_structures dep_as_such prep_structures_as nn_phrases_noun pobj_as_phrases ' +o,6503,'Feature function scaling factors m are optimized based on a maximum likelihood approach or on a direct error minimization approach ',Och,'nsubj_function_Feature amod_m_scaling nn_m_factors nsubjpass_optimized_m auxpass_optimized_are ccomp_function_optimized prep_optimized_based dep_based_on det__a amod__maximum nn__likelihood nn__approach pobj_on_ cc_on_or conj_on_on det_approach_a amod_approach_direct nn_approach_error nn_approach_minimization pobj_on_approach ' +o,6504,'(Brown et al. , 1993; Vogel et al. , 1996; Garca-Varea et al. , 2002; Ahrenberg et al. , 1998; Tiedemann, 1999; Tufis and Barbu, 2002; Melamed, 2000).',Brown,'' +o,6505,'Instead of using a single system output as the skeleton , we employ a minimum Bayes-risk decoder to select the best single system output from the merged N-best list by minimizing the BLEU loss ',Papineni,'dep_of_Instead advcl_employ_of pcomp_of_using det_output_a amod_output_single nn_output_system dobj_using_output prep_using_as det_skeleton_the pobj_as_skeleton nsubj_employ_we det_decoder_a amod_decoder_minimum nn_decoder_Bayes-risk nsubj_select_decoder aux_select_to xcomp_employ_select det_output_the amod_output_best amod_output_single nn_output_system dobj_select_output prep_output_from det_list_the amod_list_merged amod_list_N-best pobj_from_list prep_select_by pcomp_by_minimizing det_loss_the amod_loss_BLEU amod_loss_ dobj_minimizing_loss ' +o,6506,'(He et al., 2008).',He,'' +o,6507,'Dirichlet priors can be used to bias HMMs toward more skewed distributions , which is especially useful in the weakly supervised setting consideredhere ',Johnson,'amod_priors_Dirichlet nsubjpass_used_priors aux_used_can auxpass_used_be prep_used_to nn_HMMs_bias pobj_to_HMMs prep_used_toward advmod_skewed_more amod__skewed nn__distributions pobj_toward_ nsubj_useful_which cop_useful_is advmod_useful_especially rcmod__useful prep_useful_in det_consideredhere_the advmod_supervised_weakly amod_consideredhere_supervised nn_consideredhere_setting pobj_in_consideredhere ' +o,6508,'Here we used the averaged perceptron , where the weight matrix used to classify the test data is the average of all of the matrices posited during training , ie , a1 a62 a52 a49 a62 a49 a42a51a50a53a52 a1 a42 42 Multicomponent architecture Task specific and external training data are integrated with a two-component perceptron ',Collins,'advmod__Here nsubj__we dep__used det__the amod__averaged nn__perceptron advmod_average_where det_matrix_the nn_matrix_weight nsubj_average_matrix partmod_matrix_used aux_classify_to xcomp_used_classify det_data_the nn_data_test dobj_classify_data cop_average_is det_average_the advcl__average prep_average_of pobj_of_all prep_all_of det_matrices_the pobj_of_matrices partmod_matrices_posited prep_posited_during nn_a49_training dep_a49_ie amod_a49_a1 amod_a49_a62 nn_a49_a52 pobj_during_a49 amod_a1_a62 amod_a1_a49 amod_a1_a42a51a50a53a52 nsubjpass_integrated_a1 partmod_a1_a42 dep_specific_42 dep_specific_Multicomponent dep_specific_architecture dep_specific_Task dobj_a42_specific cc_specific_and amod_data_external nn_data_training conj_specific_data auxpass_integrated_are dep__integrated prep_integrated_with det_perceptron_a amod_perceptron_two-component pobj_with_perceptron ' +o,6509,'The polarity value proposed by is as follows ',Turney,'det_value_The amod_value_polarity nsubj_is_value partmod_value_proposed prep_proposed_by pobj_by_ mark_follows_as advcl_is_follows ' +o,6510,'The percentage agreement for each of the features is shown in the following table : feature percent agreement form 100 \% intentionality 749 \% awareness 935 \% safety 907 \% As advocated by , we have used the Kappa coefficient as a measure of coder agreement ',Carletta,'det_agreement_The nn_agreement_percentage nsubjpass_shown_agreement prep_agreement_for pobj_for_each prep_each_of det_features_the pobj_of_features auxpass_shown_is prep_shown_in det_table_the amod_table_following pobj_in_table nn_form_feature nn_form_percent nn_form_agreement parataxis_shown_form dep_\%_100 amod_form_\% advmod_form_intentionality num_\%_749 dep_safety_\% dep_\%_awareness num_\%_935 dep_safety_\% dep_used_safety num_\%_907 dep_safety_\% mark_advocated_As advcl_used_advocated prep_advocated_by pobj_by_ nsubj_used_we aux_used_have dep_intentionality_used det__the nn__Kappa nn__coefficient dobj_used_ prep_used_as det_measure_a pobj_as_measure prep_measure_of nn_agreement_coder pobj_of_agreement ' +o,6511,'To regularize the model we take as the final model the average of all weight vectors posited during training ',Collins,'aux_regularize_To dep_take_regularize det_model_the dobj_regularize_model nsubj_take_we prep_take_as det_model_the amod_model_final pobj_as_model det_average_the nsubj_posited_average prep_average_of det_vectors_all nn_vectors_weight pobj_of_vectors dep_take_posited prep_posited_during pobj_during_training ' +o,6512,'3 Bilingual Task : An Application for Word Alignment 31 Sentence and word alignment Bilingual alignment methods ',Brown,'nn_Task_Bilingual dep_3_Task det_Application_An dep_3_Application prep_Application_for nn_Sentence_Word nn_Sentence_Alignment num_Sentence_31 pobj_for_Sentence cc_Sentence_and conj_Sentence_word amod_methods_alignment nn_methods_Bilingual nn_methods_alignment dep_word_methods ' +o,6513,'The first is identifying words and phrases that are associated with subjectivity , for example , that think is associated with private states and that beautiful is associated with positive sentiments -LRB- eg , -RRB- ',Turney,'det_first_The nsubj_identifying_first aux_identifying_is dobj_identifying_words cc_words_and conj_words_phrases nsubjpass_associated_that auxpass_associated_are rcmod_words_associated prep_associated_with pobj_with_subjectivity prep_associated_for pobj_for_example nsubj_think_that rcmod_example_think auxpass_associated_is dep_think_associated prep_associated_with amod_states_private pobj_with_states cc_associated_and dep_associated_that nsubjpass_associated_beautiful auxpass_associated_is conj_associated_associated prep_associated_with amod_sentiments_positive pobj_with_sentiments appos_sentiments_eg dep_eg_ ' +n,6514,'METEOR was chosen since , unlike the more commonly used BLEU metric , it provides reasonably reliable scores for individual sentences ',Papineni,'nsubjpass_chosen_METEOR auxpass_chosen_was mark_provides_since prep_provides_unlike det__the advmod_used_more advmod_used_commonly amod__used amod__BLEU amod__metric pobj_unlike_ nsubj_provides_it advcl_chosen_provides advmod_reliable_reasonably amod_scores_reliable dobj_provides_scores prep_scores_for amod_sentences_individual pobj_for_sentences ' +o,6515,'The same probabilities are also included using 50 hard word classes derived from the parallel corpus using the GIZA + + mkcls utility ',Och,'det_probabilities_The amod_probabilities_same nsubjpass_included_probabilities auxpass_included_are advmod_included_also xcomp_included_using num_classes_50 amod_classes_hard nn_classes_word dobj_using_classes partmod_classes_derived prep_derived_from det_corpus_the amod_corpus_parallel pobj_from_corpus partmod_corpus_using det_utility_the amod_utility_GIZA cc_GIZA_+ conj_GIZA_+ nn_utility_mkcls dobj_using_utility ' +o,6516,'Many methods for calculating the similarity have been proposed ',Och,'amod_methods_Many nsubjpass_proposed_methods prep_methods_for pcomp_for_calculating det_similarity_the dobj_calculating_similarity aux_proposed_have auxpass_proposed_been ' +o,6517,'Approaches include word substitution systems , phrase substitution systems , and synchronous context-free grammar systems , all of which train on string pairs and seek to establish connections between source and target strings ',Koehn,'nsubj_include_Approaches nn_systems_word nn_systems_substitution nsubj__systems xcomp_include_ nn_systems_phrase nn_systems_substitution nsubj__systems ccomp_include_ cc_include_and amod_systems_synchronous amod_systems_context-free nn_systems_grammar nsubj__systems conj_include_ dep_of_all nsubj_train_of pobj_of_which ccomp__train prep_train_on nn_pairs_string pobj_on_pairs cc_train_and conj_train_seek aux_establish_to xcomp_seek_establish dobj_establish_connections prep_connections_between nn_strings_source cc_source_and conj_source_target pobj_between_strings ' +n,6518,'Our method is a natural extension of those proposed in and , and overcomes their drawbacks while retaining their advantages ',Brown,'poss_method_Our nsubj_extension_method cop_extension_is det_extension_a amod_extension_natural prep_extension_of pobj_of_those dep_those_proposed prep_proposed_in pobj_in_ cc__and conj__ cc_proposed_and conj_proposed_overcomes poss_drawbacks_their dobj_overcomes_drawbacks prep_overcomes_while pcomp_while_retaining poss_advantages_their dobj_retaining_advantages ' +n,6519,'Another consequence of not generating posthead conjunctions and punctuation as first-class words is that they 19 In fact , if punctuation occurs before the head , it is not generated at alla deficiency in the parsing model that appears to be a holdover from the deficient punctuation handling in the model of ',Collins,'det_consequence_Another nsubj_is_consequence prep_consequence_of neg_generating_not pcomp_of_generating amod_conjunctions_posthead dobj_generating_conjunctions cc_conjunctions_and conj_conjunctions_punctuation prep_generating_as amod_words_first-class pobj_as_words complm_generated_that nsubjpass_generated_they dep_occurs_19 dep_occurs_In pobj_In_fact mark_occurs_if nsubj_occurs_punctuation parataxis_generated_occurs prep_occurs_before det_head_the pobj_before_head nsubjpass_generated_it auxpass_generated_is neg_generated_not ccomp_is_generated prep_generated_at nn_deficiency_alla pobj_at_deficiency prep_generated_in det_model_the amod_model_parsing pobj_in_model nsubj_appears_that rcmod_model_appears aux_holdover_to cop_holdover_be det_holdover_a xcomp_appears_holdover prep_holdover_from det_punctuation_the amod_punctuation_deficient pobj_from_punctuation partmod_punctuation_handling prep_handling_in det_model_the pobj_in_model prep_model_of ' +o,6520,'Next , we learn our polarity classifier using positive and negative reviews taken from two movie 611 review datasets , one assembled by and the other by ourselves ',Pang,'advmod_learn_Next nsubj_learn_we poss_classifier_our amod_classifier_polarity dobj_learn_classifier xcomp_learn_using amod_reviews_positive cc_positive_and conj_positive_negative dobj_using_reviews partmod_reviews_taken prep_taken_from num_movie_two pobj_from_movie num_datasets_611 nn_datasets_review dep_movie_datasets conj_reviews_one partmod_one_assembled prep_assembled_by pobj_by_ cc_reviews_and det_other_the conj_reviews_other prep_other_by pobj_by_ourselves ' +o,6521,'Sentiment classification is a well studied problem and in many domains users explicitly 1We use the term aspect to denote properties of an object that can be rated by a user as in Snyder and Barzilay ',Turney,'nn_classification_Sentiment nsubj__classification cop__is det__a advmod_studied_well amod__studied nn__problem cc__and conj__in amod_users_many nn_users_domains pobj_in_users advmod_use_explicitly nsubj_use_1We dep__use det_aspect_the nn_aspect_term nsubj_denote_aspect aux_denote_to xcomp_use_denote dobj_denote_properties prep_properties_of det_object_an pobj_of_object nsubjpass_rated_that aux_rated_can auxpass_rated_be rcmod_object_rated prep_rated_by det_user_a pobj_by_user dep_in_as prep_user_in nn__Snyder cc_Snyder_and conj_Snyder_Barzilay pobj_in_ ' +o,6522,'In contrast , semi-supervised domain adaptation is the scenario in which , in addition to the labeled source data , we only have unlabeled and no labeled target domain data ',McClosky,'prep_scenario_In pobj_In_contrast amod__semi-supervised nn__domain nn__adaptation nsubj_scenario_ cop_scenario_is det_scenario_the rel_labeled_in dep_in_which pobj_in_in pobj_in_addition prep_addition_to det_data_the amod_data_labeled nn_data_source pobj_to_data nsubj_labeled_we advmod_labeled_only aux_labeled_have advmod_labeled_unlabeled cc_unlabeled_and conj_unlabeled_no rcmod_scenario_labeled nn_data_target nn_data_domain dobj_labeled_data ' +o,6523,'After that , we used three types of methods for performing a symmetrization of IBM models : intersection , union , and refined methods ',Och,'prep_used_After pobj_After_that nsubj_used_we num_types_three dobj_used_types prep_types_of pobj_of_methods prep_used_for pcomp_for_performing det_symmetrization_a dobj_performing_symmetrization prep_symmetrization_of nn_models_IBM pobj_of_models dep_symmetrization_intersection conj_intersection_union cc_intersection_and amod_methods_refined conj_intersection_methods ' +o,6524,'Several automatic sentence alignment approaches have been proposed based on sentence length and lexical information ',Brown,'amod_approaches_Several amod_approaches_automatic nn_approaches_sentence nn_approaches_alignment nsubjpass_proposed_approaches aux_proposed_have auxpass_proposed_been prep_proposed_based dep_based_on nn__sentence nn__length pobj_on_ cc__and amod__lexical nn__information conj__ ' +o,6525,'These parameters 1 8 are tuned by minimum error rate training on the dev sets ',Och,'det_8_These amod_8_parameters number_8_1 nsubjpass_tuned_8 auxpass_tuned_are prep_tuned_by amod__minimum nn__error nn__rate nn__training pobj_by_ prep__on det_sets_the nn_sets_dev pobj_on_sets ' +o,6526,'However , another approach is to train a separate out-of-domain parser , and use this to generate additional features on the supervised and unsupervised in-domain data ',Blitzer,'advmod_is_However det_approach_another nsubj_is_approach aux_train_to xcomp_is_train det_parser_a amod_parser_separate amod_parser_out-of-domain dobj_train_parser cc_train_and conj_train_use nsubj_generate_this aux_generate_to xcomp_use_generate amod_features_additional dobj_generate_features prep_generate_on det_data_the amod_data_supervised cc_supervised_and conj_supervised_unsupervised amod_data_in-domain pobj_on_data ' +o,6527,'To extract such word clusters we used suffix arrays proposed in Yamamoto and and the pointwise mutual information measure , see Church and Hanks -LRB- 1990 -RRB- ',Church,'aux_extract_To dep_see_extract amod_clusters_such nn_clusters_word dobj_extract_clusters nsubj_used_we rcmod_clusters_used nn_arrays_suffix nsubj_proposed_arrays ccomp_used_proposed prep_proposed_in pobj_in_Yamamoto cc_Yamamoto_and conj_Yamamoto_ cc_Yamamoto_and det_measure_the amod_measure_pointwise amod_measure_mutual nn_measure_information conj_Yamamoto_measure nsubj_see_Church cc_Church_and conj_Church_Hanks appos_Church_1990 ' +o,6528,'A third of this is syntactically parsed as part of the Penn Treebank and has dialog act annotation ',Marcus,'det_third_A nsubj_parsed_third prep_third_of pobj_of_this cop_parsed_is advmod_parsed_syntactically prep_parsed_as pobj_as_part prep_part_of det__the nn__Penn nn__Treebank pobj_of_ cc_parsed_and conj_parsed_has amod__dialog nn__act nn__annotation dobj_has_ ' +n,6529,'One conclusion that we can draw is that at present the additional word features used in looking at words more than one position away from the current do not appear to be helping the overall performance of the models ',Ratnaparkhi,'num_conclusion_One nsubj_is_conclusion complm_draw_that nsubj_draw_we aux_draw_can dep_conclusion_draw complm_appear_that prep_appear_at pobj_at_present det_features_the amod_features_additional nn_features_word nsubj_appear_features partmod_features_used prep_used_in pobj_in_ partmod__looking prep_looking_at pobj_at_words dep_than_more quantmod_one_than num_position_one dep_away_position advmod_words_away dep_away_from det_current_the pobj_from_current aux_appear_do neg_appear_not ccomp_is_appear aux_helping_to aux_helping_be xcomp_appear_helping det_performance_the amod_performance_overall dobj_helping_performance prep_performance_of det_models_the pobj_of_models ' +o,6530,'To overcome these limitations , many syntaxbased SMT models have been proposed ',Wu,'aux_overcome_To dep_proposed_overcome det_limitations_these dobj_overcome_limitations amod_models_many amod_models_syntaxbased nn_models_SMT nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been ' +o,6531,'Our test set is 3718 sentences from the English Penn treebank which were translated into German ',Marcus,'poss_set_Our nn_set_test nsubj_sentences_set cop_sentences_is num_sentences_3718 prep_sentences_from det__the nn__English nn__Penn nn__treebank pobj_from_ nsubjpass_translated_which auxpass_translated_were rcmod__translated prep_translated_into pobj_into_German ' +o,6532,'The third exploits automatic subjectivity analysis in applications such as review classification -LRB- eg , -RRB- , mining texts for product reviews -LRB- eg , -RRB- , summarization -LRB- eg , -RRB- , information extraction -LRB- eg , -RRB- , 1Note that sentiment , the focus of much recent work in the area , is a type of subjectivity , specifically involving positive or negative opinion , emotion , or evaluation ',Turney,'det_exploits_The amod_exploits_third amod_analysis_automatic nn_analysis_subjectivity dep_mining_analysis prep_analysis_in pobj_in_applications dep_as_such prep_applications_as pobj_as_review dep_mining_classification appos_classification_eg dep_eg_ dep_exploits_mining dobj_mining_texts prep_texts_for nn_reviews_product pobj_for_reviews appos_reviews_eg dep_eg_ appos_reviews_summarization appos_summarization_eg dep_eg_ nn_extraction_information appos_reviews_extraction appos_extraction_eg dep_eg_ appos_reviews_1Note complm_type_that nsubj_type_sentiment det_focus_the appos_sentiment_focus prep_focus_of amod_work_much amod_work_recent pobj_of_work prep_work_in det_area_the pobj_in_area cop_type_is det_type_a dep_1Note_type prep_type_of pobj_of_subjectivity advmod_involving_specifically partmod_reviews_involving amod_opinion_positive cc_positive_or conj_positive_negative dobj_involving_opinion conj_reviews_emotion cc_reviews_or conj_reviews_evaluation ' +o,6533,'Only recently the issue has drawn attention : present an initial analysis of the factors that influence system performance in content selection ',Nenkova,'advmod_recently_Only advmod_drawn_recently det_issue_the nsubj_drawn_issue aux_drawn_has dobj_drawn_attention advmod_present_ parataxis_drawn_present det_analysis_an amod_analysis_initial dobj_present_analysis prep_analysis_of det_factors_the pobj_of_factors nsubj_influence_that rcmod_factors_influence nn_performance_system dobj_influence_performance prep_performance_in nn_selection_content pobj_in_selection ' +o,6534,'-LRB- subjective -RRB- So far , none of the studies in sentiment detection or opinion extraction have specifically looked at the role of superlatives in these areas ',Pang,'nsubj_looked_subjective advmod_far_So advmod_subjective_far nsubj_looked_none prep_none_of det_studies_the pobj_of_studies prep_studies_in nn__sentiment nn__detection pobj_in_ cc__or nn__opinion nn__extraction conj__ aux_looked_have advmod_looked_specifically prep_looked_at det_role_the pobj_at_role prep_role_of pobj_of_superlatives prep_superlatives_in det_areas_these pobj_in_areas ' +o,6535,'It is difficult to compare these with previous work , but report that in a completely unsupervised setting , their MRF model , which uses a large set of additional features and a more complex estimation procedure , achieves an average 1-to-1 accuracy of 413 \% ',Haghighi,'nsubj_difficult_It cop_difficult_is aux_compare_to xcomp_difficult_compare dobj_compare_these prep_these_with amod_work_previous pobj_with_work cc_compare_but advmod_compare_ conj_compare_report complm_achieves_that prep_achieves_in det_setting_a advmod_unsupervised_completely amod_setting_unsupervised pobj_in_setting poss_model_their nn_model_MRF nsubj_achieves_model nsubj_uses_which rcmod_model_uses det_set_a amod_set_large dobj_uses_set prep_set_of amod_features_additional pobj_of_features cc_features_and det_procedure_a advmod_complex_more amod_procedure_complex nn_procedure_estimation conj_features_procedure ccomp_report_achieves det_accuracy_an amod_accuracy_average amod_accuracy_1-to-1 dobj_achieves_accuracy prep_accuracy_of num_\%_413 pobj_of_\% ' +o,6536,'In this paper , a new part-of-speech tagging method hased on neural networks -LRB- Net-Tagger -RRB- is presented and its performance is compared to that of a llMM-tagger and a trigrambased tagger ',Cutting,'prep_hased_In det_paper_this pobj_In_paper det_method_a amod_method_new amod_method_part-of-speech nn_method_tagging nsubj_hased_method prep_hased_on amod_networks_neural nsubjpass_presented_networks appos_networks_Net-Tagger auxpass_presented_is pcomp_on_presented cc_presented_and poss_performance_its nsubj_is_performance conj_presented_is prep_is_compared dep_compared_to pobj_to_that prep_that_of det__a amod__llMM-tagger pobj_of_ cc__and det__a amod__trigrambased nn__tagger conj__ ' +p,6537,'The BLEU metric in MT has been particularly successful ; for example MT05 , the 2005 NIST MT evaluation exercise , used BLEU-4 as the only method of evaluation ',Papineni,'det__The amod__BLEU amod__metric nsubj_successful_ prep__in pobj_in_MT aux_successful_has dep_successful_been advmod_successful_particularly prep_used_for nn_MT05_example pobj_for_MT05 det_exercise_the num_exercise_2005 nn_exercise_NIST nn_exercise_MT nn_exercise_evaluation appos_MT05_exercise parataxis_successful_used dobj_used_BLEU-4 prep_used_as det_method_the amod_method_only pobj_as_method prep_method_of pobj_of_evaluation ' +o,6538,'We determined appropriate training parameters and network size based on intermediate validation 1We used a publicly available tagger to provide the tags ',Ratnaparkhi,'nsubj_determined_We amod_parameters_appropriate nn_parameters_training dobj_determined_parameters cc_parameters_and nn_size_network conj_parameters_size prep_determined_based dep_based_on amod_1We_intermediate nn_1We_validation pobj_on_1We partmod_1We_used det__a advmod_available_publicly amod__available nn__tagger dobj_used_ aux_provide_to xcomp_used_provide det_tags_the dobj_provide_tags ' +o,6539,' s parser and its reimplementation and extension by Bikel -LRB- 2002 -RRB- have by now been applied to a variety of languages : English , Czech , German , Spanish , French , Chinese and , according to Dan Bikels web page , Arabic ',Collins,'nsubj_parser_ dep_parser_s cc_parser_and poss_reimplementation_its conj_parser_reimplementation cc_reimplementation_and conj_reimplementation_extension prep_reimplementation_by pobj_by_Bikel appos_Bikel_2002 aux_applied_have prep_applied_by pobj_by_now auxpass_applied_been rcmod_Bikel_applied prep_applied_to det_variety_a pobj_to_variety prep_variety_of pobj_of_languages nn__English dep_reimplementation_ amod__Czech appos__ amod__German appos__ amod__Spanish appos__ amod__French appos__ amod__Chinese appos__ cc_parser_and dep_parser_according dep_according_to nn_page_Dan nn_page_Bikels nn_page_web pobj_to_page amod_parser_Arabic ' +o,6540,'1 Introduction Parsing technology has come a long way since Charniak demonstrated that a simple treebank PCFG performs better than any other parser -LRB- with F175 accuracy -RRB- on parsing the WSJ Penn treebank ',Marcus,'num_technology_1 nn_technology_Introduction nn_technology_Parsing nsubj_come_technology aux_come_has det_way_a amod_way_long dobj_come_way mark_demonstrated_since nn__Charniak nsubj_demonstrated_ advcl_come_demonstrated complm_performs_that det_PCFG_a amod_PCFG_simple nn_PCFG_treebank nsubj_performs_PCFG ccomp_demonstrated_performs advmod_performs_better dep_better_than det_parser_any amod_parser_other pobj_than_parser dep_parser_with nn_accuracy_F175 pobj_with_accuracy prep_performs_on pcomp_on_parsing det_treebank_the nn_treebank_WSJ nn_treebank_Penn dobj_parsing_treebank ' +o,6541,' approached chunking by using a machine learning method ',Ramshaw,'nsubj_approached_ xcomp_approached_chunking prep_chunking_by pcomp_by_using det_machine_a dobj_using_machine partmod_machine_learning dobj_learning_method ' +o,6542,' , -RRB- , a sentence -LRB- eg , Liu et al ',Turney,'dep__ dep__ det_sentence_a dep__sentence dep_sentence_eg dep_eg_Liu cc_Liu_et conj_Liu_al ' +o,6543,'The recent approaches used pair-wise alignment algorithms based on symmetric alignments from a HMM alignment model or edit distance alignments allowing shifts ',Rosti,'det_approaches_The amod_approaches_recent nsubj_used_approaches amod_algorithms_pair-wise nn_algorithms_alignment dobj_used_algorithms prep_used_based dep_based_on amod_alignments_symmetric pobj_on_alignments prep_alignments_from det__a nn__HMM nn__alignment nn__model pobj_from_ cc_alignments_or amod_alignments_edit nn_alignments_distance conj_alignments_alignments partmod_alignments_allowing dobj_allowing_shifts ' +p,6544,'On the other hand , integrating an additional component into a baseline SMT system is notoriously tricky as evident in the research on integrating word sense disambiguation -LRB- WSD -RRB- into SMT systems : different ways of integration lead to conflicting conclusions on whether WSD helps MT performance ',Carpuat,'prep_tricky_On det_hand_the amod_hand_other pobj_On_hand dep_tricky_integrating det_component_an amod_component_additional dobj_integrating_component prep_integrating_into det_baseline_a pobj_into_baseline nn_system_SMT nsubj_tricky_system cop_tricky_is advmod_tricky_notoriously prep_tricky_as amod_systems_evident prep_evident_in det_research_the pobj_in_research prep_research_on pcomp_on_integrating nn_disambiguation_word nn_disambiguation_sense dobj_integrating_disambiguation abbrev_disambiguation_WSD prep_integrating_into pobj_into_SMT pobj_as_systems amod_ways_different parataxis_tricky_ways prep_ways_of nn_lead_integration pobj_of_lead prep_lead_to amod_conclusions_conflicting pobj_to_conclusions advmod_ways_on dep_helps_whether nsubj_helps_WSD dep_on_helps nn_performance_MT dobj_helps_performance ' +o,6545,'Antonyms often indicate the discourse relation of contrast ',Marcu,'nsubj_indicate_Antonyms advmod_indicate_often det_relation_the nn_relation_discourse dobj_indicate_relation prep_relation_of pobj_of_contrast ' +o,6546,'For example , a statistical machine translation system such as ISIs AlTemp SMT system can generate a list of n-best alternative translations given a source sentence ',Och,'prep_generate_For pobj_For_example det_system_a amod_system_statistical nn_system_machine nn_system_translation nsubj_generate_system dep_as_such prep_system_as nn__ISIs nn__AlTemp nn__SMT nn__system pobj_as_ aux_generate_can det_list_a dobj_generate_list prep_list_of amod_translations_n-best amod_translations_alternative pobj_of_translations partmod_translations_given det_sentence_a nn_sentence_source dobj_given_sentence ' +o,6547,'The per-state models in this paper are log-linear models , building upon the models in and , though some models are in fact strictly simpler ',Ratnaparkhi,'det_models_The amod_models_per-state nsubj_models_models prep_models_in det_paper_this pobj_in_paper cop_models_are amod_models_log-linear partmod_models_building prep_building_upon det_models_the pobj_upon_models prep_models_in pobj_in_ cc__and conj__ mark_are_though det_models_some nsubj_are_models advcl_models_are prep_are_in measure_simpler_fact advmod_fact_strictly pobj_in_simpler ' +o,6548,'This is the traditional approach for glass-box smoothing ',Koehn,'nsubj_approach_This cop_approach_is det_approach_the amod_approach_traditional prep_approach_for amod_smoothing_glass-box pobj_for_smoothing ' +o,6549,'The former extracts collocations within a fixed window ',Church,'det_extracts_The amod_extracts_former nsubj_collocations_extracts prep_collocations_within det_window_a amod_window_fixed pobj_within_window ' +p,6550,'This is analogous , and in a certain sense equivalent , to empirical risk minimization , which has been used successfully in related areas , such as speech recognition , language modeling , and machine translation ',Och,'nsubj_analogous_This cop_analogous_is cc_analogous_and conj_analogous_in det_equivalent_a amod_equivalent_certain nn_equivalent_sense pobj_in_equivalent prep_equivalent_to amod_minimization_empirical nn_minimization_risk pobj_to_minimization nsubjpass_used_which aux_used_has auxpass_used_been rcmod_minimization_used advmod_used_successfully prep_used_in amod_areas_related pobj_in_areas dep_as_such prep_equivalent_as nn__speech nn__recognition pobj_as_ nn__language nn__modeling conj__ cc__and nn_translation_machine conj__translation ' +o,6551,'Similarly , uses a six content word window to extract significant collocations ',Smadja,'advmod_uses_Similarly nsubj_uses_ det_window_a num_window_six nn_window_content nn_window_word nsubj_extract_window aux_extract_to xcomp_uses_extract amod_collocations_significant dobj_extract_collocations ' +o,6552,'The text was split at the sentence level , tokenized and PoS tagged , in the style of the Wall Street Journal Penn TreeBank ',Brown,'det_text_The nsubjpass_split_text auxpass_split_was prep_split_at det_level_the nn_level_sentence pobj_at_level conj_level_tokenized cc_level_and nn_tagged_PoS conj_level_tagged prep_split_in det_style_the pobj_in_style prep_style_of det__the nn__Wall nn__Street nn__Journal nn__Penn nn__TreeBank pobj_of_ ' +n,6553,'In comparison , the 2D model in Figure 2 -LRB- c -RRB- used in previous work can only model the interaction between adjacent questions ',Ding,'prep_model_In pobj_In_comparison det_model_the num_model_2D nsubj_model_model prep_model_in pobj_in_Figure num_Figure_2 dep_-LRB-_c partmod_Figure_used prep_used_in amod__previous nn__work pobj_in_ aux_model_can advmod_model_only det_interaction_the dobj_model_interaction prep_interaction_between amod_questions_adjacent pobj_between_questions ' +o,6554,'Data and Parameters To facilitate comparison with previous work , we trained our models on sections 2-21 of the WSJ section of the Penn tree-bank ',Marcus,'nsubj_facilitate_Data cc_Data_and conj_Data_Parameters aux_facilitate_To dobj_facilitate_comparison prep_comparison_with amod_work_previous pobj_with_work nsubj_trained_we dep_facilitate_trained poss_models_our dobj_trained_models prep_models_on pobj_on_sections num_sections_2-21 prep_sections_of det_section_the nn_section_WSJ pobj_of_section prep_section_of det_tree-bank_the nn_tree-bank_Penn pobj_of_tree-bank ' +o,6555,'Each linked fragment pair consists of a source-language side and a target-language side , similar to ',Wu,'det_pair_Each amod_pair_linked nn_pair_fragment nsubj_consists_pair prep_consists_of det_side_a amod_side_source-language pobj_of_side cc_side_and det_side_a amod_side_target-language conj_side_side amod_side_similar prep_similar_to ' +n,6556,'The problem is typically presented in log-space , which simplifies computations , but otherwise does not change the problem due to the monotonicity of the log function -LRB- hm = log hprimem -RRB- log p -LRB- t s -RRB- = summationdisplay m m hm -LRB- t , s -RRB- -LRB- 3 -RRB- Phrase-based models are limited to the mapping of small contiguous chunks of text ',Koehn,'det_problem_The nsubjpass_presented_problem auxpass_presented_is advmod_presented_typically prep_presented_in pobj_in_log-space nsubj_simplifies_which rcmod_log-space_simplifies dobj_simplifies_computations cc_simplifies_but advmod_simplifies_otherwise aux_change_does neg_change_not conj_simplifies_change det_problem_the iobj_change_problem amod_p_due prep_due_to det_monotonicity_the pobj_to_monotonicity prep_monotonicity_of det_function_the nn_function_log pobj_of_function nsubj_hprimem_hm dep_hprimem_= nn_hprimem_log dep_p_hprimem nn_p_log dobj_change_p nn_s_t appos_p_s dep_p_= amod_m_summationdisplay iobj_=_m nn_hm_m dobj_=_hm dep_hm_t dep_t_s dep_limited_3 amod__Phrase-based nn__models nsubjpass_limited_ auxpass_limited_are dep_presented_limited prep_limited_to det_mapping_the pobj_to_mapping prep_mapping_of amod_chunks_small dep_small_contiguous pobj_of_chunks prep_chunks_of pobj_of_text ' +o,6557,'For example , the lexicalized grammars of and Charniak -LRB- 1997 -RRB- and the statesplit grammars of Petrov et al ',Collins,'prep_grammars_For pobj_For_example dep_grammars_the amod_grammars_lexicalized nsubj_grammars_grammars prep_grammars_of pobj_of_ cc__and conj__Charniak appos_grammars_1997 cc_grammars_and det_statesplit_the conj_grammars_statesplit prep_grammars_of pobj_of_Petrov cc_Petrov_et conj_Petrov_al ' +n,6558,'Though taggers based on dependency networks , SVM , MaxEnt , CRF , and other methods may reach slightly better results , their train\\/test cycle is orders of magnitude longer ',Ratnaparkhi,'mark_reach_Though nsubj_reach_taggers prep_taggers_based dep_based_on amod__dependency nn__networks pobj_on_ nn__SVM conj__ nn__MaxEnt conj__ nn__CRF conj__ cc__and amod_methods_other conj__methods aux_reach_may advcl_orders_reach advmod_better_slightly amod_results_better dobj_reach_results poss_cycle_their amod_cycle_train\\/test nsubj_orders_cycle cop_orders_is prep_orders_of pobj_of_magnitude advmod_orders_longer ' +o,6559,'4 Related work Algorithms for retrieving collocations has been described ',Smadja,'num_Algorithms_4 amod_Algorithms_Related nn_Algorithms_work nsubjpass_described_Algorithms prep_Algorithms_for amod_collocations_retrieving pobj_for_collocations aux_described_has auxpass_described_been amod__ dobj_described_ ' +o,6560,'Part-ofspeech taggers are used in a few applications , such as speech synthesis and question answering ',Brown,'nn_taggers_Part-ofspeech nsubjpass_used_taggers auxpass_used_are prep_used_in det_applications_a amod_applications_few pobj_in_applications dep_as_such prep_applications_as nn__speech nn__synthesis pobj_as_ cc__and nn__question nn__answering conj__ ' +o,6561,'Hence our classifier evaluation omits those two word positions , leading to n2 classifications for a string of length n Table 1 shows statistics from sections 2-21 of the Penn WSJ Treebank ',Marcus,'advmod_omits_Hence poss_evaluation_our nn_evaluation_classifier nsubj_omits_evaluation det_positions_those num_positions_two nn_positions_word dobj_omits_positions xcomp_omits_leading prep_leading_to amod_classifications_n2 pobj_to_classifications prep_classifications_for det_string_a pobj_for_string prep_string_of nn_n_length pobj_of_n nsubj_shows_Table num_Table_1 dep_omits_shows dobj_shows_statistics prep_shows_from pobj_from_sections num_sections_2-21 prep_sections_of det_Treebank_the nn_Treebank_Penn nn_Treebank_WSJ pobj_of_Treebank ' +o,6562,'Minimum-error-rate training was done using Koehns implementation of minimum-error-rate model ',Och,'amod_training_Minimum-error-rate nsubjpass_done_training auxpass_done_was xcomp_done_using nn_implementation_Koehns dobj_using_implementation prep_implementation_of amod_model_ amod_model_minimum-error-rate pobj_of_model ' +o,6563,'The work reported in this paper is most closely related to work on statistical machine translation , particularly the IBM-style work on CANDIDE ',Brown,'det_work_The nsubj_related_work partmod_work_reported prep_reported_in det_paper_this pobj_in_paper cop_related_is advmod_closely_most advmod_related_closely aux_work_to xcomp_related_work prep_work_on amod_translation_statistical nn_translation_machine pobj_on_translation advmod_translation_particularly det_work_the amod_work_IBM-style dep_translation_work prep_work_on pobj_on_CANDIDE ' +o,6564,'CIT -RRB- ',Brown,'nn_-RRB-_CIT ' +o,6565,'These domains have been commonly used in prior work on summarization ',Lin,'det_domains_These nsubjpass_used_domains aux_used_have auxpass_used_been advmod_used_commonly prep_used_in amod_work_prior pobj_in_work prep_work_on pobj_on_summarization ' +o,6566,'The analyser -- and therefore the generator-includes exception lists derived from WordNet ',Marcus,'det_analyser_The nsubjpass_lists_analyser dep_exception_and dep_exception_therefore det_exception_the amod_exception_generator-includes dep_analyser_exception dep_lists_derived prep_derived_from nn__WordNet pobj_from_ ' +o,6567,'In practice , 7 - \\/ is very large and the model \'s expectation Efj can not be computed directly , so the following approximation is used : n E fj , ~ E15 -LRB- hi -RRB- p -LRB- tilhi -RRB- fj -LRB- hi , ti -RRB- i = 1 where fi -LRB- hi -RRB- is the observed probability of the history hi in the training set ',Marcus,'prep_i_In pobj_In_practice nsubj_i_7 nsubj_large_\\/ cop_large_is advmod_large_very dep_7_large cc_large_and det_model_the poss_Efj_model possessive_model_\'s nn_Efj_expectation nsubjpass_computed_Efj aux_computed_can neg_computed_not auxpass_computed_be conj_large_computed advmod_computed_directly dep_large_so det__the amod__following nn__approximation nsubjpass_used_ auxpass_used_is dep_large_used nn_fj_n nn_fj_E dep_7_fj number_E15_~ num_fj_E15 appos_fj_hi nn_fj_p appos_fj_tilhi appos_fj_fj appos_7_hi dep_hi_ti dep_1_= dep_i_1 advmod_probability_where nsubj_probability_fi appos_fi_hi cop_probability_is det_probability_the amod_probability_observed advcl_1_probability prep_probability_of det_hi_the nn_hi_history pobj_of_hi prep_hi_in det_set_the nn_set_training pobj_in_set ' +o,6568,'4 Experiment 41 Evaluation Method We evaluated each sentence compression method using word F-measures , bigram F-measures , and BLEU scores ',Papineni,'num_Evaluation_4 nn_Evaluation_Experiment num_Evaluation_41 nsubj_Method_Evaluation nsubj_evaluated_We ccomp_Method_evaluated det_method_each nn_method_sentence nn_method_compression dobj_evaluated_method partmod_method_using nn_F-measures_word dobj_using_F-measures nn_F-measures_bigram conj_F-measures_F-measures cc_F-measures_and amod_scores_BLEU conj_F-measures_scores ' +o,6569,' show that exploiting all contiguous word blocks in phrase-based alignment is better than focusing on syntactic constituents only ',Koehn,'nsubj_show_ complm_focusing_that csubj_focusing_exploiting advmod_contiguous_all amod_blocks_contiguous nn_blocks_word dobj_exploiting_blocks prep_blocks_in amod_alignment_phrase-based pobj_in_alignment aux_focusing_is advmod_focusing_better dep_better_than ccomp_show_focusing prep_focusing_on amod_constituents_syntactic pobj_on_constituents advmod_focusing_only ' +o,6570,'It is an implementation of Models 1-4 of , where each of these models produces a Viterbi alignment ',Brown,'nsubj_implementation_It cop_implementation_is det_implementation_an prep_implementation_of pobj_of_Models num_Models_1-4 prep_implementation_of pobj_of_ advmod_produces_where nsubj_produces_each prep_each_of det_models_these pobj_of_models rcmod__produces det_alignment_a nn_alignment_Viterbi dobj_produces_alignment ' +o,6571,'For regularization purposes we adopt an average perceptron which returns for each y , y = 1T summationtextTt = 1 ty , the average of all weight vectors ty posited during training ',Collins,'prep_adopt_For nn_purposes_regularization pobj_For_purposes nsubj_adopt_we det__an amod__average nn__perceptron dobj_adopt_ dobj_ty_which nsubj_ty_returns prep_returns_for det_y_each pobj_for_y partmod_y_y dep_summationtextTt_= num_summationtextTt_1T ccomp_y_summationtextTt dep_ty_= num_ty_1 rcmod__ty det_average_the appos__average prep_average_of det_ty_all nn_ty_weight nn_ty_vectors pobj_of_ty partmod_ty_posited prep_posited_during pobj_during_training ' +o,6572,'Its applications range from sentence boundary disambiguation to part-of-speech tagging , parsing and machine translation ',Ratnaparkhi,'poss_applications_Its nsubj_range_applications prep_range_from nn__sentence nn__boundary nn__disambiguation pobj_from_ prep_range_to amod__part-of-speech nn__tagging pobj_to_ xcomp_range_parsing nn__ cc__and conj__machine nn__translation dobj_parsing_ ' +o,6573,'For the named entity features , we used a fairly standard feature set , similar to those described in ',Finkel,'prep_used_For det_features_the amod_features_named nn_features_entity pobj_For_features nsubj_used_we det_set_a advmod_set_fairly amod_set_standard nn_set_feature dobj_used_set amod_set_similar prep_similar_to pobj_to_those partmod_those_described prep_described_in ' +o,6574,'In contrast , semi-supervised domain adaptation is the scenario in which , in addition to the labeled source data , we only have unlabeled and no labeled target domain data ',Blitzer,'prep_scenario_In pobj_In_contrast amod__semi-supervised nn__domain nn__adaptation nsubj_scenario_ cop_scenario_is det_scenario_the rel_labeled_in dep_in_which pobj_in_in pobj_in_addition prep_addition_to det_data_the amod_data_labeled nn_data_source pobj_to_data nsubj_labeled_we advmod_labeled_only aux_labeled_have advmod_labeled_unlabeled cc_unlabeled_and conj_unlabeled_no rcmod_scenario_labeled nn_data_target nn_data_domain dobj_labeled_data ' +o,6575,'For instance , for Maximum Entropy , I picked for the basic theory , for an application -LRB- POS tagging in this case -RRB- , and for more advanced topics such as optimization and smoothing ',Manning,'prep_picked_For pobj_For_instance prep_picked_for nn_Entropy_Maximum pobj_for_Entropy nsubj_picked_I ccomp_picked_ prep__for det_theory_the amod_theory_basic pobj_for_theory conj__ prep__for det_application_an pobj_for_application dep_application_POS amod_POS_tagging prep_POS_in det_case_this pobj_in_case cc__and conj__ prep__for advmod_advanced_more amod_topics_advanced pobj_for_topics dep_as_such prep_topics_as pobj_as_optimization cc_optimization_and conj_optimization_smoothing ' +o,6576,'On the other hand , the thesaurus-based method of Yarowsky may suffer from loss of information -LRB- since it is semi-class-based -RRB- as well as data sparseness -LRB- since H Classes used in Resnik are based on the WordNet taxonomy while classes of et al ',Brown,'prep_suffer_On det_hand_the amod_hand_other pobj_On_hand det_method_the amod_method_thesaurus-based nsubj_suffer_method prep_method_of nn__Yarowsky pobj_of_ aux_suffer_may prep_suffer_from pobj_from_loss prep_loss_of pobj_of_information mark_semi-class-based_since nsubj_semi-class-based_it cop_semi-class-based_is dep_loss_semi-class-based dep_well_as cc_loss_well dep_well_as conj_loss_data nsubj_used_sparseness prep_sparseness_since nn_Classes_H pobj_since_Classes dep_suffer_used mark_based_in nn__Resnik nsubjpass_based_ auxpass_based_are advcl_used_based prep_based_on det_taxonomy_the nn_taxonomy_WordNet pobj_on_taxonomy prep_based_while pobj_while_classes prep_classes_of pobj_of_ cc__et conj__al ' +o,6577,'Identification of Terms To-be Transliterated -LRB- TTT -RRB- must not be confused with recognition of Named Entities -LRB- NE -RRB- ',Hermjakob,'nsubjpass_confused_Identification prep_Identification_of nn_Transliterated_Terms nn_Transliterated_To-be pobj_of_Transliterated abbrev_Transliterated_TTT aux_confused_must neg_confused_not auxpass_confused_be prep_confused_with pobj_with_recognition prep_recognition_of nn_Entities_Named pobj_of_Entities abbrev_Entities_NE ' +o,6578,' who employ clusters of related words constructed by the Brown clustering algorithm for syntactic processing of texts ',Brown,'nsubj_employ_who dep__employ dobj_employ_clusters prep_clusters_of amod_words_related pobj_of_words partmod_words_constructed prep_constructed_by det__the nn__Brown nn__clustering nn__algorithm pobj_by_ prep__for amod_processing_syntactic pobj_for_processing prep_processing_of pobj_of_texts ' +o,6579,'To this extent , we cast the supersense tagging problem as a sequence labeling task and train a discriminative Hidden Markov Model -LRB- HMM -RRB- , based on that of , on the manually annotated Semcor corpus ',Collins,'prep_cast_To det_extent_this pobj_To_extent nsubj_cast_we det_problem_the amod_problem_supersense amod_problem_tagging dobj_cast_problem prep_cast_as det_task_a nn_task_sequence nn_task_labeling pobj_as_task cc_cast_and conj_cast_train det_Model_a amod_Model_discriminative nn_Model_Hidden nn_Model_Markov dobj_train_Model abbrev_Model_HMM partmod_Model_based prep_based_on pobj_on_that prep_that_of pobj_of_ prep_train_on det__the advmod_annotated_manually amod__annotated nn__Semcor nn__corpus pobj_on_ ' +o,6580,'We then propose a relatively simple yet effective method for resolving translation disambiguation using mutual information -LRB- MI -RRB- statistics obtained only from the target document collection ',Church,'nsubj_propose_We advmod_propose_then det_method_a advmod_simple_relatively amod_method_simple cc_simple_yet conj_simple_effective dobj_propose_method prep_propose_for pcomp_for_resolving nn_disambiguation_translation dobj_resolving_disambiguation xcomp_resolving_using amod_information_mutual dobj_using_information abbrev_information_MI amod_statistics_ nsubj_obtained_statistics dep_propose_obtained advmod_obtained_only prep_obtained_from det_collection_the nn_collection_target nn_collection_document pobj_from_collection ' +o,6581,'Notice that most in-context and dictionary translations of source words are bounded within the same category in a typical thesaurus such as the LLOCE and CILIN ',Brown,'complm_bounded_that advmod_in-context_most amod_translations_in-context cc_in-context_and conj_in-context_dictionary nsubjpass_bounded_translations prep_translations_of nn_words_source pobj_of_words auxpass_bounded_are dep_Notice_bounded prep_bounded_within det_category_the amod_category_same pobj_within_category prep_category_in det_thesaurus_a amod_thesaurus_typical pobj_in_thesaurus dep_as_such prep_thesaurus_as det__the nn__LLOCE pobj_as_ cc__and nn__CILIN conj__ ' +o,6582,'-LRB- HICSS-35 documents , genres also work on an intra-document , or page segment level because a single document can contain instances of multiple genres , eg , contact information , list of publications , CV , see ',Pang,'nn__HICSS-35 nsubj__ amod_documents_ dobj__documents nsubj_work_genres advmod_work_also dep__work prep_work_on det_intra-document_an pobj_on_intra-document cc_intra-document_or nn_level_page nn_level_segment conj_intra-document_level mark_contain_because det_document_a amod_document_single nsubj_contain_document aux_contain_can advcl_work_contain dobj_contain_instances prep_instances_of amod_genres_multiple pobj_of_genres nsubj_see_eg nn_information_contact appos_eg_information appos_eg_list prep_list_of pobj_of_publications appos_eg_CV dep__see acomp_see_ ' +o,6583,'The Xerox tagger comes with a set of rules that assign an unknown word a set of possible pos-tags -LRB- ie , POS-class -RRB- on the basis of its ending segment ',Cutting,'det__The nn__Xerox nn__tagger nsubj_comes_ prep_comes_with det_set_a pobj_with_set prep_set_of pobj_of_rules nsubj_assign_that rcmod_rules_assign det_word_an amod_word_unknown dobj_assign_word det_set_a dep_word_set prep_word_of amod_pos-tags_possible pobj_of_pos-tags appos_pos-tags_ie dep_ie_POS-class prep_assign_on det_basis_the pobj_on_basis prep_basis_of poss_segment_its nn_segment_ending pobj_of_segment ' +o,6584,'1 Introduction Mainstream approaches in statistical parsing are based on nondeterministic parsing techniques , usually employing some kind of dynamic programming , in combination with generative probabilistic models that provide an n-best ranking of the set of candidate analyses derived by the parser ',Collins,'num_approaches_1 nn_approaches_Introduction nn_approaches_Mainstream nsubjpass_based_approaches prep_approaches_in amod_parsing_statistical pobj_in_parsing auxpass_based_are prep_based_on amod_techniques_nondeterministic nn_techniques_parsing pobj_on_techniques advmod_employing_usually xcomp_based_employing det_kind_some dobj_employing_kind prep_kind_of amod_programming_dynamic pobj_of_programming prep_kind_in pobj_in_combination prep_combination_with amod_models_generative amod_models_probabilistic pobj_with_models nsubj_provide_that rcmod_models_provide det_ranking_an amod_ranking_n-best dobj_provide_ranking prep_ranking_of det_set_the pobj_of_set prep_set_of nn_analyses_candidate pobj_of_analyses partmod_analyses_derived prep_derived_by det_parser_the pobj_by_parser ' +o,6585,'This step can be seen as a multi-label , multi-class call classi cation problem for customer care applications ',Chu-Carroll,'det_step_This nsubjpass_seen_step aux_seen_can auxpass_seen_be prep_seen_as det_problem_a amod_problem_multi-label amod_problem_multi-class amod_problem_call nn_problem_classi nn_problem_cation pobj_as_problem prep_problem_for nn_applications_customer nn_applications_care pobj_for_applications ' +o,6586,'tile data put tbrward by ll , amshaw and Marcus ',Ramshaw,'nn_data_tile nsubj_put_data dep_put_tbrward prep_tbrward_by pobj_by_ll conj_ll_amshaw cc_ll_and nn__Marcus conj_ll_ ' +n,6587,'Several papers have looked at higher-order representations , but have not examined the equivalence of syn\\/para distributions when formalized as Markov chains ',Rapp,'amod_papers_Several nsubj_looked_papers aux_looked_have prep_looked_at amod_representations_higher-order pobj_at_representations cc_looked_but aux_examined_have neg_examined_not conj_looked_examined det_equivalence_the dobj_examined_equivalence prep_equivalence_of nn_distributions_syn\\/para pobj_of_distributions advmod_formalized_when advcl_examined_formalized prep_formalized_as nn_chains_Markov pobj_as_chains ' +p,6588,' investigated the use of concurrent parsing of parallel corpora in a transduction inversion framework , helping to resolve attachment ambiguities in one language by the coupled parsing state in the second language ',Wu,'nsubj_investigated_ det_use_the dobj_investigated_use prep_use_of amod_parsing_concurrent pobj_of_parsing prep_parsing_of amod_corpora_parallel pobj_of_corpora prep_investigated_in det_framework_a nn_framework_transduction nn_framework_inversion pobj_in_framework xcomp_investigated_helping aux_resolve_to xcomp_helping_resolve nn_ambiguities_attachment dobj_resolve_ambiguities prep_resolve_in num_language_one pobj_in_language prep_resolve_by det_state_the amod_state_coupled amod_state_parsing pobj_by_state prep_state_in det_language_the amod_language_second pobj_in_language ' +n,6589,'Of the methods we compare against , only the WordNet-based similarity measures , , and provide a method for predicting verb similarities ; our learned measure widely outperforms these methods , achieving a 136 \% F-score improvement over the LESK similarity measure ',Navigli,'prep_compare_Of det_methods_the pobj_Of_methods nsubj_compare_we prep_compare_against advmod_similarity_only det_similarity_the amod_similarity_WordNet-based nsubj_measures_similarity dep_against_measures pobj_against_ cc_compare_and nsubj_provide_ conj_compare_provide det_method_a dobj_provide_method prep_provide_for pcomp_for_predicting nn_similarities_verb dobj_predicting_similarities poss_measure_our amod_measure_learned nsubj_outperforms_measure advmod_outperforms_widely parataxis_compare_outperforms det_methods_these dobj_outperforms_methods xcomp_outperforms_achieving det_improvement_a num_\%_136 measure_F-score_\% amod_improvement_F-score dobj_achieving_improvement prep_improvement_over det_measure_the nn_measure_LESK nn_measure_similarity pobj_over_measure ' +o,6590,'In this year , CoNLL-2007 shared task focuses on multilingual dependency parsing based on ten different languages and domain adaptation for English without taking the languagespecific knowledge into consideration ',Marcus,'prep_focuses_In det_year_this pobj_In_year amod__CoNLL-2007 amod__shared nn__task nsubj_focuses_ prep_focuses_on amod_dependency_multilingual pobj_on_dependency partmod_dependency_parsing prep_parsing_based dep_based_on amod__ten amod__different nn__languages pobj_on_ cc__and nn_adaptation_domain conj__adaptation prep_parsing_for nn__English pobj_for_ prep_parsing_without pcomp_without_taking det_knowledge_the amod_knowledge_languagespecific dobj_taking_knowledge prep_taking_into pobj_into_consideration ' +o,6591,'For example , bilingual lexicographers can use bitexts to discover new cross-language lexicalization patterns ; students of foreign languages can use one half of a bitext to practice their reading skills , referring to the other half for translation when they get stuck ',Brown,'prep_use_For pobj_For_example amod_lexicographers_bilingual nsubj_use_lexicographers aux_use_can nsubj_discover_bitexts aux_discover_to xcomp_use_discover amod_patterns_new amod_patterns_cross-language nn_patterns_lexicalization dobj_discover_patterns advmod_discover_ nsubj_use_students prep_students_of amod_languages_foreign pobj_of_languages aux_use_can parataxis_use_use num_half_one dobj_use_half prep_half_of det_bitext_a pobj_of_bitext aux_practice_to infmod_bitext_practice poss_skills_their nn_skills_reading dobj_practice_skills partmod_half_referring prep_referring_to det_half_the amod_half_other pobj_to_half prep_half_for pobj_for_translation advmod_get_when nsubj_get_they advcl_referring_get amod__stuck dobj_get_ ' +o,6592,'41 Translation Modeling We can test our models utility for translation by transforming its parameters into a phrase table for the phrasal decoder Pharaoh ',Koehn,'num_Modeling_41 nn_Modeling_Translation nsubj_test_We aux_test_can dep_Modeling_test poss_utility_our nn_utility_models dobj_test_utility prep_test_for pobj_for_translation prep_test_by pcomp_by_transforming poss_parameters_its dobj_transforming_parameters prep_transforming_into det_table_a nn_table_phrase pobj_into_table prep_table_for det_Pharaoh_the amod_Pharaoh_phrasal nn_Pharaoh_decoder pobj_for_Pharaoh ' +o,6593,'1 Introduction Word alignment is an important component of a complete statistical machine translation pipeline ',Koehn,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubj_component_alignment cop_component_is det_component_an amod_component_important prep_component_of det_pipeline_a amod_pipeline_complete amod_pipeline_statistical nn_pipeline_machine nn_pipeline_translation pobj_of_pipeline ' +o,6594,'Other similar work includes the mention detection -LRB- MD -RRB- task and joint probabilistic model of coreference ','Daume III','amod_work_Other amod_work_similar nsubj_includes_work det_detection_the nn_detection_mention nn__detection abbrev__MD nn__task dobj_includes_ cc__and amod_model_joint amod_model_probabilistic conj__model prep_model_of pobj_of_coreference ' +o,6595,'have been proposed ',Marcus,'aux_been_have amod__proposed nsubj_been_ ' +o,6596,'5 Datasets For evaluation we selected two domain adaptation datasets : spam and sentiment ',Blitzer,'dep_5_Datasets prep_selected_For pobj_For_evaluation nsubj_selected_we ccomp_Datasets_selected num_datasets_two nn_datasets_domain nn_datasets_adaptation dobj_selected_datasets nn__spam dep_5_ cc__and conj__sentiment ' +p,6597,'Sentence-level subjectivity detection , where training data is easier to obtain than for positive vs negative classification , has been successfully performed using supervised statistical methods alone or in combination with a knowledgebased approach ',Pang,'amod_detection_Sentence-level nn_detection_subjectivity nsubjpass_performed_detection advmod_easier_where nn_data_training nsubj_easier_data cop_easier_is rcmod_detection_easier aux_obtain_to xcomp_easier_obtain prep_obtain_than dep_than_for amod_classification_positive amod_classification_vs amod_classification_negative pobj_for_classification aux_performed_has auxpass_performed_been advmod_performed_successfully xcomp_performed_using amod_methods_supervised amod_methods_statistical dobj_using_methods prep_using_alone dep_alone_ cc_alone_or conj_alone_in pobj_in_combination prep_combination_with det__a amod__knowledgebased nn__approach pobj_with_ ' +o,6598,'One option would be to leverage unannotated text ',Smith,'num_option_One nsubj_be_option aux_be_would prep_be_to nn_text_leverage nn_text_unannotated pobj_to_text ' +o,6599,'The problem itself has started to get attention only recently ',Dredze,'det_problem_The nsubj_started_itself aux_started_has dep_problem_started aux_get_to xcomp_started_get dobj_get_attention advmod_recently_only advmod_get_recently ' +o,6600,'Not only is this beneficial in terms of parsing complexity , but smaller rules can also improve a translation models ability to generalize to new data ',Zhang,'dep_beneficial_Not dep_Not_only cop_beneficial_is dep_beneficial_this prep_beneficial_in pobj_in_terms prep_terms_of amod_complexity_parsing pobj_of_complexity cc_beneficial_but conj_beneficial_smaller nsubj_improve_rules aux_improve_can advmod_improve_also ccomp_beneficial_improve det_ability_a nn_ability_translation nn_ability_models nsubj_generalize_ability aux_generalize_to xcomp_improve_generalize prep_generalize_to amod_data_new pobj_to_data ' +n,6601,'In order to capture the dependency relationship between lexcial heads breaks down the rules from head outwards , which prevents us from factorizing them in other ways ',Collins,'prep_breaks_In pobj_In_order aux_capture_to infmod_order_capture det_relationship_the amod_relationship_dependency dobj_capture_relationship prep_relationship_between amod_heads_lexcial pobj_between_heads advmod_breaks_ prt_breaks_down det_rules_the dobj_breaks_rules prep_rules_from pobj_from_head nsubj_breaks_outwards nsubj_prevents_which rcmod_outwards_prevents dobj_prevents_us prep_prevents_from pcomp_from_factorizing dobj_factorizing_them prep_factorizing_in amod_ways_other pobj_in_ways ' +o,6602,'task , and reported errors in the range of 26 \% are common ',Ratnaparkhi,'nsubj__task cc__and conj__reported nsubj_common_errors prep_errors_in det_range_the pobj_in_range prep_range_of num_\%_26 pobj_of_\% cop_common_are ccomp_reported_common ' +o,6603,'The natural next step in sentence alignment is to account for word ordering in the translation model , eg , the models described in could be used ',Brown,'det_step_The amod_step_natural amod_step_next nsubj_is_step prep_step_in nn_alignment_sentence pobj_in_alignment ccomp_used_is aux_account_to xcomp_is_account prep_account_for pobj_for_word partmod_word_ordering prep_ordering_in det_model_the nn_model_translation pobj_in_model nsubjpass_used_eg det_models_the nsubjpass_used_models partmod_models_described prep_described_in pobj_in_ aux_used_could auxpass_used_be ' +o,6604,'It is interesting to constrast this method with the ` parse-parse-match \' approaches that have been reported recently for producing parallel bracketed corpora ',Brown,'nsubj_interesting_It cop_interesting_is aux_constrast_to xcomp_interesting_constrast det_method_this dobj_constrast_method prep_constrast_with det_approaches_the amod_approaches_parse-parse-match pobj_with_approaches nsubjpass_reported_that aux_reported_have auxpass_reported_been rcmod_approaches_reported advmod_reported_recently prep_reported_for pcomp_for_producing amod__parallel amod__bracketed nn__corpora dobj_producing_ ' +o,6605,'Suhm and Waibel and Eckert , Gallwitz , and Niemann each condition a recognizer LM on left-to-right DA predictions and are able to 366 Stolcke et al Dialogue Act Modeling show reductions in word error rate of 1 \% on task-oriented corpora ',Berger,'nsubj_LM_Suhm cc_Suhm_and nn__Waibel conj_Suhm_ cc_Suhm_and nn_Gallwitz_Eckert conj_Suhm_Gallwitz cc_Gallwitz_and nn__Niemann conj_Gallwitz_ det_condition_each dep_Gallwitz_condition det_recognizer_a dep_Gallwitz_recognizer prep_LM_on amod_predictions_left-to-right nn_predictions_DA pobj_on_predictions cc_LM_and cop_able_are conj_LM_able prep_able_to num_reductions_366 nn_reductions_Stolcke nn_reductions_et nn_reductions_al nn_reductions_Dialogue nn_reductions_Act nn_reductions_Modeling nn_reductions_show pobj_to_reductions prep_reductions_in nn_rate_word nn_rate_error pobj_in_rate prep_rate_of num_\%_1 pobj_of_\% prep_\%_on amod_corpora_task-oriented pobj_on_corpora ' +n,6606,'Besides , our model , as being linguistically motivated , is also more expressive than the formally syntax-based models of Chiang and ',Wu,'advmod_expressive_Besides poss_model_our nsubj_expressive_model prep_model_as cop_motivated_being advmod_motivated_linguistically pcomp_as_motivated cop_expressive_is advmod_expressive_also advmod_expressive_more prep_expressive_than det_models_the advmod_syntax-based_formally amod_models_syntax-based pobj_than_models prep_models_of nn_and_Chiang nn_and_ pobj_of_and ' +o,6607,'41 Corpora Sentence compression systems have been tested on product review data from the Ziff-Davis -LRB- ZD , henceforth -RRB- Corpus by Knight and Marcu , general news articles by and biomedical articles ',Clarke,'num_systems_41 nn_systems_Corpora nn_systems_Sentence nn_systems_compression nsubjpass_tested_systems aux_tested_have auxpass_tested_been prep_tested_on nn_data_product nn_data_review pobj_on_data prep_data_from det_Corpus_the nn_Corpus_Ziff-Davis appos_Corpus_ZD dep_ZD_henceforth pobj_from_Corpus prep_tested_by nn__Knight cc_Knight_and conj_Knight_Marcu pobj_by_ amod_articles_general nn_articles_news appos__articles prep_tested_by nn__ cc__and conj__biomedical nn__articles pobj_by_ ' +o,6608,'The methods for calculating relative frequencies and lexical weights are also adapted for the weighted matrix case ',Koehn,'det_methods_The nsubjpass_adapted_methods prep_methods_for pcomp_for_calculating amod__relative nn__frequencies dobj_calculating_ cc__and amod__lexical nn__weights conj__ auxpass_adapted_are advmod_adapted_also prep_adapted_for det_case_the amod_case_weighted nn_case_matrix pobj_for_case ' +o,6609,'For instance , word alignment models are often trained using the GIZA + + toolkit ; error minimizing training criteria such as the Minimum Error Rate Training are employed in order to learn feature function weights for log-linear models ; and translation candidates are produced using phrase-based decoders in combination with n-gram language models ',Och,'prep_trained_For pobj_For_instance nn_models_word nn_models_alignment nsubjpass_trained_models auxpass_trained_are advmod_trained_often xcomp_trained_using det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__toolkit dobj_using_ nsubjpass_employed_error partmod_error_minimizing nn_criteria_training dobj_minimizing_criteria dep_as_such prep_criteria_as det__the nn__Minimum nn__Error nn__Rate nn__Training pobj_as_ auxpass_employed_are parataxis_trained_employed prep_employed_in pobj_in_order aux_learn_to xcomp_employed_learn nn_weights_feature nn_weights_function dobj_learn_weights prep_learn_for amod_models_log-linear pobj_for_models cc_trained_and nn_candidates_translation nsubjpass_produced_candidates auxpass_produced_are conj_trained_produced xcomp_produced_using amod_decoders_phrase-based dobj_using_decoders partmod_decoders_ prep__in pobj_in_combination prep__with amod_models_n-gram nn_models_language pobj_with_models advmod__ ' +o,6610,'Both training and testing sentences were processed using Collins parser to generate parse-tree automatically ',Collins,'det_sentences_Both nn_sentences_training cc_training_and conj_training_testing nsubjpass_processed_sentences auxpass_processed_were xcomp_processed_using nn__Collins nn__parser nsubj_generate_ aux_generate_to xcomp_using_generate dobj_generate_parse-tree advmod_generate_automatically ' +o,6611,'This linear model is learned using a variant of the incremental perceptron algorithm ',Collins,'det_model_This amod_model_linear nsubjpass_learned_model auxpass_learned_is xcomp_learned_using det_variant_a dobj_using_variant prep_variant_of det_algorithm_the amod_algorithm_incremental nn_algorithm_perceptron pobj_of_algorithm ' +o,6612,'32 Translation performance For the experiments reported in this section , we used feature weights trained with minimum error rate training Because MERT ignores the denominator in Equation 1 , it is invariant with respect to the scale of the weight vector the Moses implementation simply normalises the weight vector it finds by its lscript1-norm ',Och,'num_performance_32 nn_performance_Translation prep_used_For det_experiments_the pobj_For_experiments partmod_experiments_reported prep_reported_in det_section_this pobj_in_section nsubj_used_we dep_performance_used nn_weights_feature nsubj__weights partmod_weights_trained prep_trained_with amod_training_minimum nn_training_error nn_training_rate pobj_with_training ccomp_used_ mark_ignores_Because nsubj_ignores_MERT advcl_invariant_ignores det_denominator_the dobj_ignores_denominator prep_denominator_in pobj_in_Equation num_Equation_1 nsubj_invariant_it cop_invariant_is ccomp__invariant prep_invariant_with pobj_with_respect prep_invariant_to det_scale_the pobj_to_scale prep_scale_of det_vector_the nn_vector_weight pobj_of_vector det_implementation_the nn_implementation_Moses nsubj_normalises_implementation advmod_normalises_simply dep_performance_normalises det_vector_the nn_vector_weight dobj_normalises_vector nsubj_finds_it rcmod_vector_finds prep_finds_by poss_lscript1-norm_its pobj_by_lscript1-norm ' +o,6613,' , Pedersen , Yarowsky and Florian -RRB- as well as maximum entropy models -LRB- eg , Dang and Palmer , and Manning -RRB- ',Klein,'nn__Pedersen appos__ appos__Yarowsky cc_Yarowsky_and nn__Florian conj_Yarowsky_ dep_well_as cc_Yarowsky_well dep_well_as amod_models_maximum nn_models_entropy conj_Yarowsky_models dep_models_eg conj_eg_Dang cc_eg_and nn__Palmer nn__ conj__ cc__and conj__Manning conj_eg_ ' +o,6614,'Starting with bilingualphrasepairsextractedfromautomatically aligned parallel text , these PSCFG approaches augment each contiguous -LRB- in source and target words -RRB- phrase pair with a left-hand-side symbol -LRB- like the VP in the example above -RRB- , and perform a generalization procedure to form rules that include nonterminal symbols ',Och,'prep_approaches_Starting dep_Starting_with advmod_aligned_bilingualphrasepairsextractedfromautomatically amod__aligned amod__parallel nn__text pobj_with_ det_PSCFG_these nsubj_approaches_PSCFG xcomp_approaches_augment det_pair_each amod_pair_contiguous dep_contiguous_in nn_words_source cc_source_and conj_source_target pobj_in_words nn_pair_phrase dobj_augment_pair prep_augment_with det_symbol_a amod_symbol_left-hand-side pobj_with_symbol dep_symbol_like det_VP_the pobj_like_VP prep_VP_in det_example_the pobj_in_example advmod_example_above cc_augment_and conj_augment_perform det_procedure_a nn_procedure_generalization dobj_augment_procedure aux_form_to xcomp_augment_form dobj_form_rules nsubj_include_that rcmod_rules_include amod_symbols_nonterminal dobj_include_symbols ' +o,6615,'The IBM model 1 is used to find an initial estimate of the translation probabilities ',Brown,'det__The nn__IBM nn__model num__1 nsubjpass_used_ auxpass_used_is aux_find_to xcomp_used_find det_estimate_an amod_estimate_initial dobj_find_estimate prep_estimate_of det_probabilities_the nn_probabilities_translation pobj_of_probabilities ' +o,6616,'The accuracy of the generator outputs was evaluated by the BLEU score , which is commonly used for the evaluation of machine translation and recently used for the evaluation of generation ',Papineni,'det_accuracy_The nsubjpass_evaluated_accuracy prep_accuracy_of det_outputs_the nn_outputs_generator pobj_of_outputs auxpass_evaluated_was prep_evaluated_by det__the amod__BLEU nn__score pobj_by_ nsubjpass_used_which auxpass_used_is advmod_used_commonly rcmod__used prep_used_for det_evaluation_the pobj_for_evaluation prep_evaluation_of nn_translation_machine pobj_of_translation cc_used_and advmod_used_recently conj_used_used prep_used_for det_evaluation_the pobj_for_evaluation prep_evaluation_of nn__generation pobj_of_ ' +o,6617,'It us widely acknowledged that word sense d~samblguatmn (WSD) us a central problem m natural language processing In order for computers to be able to understand and process natural language beyond simple keyword matching, the problem of d~samblguatmg word sense, or dlscermng the meamng of a word m context, must be effectively dealt with Advances in WSD v, ill have slgmficant Impact on apphcatlons hke information retrieval and machine translation For natural language subtasks hke part-of-speech tagging or s)ntactm parsing, there are relatlvely well defined and agreed-upon cnterm of what it means to have the \'correct\' part of speech or syntactic structure assigned to a word or sentence For instance, the Penn Treebank corpus (Marcus et al, 1993) pro~ide~,t large repo.~tory of texts annotated w~th partof-speech and s}ntactm structure mformatlon Tv.o independent human annotators can achieve a high rate of agreement on assigning part-of-speech tags to words m a g~ven sentence Unfortunately, th~s us not the case for word sense assignment F~rstly, it is rarely the case that any two dictionaries will have the same set of sense defimtmns for a g~ven word Different d~ctlonanes tend to carve up the \'semantic space\' m a different way, so to speak Secondly, the hst of senses for a word m a typical dmtmnar~ tend to be rather refined and comprehensive This is especmlly so for the commonly used words which have a large number of senses The sense dustmctmn between the different senses for a commonly used word m a d~ctmnary hke WoRDNET (Miller, 1990) tend to be rather fine Hence, two human annotators may genuinely dusagree m their sense assignment to a word m context The agreement rate between human annotators on word sense assignment us an Important concern for the evaluatmn of WSD algorithms One would prefer to define a dusamblguatlon task for which there us reasonably hlgh agreement between human annotators The agreement rate between human annotators will then form the upper ceiling against whmh to compare the performance of WSD algorithms For instance, the SENSEVAL exerclse has performed a detaded study to find out the raterannotator agreement among ~ts lexicographers taggrog the word senses (Kllgamff, 1998c, Kllgarnff, 1998a, Kflgarrlff, 1998b) 2 A Case Study In this-paper, we examine the ~ssue of raterannotator agreement by comparing the agreement rate of human annotators on a large sense-tagged corpus of more than 30,000 instances of the most frequently occurring nouns and verbs of Enghsh This corpus is the intersection of the WORDNET Semcor corpus (Miller et al, 1993) and the DSO corpus (Ng and Lee, 1996, Ng, 1997), which has been independently tagged wlth the refined senses of WORDNET by two separate groups of human annotators The Semcor corpus us a subset of the Brown corpus tagged with ~VoRDNET senses, and consists of more than 670,000 words from 352 text files Sense taggmg was done on the content words (nouns, ~erbs, adjectives and adverbs) m this subset The DSO corpus consists of sentences drawn from the Brown corpus and the Wall Street Journal For each word w from a hst of 191 frequently occurring words of Enghsh (121 nouns and 70 verbs), sentences containing w (m singular or plural form, and m its various reflectional verb form) are selected and each word occurrence w ~s tagged w~th a sense from WoRDNET There ~s a total of about 192,800 sentences in the DSO corpus m which one word occurrence has been sense-tagged m each sentence The intersection of the Semcor corpus and the DSO corpus thus consists of Brown corpus sentences m which a word occurrence w is sense-tagged m each sentence, where w Is one of.the 191 frequently oc-,currmg English nouns or verbs Since this common pomon has been sense-tagged by two independent groups of human annotators, ~t serves as our data set for investigating inter-annotator agreement in this paper 3 Sentence Matching To determine the extent of inter-annotator agreement, the first step ~s to match each sentence m Semcor to its corresponding counterpart In the DSO corpus This step ~s comphcated by the following factors 1 Although the intersected portion of both corpora came from Brown corpus, they adopted different tokemzatmn convention, and segmentartan into sentences differed sometimes 2 The latest versmn of Semcor makes use of the senses from WORDNET 1 6, whereas the senses used m the DSO corpus were from WoRDNET 15 1 To match the sentences, we first converted the senses m the DSO corpus to those of WORDNET 1 6 We ignored all sentences m the DSO corpus m which a word is tagged with sense 0 or -1 (A word is tagged with sense 0 or -1 ff none of the given senses m WoRDNFT applies ) 4, sentence from Semcor is considered to match one from the DSO corpus ff both sentences are exactl) ldent~cal or ff the~ differ only m the pre~ence or absence of the characters \' (permd) or -\' (hyphen) For each remaining Semcor sentence, taking into account word ordering, ff 75\% or more of the words m the sentence match those in a DSO corpus sentence, then a potential match ~s recorded These i -kctua[ly, the WORD~q\'ET senses used m the DSO corpus were from a shght variant of the official WORDNE\'I 1 5 release Th~s ssas brought to our attention after the pubhc release of the DSO corpus potential matches are then manually verffied to ensure that they are true matches and to ~eed out any false matches Using this method of matching, a total of 13,188 sentence-palrs contasnmg nouns and 17,127 sentence-pa~rs containing verbs are found to match from both corpora, ymldmg 30,315 sentences which form the intersected corpus used m our present study 4 The Kappa Statistic Suppose there are N sentences m our corpus where each sentence contains the word w Assume that w has M senses Let 4 be the number of sentences which are assigned identical sense b~ two human annotators Then a simple measure to quantify the agreement rate between two human annotators Is Pc, where Pc, = A/N The drawback of this simple measure is that it does not take into account chance agreement between two annotators The Kappa statistic a (Cohen, 1960) is a better measure of rater-annotator agreement which takes into account the effect of chance agreement It has been used recently w~thm computatmnal hngu~stlcs to measure raterannotator agreement (Bruce and Wmbe, 1998, Carletta, 1996, Veroms, 1998) Let Cj be the sum of the number of sentences which have been assigned sense 3 by annotator 1 and the number of sentences whmh have been assigned sense 3 by annotator 2 Then P~-P~ 1-P~ where M j=l and Pe measures the chance agreement between two annotators A Kappa ~alue of 0 indicates that the agreement is purely due to chance agreement, whereas a Kappa ~alue of 1 indicates perfect agreement A Kappa ~alue of 0 8 and above is considered as mdmatmg good agreement (Carletta, 1996) Table 1 summarizes the inter-annotator agreement on the mtersected corpus The first (becond) row denotes agreement on the nouns (xerbs), wh~le the lass row denotes agreement on all words combined The a~erage ~ reported m the table is a s~mpie average of the individual ~ value of each word The agreement rate on the 30,315 sentences as measured by P= is 57\% This tallies with the figure reported ~n our earlier paper (Ng and Lee, 1996) where we performed a quick test on a subset of 5,317 sentences,n the intersection of both the Semcor corpus and the DSO corpus 10 [] mm m m m m m mm m m m m mm m m m Type Num of v, ords A N [ P~ Avg Nouns 121 7,676 13,188 I 0 582 0 300 Verbs 70 9,520 17,127 I 0 555 0 347 All I 191 I 17,196 30,315 I 056T 0317 Table 1 Raw inter-annotator agreement 5 Algorithm Since the rater-annotator agreement on the intersected corpus is not high, we would like to find out how the agreement rate would be affected if different sense classes were in use In this section, we present a greedy search algorithm that can automatmalb derive coarser sense classes based on the sense tags assigned by two human annotators The resulting derived coarse sense classes achmve a higher agreement rate but we still maintain as many of the original sense classes as possible The algorithm is given m Figure 1 The algorithm operates on a set of sentences where each sentence contains an occurrence of the word w whmh has been sense-tagged by two human annotators At each Iteration of the algorithm, tt finds the pair of sense classes Ct and Cj such that merging these two sense classes results in the highest t~ value for the resulting merged group of sense classes It then proceeds to merge Cz and C~ Thin process Is repeated until the ~ value reaches a satisfactory value ~,~t,~, which we set as 0 8 Note that this algorithm is also applicable to deriving any coarser set of classes from a refined set for any NLP tasks in which prior human agreement rate may not be high enough Such NLP tasks could be discourse tagging, speech-act categorization, etc 6 Results For each word w from the list of 121 nouns and 70 verbs, ~e applied the greedy search algorithm to each set of sentences in the intersected corpus contaming w For a subset of 95 words (53 nouns and 42 verbs), the algorithm was able to derive a coarser set of 2 or more senses for each of these 95 words such that the resulting Kappa ~alue reaches 0 8 or higher For the other 96 words, m order for the Kappa value to reach 0 8 or higher, the algorithm collapses all senses of the ~ord to a single (trivial) class Table 2 and 3 summarizes the results for the set of 53 nouns and 42 ~erbs, respectively Table 2 md~cates that before the collapse of sense classes, these 53 nouns have an average of 7 6 senses per noun There is a total of 5,339 sentences in the intersected corpus containing these nouns, of which 3,387 sentences were assigned the same sense by the two groups of human annotators The average Kappa statistic (computed as a simple average of the Kappa statistic of ~he mdlwdual nouns) is 0 463 After the collapse of sense classes by the greedy search algorithm, the average number of senses per noun for these 53 nouns drops to 40 Howe~er, the number of sentences which have been asmgned the same coarse sense by the annotators increases to 5,033 That is, about 94 3\% of the sentences have been assigned the same coarse sense, and that the average Kappa statistic has improved to 0 862, mgmfymg high rater-annotator agreement on the derived coarse senses Table3 gl~es the analogous figures for the 42 verbs, agmn mdmatmg that high agreement is achieved on the coarse sense classes den~ed for verbs 7 Discussion Our findings on rater-annotator agreement for word sense tagging indicate that for average language users, it is quite dl~cult to achieve high agreement when they are asked to assign refned sense tags (such as those found in WORDNET) given only the scanty definition entries m the WORDNET dlctionary and a few or no example sentences for the usage of each word sense Thin observation agrees wlth that obtmned m a recent study done by (Veroms, 1998), where the agreement on sense-tagging by naive users was also not hlgh Thus It appears that an average language user is able to process language wlthout needing to perform the task of dlsamblguatmg word sense to a very fine-grained resolutmn as formulated m a tradltlonal dmtlonary In contrast, expert lexicographers tagged the ~ ord sense in the sentences used m the SENSEVAL exerclse, where high rater-annotator agreement was reported There are also fuller dlctlonary entries m the HECTOR dlctlonary used and more e ~* then ~\' +~(C~,,C~_t), z* +~, ~* +end for merge the sense class C,.',Carletta,'' +o,6618,'We use the IBM Model 1 and the Hidden Markov Model -LRB- HMM , -RRB- to estimate the alignment model ',Brown,'nsubj_use_We det__the nn__IBM nn__Model num__1 nsubj_estimate_ cc__and det_Model_the nn_Model_Hidden nn_Model_Markov conj__Model appos_Model_HMM dep_HMM_ aux_estimate_to xcomp_use_estimate det_model_the amod_model_alignment dobj_estimate_model ' +o,6619,'We provide results using a range of automatic evaluation metrics : BLEU , Precision and Recall , and Wordand Sentence Error Rates ',Papineni,'nsubj_provide_We dobj_provide_results partmod_results_using det_range_a dobj_using_range prep_range_of amod_metrics_automatic nn_metrics_evaluation pobj_of_metrics nn__BLEU dep_results_ conj__Precision cc__and nn__Recall conj__ cc_results_and nn_Rates_Wordand nn_Rates_Sentence nn_Rates_Error conj_results_Rates ' +o,6620,'ROUGE-L and ROUGE-1 are supposed to be appropriate for the headline gener853 ation task ',Lin,'nsubjpass_supposed_ROUGE-L cc_ROUGE-L_and conj_ROUGE-L_ROUGE-1 auxpass_supposed_are aux_appropriate_to cop_appropriate_be xcomp_supposed_appropriate prep_appropriate_for det_task_the nn_task_headline nn_task_gener853 nn_task_ation pobj_for_task ' +o,6621,'The common types of features include contextual , co-occurrence , and syntactic dependency ',Lin,'det_types_The amod_types_common nsubj_include_types prep_types_of pobj_of_features amod__contextual dobj_include_ amod__co-occurrence conj__ cc__and amod_dependency_syntactic conj__dependency ' +o,6622,'32 Results and Discussion The BLEU scores for 10 direct translations and 4 sets of heuristic selections 4Admittedly , in typical instances of such chains , English would appear earlier ',Papineni,'num_Results_32 nsubj_appear_Results cc_Results_and conj_Results_Discussion det_4Admittedly_The num_scores_BLEU measure__scores amod_4Admittedly_ prep__for num_translations_10 amod_translations_direct pobj_for_translations cc_translations_and num_sets_4 conj_translations_sets prep_sets_of amod_selections_heuristic pobj_of_selections dep_Discussion_4Admittedly prep_appear_in amod_instances_typical pobj_in_instances prep_instances_of amod_chains_such pobj_of_chains nsubj_appear_English aux_appear_would advmod_appear_earlier ' +o,6623,'Furthermore , WASP1 + + employs minimum error rate training to directly optimize the evaluation metrics ',Och,'advmod_employs_Furthermore nn_+_WASP1 nn_+_+ nsubj_employs_+ amod__minimum nn__error nn__rate nn__training dobj_employs_ aux_optimize_to advmod_optimize_directly xcomp_employs_optimize det_metrics_the nn_metrics_evaluation dobj_optimize_metrics ' +o,6624,'6 The Experiments To investigate the e ects of lookahead on our family of deterministic parsers , we ran empirical experiments on the standard the Penn Treebank datasets ',Marcus,'det_Experiments_The nsubj_investigate_Experiments aux_investigate_To dep_ran_investigate det_ects_the amod_ects_e dobj_investigate_ects prep_ects_of pobj_of_lookahead prep_investigate_on poss_family_our pobj_on_family prep_family_of amod_parsers_deterministic pobj_of_parsers nsubj_ran_we dep_6_ran amod_experiments_empirical dobj_ran_experiments prep_ran_on det_standard_the pobj_on_standard det__the nn__Penn nn__Treebank nsubj_datasets_ dep_6_datasets ' +o,6625,'We have already shown in Section 3 how to solve -LRB- a -RRB- ; here we avoid -LRB- b -RRB- by maximizing conditional likelihood , marginalizing out the hidden variable , denotedz : max vector summationdisplay x , y p -LRB- x , y -RRB- log summationdisplay z pvector -LRB- y , z x -RRB- -LRB- 17 -RRB- This sort of conditional training with hidden variables was carried out by , for example , in reranking ; it is related to the information bottleneck method and contrastive estimation ',Koo,'nsubj_shown_We aux_shown_have advmod_shown_already prep_shown_in pobj_in_Section num_Section_3 advmod_carried_how aux_solve_to csubjpass_carried_solve dep_-RRB-_a dobj_solve_here nsubj_avoid_we rcmod_here_avoid dep_-LRB-_b prep_solve_by pcomp_by_maximizing amod_likelihood_conditional dobj_maximizing_likelihood dep_maximizing_marginalizing prt_marginalizing_out det_variable_the amod_variable_hidden dobj_marginalizing_variable dep_maximizing_denotedz nn_vector_max nsubj_summationdisplay_vector ccomp_denotedz_summationdisplay nsubj_log_x nn_p_y appos_x_p dep_p_x appos_x_y ccomp_summationdisplay_log advmod_log_summationdisplay amod_pvector_z dobj_log_pvector appos_pvector_y nn_x_z dep_y_x appos_pvector_17 det_sort_This dobj_maximizing_sort prep_sort_of amod_training_conditional pobj_of_training prep_training_with amod_variables_hidden pobj_with_variables auxpass_carried_was ccomp_shown_carried prt_carried_out prep_carried_by pobj_by_ prep_carried_for pobj_for_example prep_example_in pobj_in_reranking nsubjpass_related_it auxpass_related_is parataxis_shown_related prep_related_to det__the nn__information nn__bottleneck nn__method pobj_to_ cc__and nn__contrastive nn__estimation conj__ ' +o,6626,'This paper continues a line of research on online discriminative training , extending that of Watanabe et al ',Tillmann,'det_paper_This nsubj_continues_paper det_line_a dobj_continues_line prep_line_of pobj_of_research prep_research_on amod__online amod__discriminative nn__training pobj_on_ xcomp_continues_extending dobj_extending_that prep_that_of pobj_of_Watanabe cc_Watanabe_et conj_Watanabe_al ' +o,6627,'For comparison , we use the MT training program , GIZA + + , the phrase-base decoder , Pharaoh , and the wordbased decoder , Rewrite ',Koehn,'prep_use_For pobj_For_comparison nsubj_use_we det_program_the nn_program_MT nn_program_training dobj_use_program nn__GIZA nn__+ nn__+ conj_program_ det_decoder_the amod_decoder_phrase-base conj_program_decoder nn__Pharaoh conj_program_ cc_program_and det_decoder_the amod_decoder_wordbased conj_program_decoder nn__Rewrite appos_decoder_ ' +o,6628,'5 Comparison with related work Preliminary work on SF extraction from coq ~ ora was done by and ',Dunning,'num_Comparison_5 nsubjpass_done_Comparison prep_Comparison_with amod_work_related nn_work_work amod_work_Preliminary pobj_with_work prep_work_on nn_extraction_SF pobj_on_extraction prep_extraction_from nn_ora_coq nn_ora_~ pobj_from_ora auxpass_done_was prep_done_by pobj_by_ cc__and conj__ ' +o,6629,'Recentworkconsidersadamagedtagdictionary by assuming that tags are known only for words that occur more than once or twice ',Johnson,'prep_Recentworkconsidersadamagedtagdictionary_by pcomp_by_assuming complm_known_that nsubjpass_known_tags auxpass_known_are ccomp_assuming_known advmod_known_only prep_known_for pobj_for_words nsubj_occur_that rcmod_words_occur advmod_once_more dep_once_than advmod_occur_once cc_Recentworkconsidersadamagedtagdictionary_or advmod_Recentworkconsidersadamagedtagdictionary_twice ' +o,6630,'An existing method to combine multiple parsing algorithms is the ensemble approach , which was reported to be useful in improving dependency parsing ',Hall,'det_method_An amod_method_existing nsubj__method aux_combine_to infmod_method_combine amod_algorithms_multiple amod_algorithms_parsing dobj_combine_algorithms cop__is det__the nn__ensemble nn__approach nsubjpass_reported_which auxpass_reported_was rcmod__reported aux_useful_to cop_useful_be xcomp_reported_useful prep_useful_in pcomp_in_improving amod_parsing_dependency dobj_improving_parsing ' +o,6631,'33 BLEU Score The BLEU score measures the agreement between a hypothesiseI1 generated by the MT system and a reference translation eI1 ',Papineni,'number_BLEU_33 num_Score_BLEU det__The amod__BLEU nn__score nsubj_measures_ dep_Score_measures det_agreement_the dobj_measures_agreement prep_agreement_between det_hypothesiseI1_a pobj_between_hypothesiseI1 partmod_hypothesiseI1_generated prep_generated_by det_system_the nn_system_MT pobj_by_system cc_system_and det_eI1_a nn_eI1_reference nn_eI1_translation conj_system_eI1 ' +o,6632,'As far as the log-linear combination of float features is concerned , similar training procedures have been proposed in ',Och,'mark_concerned_As advmod_as_far advmod_concerned_as det_combination_the amod_combination_log-linear nsubj_concerned_combination prep_combination_of nn_features_float pobj_of_features cop_concerned_is advcl_proposed_concerned amod_procedures_similar nn_procedures_training nsubjpass_proposed_procedures aux_proposed_have auxpass_proposed_been prep_proposed_in ' +o,6633,'Proceedings of the 40th Annual Meeting of the Association for , a number of other algorithms have been developed ',Brown,'nsubjpass_developed_Proceedings prep_Proceedings_of det_Meeting_the amod_Meeting_40th amod_Meeting_Annual pobj_of_Meeting prep_Meeting_of det_Association_the pobj_of_Association prep_Association_for pobj_for_ det_number_a nsubjpass_developed_number prep_number_of amod_algorithms_other pobj_of_algorithms aux_developed_have auxpass_developed_been ' +o,6634,'However , most of them do not build a NEs resource but exploit external gazetteers , ',Cucerzan,'advmod_build_However advmod_build_most dep_most_of pobj_of_them aux_build_do neg_build_not det_resource_a nn_resource_NEs dobj_build_resource cc_build_but conj_build_exploit amod_gazetteers_external dobj_exploit_gazetteers advmod_exploit_ ' +o,6635,'Then P -LRB- eI1jfj1 -RRB- = summationtextaI 1 P -LRB- eI1 , aI1jfj1 -RRB- ',Marcus,'advmod_P_Then nsubj_summationtextaI_eI1jfj1 dep_summationtextaI_= ccomp_P_summationtextaI num_P_1 nsubj_P_P appos_P_eI1 dep_eI1_aI1jfj1 partmod_P_ ' +o,6636,'to estimale a model -LRB- clustering words -RRB- , and measured the I -LRB- L distancd ~ between ` l ` he K -RRB- , distance -LRB- relative Clt , l : Opy -RRB- , which is widely used in information theory and sta , tist , ics , is a , nleasur ,2 of ` dista , n -LRB- : c \' l -RRB- ~ -LRB- , wcen two distributions 52 Experiment 2 : Qualitative Evaluation We extracted roughly 180,000 case fl : anles from the bracketed WSJ -LRB- Wall Street Journal -RRB- corpus of the Penn Tree Bank as co-occurrence data ',Marcus,'aux_estimale_to det_model_a dobj_estimale_model amod_words_clustering appos_model_words cc_estimale_and conj_estimale_measured det_I_the dobj_measured_I nn_~_L amod_~_distancd dep_I_~ prep_~_between pcomp_between_l dobj_l_he dep_l_K appos_I_distance amod_Clt_relative appos_distance_Clt nn_Opy_l dep_Clt_Opy nsubj_used_which cop_used_is advmod_used_widely rcmod_I_used prep_used_in nn_theory_information pobj_in_theory cc_theory_and conj_theory_sta nsubj_a_tist appos_tist_ics cop_a_is ccomp_used_a appos_a_nleasur num_nleasur_,2 prep_nleasur_of pobj_of_dista appos_dista_n dep_n_c amod_~_l nn_Qualitative_~ dep_Experiment_wcen num_distributions_two dep_wcen_distributions nn_Experiment_52 dep_~_Experiment dep_Experiment_2 prep_c_Qualitative dep_Qualitative_Evaluation nsubj_extracted_We rcmod_Evaluation_extracted quantmod_180,000_roughly num_fl_180,000 nn_fl_case dobj_extracted_fl dep_measured_anles dep_measured_from det_corpus_the amod_corpus_bracketed nn_corpus_WSJ nn_Journal_Wall nn_Journal_Street appos_corpus_Journal pobj_from_corpus prep_corpus_of det__the nn__Penn nn__Tree nn__Bank pobj_of_ prep_corpus_as amod_data_co-occurrence pobj_as_data ' +o,6637,'http://ducnistgov 2004 Journal of the Association for Computing Machinery 16 264 -- 285 , Message Understanding Conferences -LRB- MUC -RRB- , TIPSTER SUMMAC Text Summarization Evaluation , Document Understanding Conference -LRB- DUC -RRB- , and Text Summarization Challenge -LRB- TSC -RRB- , have attested the importance of this topic ',Marcus,'nn_Journal_http://ducnistgov num_Journal_ num_Journal_ num_Journal_2004 nn_Journal_ nn_Journal_ nsubj_attested_Journal prep_Journal_of det_Association_the pobj_of_Association prep_Association_for nn__Computing nn__Machinery nn__ nn__ num__16 pobj_for_ number_264_ num__264 number__285 num__ nn__ nn__ dep__ nn__Message nn__Understanding nn__Conferences abbrev__MUC conj__ nn__TIPSTER nn__SUMMAC nn__Text nn__Summarization nn__Evaluation conj__ nn__Document nn__Understanding nn__Conference abbrev__DUC conj__ cc__and nn__Text nn__Summarization nn__Challenge abbrev__TSC conj__ aux_attested_have det_importance_the dobj_attested_importance prep_importance_of det_topic_this pobj_of_topic ' +o,6638,'The row labeled Precision shows the precision of the extracted information -LRB- ie , how many entries are correct , according to a human annotator -RRB- estimated by random sampling and manual evaluation of 1 \% of the data for each table , similar to ',Fleischman,'det_row_The nsubj_labeled_row nsubj_shows_Precision ccomp_labeled_shows det_precision_the dobj_shows_precision prep_precision_of det_information_the amod_information_extracted pobj_of_information dep_precision_ie advmod_many_how dep_correct_many nsubj_correct_entries cop_correct_are ccomp_ie_correct prep_ie_according dep_according_to det_annotator_a amod_annotator_human pobj_to_annotator partmod_precision_estimated prep_estimated_by amod_sampling_random pobj_by_sampling cc_sampling_and nn_evaluation_manual conj_sampling_evaluation prep_sampling_of num_\%_1 pobj_of_\% prep_\%_of det_data_the pobj_of_data prep_data_for det_table_each pobj_for_table amod_precision_similar prep_similar_to ' +o,6639,' , -RRB- ',Pang,'appos__ ' +o,6640,'We use the union , re ned and intersection heuristics de ned in which are used in conjunction with IBM Model 4 as the baseline in virtually all recent work on word alignment ',Och,'nsubj_use_We det_union_the dobj_use_union nn_ned_re conj_union_ned cc_union_and nn_heuristics_intersection conj_union_heuristics dep_union_de dep_union_ned prep_ned_in pobj_in_ nsubjpass_used_which auxpass_used_are rcmod_ned_used prep_used_in pobj_in_conjunction prep_used_with nn_Model_IBM pobj_with_Model num_Model_4 prep_used_as det_baseline_the pobj_as_baseline prep_baseline_in advmod_work_virtually det_work_all amod_work_recent pobj_in_work prep_work_on nn_alignment_word pobj_on_alignment ' +o,6641,'This algorithm is proved to converge -LRB- ie , there are no more updates -RRB- in the separable case 1 Thatis , ifthereexistweightvectorU -LRB- with U = 1 -RRB- , -LRB- -RRB- 0 -RRB- , and R -LRB- -RRB- 0 -RRB- that satisfy : i , y Y xi -LRB- xi , yi -RRB- U -LRB- xi , y -RRB- U , i , y Y xi -LRB- xi , yi -RRB- -LRB- xi , y -RRB- R , the number of updates is at most R2\\/2 ',Collins,'det_algorithm_This nsubjpass_proved_algorithm auxpass_proved_is aux_converge_to xcomp_proved_converge nsubj_are_ie expl_are_there parataxis_proved_are advmod_updates_no advmod_no_more acomp_are_updates prep_proved_in det_case_the amod_case_separable pobj_in_case partmod_case_ num_Thatis_1 dobj__Thatis amod_Thatis_ifthereexistweightvectorU dep_ifthereexistweightvectorU_with pobj_with_U dep_1_= rcmod_U_1 num_-RRB-_0 cc_case_and conj_case_R num_-RRB-_0 nsubj_satisfy_that rcmod_R_satisfy dep_y_i dep_case_y nn_U_Y nn_U_xi dep_U_xi dep_xi_yi prep_y_U dep_U_xi dep_xi_y dep_U_U dep_xi_i nn_xi_y nn_xi_Y appos_U_xi dep_xi_xi dep_xi_yi dep_xi_xi dep_xi_y nsubj_is_R det_number_the appos_R_number prep_number_of pobj_of_updates rcmod_xi_is prep_is_at amod_R2\\/2_most pobj_at_R2\\/2 ' +o,6642,' , Mihalcea and Moldovan -LRB- 2000 -RRB- , and Mihalcea -LRB- 2002 -RRB- have made further research to obtain large corpus of higher quality from an initial seed corpus ',Yarowsky,'nsubj_made_ conj__Mihalcea cc__and conj__Moldovan appos__2000 cc__and conj__Mihalcea appos_Mihalcea_2002 aux_made_have amod_research_further dobj_made_research aux_obtain_to xcomp_made_obtain amod_corpus_large dobj_obtain_corpus prep_corpus_of amod_quality_higher pobj_of_quality prep_quality_from det_corpus_an amod_corpus_initial nn_corpus_seed pobj_from_corpus ' +o,6643,'There are many POS taggers developed using different techniques for many major languages such as transformation-based error-driven learning , decision trees , Markov model , maximum entropy methods etc for English ',Cutting,'expl_are_There amod_taggers_many dep_taggers_POS nsubj_are_taggers partmod_taggers_developed xcomp_developed_using amod_techniques_different dobj_using_techniques prep_using_for amod_languages_many amod_languages_major pobj_for_languages dep_as_such prep_languages_as amod__transformation-based amod__error-driven nn__learning pobj_as_ nn__decision nn__trees appos__ nn__Markov nn__model appos__ amod_methods_maximum nn_methods_entropy appos__methods partmod_methods_ dobj__etc prep__for pobj_for_English ' +o,6644,'The list is obtained by first extracting the phrases with - TMP function tags from the PennTree bank , and taking the words in these phrases ',Marcus,'det_list_The nsubjpass_obtained_list auxpass_obtained_is prep_obtained_by advmod_extracting_first pcomp_by_extracting det_phrases_the dobj_extracting_phrases prep_phrases_with nn_tags_TMP nn_tags_function pobj_with_tags prep_tags_from det_bank_the nn_bank_PennTree pobj_from_bank cc_extracting_and conj_extracting_taking det_words_the dobj_taking_words prep_taking_in det_phrases_these pobj_in_phrases ' +o,6645,'These include scripts for creating alignments from a parallel corpus , creating phrase tables and language models , binarizing phrase tables , scripts for weight optimization using MERT , and testing scripts ',Och,'nsubj_include_These dobj_include_scripts prep_include_for pcomp_for_creating dobj_creating_alignments prep_creating_from det_corpus_a amod_corpus_parallel pobj_from_corpus conj_creating_creating nn_tables_phrase dobj_creating_tables cc_tables_and nn_models_language conj_tables_models conj_creating_binarizing nn_tables_phrase dobj_binarizing_tables appos_tables_scripts prep_scripts_for nn_optimization_weight pobj_for_optimization partmod_optimization_using nn__MERT dobj_using_ cc_creating_and conj_creating_testing dobj_testing_scripts ' +o,6646,'The standard solution is to approximate the maximum probability translation using a single derivation ',Koehn,'det_solution_The amod_solution_standard nsubj_is_solution aux_approximate_to xcomp_is_approximate det_translation_the amod_translation_maximum nn_translation_probability dobj_approximate_translation partmod_translation_using det_derivation_a amod_derivation_single dobj_using_derivation ' +p,6647,'Properly calculated BLEU scores have been shown to correlate reliably with human judgments ',Papineni,'amod_scores_Properly amod_scores_calculated nn_scores_BLEU nsubjpass_shown_scores aux_shown_have auxpass_shown_been aux_correlate_to xcomp_shown_correlate advmod_correlate_reliably prep_correlate_with amod_judgments_human pobj_with_judgments ' +o,6648,'It is known that PMI gives undue importance to low frequency events , therefore the evaluation considers only pairs of genes that occur at least 5 times in the whole corpus ',Dunning,'nsubjpass_known_It auxpass_known_is complm_gives_that nsubj_gives_PMI ccomp_known_gives amod_importance_undue dobj_gives_importance prep_gives_to amod_events_low nn_events_frequency pobj_to_events nsubj_considers_ advmod_considers_therefore det_evaluation_the nsubj_considers_evaluation dep_known_considers amod_pairs_only dobj_considers_pairs prep_pairs_of pobj_of_genes nsubj_occur_that rcmod_genes_occur quantmod_times_at dep_at_least number_times_5 dobj_occur_times prep_times_in det_corpus_the amod_corpus_whole pobj_in_corpus ' +o,6649,'To model p -LRB- t , a s -RRB- , we use a standard loglinear approach : p -LRB- t , a s -RRB- exp bracketleftBiggsummationdisplay i ifi -LRB- s , t , a -RRB- bracketrightBigg where each fi -LRB- s , t , a -RRB- is a feature function , and weights i are set using Ochs algorithm to maximize the systems BLEU score on a development corpus ',Och,'aux_model_To dep_use_model dobj_model_p dep_p_t det_s_a appos_t_s nsubj_use_we det_approach_a amod_approach_standard nn_approach_loglinear dobj_use_approach nsubj_exp_p dep_p_t det_s_a appos_t_s parataxis_use_exp advmod_i_bracketleftBiggsummationdisplay amod_ifi_i dobj_exp_ifi appos_ifi_s dep_s_t dep_s_a xcomp_exp_bracketrightBigg advmod_function_where det_fi_each nsubj_function_fi appos_fi_s dep_s_t dep_s_a cop_function_is det_function_a nn_function_feature advcl_bracketrightBigg_function cc_use_and nn_i_weights nsubjpass_set_i auxpass_set_are conj_use_set xcomp_set_using nn__Ochs nn__algorithm nsubj_maximize_ aux_maximize_to xcomp_using_maximize det_systems_the nsubj_BLEU_systems dep_maximize_BLEU nn__score dobj_BLEU_ prep_BLEU_on det_corpus_a nn_corpus_development pobj_on_corpus ' +o,6650,'Context extraction begins with a Maximum Entropy POS tagger and chunker ',Ratnaparkhi,'amod_extraction_Context nsubj_begins_extraction prep_begins_with det_tagger_a nn_tagger_Maximum nn_tagger_Entropy dep_tagger_POS pobj_with_tagger cc_tagger_and conj_tagger_chunker ' +o,6651,'For each cell in the contingency table , the expected counts are : mi j = ni + n + jn + + The measures are calculated as : 2 = i ; j -LRB- ni j mi j -RRB- 2 mi j LL = 2 i ; j log2 n 2i j mi j Log-likelihood ratios are more appropriate for sparse data than chi-square ',Dunning,'prep_j_For det_cell_each pobj_For_cell prep_cell_in det_table_the nn_table_contingency pobj_in_table det_counts_the amod_counts_expected nsubj_j_counts cop_j_are nn_j_mi rcmod_j_= nn_n_ni nn_n_+ dobj_=_n cc_jn_+ amod_n_jn cc_jn_+ conj_jn_+ det_measures_The nsubjpass_calculated_measures auxpass_calculated_are rcmod_n_calculated prep_calculated_as pobj_as_ dep_j_2 dep_i_= rcmod_2_i dep_j_j nn_j_ni nn_j_j nn_j_mi appos_j_j num_LL_2 nn_LL_mi nn_LL_j dep_j_LL dep_appropriate_= num_i_2 dobj_=_i nn_j_j nn_j_log2 nn_j_n nn_j_2i nn_j_j nn_j_mi dep_i_j amod_ratios_Log-likelihood nsubj_appropriate_ratios advmod_appropriate_ cop_appropriate_are advmod_appropriate_more dep_j_appropriate prep_appropriate_for amod_data_sparse pobj_for_data prep_data_than pobj_than_chi-square ' +o,6652,'1 Introduction During the last few years , SMT systems have evolved from the original word-based approach to phrase-based translation systems ',Brown,'num_Introduction_1 prep_evolved_During det_years_the amod_years_last amod_years_few pobj_During_years nn_systems_SMT nsubj_evolved_systems aux_evolved_have dep_Introduction_evolved prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep__to amod__phrase-based nn__translation nn__systems pobj_to_ ' +o,6653,'Minimum error rate training -LRB- MERT -RRB- with respect to BLEU score was used to tune the decoders parameters , and performed using the technique proposed in ',Och,'nn_training_Minimum nn_training_error nn_training_rate nsubjpass_used_training abbrev_training_MERT prep_training_with pobj_with_respect prep_respect_to amod_score_BLEU pobj_to_score auxpass_used_was aux_tune_to xcomp_used_tune det_parameters_the amod_parameters_decoders dobj_tune_parameters cc_used_and conj_used_performed xcomp_performed_using det_technique_the dobj_using_technique partmod_technique_proposed prep_proposed_in ' +o,6654,'Head word -LRB- and its part-of-speech tag -RRB- of the constituent After POS tagging , a syntactic parser was then used to obtain the parse tree for the sentence ',Collins,'nn_word_Head nsubjpass_used_word cc_word_and poss_tag_its amod_tag_part-of-speech conj_word_tag prep_word_of det_constituent_the pobj_of_constituent prep_used_After dep_tagging_POS pobj_After_tagging det__a amod__syntactic nn__parser nsubjpass_used_ auxpass_used_was advmod_used_then aux_obtain_to xcomp_used_obtain det_tree_the amod_tree_parse dobj_obtain_tree prep_tree_for det_sentence_the pobj_for_sentence ' +o,6655,'Otherwise they are generated along with the words using the same approach as in ',Collins,'dep_Otherwise_they auxpass_generated_are rcmod_they_generated prt_generated_along prep_generated_with det_words_the nsubj_using_words pcomp_with_using det_approach_the amod_approach_same dobj_using_approach prep_using_as dep_as_in ' +o,6656,'We use the maximum entropy tagging method described in for the experiments , which is a variant of modified to use HMM state features ',Ratnaparkhi,'nsubj_use_We det_method_the amod_method_maximum amod_method_entropy nn_method_tagging dobj_use_method partmod_method_described prep_described_in pobj_in_ prep_described_for det_experiments_the pobj_for_experiments nsubj_variant_which cop_variant_is det_variant_a rcmod_experiments_variant prep_variant_of pobj_of_ partmod__modified aux_use_to xcomp_modified_use nn_features_HMM nn_features_state dobj_use_features ' +p,6657,'The technique is based on word class models , pioneered by , which hierarchically 151 CoNLL03 CoNLL03 MUC7 MUC7 Web Component Test data Dev data Dev Test pages 1 -RRB- Baseline 8365 8925 7472 7128 7141 2 -RRB- -LRB- 1 -RRB- + Gazetteer Match 8722 9161 8583 8043 7446 3 -RRB- -LRB- 1 -RRB- + Word Class Model 8682 9085 8025 7988 7226 4 -RRB- All External Knowledge 8855 9249 8450 8323 7444 Table 4 : Utility of external knowledge ',Brown,'det_technique_The nsubjpass_based_technique auxpass_based_is prep_based_on nn_models_word nn_models_class pobj_on_models amod_pages_pioneered prep_pioneered_by pobj_by_ dep_151_which advmod_151_hierarchically rcmod__151 nn_Test_CoNLL03 nn_Test_CoNLL03 nn_Test_MUC7 nn_Test_MUC7 nn_Test_Web nn_Test_Component nn_Test_Test nn_Test_data nn_Test_Dev nn_Test_data nn_Test_Dev nsubj_151_Test appos_models_pages dep_Baseline_1 dep_pages_Baseline num_Baseline_8365 prep_Baseline_8925 number_7128_7472 amod_8925_7128 dep_7128_7141 dep_7141_2 dep_Gazetteer_1 nn_Gazetteer_+ dep_8925_Gazetteer prep_Gazetteer_Match number_9161_8722 num_8043_9161 number_8043_8583 pobj_Match_8043 num_3_7446 dep_Gazetteer_3 dep_Model_1 nn_Model_+ nn_Model_Word nn_Model_Class dep_8925_Model num_Model_8682 number_8025_9085 num_7226_8025 number_7226_7988 prep_Model_7226 dep_Knowledge_4 det_Knowledge_All amod_Knowledge_External dep_7226_Knowledge number_9249_8855 num_Knowledge_9249 partmod_Knowledge_8450 number_7444_8323 num_Table_7444 dobj_8450_Table num_Table_4 dep_Model_Utility prep_Utility_of amod_knowledge_external pobj_of_knowledge ' +p,6658,'1 Introduction During the last few years , SMT systems have evolved from the original word-based approach to phrase-based translation systems ',Koehn,'num_Introduction_1 prep_evolved_During det_years_the amod_years_last amod_years_few pobj_During_years nn_systems_SMT nsubj_evolved_systems aux_evolved_have dep_Introduction_evolved prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep__to amod_systems_phrase-based nn_systems_translation pobj_to_systems ' +o,6659,'The PropBank superimposes an annotation of semantic predicate-argument structures on top of the Penn Treebank -LRB- PTB -RRB- ',Marcus,'det_PropBank_The nsubj_superimposes_PropBank det_annotation_an dobj_superimposes_annotation prep_annotation_of amod_structures_semantic amod_structures_predicate-argument pobj_of_structures prep_superimposes_on pobj_on_top prep_top_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank abbrev_Treebank_PTB ' +o,6660,'3 Variational Bayes for ITG Goldwater and Griffiths and show that modifying an HMM to include a sparse prior over its parameters and using Bayesian estimation leads to improved accuracy for unsupervised part-of-speech tagging ',Johnson,'num_Bayes_3 nn_Bayes_Variational nsubj_show_Bayes prep_Bayes_for nn_Goldwater_ITG pobj_for_Goldwater cc_Goldwater_and nn__Griffiths conj_Goldwater_ cc_Goldwater_and nn__ conj_Goldwater_ complm_leads_that csubj_leads_modifying det_HMM_an dobj_modifying_HMM aux_include_to infmod_HMM_include det_sparse_a dobj_include_sparse dep_over_prior prep_include_over poss_parameters_its pobj_over_parameters cc_modifying_and conj_modifying_using amod_estimation_Bayesian dobj_using_estimation ccomp_show_leads prep_leads_to amod_accuracy_improved pobj_to_accuracy prep_accuracy_for amod_tagging_unsupervised amod_tagging_part-of-speech pobj_for_tagging ' +o,6661,'As a measure of association , we use the loglikelihood-ratio statistic recommended by , which is the same statistic used by Melamed to initialize his models ',Dunning,'prep_use_As det_measure_a pobj_As_measure prep_measure_of pobj_of_association nsubj_use_we det_statistic_the amod_statistic_loglikelihood-ratio nsubj_recommended_statistic ccomp_use_recommended prep_recommended_by pobj_by_ nsubj_statistic_which cop_statistic_is det_statistic_the amod_statistic_same rcmod__statistic partmod_statistic_used prep_used_by pobj_by_Melamed aux_initialize_to xcomp_used_initialize poss_models_his dobj_initialize_models ' +o,6662,'Since that time , however , increasingly large amounts of language model training data have become available ranging from approximately one billion words -LRB- the Gigaword corpora from the Linguistic Data Consortium -RRB- to trillions of words ',Brants,'prep_become_Since det_time_that pobj_Since_time advmod_become_however advmod_large_increasingly amod_amounts_large nsubj_become_amounts prep_amounts_of nn_data_language nn_data_model nn_data_training pobj_of_data aux_become_have dep_become_available xcomp_become_ranging prep_ranging_from quantmod_billion_approximately number_billion_one num_words_billion pobj_from_words det_corpora_the nn_corpora_Gigaword dep_words_corpora prep_corpora_from det_Consortium_the nn_Consortium_Linguistic nn_Consortium_Data pobj_from_Consortium prep_ranging_to pobj_to_trillions prep_trillions_of pobj_of_words ' +o,6663,'This approach , however , does not have a theoretical guarantee on optimality unless certain nontrivial conditions are satisfied ',Abney,'det_approach_This nsubj_guarantee_approach advmod_approach_however aux_guarantee_does neg_guarantee_not dep_guarantee_have det_guarantee_a amod_guarantee_theoretical prep_guarantee_on pobj_on_optimality mark_satisfied_unless amod_conditions_certain amod_conditions_nontrivial nsubjpass_satisfied_conditions auxpass_satisfied_are advcl_guarantee_satisfied ' +o,6664,'Most prior work on the speci c problem of categorizing expressly opinionated text has focused on the binary distinction of positive vs negative ',Turney,'advmod_work_Most advmod_Most_prior prep_work_on det_speci_the pobj_on_speci dep_focused_c nsubj_focused_problem prep_problem_of pcomp_of_categorizing advmod_categorizing_expressly amod_text_opinionated dobj_categorizing_text aux_focused_has dep_work_focused prep_focused_on det_distinction_the amod_distinction_binary pobj_on_distinction prep_distinction_of amod_negative_positive amod_negative_vs pobj_of_negative ' +o,6665,'This is the scenario considered by for POS tagging : how to construct an accurate tagger given a set of tags and a few example words for each of those tags ',Haghighi,'nsubj_scenario_This cop_scenario_is det_scenario_the partmod_scenario_considered prep_considered_by pobj_by_ prep__for dep_tagging_POS pobj_for_tagging advmod_construct_how aux_construct_to parataxis_scenario_construct det_tagger_an amod_tagger_accurate dobj_construct_tagger partmod_tagger_given det_set_a iobj_given_set prep_set_of pobj_of_tags cc_tags_and det_example_a amod_example_few conj_tags_example dobj_given_words prep_words_for pobj_for_each prep_each_of det_tags_those pobj_of_tags ' +o,6666,'The pchemtb-closed shared task is used to illustrate our models ',Marcus,'det__The amod__pchemtb-closed amod__shared nn__task nsubjpass_used_ auxpass_used_is aux_illustrate_to xcomp_used_illustrate poss_models_our dobj_illustrate_models ' +o,6667,'1 Introduction The field of sentiment classification has received considerable attention from researchers in recent years ',Turney,'num_Introduction_1 det_field_The nsubj_received_field prep_field_of nn_classification_sentiment pobj_of_classification aux_received_has dep_Introduction_received amod_attention_considerable dobj_received_attention prep_received_from pobj_from_researchers prep_researchers_in amod_years_recent pobj_in_years ' +p,6668,'Recently , have successfully constructed high quality and high coverage gazetteers from Wikipedia ',Kazama,'advmod_constructed_Recently nsubj_constructed_ aux_constructed_have advmod_constructed_successfully amod_quality_high dobj_constructed_quality cc_quality_and amod_gazetteers_high nn_gazetteers_coverage conj_quality_gazetteers prep_constructed_from pobj_from_Wikipedia ' +o,6669,'This paper presents an empirical study measuring the effectiveness of our evaluation functions at selecting training sentences from the Wall Street Journal -LRB- WSJ -RRB- corpus for inducing grammars ',Marcus,'det_paper_This nsubj_presents_paper det_study_an amod_study_empirical dobj_presents_study partmod_study_measuring det_effectiveness_the dobj_measuring_effectiveness prep_effectiveness_of poss_functions_our nn_functions_evaluation pobj_of_functions prep_measuring_at pcomp_at_selecting nn_sentences_training dobj_selecting_sentences prep_selecting_from det__the nn_Journal_Wall nn_Journal_Street nn__Journal abbrev_Journal_WSJ nn__corpus pobj_from_ prep__for amod_grammars_inducing pobj_for_grammars ' +p,6670,'1 Introduction The maximum entropy model has attained great popularity in the NLP field due to its power , robustness , and successful performance in various NLP tasks ',Ratnaparkhi,'num_Introduction_1 det__The amod__maximum amod__entropy nn__model nsubj_attained_ aux_attained_has dep_Introduction_attained amod_popularity_great dobj_attained_popularity prep_popularity_in det_field_the nn_field_NLP pobj_in_field dep_to_due prep_attained_to poss_power_its pobj_to_power conj_power_robustness cc_power_and amod_performance_successful conj_power_performance prep_attained_in amod_tasks_various nn_tasks_NLP pobj_in_tasks ' +o,6671,'We then rank-order the P X|Y MI XY M Z Pr Z|Y MI ZY G092log [P X P Y P X P Y ] f Y [P XY P XY ] f XY [P XY P XY ] f XY M iG13X,X} jG13Y,Y} (f ij G09 ij ) 2 ij f XY G09 XY XY (1G09( XY /N)) f XY G09 XY f XY (1G09(f XY /N)) Table 1: Probabilistic Approaches METHOD FORMULA Frequency (Guiliano, 1964) f XY Pointwise Mutual Information (MI) (Fano, 1961; Church and Hanks, 1990) log (P / PP) 2XY XY Selectional Association (Resnik, 1996) Symmetric Conditional Probability (Ferreira and Pereira, 1999) P / PP XY X Y 2 Dice Formula (Dice, 1945) 2 f / (f +f ) XY X Y Log-likelihood (Dunning, 1993; (Daille, 1996).',Dunning,'' +p,6672,'The implementation of MEBA was strongly influenced by the notorious five IBM models described in ',Brown,'det_implementation_The nsubjpass_influenced_implementation prep_implementation_of pobj_of_MEBA auxpass_influenced_was advmod_influenced_strongly prep_influenced_by det_models_the amod_models_notorious num_models_five nn_models_IBM pobj_by_models partmod_models_described prep_described_in ' +o,6673,'6 Related work Evidence from the surrounding context has been used previously to determine if the current sentence should be subjective\\/objective and adjacency pair information has been used to predict congressional votes ',Pang,'num_Evidence_6 amod_Evidence_Related nn_Evidence_work nsubjpass_used_Evidence prep_Evidence_from det_context_the amod_context_surrounding pobj_from_context aux_used_has auxpass_used_been advmod_used_previously aux_determine_to xcomp_used_determine mark__if det_sentence_the amod_sentence_current nsubj__sentence aux__should cop__be amod__subjective\\/objective advcl_determine_ cc__and amod_information_adjacency nn_information_pair nsubjpass_used_information aux_used_has auxpass_used_been conj__used aux_predict_to xcomp_used_predict amod_votes_congressional dobj_predict_votes advmod_predict_ ' +o,6674,'The topic signatures are automatically generated for each specific term by computing the likelihood ratio -LRB- - score -RRB- between two hypotheses ',Dunning,'det_signatures_The nn_signatures_topic nsubjpass_generated_signatures auxpass_generated_are advmod_generated_automatically prep_generated_for det_term_each amod_term_specific pobj_for_term prep_generated_by pcomp_by_computing det_ratio_the nn_ratio_likelihood dobj_computing_ratio appos_ratio_score prep_ratio_between num_hypotheses_two pobj_between_hypotheses ' +o,6675,'Based on the word alignment results , if the aligned target words of any two adjacent foreign linguistic phrases can also be formed into two valid adjacent phrase according to constraints proposed in the phrase extraction algorithm by , they will be extracted as a reordering training sample ',Och,'prep_extracted_Based dep_Based_on det_results_the nn_results_word nn_results_alignment pobj_on_results mark_formed_if det_words_the amod_words_aligned nn_words_target nsubjpass_formed_words prep_words_of det_phrases_any measure_adjacent_two amod_phrases_adjacent amod_phrases_foreign amod_phrases_linguistic pobj_of_phrases aux_formed_can advmod_formed_also auxpass_formed_be dep_extracted_formed prep_formed_into num_phrase_two amod_phrase_valid amod_phrase_adjacent pobj_into_phrase prep_formed_according dep_according_to pobj_to_constraints partmod_constraints_proposed prep_proposed_in det_algorithm_the nn_algorithm_phrase nn_algorithm_extraction pobj_in_algorithm prep_proposed_by pobj_by_ nsubjpass_extracted_they aux_extracted_will auxpass_extracted_be prep_extracted_as det_sample_a nn_sample_reordering nn_sample_training pobj_as_sample ' +o,6676,'These wordbased models are used to find the latent wordalignments between bilingual sentence pairs , from which a weighted string transducer can be induced -LRB- either finite state or synchronous context free grammar -RRB- ',Koehn,'det_models_These amod_models_wordbased nsubjpass_used_models auxpass_used_are aux_find_to xcomp_used_find det_wordalignments_the amod_wordalignments_latent dobj_find_wordalignments prep_wordalignments_between amod_pairs_bilingual nn_pairs_sentence pobj_between_pairs prep_used_from dobj_induced_which det_transducer_a amod_transducer_weighted nn_transducer_string nsubjpass_induced_transducer aux_induced_can auxpass_induced_be pcomp_from_induced dep_induced_either amod__finite nn__state dep_either_ cc__or amod__synchronous nn__context amod__free nn__grammar conj__ ' +o,6677,'These heuristics are extensions of those developed for phrase-based models , and involve symmetrising two directional word alignments followed by a projection step which uses the alignments to find a mapping between source words and nodes in the target parse trees ',Koehn,'det_heuristics_These nsubj_extensions_heuristics cop_extensions_are prep_extensions_of det__those amod__developed prep_developed_for amod_models_phrase-based pobj_for_models pobj_of_ cc_extensions_and conj_extensions_involve xcomp_involve_symmetrising num_alignments_two amod_alignments_directional nn_alignments_word dobj_symmetrising_alignments partmod_alignments_followed prep_followed_by det_step_a nn_step_projection pobj_by_step nsubj_uses_which rcmod_step_uses det_alignments_the nsubj_find_alignments aux_find_to xcomp_uses_find det_mapping_a dobj_find_mapping prep_mapping_between nn_words_source pobj_between_words cc_words_and conj_words_nodes prep_find_in det__the nn__target nn__parse nn__trees pobj_in_ ' +o,6678,'Iterating between these two 1 Note that these problems are associated with corpus-based approaches in general , and have been identified by a number of researchers ',Yarowsky,'prep_Iterating_between dep_1_these num_1_two pobj_between_1 dobj_Iterating_Note complm_associated_that det_problems_these nsubjpass_associated_problems auxpass_associated_are dep_Note_associated prep_associated_with amod_approaches_corpus-based pobj_with_approaches prep_approaches_in pobj_in_general cc_associated_and aux_identified_have auxpass_identified_been conj_associated_identified prep_identified_by det_number_a pobj_by_number prep_number_of pobj_of_researchers ' +o,6679,'The parameters of the MT system were optimized on MTEval02 data using minimum error rate training ',Och,'det_parameters_The nsubjpass_optimized_parameters prep_parameters_of det_system_the nn_system_MT pobj_of_system auxpass_optimized_were prep_optimized_on nn_data_MTEval02 pobj_on_data xcomp_optimized_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +o,6680,'Our starting point is the work done by Zettlemoyer and Collins on parsing using relaxed CCG grammars -LRB- ZC07 -RRB- ',Zettlemoyer,'poss_point_Our amod_point_starting nsubj_work_point cop_work_is det_work_the partmod_work_done prep_done_by pobj_by_Zettlemoyer cc_Zettlemoyer_and conj_Zettlemoyer_Collins prep_done_on pcomp_on_parsing xcomp_parsing_using nn__relaxed nn__CCG nn__grammars dobj_using_ abbrev__ZC07 ' +o,6681,'Note that unlike the constructions in and no errors are possible for ngrams stored in the model ',Talbot,'complm_possible_that prep_possible_unlike det_constructions_the pobj_unlike_constructions prep_constructions_in pobj_in_ cc__and conj__ det_errors_no nsubj_possible_errors cop_possible_are ccomp_Note_possible prep_possible_for pobj_for_ngrams partmod_ngrams_stored prep_stored_in det_model_the pobj_in_model ' +p,6682,'Annotated reference corpora , such as the Brown Corpus , the Penn Treebank , and the BNC , have helped both the development of English computational linguistics tools and English corpus linguistics ',Marcus,'amod_corpora_Annotated nn_corpora_reference nsubj_helped_corpora dep_as_such prep_corpora_as det__the nn__Brown nn__Corpus pobj_as_ det__the nn__Penn nn__Treebank conj__ cc__and det__the nn__BNC conj__ aux_helped_have predet_development_both det_development_the dobj_helped_development prep_development_of nn_tools_English amod_tools_computational nn_tools_linguistics pobj_of_tools cc_tools_and amod_linguistics_English nn_linguistics_corpus conj_tools_linguistics ' +o,6683,'That is a significant shortcoming , because in many domains , hard or soft global constraints on the label sequence are motivated by common sense : For named entity recognition , a phrase that appears multiple times should tend to get the same label each time ',Finkel,'nsubj_shortcoming_That cop_shortcoming_is det_shortcoming_a amod_shortcoming_significant mark_motivated_because prep_motivated_in amod_domains_many pobj_in_domains amod_constraints_hard cc_hard_or conj_hard_soft amod_constraints_global nsubjpass_motivated_constraints prep_constraints_on det_sequence_the nn_sequence_label pobj_on_sequence auxpass_motivated_are advcl_shortcoming_motivated prep_motivated_by amod_sense_common pobj_by_sense prep_tend_For amod_recognition_named nn_recognition_entity pobj_For_recognition det_phrase_a nsubj_tend_phrase nsubj_times_that cop_times_appears amod_times_multiple rcmod_phrase_times aux_tend_should parataxis_shortcoming_tend aux_get_to xcomp_tend_get det_label_the amod_label_same dobj_get_label det_time_each tmod_get_time ' +o,6684,' proposed to eliminate objective sentences before the sentiment classification of documents ',Pang,'nsubj_proposed_ aux_eliminate_to xcomp_proposed_eliminate amod_sentences_objective dobj_eliminate_sentences prep_eliminate_before det_classification_the nn_classification_sentiment pobj_before_classification prep_classification_of pobj_of_documents ' +o,6685,'Model weights were also trained following ',Och,'nn_weights_Model nsubjpass_trained_weights auxpass_trained_were advmod_trained_also xcomp_trained_following ' +o,6686,'By contrast , in the training method proposed by , the discriminative function f -LRB- x ; w -RRB- is estimated to maximize the F 1 - score of training dataset D This training method employs an approximate form of the F 1 - score obtained by using a logistic function ',Jansche,'prep_function_By pobj_By_contrast prep_function_in det_method_the nn_method_training pobj_in_method partmod_method_proposed prep_proposed_by pobj_by_ det_discriminative_the nsubj_function_discriminative nsubjpass_estimated_f dep_f_x dep_x_w auxpass_estimated_is ccomp_function_estimated aux_maximize_to xcomp_estimated_maximize det_F_the dobj_maximize_F dep_F_1 dep_F_score prep_score_of nn_D_training nn_D_dataset pobj_of_D det_method_This nn_method_training nsubj_employs_method rcmod_D_employs det_form_an amod_form_approximate dobj_employs_form prep_form_of det_F_the pobj_of_F num_F_1 dep_form_score partmod_score_obtained prep_obtained_by pcomp_by_using det_function_a amod_function_logistic dobj_using_function ' +o,6687,'Most of the reported work on paraphrase generation from arbitrary input sentences uses machine learning techniques trained on sentences that are known or can be inferred to be paraphrases of each other ',Barzilay,'nsubjpass_uses_Most prep_Most_of det_work_the amod_work_reported pobj_of_work prep_work_on amod_generation_paraphrase pobj_on_generation prep_generation_from amod_sentences_arbitrary nn_sentences_input pobj_from_sentences dobj_uses_machine partmod_machine_learning dobj_learning_techniques partmod_techniques_trained prep_trained_on pobj_on_sentences nsubjpass_known_that auxpass_known_are rcmod_sentences_known cc_uses_or aux_inferred_can auxpass_inferred_be conj_uses_inferred aux_paraphrases_to cop_paraphrases_be purpcl_inferred_paraphrases prep_paraphrases_of det_other_each pobj_of_other ' +o,6688,'The resulting training procedure is analogous to the one presented in and ',Brown,'det_procedure_The amod_procedure_resulting nn_procedure_training nsubj_analogous_procedure cop_analogous_is prep_analogous_to det_one_the pobj_to_one partmod_one_presented prep_presented_in pobj_in_ cc__and conj__ ' +o,6689,'He uses a specic reliability statistic , , for his measurements , but implicitly assumes kappa-like metrics are similar enough in practice for the rule of thumb to apply to them as wellA detailed discussion on the differences and similarities of these , and other , measures is provided by Krippendorff ; in this article we will use Cohens to investigate the value of the 08 reliability cut-off for computational linguistics ',Carletta,'nsubj_uses_He det_statistic_a amod_statistic_specic dep_specic_reliability dobj_uses_statistic prep_statistic_for poss_measurements_his pobj_for_measurements cc_uses_but nsubj_assumes_ advmod_assumes_implicitly conj_uses_assumes amod_metrics_kappa-like nsubj_discussion_metrics dep_discussion_are amod_discussion_similar advmod_similar_enough prep_similar_in pobj_in_practice prep_practice_for det_rule_the pobj_for_rule prep_rule_of pobj_of_thumb aux_apply_to xcomp_similar_apply prep_apply_to pobj_to_them prep_apply_as pobj_as_wellA amod_discussion_detailed ccomp_assumes_discussion prep_discussion_on det_differences_the pobj_on_differences cc_differences_and conj_differences_similarities prep_differences_of pobj_of_these cc_assumes_and advmod_provided_other nsubjpass_provided_measures auxpass_provided_is conj_assumes_provided prep_provided_by nn__Krippendorff pobj_by_ prep_use_in det_article_this pobj_in_article nsubj_use_we aux_use_will parataxis_uses_use nn__Cohens nsubj_investigate_ aux_investigate_to xcomp_use_investigate det_value_the dobj_investigate_value prep_value_of det_cut-off_the num_cut-off_08 amod_cut-off_reliability pobj_of_cut-off prep_investigate_for amod_linguistics_computational pobj_for_linguistics ' +o,6690,'We used \\* TH \\* = 3 following \' a very rough rule of thumb \' used for word-based mutual information in ',Church,'nsubj_used_We nn_\\*_\\* nn_\\*_TH nsubj_3_\\* dep_3_= ccomp_used_3 prep_used_following det_rule_a advmod_rough_very amod_rule_rough pobj_following_rule prep_rule_of pobj_of_thumb partmod_rule_used prep_used_for amod_information_word-based amod_information_mutual pobj_for_information prep_used_in ' +o,6691,'The query tions , the syntax , semantics , and abstract knowledge representation have type declarations which help to detect malformed representations ',Crouch,'det_tions_The amod_tions_query nsubj_have_tions det_syntax_the conj_tions_syntax conj_tions_semantics cc_tions_and amod_representation_abstract nn_representation_knowledge conj_tions_representation nn__type nn__declarations dobj_have_ nsubj_help_which rcmod__help aux_detect_to xcomp_help_detect amod_representations_malformed dobj_detect_representations ' +o,6692,'Also relevant is previous work that applied machine learning approaches to MT evaluation , both with human references and without ',Liu,'advmod_relevant_Also cop_work_is amod_work_previous dep_relevant_work nsubj_applied_that rcmod_work_applied nn_approaches_machine nn_approaches_learning dobj_applied_approaches prep_applied_to nn_evaluation_MT pobj_to_evaluation preconj_with_both prep_applied_with amod__human nn__references pobj_with_ cc_with_and conj_with_without pobj_without_ ' +o,6693,'Finally , recent efforts have also looked at transfer learning mechanisms for sentiment analysis , eg , see ',Blitzer,'advmod_looked_Finally amod_efforts_recent nsubj_looked_efforts aux_looked_have advmod_looked_also prep_looked_at pobj_at_transfer partmod_transfer_learning dobj_learning_mechanisms mark_see_for nn_analysis_sentiment nsubj_see_analysis appos_analysis_eg advcl_learning_see ' +o,6694,'31 Data The English data set consists of the Wall Street Journal sections 2-24 of the Penn treebank , converted to dependency format ',Marcus,'num_set_31 nn_set_Data nn_set_The nn_set_English nn_set_data nsubj_consists_set prep_consists_of det_2-24_the nn_2-24_Wall nn_2-24_Street nn_2-24_Journal nn_2-24_sections pobj_of_2-24 prep_2-24_of det__the nn__Penn nn__treebank pobj_of_ partmod_2-24_converted prep_converted_to amod_format_dependency pobj_to_format ' +p,6695,'1 Introduction Syntactically annotated corpora like the Penn Treebank , the NeGra corpus or the statistically dismnbiguated parses in provide a wealth of intbrmation , which can only be exploited with an adequate query language ',Marcus,'num_corpora_1 nn_corpora_Introduction nn_corpora_Syntactically amod_corpora_annotated nsubj_provide_corpora prep_corpora_like det__the nn__Penn nn__Treebank pobj_like_ det__the nn__NeGra nn__corpus conj__ cc__or det_parses_the advmod_dismnbiguated_statistically amod_parses_dismnbiguated conj__parses prep_parses_in pobj_in_ det_wealth_a dobj_provide_wealth prep_wealth_of pobj_of_intbrmation nsubjpass_exploited_which aux_exploited_can advmod_exploited_only auxpass_exploited_be rcmod_intbrmation_exploited prep_exploited_with det_language_an amod_language_adequate nn_language_query pobj_with_language ' +o,6696,' lter training instances based on Part-of-Speech -LRB- POS -RRB- tags , and Soricut and Marcu use syntactic features to identify sentence-internal RST structure ',Marcu,'amod_instances_ nn_instances_lter nn_instances_training nsubj_based_instances prep_based_on amod_tags_Part-of-Speech appos_tags_POS pobj_on_tags cc_based_and nn__Soricut cc_Soricut_and conj_Soricut_Marcu nsubj_use_ conj_based_use amod_features_syntactic nsubj_identify_features aux_identify_to xcomp_use_identify amod_structure_sentence-internal nn_structure_RST dobj_identify_structure ' +o,6697,'Finally , knowledge of polarity can be combined with corpus-based collocation extraction methods to automatically produce entries for the lexical functions used in MeaningText Theory for text generation ',Smadja,'advmod_combined_Finally nsubjpass_combined_knowledge prep_knowledge_of pobj_of_polarity aux_combined_can auxpass_combined_be prep_combined_with amod_methods_corpus-based nn_methods_collocation nn_methods_extraction pobj_with_methods partmod_methods_ aux_produce_to advmod_produce_automatically xcomp__produce dobj_produce_entries prep_entries_for det_functions_the amod_functions_lexical pobj_for_functions partmod_functions_used prep_used_in nn__MeaningText nn__Theory pobj_in_ prep__for nn_generation_text pobj_for_generation ' +o,6698,'Our evaluation metrics are BLEU and NIST , which are to perform caseinsensitive matching of n-grams up to n = 4 ',Papineni,'poss_metrics_Our nn_metrics_evaluation nsubj_BLEU_metrics aux_BLEU_are dobj_BLEU_ cc__and conj__NIST nsubj_are_which rcmod__are aux_perform_to xcomp_are_perform amod_matching_caseinsensitive dobj_perform_matching prep_matching_of pobj_of_n-grams advmod_perform_up dep_up_to pcomp_to_n dep_n_= dobj_n_4 ' +o,6699,'Part-of-speech features Based on the lexical categories produced by GATE , each token xi is classified into one of a set of coarse part-of-speech tags : noun , verb , adverb , wh-word , determiner , punctuation , etc We do the same for neighboring words in a -LRB- 2 , +2 -RRB- window in order to assist noun phrase segmentation ',Pang,'amod_features_Part-of-speech prep_classified_Based dep_Based_on det_categories_the amod_categories_lexical pobj_on_categories partmod_categories_produced prep_produced_by nn__GATE pobj_by_ det_xi_each amod_xi_token nsubjpass_classified_xi auxpass_classified_is dep_features_classified prep_classified_into pobj_into_one prep_one_of det_set_a pobj_of_set prep_set_of amod_tags_coarse amod_tags_part-of-speech pobj_of_tags nn_punctuation_noun dep_punctuation_verb dep_punctuation_adverb dep_punctuation_wh-word dep_punctuation_determiner nn_tags_punctuation dep_tags_etc nsubj_do_We dep_features_do det_window_the amod_window_same prep_same_for amod_words_neighboring pobj_for_words prep_words_in dep_2_a pobj_in_2 appos_2_+2 nsubj_assist_window prep_window_in pobj_in_order aux_assist_to xcomp_do_assist amod_segmentation_noun nn_segmentation_phrase dobj_assist_segmentation ' +o,6700,'Unsupervised Learning : Results To test the effectiveness of the above unsupervised learning algorithm , we ran a number of experiments using two different corpora and part of speech tag sets : the Penn Treebank Wall Street Journal Corpus and the original Brown Corpus ',Marcus,'nn_Learning_Unsupervised dep_Learning_Results aux_test_To infmod_Results_test det_effectiveness_the dobj_test_effectiveness prep_effectiveness_of det_unsupervised_the amod_unsupervised_above pobj_of_unsupervised dep_ran_learning dobj_learning_algorithm nsubj_ran_we dep_Learning_ran det_number_a dobj_ran_number prep_number_of pobj_of_experiments partmod_experiments_using num_corpora_two amod_corpora_different dobj_using_corpora cc_corpora_and conj_corpora_part prep_part_of nn_sets_speech nn_sets_tag pobj_of_sets det__the nn__Penn nn__Treebank nn__Wall nn__Street nn__Journal nn__Corpus dep_corpora_ cc_corpora_and det__the amod__original nn__Brown nn__Corpus conj_corpora_ ' +o,6701,'33 Language Model We estimate P -LRB- s -RRB- using n-gram LMs trained on data from the Web , using Stupid Backoff ',Brants,'num_Model_33 nn_Model_Language nsubj_estimate_We dep_Model_estimate nsubj_using_P dep_-LRB-_s dep_estimate_using nn_LMs_n-gram nsubj_trained_LMs ccomp_using_trained prep_trained_on pobj_on_data prep_trained_from det_Web_the pobj_from_Web xcomp_trained_using nn_Backoff_Stupid dobj_using_Backoff ' +p,6702,'Similarly , Structural Correspondence Learning has proven to be successful for the two tasks examined , PoS tagging and Sentiment Classification ',Blitzer,'advmod_proven_Similarly amod__Structural nn__Correspondence nn__Learning nsubj_proven_ aux_proven_has aux_successful_to cop_successful_be xcomp_proven_successful mark_examined_for det_tasks_the num_tasks_two nsubj_examined_tasks advcl_successful_examined nsubj_tagging_PoS dep_examined_tagging cc_tagging_and conj_tagging_Sentiment dobj_tagging_Classification ' +o,6703,'We use these tuples to calculate a balanced f-score against the gold alignment tuples4 Method Dict size f-score Gold 28 1000 Monotone 39 689 IBM-1 30 803 IBM-4 29 869 IP 28 959 The last line shows an average f-score over the 8 tied IP solutions ',Brown,'nsubj_use_We det_tuples_these nsubj_calculate_tuples aux_calculate_to xcomp_use_calculate det_f-score_a amod_f-score_balanced nsubj_959_f-score prep_f-score_against det_tuples4_the amod_tuples4_gold nn_tuples4_alignment pobj_against_tuples4 nn_959_Method nn_959_Dict nn_959_size amod_959_f-score nn_959_Gold num_959_28 num_959_1000 nn_959_Monotone num_959_39 num_959_689 nn_959_IBM-1 num_959_ num_959_30 num_959_803 nn_959_IBM-4 num_959_ num_959_29 num_959_869 nn_959_IP num_959_28 xcomp_calculate_959 det_line_The amod_line_last nsubj_shows_line dep_use_shows det_f-score_an amod_f-score_average dobj_shows_f-score prep_f-score_over det_solutions_the num_solutions_8 amod_solutions_tied nn_solutions_IP pobj_over_solutions ' +o,6704,'We split the treebank into training -LRB- sections 0-18 -RRB- , development and test -LRB- sections 22-24 -RRB- as in ',Collins,'nsubj_split_We det_treebank_the dobj_split_treebank prep_split_into pobj_into_training number_0-18_sections dep_training_0-18 nn__development conj_training_ cc_training_and conj_training_test number_22-24_sections dep_test_22-24 prep_split_as dep_as_in ' +o,6705,'The a0 coefficient is computed as follows: a0 a47 a1a32a2 a9 a1 a30 a68 a9 a1a32a30 Carletta (1996) reports that content analysis researchers generally think of a0a34a33 a49a36a35a37 as good reliability, with a49a36a35a38a40a39a37a41 a0 a41a25a49a36a35a37 allowing tentative conclusions to be drawn. All that remains is to define the chance agreement probability a1 a30 . Let a1a32a41 a1 a30 a7 and a1a32a42 a1 a30 a7 be the fraction of utterances that begin or end one or more segments in segmentation a30 respectively.',Carletta,'' +o,6706,' proposed such a method for word sense disambiguation , which we refer to as monolingual bootstrapping ',Yarowsky,'nsubj_proposed_ predet_method_such det_method_a dobj_proposed_method prep_proposed_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation rel_refer_which nsubj_refer_we rcmod_disambiguation_refer prep_refer_to advmod_monolingual_as amod_bootstrapping_monolingual pobj_to_bootstrapping ' +o,6707,'The COlllillOil poini ; s regarding collocations appear to be , as suggestsl : they are m ` bil ; rary -LRB- it is nol ; clear why to ` Bill through \' means to ` fail \' -RRB- , th -LRB- \' y are domain-dependent -LRB- ` interest rate \' , ` stock market \' -RRB- , t ; hey are recurrenl ; and cohesive lo ~ xical clusters : the presence of one of the ',Smadja,'det_poini_The nn_poini_COlllillOil nsubj_regarding_s dep_poini_regarding nsubj_appear_collocations ccomp_regarding_appear aux_be_to xcomp_appear_be prep_appear_as amod_suggestsl_ pobj_as_suggestsl nsubjpass_m_they auxpass_m_are parataxis_regarding_m dobj_m_bil nsubj_t_rary nsubj_nol_it cop_nol_is dep_rary_nol csubj_means_clear advmod_Bill_why aux_Bill_to ccomp_clear_Bill prt_Bill_through dep_nol_means aux_fail_to xcomp_means_fail appos_rary_th dep_th_y cop_domain-dependent_are rcmod_y_domain-dependent nn_rate_interest dep_domain-dependent_rate nn_market_stock appos_rate_market parataxis_regarding_t nsubj_recurrenl_hey cop_recurrenl_are parataxis_regarding_recurrenl cc_regarding_and amod_lo_cohesive nsubj_~_lo conj_regarding_~ amod_clusters_xical dobj_~_clusters det_presence_the dep_clusters_presence prep_presence_of pobj_of_one prep_presence_of pobj_of_the ' +o,6708,' and classified sentiment polarity of reviews at the document level ',Turney,'amod_polarity_ cc__and conj__ amod_polarity_ amod_polarity_classified nn_polarity_sentiment prep_polarity_of pobj_of_reviews prep_reviews_at det_level_the nn_level_document pobj_at_level ' +o,6709,'Penn Treebank corpus sections 0-20 were used for training , sections 2124 for testing ',Marcus,'nn_0-20_Penn nn_0-20_Treebank nn_0-20_corpus amod_0-20_ nn_0-20_sections nsubjpass_used_0-20 auxpass_used_were prep_used_for pobj_for_training appos_training_sections num_sections_2124 prep_sections_for pobj_for_testing ' +o,6710,' introduced five statistical translation models -LRB- IBM Models 1 5 -RRB- ',Brown,'advmod_introduced_ num_models_five amod_models_statistical nn_models_translation nsubj_introduced_models nn_Models_IBM dep_models_Models num_5_1 dep_Models_5 ' +o,6711,'In particular , Abney defines a function K that is an upper bound on the negative log-likelihood , and shows his bootstrapping algorithms locally minimize K We now present a generalization of Abneys K function and relate it to another semi-supervised learning technique , entropy regularization ',Jiao,'prep_defines_In pobj_In_particular nsubj_defines_Abney det_K_a nn_K_function dobj_defines_K nsubj_regularization_that cop_regularization_is det_regularization_an amod_regularization_upper amod_regularization_bound prep_bound_on det_technique_the amod_technique_negative amod_technique_log-likelihood cc_log-likelihood_and conj_log-likelihood_shows poss_algorithms_his amod_algorithms_bootstrapping dobj_shows_algorithms advmod_shows_locally dep_shows_minimize dobj_minimize_K nsubj_present_We advmod_present_now rcmod_K_present det_generalization_a dobj_present_generalization prep_generalization_of nn_function_Abneys nn_function_K pobj_of_function cc_present_and conj_present_relate dobj_relate_it prep_relate_to det_learning_another amod_learning_semi-supervised pobj_to_learning pobj_on_technique advmod_technique_entropy rcmod_K_regularization ' +o,6712,'Second , the word alignment is refined by a grow-diag-final heuristic ',Koehn,'advmod_refined_Second det_alignment_the nn_alignment_word nsubjpass_refined_alignment auxpass_refined_is prep_refined_by det_heuristic_a amod_heuristic_grow-diag-final pobj_by_heuristic ' +o,6713,'The word alignment is computed using GIZA + +2 for the selected 73,597 sentence pairs in the FBIS corpus in both directions and then combined using union and heuristic diagonal growing ',Koehn,'det_alignment_The nn_alignment_word nsubjpass_computed_alignment auxpass_computed_is xcomp_computed_using nn_+_GIZA dobj_using_+ num_+_+2 prep_using_for det_pairs_the amod_pairs_selected num_pairs_73,597 nn_pairs_sentence pobj_for_pairs prep_pairs_in det_corpus_the nn_corpus_FBIS pobj_in_corpus prep_using_in det_directions_both pobj_in_directions cc_computed_and advmod_computed_then conj_computed_combined xcomp_combined_using dobj_using_union cc_union_and amod_diagonal_heuristic conj_union_diagonal amod_diagonal_growing ' +o,6714,'There has of course been a large amount of work on the more general problem of word-sense disambiguation , eg , ',Yarowsky,'expl_been_There aux_been_has prep_been_of pobj_of_course det_amount_a amod_amount_large dobj_been_amount prep_amount_of pobj_of_work prep_been_on det_problem_the advmod_general_more amod_problem_general pobj_on_problem prep_problem_of amod_disambiguation_word-sense pobj_of_disambiguation appos_disambiguation_eg nn__ appos_eg_ ' +o,6715,' estimates a POS tagging error rate of 3 \% in the Treebank ',Ratnaparkhi,'amod_estimates_ det_rate_a dep_rate_POS nn_rate_tagging nn_rate_error dep_estimates_rate prep_rate_of num_\%_3 pobj_of_\% prep_\%_in det_Treebank_the pobj_in_Treebank ' +o,6716,'Word correspondence was further developed in IBM Model-1 for statistical machine translation ',Brown,'amod_correspondence_Word nsubjpass_developed_correspondence auxpass_developed_was advmod_developed_further prep_developed_in nn__IBM nn__Model-1 pobj_in_ prep__for amod_translation_statistical nn_translation_machine pobj_for_translation ' +o,6717,'The features that define the constraints on the model are obtained by instantiation of feature templates as in ',Ratnaparkhi,'det_features_The nsubjpass_obtained_features nsubj_define_that rcmod_features_define det_constraints_the dobj_define_constraints prep_define_on det_model_the pobj_on_model auxpass_obtained_are prep_obtained_by pobj_by_instantiation prep_instantiation_of nn_templates_feature pobj_of_templates prep_obtained_as dep_as_in ' +o,6718,'Step Description mean stddev \% 15 Sample 15s 007s 07 \% 16 Extraction 382s 013s 186 \% 17 Build tree 1276s 2760s 623 \% 18 Percolation 314s 491s 153 \% 1911 Leaf updates 62s 175s 30 \% 1511 Total 2049s 326s 1000 \% 2004 -RRB- ,10 the only one that we were able to train and test under exactly the same experimental conditions -LRB- including the use of POS tags from -RRB- ',Ratnaparkhi,'nn_Description_Step partmod_Description_mean num_\%_stddev iobj_mean_\% num_007s_15 nn_007s_Sample amod_007s_15s dobj_mean_007s num_\%_07 dobj_mean_\% num_013s_16 nn_013s_Extraction amod_013s_382s dobj_mean_013s num_\%_186 dobj_mean_\% num_2760s_17 nn_2760s_Build nn_2760s_tree nn_2760s_1276s dobj_mean_2760s num_\%_623 dobj_mean_\% num_491s_18 nn_491s_Percolation amod_491s_314s dobj_mean_491s num_\%_153 dobj_mean_\% num_updates_1911 amod_updates_Leaf dobj_mean_updates amod_175s_62s dobj_mean_175s num_\%_30 dobj_mean_\% num_326s_1511 amod_326s_Total amod_326s_2049s dobj_mean_326s num_\%_1000 dobj_mean_\% prep_mean_2004 pobj_2004_,10 det_one_the amod_one_only dep_Description_one complm_able_that nsubj_able_we cop_able_were dep_one_able aux_train_to xcomp_able_train cc_train_and conj_train_test prep_test_under advmod_conditions_exactly det_conditions_the amod_conditions_same amod_conditions_experimental pobj_under_conditions dep_conditions_including det_use_the pobj_including_use prep_use_of dep_tags_POS pobj_of_tags prep_tags_from pobj_from_ ' +o,6719,'Considerations of sentence fluency are also key in sentence simplification , sentence compression , text re-generation for summarization and headline generation ','Daume III','nsubj_key_Considerations prep_Considerations_of nn_fluency_sentence pobj_of_fluency cop_key_are advmod_key_also prep_key_in nn__sentence nn__simplification pobj_in_ nn__sentence nn__compression appos__ nn_re-generation_text attr_key_re-generation prep_re-generation_for nn__summarization pobj_for_ cc__and nn__headline nn__generation conj__ ' +o,6720,'Movie and product reviews have been the main focus of many of the recent studies in this area ',Pang,'nn_reviews_Movie cc_Movie_and conj_Movie_product nsubj_been_reviews aux_been_have det_focus_the amod_focus_main dobj_been_focus prep_focus_of pobj_of_many prep_many_of det_studies_the amod_studies_recent pobj_of_studies prep_been_in det_area_this pobj_in_area ' +o,6721,'We compute log-likelihood significance between features and target nouns -LRB- as in -RRB- and keep only the most significant 200 features per target word ',Dunning,'nsubj_compute_We amod_significance_log-likelihood dobj_compute_significance prep_compute_between pobj_between_features cc_features_and nn_nouns_target conj_features_nouns dep_in_as dep_compute_in pobj_in_ cc_compute_and conj_compute_keep advmod_keep_only dep_significant_the advmod_significant_most dep_only_significant num_significant_200 dobj_keep_features prep_features_per nn_word_target pobj_per_word ' +p,6722,'A notable exception is the work of ',Kim,'det_exception_A amod_exception_notable nsubj_work_exception cop_work_is det_work_the prep_work_of ' +o,6723,'This algorithm and its many variants are widely used in the computational linguistics community ',Collins,'det_algorithm_This nsubjpass_used_algorithm cc_algorithm_and poss_variants_its amod_variants_many conj_algorithm_variants auxpass_used_are advmod_used_widely prep_used_in det_community_the amod_community_computational nn_community_linguistics pobj_in_community ' +o,6724,'Consequently , we abstract away from specifying a distribution by allowing the user to assign labels to features -LRB- cf , Druck et al ',Haghighi,'advmod_abstract_Consequently nsubj_abstract_we prt_abstract_away prep_abstract_from pcomp_from_specifying det_distribution_a dobj_specifying_distribution prep_specifying_by pcomp_by_allowing det_user_the dobj_allowing_user aux_assign_to infmod_user_assign dobj_assign_labels prep_assign_to pobj_to_features nn__cf dep_features_ dep__Druck cc_Druck_et conj_Druck_al ' +p,6725,'Inversion Transduction Grammar -LRB- ITG -RRB- and Syntax-Directed Translation Schema -LRB- SDTS -RRB- lack both of these properties ',Wu,'nn_Grammar_Inversion nn_Grammar_Transduction abbrev_Grammar_ITG amod__ cc__and conj__Syntax-Directed nn__Translation nn__Schema abbrev__SDTS nsubj_lack_ dep_Grammar_lack dobj_lack_both prep_both_of det_properties_these pobj_of_properties ' +p,6726,'Introduction Michael parsing models have been quite influential in the field of natural language processing ',Collins,'nn_models_Introduction nn_models_Michael nn_models_ nn_models_parsing nsubj_influential_models aux_influential_have dep_influential_been advmod_influential_quite prep_influential_in det_field_the pobj_in_field prep_field_of amod_processing_natural nn_processing_language pobj_of_processing ' +o,6727,'We adopted the stop condition suggested in et al 1996 the maximization of the likelihood on a cross-validation set of samples which is unseen at the parameter esti ~ _ tion ',Berger,'nsubj_adopted_We det_condition_the nn_condition_stop nsubj_suggested_condition ccomp_adopted_suggested prep_suggested_in nn_al_ nn_al_et pobj_in_al num_al_1996 det_maximization_the dobj_suggested_maximization prep_maximization_of det_likelihood_the pobj_of_likelihood prep_likelihood_on det_set_a amod_set_cross-validation pobj_on_set prep_set_of pobj_of_samples nsubj_unseen_which cop_unseen_is rcmod_samples_unseen prep_unseen_at det_tion_the nn_tion_parameter nn_tion_esti nn_tion_~ nn_tion__ pobj_at_tion ' +o,6728,'Automatic NE transliteration is an important component in many cross-language applications , such as Cross-Lingual Information Retrieval -LRB- CLIR -RRB- and Machine Translation -LRB- MT -RRB- ',Hermjakob,'nn_transliteration_Automatic nn_transliteration_NE nsubj_component_transliteration cop_component_is det_component_an amod_component_important prep_component_in amod_applications_many amod_applications_cross-language pobj_in_applications dep_as_such prep_component_as nn_Retrieval_Cross-Lingual nn_Retrieval_Information pobj_as_Retrieval abbrev_Retrieval_CLIR cc_Retrieval_and nn_Translation_Machine conj_Retrieval_Translation abbrev_Translation_MT ' +o,6729,'These are the same distributions that are needed by previous POS-based language models -LRB- Equation 5 -RRB- and POS taggers ',Marcus,'nsubj_distributions_These cop_distributions_are det_distributions_the amod_distributions_same nsubjpass_needed_that auxpass_needed_are rcmod_distributions_needed prep_needed_by amod_models_previous amod_models_POS-based nn_models_language pobj_by_models appos_models_Equation num_Equation_5 cc_models_and dep__POS nn__taggers conj_models_ ' +o,6730,'73 ment and phrase-extraction heuristics described in , minimum-error-rate training , a trigram language model with KneserNey smoothing trained with SRILM on the English side of the training data , and Moses to decode ',Och,'num_ment_73 nsubj_described_ment cc_ment_and amod_heuristics_phrase-extraction conj_ment_heuristics prep_described_in amod__ amod__minimum-error-rate nn__training pobj_in_ det_model_a nn_model_trigram nn_model_language nsubj_smoothing_model prep_model_with pobj_with_KneserNey conj_described_smoothing amod_data_trained prep_trained_with nn__SRILM pobj_with_ prep__on det_side_the amod_side_English pobj_on_side prep_side_of det_training_the pobj_of_training dobj_smoothing_data cc_described_and nsubj__Moses conj_described_ aux_decode_to xcomp__decode ' +o,6731,'Unlike probabilistic parsing , proposed by , \\* also a staff member of Matsushita Electric Industrial Co , Ltd , Shinagawa , Tokyo , JAPAN ',Brown,'dep_proposed_Unlike amod_parsing_probabilistic pobj_Unlike_parsing prep_proposed_by pobj_by_ dep_member_\\* advmod_member_also det_member_a nn_member_staff dep_proposed_member prep_member_of nn_Tokyo_Matsushita nn_Tokyo_Electric nn_Tokyo_Industrial nn_Tokyo_Co appos_Tokyo_Ltd appos_Tokyo_Shinagawa pobj_of_Tokyo dep_member_JAPAN ' +o,6732,'In order to minimize the number of decision errors at the sentence level , we have to choose the sequence of target words eI1 according to the equation : eI1 = argmax eI1 n Pr -LRB- eI1jfJ1 -RRB- o = argmax eI1 n Pr -LRB- eI1 -RRB- Pr -LRB- fJ1 jeI1 -RRB- o : Here , the posterior probability Pr -LRB- eI1jfJ1 -RRB- is decomposed into the language model probability Pr -LRB- eJ1 -RRB- and the string translation probability Pr -LRB- fJ1 jeI1 -RRB- ',Brown,'mark_minimize_In dep_minimize_order aux_minimize_to dep_have_minimize det_number_the dobj_minimize_number prep_number_of nn_errors_decision pobj_of_errors prep_minimize_at det_level_the nn_level_sentence pobj_at_level nsubj_have_we aux_choose_to xcomp_have_choose det_sequence_the dobj_choose_sequence prep_sequence_of nn_eI1_target nn_eI1_words pobj_of_eI1 prep_choose_according dep_according_to det__the nn__equation pobj_to_ advmod_=_eI1 dep_choose_= amod_n_argmax amod_n_eI1 dobj_=_n nn_o_Pr nn_o_eI1jfJ1 nsubj_o_o dep_o_= nn_eI1_argmax iobj_=_eI1 nn_Pr_n nn_Pr_Pr appos_Pr_eI1 dobj_=_Pr amod_jeI1_fJ1 appos_Pr_jeI1 dep_=_o advmod_decomposed_Here det_Pr_the nn_Pr_posterior nn_Pr_probability nsubjpass_decomposed_Pr appos_Pr_eI1jfJ1 auxpass_decomposed_is parataxis_have_decomposed prep_decomposed_into det_Pr_the nn_Pr_language nn_Pr_model nn_Pr_probability pobj_into_Pr appos_Pr_eJ1 cc_Pr_and det_Pr_the nn_Pr_string nn_Pr_translation nn_Pr_probability conj_Pr_Pr amod_jeI1_fJ1 appos_Pr_jeI1 ' +o,6733,'Recently , presented an unsupervised approach to coreference resolution , which mined the co-referring NP pairs with similar predicatearguments from a large corpus using a bootstrapping method ',Bean,'advmod_presented_Recently nsubj_presented_ det_approach_an amod_approach_unsupervised dobj_presented_approach prep_presented_to amod_resolution_coreference pobj_to_resolution nsubj_mined_which rcmod_resolution_mined det_pairs_the amod_pairs_co-referring nn_pairs_NP dobj_mined_pairs prep_mined_with amod_predicatearguments_similar pobj_with_predicatearguments prep_mined_from det_corpus_a amod_corpus_large pobj_from_corpus partmod_corpus_using det_method_a amod_method_bootstrapping dobj_using_method ' +o,6734,'And 20NG is a collection of approximately 20,000 20-category documents 1 In sentiment text classification , we also use two data sets : one is the widely used Cornell movie-review dataset2 and one dataset from product reviews of domain DVD3 ',Blitzer,'cc_collection_And nsubj_collection_20NG cop_collection_is det_collection_a prep_collection_of quantmod_20,000_approximately num_documents_20,000 amod_documents_20-category pobj_of_documents dep_documents_1 prep_use_In nn_classification_sentiment nn_classification_text pobj_In_classification nsubj_use_we advmod_use_also dep_collection_use num_data_two nsubj_sets_data ccomp_use_sets nsubj__one cop__is det__the advmod_used_widely amod__used nn__Cornell amod__movie-review nn__dataset2 ccomp_sets_ cc__and num_dataset_one conj__dataset prep_dataset_from nn_reviews_product pobj_from_reviews prep_reviews_of nn_DVD3_domain pobj_of_DVD3 ' +o,6735,'Adaptations to the algorithms in the presence of ngram LMs are discussed in ',Huang,'nsubjpass_discussed_Adaptations prep_Adaptations_to det_algorithms_the pobj_to_algorithms prep_algorithms_in det_presence_the pobj_in_presence prep_presence_of nn_LMs_ngram pobj_of_LMs auxpass_discussed_are prep_discussed_in ' +p,6736,'We do not completely rule out the possibility that some more sophisticated , ontologically promiscuous , first-order analysis -LRB- perhaps along the lines of -RRB- might account for these kinds of monotonicity inferences ',Hobbs,'nsubj_rule_We aux_rule_do neg_rule_not advmod_rule_completely prt_rule_out det_possibility_the dobj_rule_possibility complm_account_that det_analysis_some advmod_sophisticated_more amod_analysis_sophisticated advmod_sophisticated_ontologically dep_sophisticated_promiscuous amod_analysis_first-order nsubj_account_analysis dep_along_perhaps dep_analysis_along det_lines_the pobj_along_lines prep_lines_of pobj_of_ aux_account_might ccomp_rule_account prep_account_for det_kinds_these pobj_for_kinds prep_kinds_of amod_inferences_monotonicity pobj_of_inferences ' +o,6737,'By core phrases , we mean the kind of nonrecursive simplifications of the NP and VP that in the literature go by names such as noun\\/verb groups or chunks , and base NPs ',Ramshaw,'prep_mean_By amod_phrases_core pobj_By_phrases nsubj_mean_we det_kind_the dobj_mean_kind prep_kind_of amod_simplifications_nonrecursive pobj_of_simplifications prep_simplifications_of det_NP_the pobj_of_NP cc_NP_and conj_NP_VP nsubj_go_that prep_go_in det_literature_the pobj_in_literature rcmod_kind_go prep_go_by pobj_by_names dep_as_such prep_names_as amod__noun\\/verb nn__groups pobj_as_ cc__or conj__chunks cc_kind_and nn_NPs_base conj_kind_NPs ' +o,6738,'P -LRB- d -RRB- P L -LRB- d -RRB- -LRB- 4 -RRB- Statistical approaches to language modeling have been used in much NLP research , such as machine translation and speech recognition ',Brown,'nsubj_L_P dep_L_d amod_L_P appos_L_d dep_L_4 nn_approaches_Statistical nsubjpass_used_approaches prep_approaches_to nn_modeling_language pobj_to_modeling aux_used_have auxpass_used_been dep_L_used prep_used_in amod_research_much nn_research_NLP pobj_in_research dep_as_such prep_research_as nn__machine nn__translation pobj_as_ cc__and nn__speech nn__recognition conj__ ' +p,6739,'It is promising to optimize the model parameters directly with respect to AER as suggested in statistical machine translation ',Och,'nsubj_promising_It aux_promising_is aux_optimize_to xcomp_promising_optimize det_parameters_the nn_parameters_model dobj_optimize_parameters advmod_optimize_directly prep_optimize_with pobj_with_respect aux_AER_to infmod_respect_AER mark_suggested_as advcl_AER_suggested prep_suggested_in amod_translation_statistical nn_translation_machine pobj_in_translation ' +o,6740,'The agreement on identifying the boundaries of units , using the kappa statistic discussed in , was = 9 -LRB- for two annotators and 500 units -RRB- ; the agreement on features -LRB- two annotators and at least 200 units -RRB- was as follows : utype : = 76 ; verbed : = 9 ; nite : = 81 ',Carletta,'det_agreement_The prep_agreement_on pcomp_on_identifying det_boundaries_the dobj_identifying_boundaries prep_boundaries_of pobj_of_units dep_identifying_using det_statistic_the nn_statistic_kappa dobj_using_statistic dep_agreement_discussed prep_discussed_in pobj_in_ aux_=_was dep__= dobj_=_9 dep_discussed_for num_annotators_two pobj_for_annotators cc_annotators_and num_units_500 conj_annotators_units det_agreement_the nsubj_was_agreement prep_agreement_on pobj_on_features dep_features_two dep_two_annotators cc_annotators_and quantmod_200_at dep_at_least num_units_200 conj_annotators_units parataxis_discussed_was mark_follows_as advcl_was_follows parataxis_discussed_utype dep_utype_= dobj_=_76 parataxis_discussed_verbed parataxis_discussed_= dobj_=_9 parataxis_discussed_nite dep_nite_= dobj_=_81 ' +o,6741,'(Och et al., 1999; Koehn et al., 2003; Liang et al., 2006).',Koehn,'' +o,6742,'The more recent set of techniques includes mult iplicative weightupdate algorithms , latent semantic analysis , transformation-based learning , differential grammars , decision lists , and a variety of Bayesian classifiers ',Yarowsky,'det_set_The advmod_recent_more amod_set_recent nsubj_includes_set prep_set_of pobj_of_techniques amod__mult nn__iplicative nn__weightupdate nn__algorithms dobj_includes_ amod__latent amod__semantic nn__analysis conj__ amod__transformation-based nn__learning conj__ amod__differential nn__grammars conj__ nn__decision nn__lists conj__ cc__and det_variety_a conj__variety prep_variety_of amod__Bayesian nn__classifiers pobj_of_ ' +o,6743,'Please note that our approach is very different from other approaches to context dependent rule selection such as and ',He,'dep_note_Please complm_different_that poss_approach_our nsubj_different_approach cop_different_is advmod_different_very ccomp_note_different prep_different_from amod_approaches_other pobj_from_approaches prep_different_to pobj_to_context partmod_context_dependent nn_selection_rule dobj_dependent_selection dep_as_such prep_selection_as pobj_as_ cc_dependent_and ' +o,6744,' proposed a word similarity measure based on the distributio nal pattern of words which allows to construct a thesaurus using a parsed corpus ',Lin,'nsubj_proposed_ det_measure_a nn_measure_word nn_measure_similarity dobj_proposed_measure prep_proposed_based dep_based_on det_pattern_the amod_pattern_distributio nn_pattern_nal pobj_on_pattern prep_pattern_of pobj_of_words nsubj_allows_which rcmod_words_allows aux_construct_to xcomp_allows_construct det_thesaurus_a dobj_construct_thesaurus partmod_thesaurus_using det_corpus_a amod_corpus_parsed dobj_using_corpus ' +o,6745,'The data was seglnented into baseNP parts and non-lmseNP t -RRB- arts ill a similar fitshion as the data used 1 -RRB- y ',Ramshaw,'det_data_The auxpass_seglnented_was rcmod_data_seglnented prep_seglnented_into nn_parts_baseNP pobj_into_parts cc_parts_and nn_arts_non-lmseNP nn_arts_t conj_parts_arts amod_y_ill det_fitshion_a amod_fitshion_similar dep_ill_fitshion prep_fitshion_as det_data_the pobj_as_data amod_fitshion_used tmod_used_1 dep_data_y ' +o,6746,'We use maximum entropy model (Berger et al. , 1996) for both the mention-pair model (9) and the entity-mention model (8): a83a84a1a86a85a88a87 a43 a44 a71 a43 a16 a5a13a7 a55a35a34a23a36 a6a35a37 a6a39a38a40a6a42a41 a31a44a43a3a45a31 a6 a45a46a48a47a24a49 a50 a1 a43 a44 a71 a43 a16 a5 a71 (10) a83a84a1a4a85 a87 a55 a81 a71 a43 a16 a5a13a7 a55a35a34 a36 a6 a37 a6a39a38a40a6a42a41 a11a7a32 a45a31 a6 a45a46a48a47 a49 a50 a1 a55a39a81 a71 a43 a16 a5 a71 (11) wherea57 a16 a1a51a8 a71a52a8 a71a90a85a73a5 is a feature and a53 a16 is its weight; a50 a1a33a8 a71a54a8a5 is a normalizing factor to ensure that (10) or (11) is a probability.',Berger,'' +o,6747,'By core phrases , we mean the kind of nonrecursive simplifications of the NP and VP that in the literature go by names such as noun\\/verb groups or chunks , and base NPs ',Marcus,'prep_mean_By amod_phrases_core pobj_By_phrases nsubj_mean_we det_kind_the dobj_mean_kind prep_kind_of amod_simplifications_nonrecursive pobj_of_simplifications prep_simplifications_of det_NP_the pobj_of_NP cc_NP_and conj_NP_VP nsubj_go_that prep_go_in det_literature_the pobj_in_literature rcmod_kind_go prep_go_by pobj_by_names dep_as_such prep_names_as amod__noun\\/verb nn__groups pobj_as_ cc__or conj__chunks cc_kind_and nn_NPs_base conj_kind_NPs ' +o,6748,'For our contrast submission , we rescore the first-pass translation lattices with a large zero-cutoff stupid-backoff language model estimated over approximately five billion words of newswire text ',Brants,'prep_rescore_For poss_submission_our nn_submission_contrast pobj_For_submission nsubj_rescore_we det_lattices_the amod_lattices_first-pass nn_lattices_translation dobj_rescore_lattices prep_rescore_with det_model_a amod_model_large amod_model_zero-cutoff amod_model_stupid-backoff amod_model_ nn_model_language pobj_with_model partmod_model_estimated prep_estimated_over quantmod_billion_approximately number_billion_five num_words_billion pobj_over_words prep_words_of nn_text_newswire pobj_of_text ' +o,6749,'However , these unsupervised methodologies show a major drawback by extracting quasi-exact or even exact match pairs of sentences as they rely on classical string similarity measures such as the Edit Distance in the case of and Word N-gram Overlap for ',Barzilay,'advmod_show_However det_methodologies_these amod_methodologies_unsupervised nsubj_show_methodologies det_drawback_a amod_drawback_major dobj_show_drawback prep_show_by pcomp_by_extracting amod_pairs_quasi-exact cc_quasi-exact_or advmod_exact_even conj_quasi-exact_exact nn_pairs_match dobj_extracting_pairs prep_pairs_of pobj_of_sentences mark_rely_as nsubj_rely_they advcl_extracting_rely prep_rely_on amod_measures_classical nn_measures_string nn_measures_similarity pobj_on_measures dep_as_such prep_measures_as det_Distance_the nn_Distance_Edit pobj_as_Distance prep_Distance_in det_case_the pobj_in_case prep_case_of pobj_of_ cc__and nn_Overlap_Word nn_Overlap_N-gram conj__Overlap prep_rely_for ' +o,6750,'We have adopted the evaluation method of : compare the generated hypernyms with hypernyms present in a lexical resource , in our case the Dutch part of EuroWordNet ',Snow,'nsubj_adopted_We aux_adopted_have det_method_the nn_method_evaluation dobj_adopted_method prep_method_of pcomp_of_ ccomp__compare det_hypernyms_the amod_hypernyms_generated dobj_compare_hypernyms prep_compare_with pobj_with_hypernyms amod_hypernyms_present prep_present_in det_resource_a amod_resource_lexical pobj_in_resource prep_hypernyms_in poss_case_our pobj_in_case det_part_the amod_part_Dutch dep_compare_part prep_part_of nn__EuroWordNet pobj_of_ ' +o,6751,'F-me. 1 CBC-NER system M 71.67 23.47 35.36CBC-NER system A 70.66 32.86 44.86 2 XIP NER 77.77 56.55 65.48 XIP + CBC M 78.41 60.26 68.15 XIP + CBC A 76.31 60.48 67.48 3 Stanford NER 67.94 68.01 67.97 Stanford + CBC M 69.40 71.07 70.23 Stanford + CBC A 70.09 72.93 71.48 4 GATE NER 63.30 56.88 59.92 GATE + CBC M 66.43 61.79 64.03 GATE + CBC A 66.51 63.10 64.76 5 Stanford + XIP 72.85 75.87 74.33 Stanford + XIP + CBC M 72.94 77.70 75.24 Stanford + XIP + CBC A 73.55 78.93 76.15 6 GATE + XIP 69.38 66.04 67.67 GATE + XIP + CBC M 69.62 67.79 68.69 GATE + XIP + CBC A 69.87 69.10 69.48 7 GATE + Stanford 63.12 69.32 66.07 GATE + Stanford + CBC M 65.09 72.05 68.39 GATE + Stanford + CBC A 65.66 73.25 69.25 Table 1: Results given by different hybrid NER systems and coupled with the CBC-NER system corpora (CoNLL, MUC6, MUC7 and ACE): ner-eng-ie.crf-3-all2008-distsim.ser.gz (Finkel et al., 2005) (line 3 in Table 1), GATE NER or in short GATE (Cunningham et al., 2002) (line 4 in Table 1), and several hybrid systems which are given by the combination of pairs taken among the set of the three last-mentioned NER systems (lines 5 to 7 in Table 1).',Finkel,'' +o,6752,' and and showed that the MatrixTree Theorem can be used to train edge-factored log-linearmodelsofdependencyparsing ',Smith,'nsubj_showed_ cc__and conj__ cc__and nn__ conj__ complm_used_that det_Theorem_the nn_Theorem_MatrixTree nsubjpass_used_Theorem aux_used_can auxpass_used_be ccomp_showed_used aux_train_to purpcl_used_train amod_log-linearmodelsofdependencyparsing_edge-factored dobj_train_log-linearmodelsofdependencyparsing ' +o,6753,'These rules can be learned from a parallel corpus using English parsetrees , Chinese strings , and word alignment ',Galley,'det_rules_These nsubjpass_learned_rules aux_learned_can auxpass_learned_be prep_learned_from det_corpus_a amod_corpus_parallel pobj_from_corpus partmod_corpus_using amod_parsetrees_English dobj_using_parsetrees amod_strings_Chinese conj_parsetrees_strings cc_parsetrees_and nn_alignment_word conj_parsetrees_alignment ' +o,6754,'Amount of works have been done on sentimental classification in different levels ',Pang,'nsubjpass_done_Amount prep_Amount_of pobj_of_works aux_done_have auxpass_done_been prep_done_on amod_classification_sentimental pobj_on_classification prep_classification_in amod_levels_different pobj_in_levels ' +o,6755,'The suffixes C \\* and V \\* denote the models using incomplete skip-chain edges and vertical sequential edges proposed in , as shown in Figures 2 -LRB- a -RRB- and 2 -LRB- c -RRB- ',Ding,'det_C_The amod_C_suffixes rcmod_C_\\* cc_\\*_and conj_\\*_V dep_denote_\\* dep_models_denote det_models_the dep_C_models partmod_models_using amod_edges_incomplete amod_edges_skip-chain dobj_using_edges cc_models_and amod_edges_vertical amod_edges_sequential conj_models_edges partmod_edges_proposed prep_proposed_in pobj_in_ mark_shown_as advcl_proposed_shown prep_shown_in pobj_in_Figures num_Figures_2 dep_-LRB-_a cc_Figures_and conj_Figures_2 dep_-LRB-_c ' +o,6756,'The model scaling factors 1 , , 5 and the word and phrase penalties are optimized with respect to some evaluation criterion such as BLEU score ',Och,'det_factors_The amod_factors_model nn_factors_scaling nsubjpass_optimized_factors dep_factors_1 conj_factors_5 cc_factors_and det_penalties_the nn_penalties_word cc_word_and conj_word_phrase conj_factors_penalties auxpass_optimized_are prep_optimized_with pobj_with_respect prep_optimized_to det__some nn__evaluation nn__criterion pobj_to_ dep_as_such prep__as amod_score_BLEU pobj_as_score ' +n,6757,'String alignment with synchronous grammars is quite expensive even for simple synchronous formalisms like ITG but Duchi et al ',Wu,'nn_alignment_String nsubj_expensive_alignment prep_alignment_with amod_grammars_synchronous pobj_with_grammars cop_expensive_is advmod_expensive_quite advmod_expensive_even prep_expensive_for amod_formalisms_simple amod_formalisms_synchronous pobj_for_formalisms prep_formalisms_like nn__ITG pobj_like_ cc__but nn_al_Duchi nn_al_et conj__al ' +o,6758,'This makes it suitable for discriminative SMT training , which is still a challenge for large parameter sets ',Tillmann,'nsubj_makes_This nsubj_suitable_it xcomp_makes_suitable prep_suitable_for amod_training_discriminative nn_training_SMT pobj_for_training nsubj_challenge_which cop_challenge_is advmod_challenge_still det_challenge_a rcmod_training_challenge prep_challenge_for amod_sets_large nn_sets_parameter pobj_for_sets ' +o,6759,'There have been many statistical measures which estimate co-occurrence and the degree of association in previous researches , such as mutual information , t-score , dice matrix ',Smadja,'expl_been_There aux_been_have amod_measures_many amod_measures_statistical dobj_been_measures nsubj_estimate_which rcmod_measures_estimate acomp_estimate_co-occurrence cc_measures_and det_degree_the conj_measures_degree prep_degree_of pobj_of_association prep_association_in amod_researches_previous pobj_in_researches dep_as_such conj_measures_as amod__mutual nn__information pobj_as_ amod__t-score appos__ nn_matrix_dice appos__matrix ' +o,6760,' 4 ',Yates,'num__4 ' +n,6761,'1 Introduction Phrase-based systems , flat and hierarchical alike , have achieved a much better translation coverage than wordbased ones , but untranslated words remain a major problem in SMT ',Koehn,'num_systems_1 nn_systems_Introduction amod_systems_Phrase-based nsubj_achieved_systems amod__flat cc_flat_and conj_flat_hierarchical amod__alike appos_systems_ aux_achieved_have det_coverage_a advmod_better_much amod_coverage_better nn_coverage_translation dobj_achieved_coverage mark__than amod_ones_wordbased nsubj__ones advcl_achieved_ cc__but conj__untranslated nsubj_problem_words cop_problem_remain det_problem_a amod_problem_major ccomp_untranslated_problem prep_problem_in pobj_in_SMT ' +o,6762,'45 Consistency of Annotations In order to assess the consistency of annotation , we follow in using Cohen \'s ~ , a chancecorrected measure of inter-rater agreement ',Carletta,'num_Consistency_45 prep_Consistency_of pobj_of_Annotations mark_assess_In dep_assess_order aux_assess_to dep_Consistency_assess det_consistency_the dobj_assess_consistency prep_consistency_of pobj_of_annotation nsubj_follow_we dep_assess_follow acomp_follow_ prep__in pcomp_in_using poss_~_Cohen possessive_Cohen_\'s dobj_using_~ det_measure_a amod_measure_chancecorrected appos_~_measure prep_measure_of amod_agreement_inter-rater pobj_of_agreement ' +o,6763,'Currently , machine learning methods and combinations of classifiers have been popular ',Yarowsky,'advmod_popular_Currently nsubj_popular_machine partmod_machine_learning dobj_learning_methods num_methods_ cc_methods_and conj_methods_combinations prep_methods_of amod__classifiers pobj_of_ aux_popular_have dep_popular_been ' +o,6764,' used transformation based learning using a large annotated corpus for English ',Ramshaw,'num_transformation_ amod_transformation_used dep_based_transformation xcomp_based_learning xcomp_learning_using det_corpus_a amod_corpus_large amod_corpus_annotated dobj_using_corpus prep_using_for pobj_for_English ' +o,6765,'In addition , corpus-based stochastic modelling of lexical patterns may provide information about word sense frequency of the kind advocated since ',Marcus,'prep_provide_In pobj_In_addition amod_modelling_corpus-based amod_modelling_stochastic nsubj_provide_modelling prep_modelling_of amod__lexical nn__patterns pobj_of_ aux_provide_may dobj_provide_information prep_provide_about nn_frequency_word nn_frequency_sense pobj_about_frequency prep_frequency_of det_kind_the pobj_of_kind partmod_kind_advocated prep_advocated_since pobj_since_ ' +o,6766,'1 Introduction In the community of sentiment analysis , transferring a sentiment classifier from one source domain to another target domain is still far from a trivial work , because sentiment expression often behaves with strong domain-specific nature ',Turney,'num_Introduction_1 dep_Introduction_In det_community_the pobj_In_community prep_community_of nn__sentiment nn__analysis pobj_of_ partmod_community_transferring det_classifier_a nn_classifier_sentiment dobj_transferring_classifier prep_transferring_from num_domain_one nn_domain_source pobj_from_domain prep_domain_to det_domain_another nn_domain_target pobj_to_domain rcmod_domain_is advmod_is_still advmod_is_far dep_far_from det_work_a amod_work_trivial pobj_from_work mark_behaves_because nn_expression_sentiment nsubj_behaves_expression advmod_behaves_often advcl_is_behaves prep_behaves_with amod_nature_strong amod_nature_domain-specific pobj_with_nature ' +o,6767,'1 Introduction Word compositions have long been a concern in lexicography , and now as a specific kind of lexical knowledge , it has been shown that they have an important role in many areas in natural language processing , eg , parsing , generation , lexicon building , word sense disambiguation , and information retrieving , et ',Yarowsky,'dep_been_1 nn_compositions_Introduction nn_compositions_Word nsubj_been_compositions aux_been_have advmod_been_long det_concern_a dobj_been_concern prep_concern_in amod__lexicography pobj_in_ cc_been_and advmod_shown_now prep_shown_as det_kind_a amod_kind_specific pobj_as_kind prep_kind_of amod_knowledge_lexical pobj_of_knowledge nsubjpass_shown_it aux_shown_has auxpass_shown_been conj_been_shown complm_have_that nsubj_have_they ccomp_shown_have det_role_an amod_role_important dobj_have_role prep_role_in amod_areas_many pobj_in_areas prep_areas_in amod_building_natural nn_building_language nn_building_processing dep_building_eg dep_building_parsing dep_building_generation amod_building_lexicon pobj_in_building nn_disambiguation_word nn_disambiguation_sense conj_building_disambiguation cc_building_and nn_retrieving_information conj_building_retrieving dep_building_et ' +n,6768,'Such a quasi-syntactic structure can naturally capture the reordering of phrases that is not directly modeled by a conventional phrase-based approach ',Koehn,'predet_structure_Such det_structure_a amod_structure_quasi-syntactic nsubj_capture_structure aux_capture_can advmod_capture_naturally det_reordering_the dobj_capture_reordering prep_reordering_of pobj_of_phrases nsubjpass_modeled_that auxpass_modeled_is neg_modeled_not advmod_modeled_directly rcmod_phrases_modeled prep_modeled_by det_approach_a amod_approach_conventional amod_approach_phrase-based pobj_by_approach ' +o,6769,'6 Discussion Lack of interannotator agreement presents a significant problem in annotation efforts ',Marcus,'num_Lack_6 nn_Lack_Discussion nsubj_presents_Lack prep_Lack_of nn_agreement_interannotator pobj_of_agreement det_problem_a amod_problem_significant dobj_presents_problem prep_problem_in nn_efforts_annotation pobj_in_efforts ' +o,6770,' has used a few seeds and untagged sentences in a bootstrapping algorithm based on decision lists ',Yarowsky,'nsubj_used_ aux_used_has det_seeds_a amod_seeds_few dobj_used_seeds cc_seeds_and amod_sentences_untagged conj_seeds_sentences prep_used_in det_algorithm_a amod_algorithm_bootstrapping pobj_in_algorithm prep_algorithm_based dep_based_on nn_lists_decision pobj_on_lists ' +o,6771,' 1990 -RRB- ',Hindle,'dep__1990 ' +o,6772,'The fact that different authors use different versions of the same gold standard to evaluate similar experiments -LRB- eg Goldwater & Griffiths versus -RRB- supports this claim ',Johnson,'det_fact_The nsubj_supports_fact complm_use_that amod_authors_different nsubj_use_authors dep_fact_use amod_versions_different nsubj_evaluate_versions prep_versions_of det_standard_the amod_standard_same nn_standard_gold pobj_of_standard aux_evaluate_to xcomp_use_evaluate amod_experiments_similar dobj_evaluate_experiments nn_Goldwater_eg dep_fact_Goldwater cc_Goldwater_& nn__Griffiths conj_Goldwater_ cc_Goldwater_versus nn__ conj_Goldwater_ det_claim_this dobj_supports_claim ' +p,6773,'Recently , suggest to mine semantic relatedness from Wikipedia , which can deal with the data sparseness problem suffered by using WordNet ',Ponzetto,'advmod_suggest_Recently nsubj_suggest_ aux_mine_to xcomp_suggest_mine amod_relatedness_semantic dobj_mine_relatedness prep_mine_from pobj_from_Wikipedia nsubj_deal_which aux_deal_can rcmod_Wikipedia_deal prep_deal_with det_problem_the nn_problem_data nn_problem_sparseness pobj_with_problem partmod_problem_suffered prep_suffered_by pcomp_by_using dobj_using_WordNet ' +o,6774,'As is common , the treebank is first transformed in various ways , in order to give an accurate PCFG ',Collins,'mark__As cop__is amod__common advcl_transformed_ det_treebank_the nsubjpass_transformed_treebank auxpass_transformed_is advmod_transformed_first prep_transformed_in amod_ways_various pobj_in_ways mark_give_in dep_give_order aux_give_to purpcl_transformed_give det_PCFG_an amod_PCFG_accurate dobj_give_PCFG ' +o,6775,'The second type has clear interpretation as a probability model , but no criteria to determine the number of clusters ',Brown,'det_type_The amod_type_second nsubj_has_type amod_interpretation_clear dobj_has_interpretation prep_interpretation_as det_model_a nn_model_probability pobj_as_model cc_has_but det_criteria_no conj_has_criteria aux_determine_to xcomp_criteria_determine det_number_the dobj_determine_number prep_number_of pobj_of_clusters ' +o,6776,'It is also related to -LRB- log - -RRB- linear models described in Berger , Della Pietra , and Della Pietra , Xue ; , and Peng , Feng , and McCallum -LRB- 2004 -RRB- ',Och,'nsubjpass_related_It auxpass_related_is advmod_related_also prep_related_to dep_-LRB-_log amod_models_linear pobj_to_models partmod_models_described prep_described_in pobj_in_Berger nn_Pietra_Della conj_Berger_Pietra cc_Berger_and nn__Della nn__Pietra conj_Berger_ nn__Xue conj_models_ nn_Feng_ nn_Feng_ dep_Peng_and dep_Feng_Peng dep__Feng cc_models_and conj_models_McCallum appos_McCallum_2004 ' +o,6777,' describe a perceptron style algorithm for training millions of features ',Tillmann,'nsubj_describe_ det_algorithm_a nn_algorithm_perceptron nn_algorithm_style dobj_describe_algorithm prep_algorithm_for nn_millions_training pobj_for_millions prep_millions_of pobj_of_features ' +o,6778,'For each co-occurring pair of word types u and v , these likelihoods are initially set proportional to their co-occurrence frequency n -LRB- u , v -RRB- and inversely proportional to their marginal frequencies n -LRB- u -RRB- and n -LRB- v -RRB- z , following 2 ',Dunning,'prep_set_For det_pair_each amod_pair_co-occurring pobj_For_pair prep_pair_of nn_u_word nn_u_types pobj_of_u cc_u_and conj_u_v det_likelihoods_these nsubjpass_set_likelihoods auxpass_set_are advmod_set_initially amod_proportional_proportional dep_proportional_to poss_n_their amod_n_co-occurrence nn_n_frequency pobj_to_n appos_n_u dep_u_v cc_to_and conj_to_inversely dobj_set_proportional prep_proportional_to poss_n_their amod_n_marginal amod_n_frequencies pobj_to_n appos_n_u cc_n_and conj_n_n appos_n_v dep_proportional_z prep_set_following pobj_following_ num__2 ' +o,6779,'The optimal weights for the different columns can then be assigned with the help of minimum error rate training ',Och,'det_weights_The amod_weights_optimal nsubjpass_assigned_weights prep_weights_for det_columns_the amod_columns_different pobj_for_columns aux_assigned_can advmod_assigned_then auxpass_assigned_be prep_assigned_with det_help_the pobj_with_help prep_help_of amod_training_minimum nn_training_error nn_training_rate pobj_of_training ' +p,6780,'2 Previous Work So far , Structural Correspondence Learning has been applied successfully to PoS tagging and Sentiment Analysis ',Blitzer,'num_Work_2 amod_Work_Previous dep_far_So advmod_applied_far amod_Learning_Structural nn_Learning_Correspondence nsubjpass_applied_Learning aux_applied_has auxpass_applied_been dep_Work_applied advmod_applied_successfully aux_PoS_to xcomp_applied_PoS nn_Analysis_tagging cc_tagging_and conj_tagging_Sentiment dobj_PoS_Analysis ' +o,6781,'They have been employed in word sense disambiguation , automatic construction of bilingual dictionaries , and inducing statistical machine translation models ',Koehn,'nsubjpass_employed_They aux_employed_have auxpass_employed_been prep_employed_in nn__word nn__sense nn__disambiguation pobj_in_ amod_construction_automatic appos__construction prep_construction_of amod_dictionaries_bilingual pobj_of_dictionaries amod_construction_ cc__and conj__inducing amod_models_statistical nn_models_machine nn_models_translation dobj_inducing_models ' +p,6782,'The well-known BLEU is based on the number of common n-grams between the translation hypothesis and human reference translations of the same sentence ',Papineni,'det__The amod__well-known nn__BLEU nsubjpass_based_ auxpass_based_is prep_based_on det_number_the pobj_on_number prep_number_of amod_n-grams_common pobj_of_n-grams prep_n-grams_between det_hypothesis_the nn_hypothesis_translation pobj_between_hypothesis cc_n-grams_and amod_translations_human nn_translations_reference conj_n-grams_translations prep_translations_of det_sentence_the amod_sentence_same pobj_of_sentence ' +o,6783,'53 Experimental setup We used the Stanford Parser for both languages , Penn English Treebank and Penn Arabic Treebank set ',Marcus,'num_setup_53 nn_setup_Experimental nsubj_used_We dep_setup_used det__the nn__Stanford nn__Parser dobj_used_ prep__for det_languages_both pobj_for_languages nn__Penn nn__English nn__Treebank conj_languages_ cc_languages_and nn__Penn nn__Arabic nn__Treebank nn__set conj_languages_ ' +o,6784,' and Chiang , in terms of what alignments they induce , has been discussed in Wu and et al ',Wellington,'nn__ cc__and conj__Chiang nsubjpass_discussed_ prep__in pobj_in_terms prep_terms_of nsubj_alignments_what pcomp_of_alignments nsubj_induce_they ccomp_alignments_induce aux_discussed_has auxpass_discussed_been prep_discussed_in nn__Wu pobj_in_ cc__and conj__ cc__et conj__al ' +o,6785,'is combined with -LRB- -RRB- E jiT ,1 + to be aligned with -LRB- -RRB- F nmT , , then -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- ATTCNTATTr E K E i FEF jinmjinm , Pr , P ,1 -RRB- , -LRB- , -RRB- , -LRB- -RRB- ,1 -LRB- + = where K is the degree of EiN Finally , the node translation probability is modeled as -LRB- -RRB- -LRB- -RRB- -LRB- -RRB- tNtNlNlNNN EiFlEiFlEjFl PrPrPr And the text translation probability -LRB- -RRB- EF ttPr is model using IBM model I ',Brown,'auxpass_combined_is mark_EF_with nsubj_jiT_E dep_EF_jiT number_+_,1 dobj_jiT_+ aux_aligned_to auxpass_aligned_be xcomp_jiT_aligned prep_aligned_with nn_nmT_F pobj_with_nmT appos_nmT_then nn_jinmjinm_ATTCNTATTr nn_jinmjinm_E nn_jinmjinm_K nn_jinmjinm_E nn_jinmjinm_i nn_jinmjinm_FEF dep_-RRB-_jinmjinm appos_jinmjinm_Pr appos_jinmjinm_P num_P_,1 nsubj_EF_,1 dep_,1_+ ccomp_+_= advmod_degree_where nsubj_degree_K cop_degree_is det_degree_the dep_modeled_degree prep_degree_of pobj_of_EiN advmod_degree_Finally det_probability_the amod_probability_node nn_probability_translation nsubjpass_modeled_probability auxpass_modeled_is ccomp_=_modeled prep_modeled_as nn_PrPrPr_tNtNlNlNNN nn_PrPrPr_EiFlEiFlEjFl dep_-LRB-_PrPrPr cc_PrPrPr_And det_probability_the nn_probability_text nn_probability_translation conj_PrPrPr_probability advcl_combined_EF nsubj_model_ttPr cop_model_is ccomp_EF_model partmod_model_using nn_model_IBM dobj_using_model nsubjpass_combined_I ' +o,6786,'Like , the decoder is the same for both the perceptron and the log-linear parsing models ; the only change is the method for setting the weights ',Collins,'prep_same_Like pobj_Like_ det_decoder_the nsubj_same_decoder cop_same_is det_same_the prep_same_for preconj_perceptron_both det_perceptron_the pobj_for_perceptron cc_perceptron_and det_models_the amod_models_log-linear amod_models_parsing conj_perceptron_models det_change_the amod_change_only nsubj_method_change cop_method_is det_method_the parataxis_same_method prep_method_for pcomp_for_setting det_weights_the dobj_setting_weights ' +o,6787,'The reordered sentence is then re-tokenized to be consistent with the baseline system , which uses a different tokenization scheme that is more friendly to the MT system3 We use BLEU scores as the performance measure in our evaluation ',Papineni,'det_sentence_The amod_sentence_reordered nsubj_re-tokenized_sentence cop_re-tokenized_is advmod_re-tokenized_then aux_consistent_to cop_consistent_be xcomp_re-tokenized_consistent prep_consistent_with det_system_the amod_system_baseline pobj_with_system nsubj_uses_which rcmod_system_uses det_scheme_a amod_scheme_different nn_scheme_tokenization dobj_uses_scheme nsubj_friendly_that cop_friendly_is advmod_friendly_more rcmod_scheme_friendly prep_friendly_to det_system3_the nn_system3_MT pobj_to_system3 nsubj_use_We dep_re-tokenized_use amod_scores_BLEU dobj_use_scores prep_use_as det_measure_the nn_measure_performance pobj_as_measure prep_measure_in poss_evaluation_our pobj_in_evaluation ' +o,6788,'Extensions to Hiero Several authors describe extensions to Hiero , to incorporate additional syntactic information , or to combine it with discriminative latent models ',Shen,'nsubj_describe_Extensions prep_Extensions_to nn_authors_Hiero amod_authors_Several pobj_to_authors dobj_describe_extensions prep_describe_to pobj_to_Hiero aux_incorporate_to dep_describe_incorporate amod__additional amod__syntactic nn__information dobj_incorporate_ cc_incorporate_or aux_combine_to conj_incorporate_combine dobj_combine_it prep_combine_with amod__discriminative amod__latent nn__models pobj_with_ ' +p,6789,'They are latent variable models which are not tractable to compute exactly , but two approximations exist which have been shown to be effective for constituent parsing ',Titov,'nsubj_models_They cop_models_are amod_models_latent amod_models_variable nsubj_tractable_which cop_tractable_are neg_tractable_not rcmod_models_tractable aux_compute_to xcomp_tractable_compute advmod_compute_exactly cc_models_but num_approximations_two nsubj_exist_approximations conj_models_exist nsubjpass_shown_which aux_shown_have auxpass_shown_been ccomp_exist_shown aux_effective_to cop_effective_be xcomp_shown_effective prep_effective_for amod_parsing_constituent pobj_for_parsing ' +o,6790,'The highest BLEU score was chosen as the optimization criterion ',Papineni,'det_BLEU_The amod_BLEU_highest nsubj_score_BLEU nsubjpass_chosen_ auxpass_chosen_was ccomp_score_chosen prep_chosen_as det_criterion_the nn_criterion_optimization pobj_as_criterion ' +o,6791,'We found that the deletion of lead parts did not occur very often in our summary , unlike the case of ',Jing,'nsubj_found_We complm_occur_that det_deletion_the nsubj_occur_deletion prep_deletion_of amod_parts_lead pobj_of_parts aux_occur_did neg_occur_not ccomp_found_occur advmod_often_very advmod_occur_often prep_occur_in poss_summary_our pobj_in_summary prep_occur_unlike det_case_the pobj_unlike_case prep_case_of ' +o,6792,'The software also required GIZA + + word alignment too ',Och,'det_software_The nsubj_required_software advmod_software_also amod_alignment_GIZA cc_GIZA_+ conj_GIZA_+ nn_alignment_word dobj_required_alignment advmod_required_too ' +o,6793,'This result supports the intuition in that correlation at segment level is necessary to ensure the reliability of metrics in different situations ',Banerjee,'det_result_This nsubj_supports_result det_intuition_the dobj_supports_intuition prep_supports_in pcomp_in_ complm_necessary_that nsubj_necessary_correlation prep_correlation_at nn_level_segment pobj_at_level cop_necessary_is ccomp__necessary aux_ensure_to xcomp_necessary_ensure det_reliability_the dobj_ensure_reliability prep_reliability_of pobj_of_metrics prep_metrics_in amod_situations_different pobj_in_situations ' +o,6794,'As the strength of relevance between a target compound noun t and its co-occurring word r , the feature value of r , w -LRB- t ; r -RRB- is deflned by the log likelihood ratio 1 as follows ',Dunning,'prep_deflned_As det_strength_the pobj_As_strength prep_strength_of pobj_of_relevance prep_relevance_between det_t_a nn_t_target nn_t_compound nn_t_noun pobj_between_t cc_t_and poss_r_its amod_r_co-occurring nn_r_word conj_t_r det_value_the nn_value_feature nsubjpass_deflned_value prep_value_of pobj_of_r appos_r_w nn_r_t appos_w_r auxpass_deflned_is prep_deflned_by det_likelihood_the nn_likelihood_log pobj_by_likelihood nn__ratio nsubj_follows_ dep_as_1 advmod_follows_as dep_deflned_follows ' +o,6795,'Hindle uses the observed frequencies within a specific syntactic pattern -LRB- subject\\/verb , and verb\\/object -RRB- to derive a cooccu , -RRB- rence score which is an estimate of mutual information ',Church,'advmod_uses_Hindle det_frequencies_the amod_frequencies_observed dobj_uses_frequencies prep_uses_within det_score_a amod_score_specific dep_specific_syntactic nn_score_pattern dep_cooccu_subject\\/verb cc_subject\\/verb_and conj_subject\\/verb_verb\\/object prep_subject\\/verb_to pobj_to_derive det_cooccu_a dep_score_cooccu amod_score_rence pobj_within_score nsubj_estimate_which cop_estimate_is det_estimate_an dep_uses_estimate prep_estimate_of amod_information_mutual pobj_of_information ' +o,6796,'The kappa statistic for identifying question segments is 068 , and for linking question and answer segments given a question segment is 081 ',Carletta,'det_statistic_The nn_statistic_kappa nsubj__statistic prep__for pcomp_for_identifying nn_segments_question nsubj_068_segments cop_068_is ccomp_identifying_068 cc_for_and conj_for_for pcomp_for_linking nn_segments_question cc_question_and conj_question_answer dobj_linking_segments prep_linking_given det_segment_a nn_segment_question nsubj_081_segment cop_081_is dep_given_081 ' +o,6797,'As a unified approach , we augment the SDIG by adding all the possible word pairs -LRB- , -RRB- ji fe as a parallel ET pair and using the IBM Model 1 word to word translation probability as the ET translation probability ',Brown,'prep_augment_As det_approach_a amod_approach_unified pobj_As_approach nsubj_augment_we det_SDIG_the dobj_augment_SDIG prep_augment_by pcomp_by_adding predet_pairs_all det_pairs_the amod_pairs_possible nn_pairs_word nn_fe_pairs nn_fe_ji dobj_adding_fe prep_fe_as det_pair_a amod_pair_parallel nn_pair_ET pobj_as_pair cc_adding_and conj_adding_using det_word_the nn_word_IBM nn_word_Model num_word_1 nn_word_ dobj_using_word prep_using_to nn_probability_word nn_probability_translation pobj_to_probability prep_probability_as det_probability_the nn_probability_ET nn_probability_translation pobj_as_probability ' +o,6798,'Specifically , the following information can be either automatically identified or manually annotated : Syntactic structures automatically identified from a parser ; Semantic roles of entities in the question ; Discourse roles either manually annotated or identified by rules that map directly from semantic roles to discourse roles ',Collins,'advmod_identified_Specifically det_information_the amod_information_following nsubjpass_identified_information aux_identified_can auxpass_identified_be advmod_identified_either advmod_identified_automatically cc_identified_or advmod_identified_manually conj_identified_annotated amod_structures_Syntactic dobj_annotated_structures advmod_identified_automatically partmod_structures_identified prep_identified_from det__a nn__parser pobj_from_ amod_roles_Semantic dep_structures_roles prep_roles_of pobj_of_entities prep_entities_in det__the nn__question pobj_in_ amod_roles_Discourse dep_structures_roles advmod_annotated_either advmod_annotated_manually partmod_roles_annotated cc_annotated_or conj_annotated_identified prep_annotated_by pobj_by_rules nsubj_map_that rcmod_rules_map advmod_map_directly prep_map_from amod_roles_semantic pobj_from_roles prep_map_to nn_roles_discourse pobj_to_roles ' +o,6799,'In our experiments we use the same definition of structural locality as was proposed for the ISBN dependency parser in ',Titov,'prep_definition_In poss_experiments_our pobj_In_experiments nsubj_definition_we dep_definition_use det_definition_the amod_definition_same prep_definition_of amod_locality_structural pobj_of_locality mark_proposed_as auxpass_proposed_was advcl_definition_proposed prep_proposed_for det_parser_the nn_parser_ISBN nn_parser_dependency pobj_for_parser prep_proposed_in ' +o,6800,'The labeled corpus is the Penn Wall Street Journal treebank ',Marcus,'det_corpus_The amod_corpus_labeled nsubj_treebank_corpus cop_treebank_is det_treebank_the nn_treebank_Penn nn_treebank_Wall nn_treebank_Street nn_treebank_Journal ' +o,6801,'1 Introduction Sentiment analysis of text documents has received considerable attention recently ',Turney,'num_analysis_1 nn_analysis_Introduction nn_analysis_Sentiment nsubj_received_analysis prep_analysis_of nn_documents_text pobj_of_documents aux_received_has amod_attention_considerable dobj_received_attention advmod_received_recently ' +o,6802,'The phrases in the translations were located using standard phrase extraction techniques ',Koehn,'det_phrases_The nsubjpass_located_phrases prep_phrases_in det_translations_the pobj_in_translations auxpass_located_were xcomp_located_using amod_techniques_standard nn_techniques_phrase nn_techniques_extraction dobj_using_techniques ' +o,6803,'The production rules in ITGs are of the following form , with a notation similar to what is typically used for SDTSs and SCFGs in the right column : A -LRB- BC -RRB- A B1C2 , B1C2 A BC A B1C2 , C2B1 A e f A e , f A e A e , A f A , f It is important to note that RHSs of production rules have at most one source-side and one targetside terminal symbol ',Wu,'det_production_The nsubj_are_rules prep_rules_in pobj_in_ITGs ccomp_production_are prep_are_of det__the amod__following nn__form pobj_of_ prep_are_with det_notation_a pobj_with_notation amod_notation_similar prep_similar_to nsubjpass_used_what auxpass_used_is advmod_used_typically dep_A_used prep_used_for pobj_for_SDTSs cc_SDTSs_and conj_SDTSs_SCFGs prep_SDTSs_in det_column_the amod_column_right pobj_in_column pobj_to_A abbrev_A_BC det_B1C2_A nsubj_C2B1_B1C2 nn_BC_B1C2 nn_BC_A appos_B1C2_BC det_B1C2_A dep_BC_B1C2 rcmod_A_C2B1 dobj_C2B1_A dep_A_e dep_e_f det_e_A dep_A_e dep_A_f det_A_A amod_A_e dep_production_A dep_important_e det_A_A amod_A_f dep_e_A dep_important_f nsubj_important_It cop_important_is rcmod_A_important aux_note_to xcomp_important_note complm_have_that nsubj_have_RHSs prep_RHSs_of nn_rules_production pobj_of_rules ccomp_note_have quantmod_one_at dep_at_most num_source-side_one dobj_have_source-side cc_source-side_and num_symbol_one amod_symbol_targetside amod_symbol_terminal conj_source-side_symbol ' +o,6804,' applied the distributional similarity proposed by to coordination disambiguation ',Lin,'nsubj_applied_ det_similarity_the amod_similarity_distributional nsubj_proposed_similarity ccomp_applied_proposed prep_proposed_by pobj_by_ prep_proposed_to nn_disambiguation_coordination pobj_to_disambiguation ' +o,6805,'We used the Wall Street Journal -LRB- WSJ -RRB- part of the Penn Treebank , where extraction is represented by co-indexing an empty terminal element -LRB- henceforth EE -RRB- to its antecedent ',Marcus,'nsubj_part_We dep_part_used det_part_the nn_Journal_Wall nn_Journal_Street nn_part_Journal abbrev_Journal_WSJ prep_part_of det__the nn__Penn nn__Treebank pobj_of_ advmod_represented_where nsubjpass_represented_extraction auxpass_represented_is advcl_part_represented prep_represented_by pcomp_by_co-indexing det_element_an amod_element_empty dep_empty_terminal dobj_co-indexing_element nn_EE_henceforth abbrev_element_EE prep_co-indexing_to poss_antecedent_its pobj_to_antecedent ' +o,6806,'Li and Roth demonstrated that their shallow parser , trained to label shallow constituents along the lines of the well-known CoNLL2000 task , outperformed the Collins parser in correctly identifying these constituents in the Penn Wall Street Journal -LRB- WSJ -RRB- Treebank ',Marcus,'cc_Li_and conj_Li_Roth partmod_Li_demonstrated complm_outperformed_that poss_parser_their nn_parser_shallow nsubj_outperformed_parser partmod_parser_trained aux_label_to xcomp_trained_label nn_constituents_shallow dobj_label_constituents prep_label_along det_lines_the pobj_along_lines prep_lines_of det__the amod__well-known nn__CoNLL2000 nn__task pobj_of_ ccomp_demonstrated_outperformed det_parser_the nn_parser_Collins dobj_outperformed_parser prep_outperformed_in advmod_identifying_correctly pcomp_in_identifying det_constituents_these dobj_identifying_constituents prep_identifying_in det_Wall_the nn_Wall_Penn pobj_in_Wall nn_Treebank_Street nn_Treebank_Journal abbrev_Treebank_WSJ dep_Li_Treebank ' +o,6807,'For every class the weights of the active features are combined and the best scoring class is chosen ',Berger,'det_class_every pobj_For_class det_weights_the nsubjpass_combined_weights prep_weights_of det_features_the amod_features_active pobj_of_features auxpass_combined_are rcmod_class_combined cc_combined_and det_class_the amod_class_best nn_class_scoring nsubjpass_chosen_class auxpass_chosen_is conj_combined_chosen ' +n,6808,'Although generating training examples in advance without a working parser is much faster than using inference , our training time can probably be decreased further by choosing a parsing strategy with a lower branching factor ',Collins,'mark_faster_Although csubj_faster_generating nn_examples_training dobj_generating_examples prep_examples_in pobj_in_advance prep_generating_without det__a amod__working nn__parser pobj_without_ cop_faster_is advmod_faster_much advcl_decreased_faster prep_faster_than pcomp_than_using amod__inference dobj_using_ poss_time_our nn_time_training nsubjpass_decreased_time aux_decreased_can advmod_decreased_probably auxpass_decreased_be advmod_decreased_further prep_decreased_by pcomp_by_choosing det_strategy_a amod_strategy_parsing dobj_choosing_strategy prep_strategy_with det_factor_a amod_factor_lower amod_factor_branching pobj_with_factor ' +o,6809,'We then built separate directed word alignments for EnglishX andXEnglish -LRB- X -LCB- Indonesian , Spanish -RCB- -RRB- using IBM model 4 , combined them using the intersect + grow heuristic , and extracted phrase-level translation pairs of maximum length seven using the alignment template approach ',Och,'nsubj_built_We advmod_built_then amod_alignments_separate amod_alignments_directed nn_alignments_word dobj_built_alignments prep_built_for nn_andXEnglish_EnglishX pobj_for_andXEnglish dep_andXEnglish_X appos_andXEnglish_Indonesian dep_Indonesian_Spanish partmod_andXEnglish_using nn_model_IBM nsubj__model num__4 xcomp_using_ partmod__combined nsubj_using_them dep_combined_using det__the amod__intersect cc_intersect_+ conj_intersect_grow amod__heuristic dobj_using_ cc_built_and conj_built_extracted amod_pairs_phrase-level nn_pairs_translation dobj_extracted_pairs prep_pairs_of nn_length_maximum pobj_of_length num_length_seven xcomp_built_using det_approach_the amod_approach_alignment nn_approach_template dobj_using_approach ' +o,6810,'In this paper , we adopt Stanford Maximum Entropy implementation in our experiments ',Manning,'prep_adopt_In det_paper_this pobj_In_paper nsubj_adopt_we nn_implementation_Stanford nn_implementation_Maximum nn_implementation_Entropy nn_implementation_ dobj_adopt_implementation prep_implementation_in poss_experiments_our pobj_in_experiments ' +o,6811,'In the nal step , we score our translations with 4-gram BLEU ',Papineni,'prep_score_In det_step_the nn_step_nal pobj_In_step nsubj_score_we poss_translations_our dobj_score_translations prep_score_with amod_BLEU_4-gram pobj_with_BLEU ' +o,6812,'This can be done by smoothing the observed frequencies 7 or by class-based methods ',Dagan,'nsubjpass_done_This aux_done_can auxpass_done_be prep_done_by pcomp_by_smoothing det_frequencies_the amod_frequencies_observed nsubj__frequencies dep__7 xcomp_smoothing_ cc_by_or conj_by_by amod_methods_class-based pobj_by_methods ' +o,6813,'A number of knowledge-rich and knowledge-poor methods have been proposed for recognizing when words are similar ',Hindle,'det_number_A nsubjpass_proposed_number prep_number_of amod__knowledge-rich pobj_of_ cc__and amod_methods_knowledge-poor nn_methods_ conj__methods aux_proposed_have auxpass_proposed_been prep_proposed_for pcomp_for_recognizing advmod_similar_when nsubj_similar_words cop_similar_are advcl_recognizing_similar ' +p,6814,'Here we choose to work with stupid backoff smoothing since this is significantly more efficient to train and deploy in a distributed framework than a contextdependent smoothing scheme such as Kneser-Ney ',Brants,'nsubj_choose_we rcmod_Here_choose aux_work_to xcomp_choose_work prep_work_with amod_backoff_stupid pobj_with_backoff dep_Here_smoothing advmod_smoothing_ mark_efficient_since nsubj_efficient_this cop_efficient_is advmod_efficient_significantly advmod_efficient_more advcl_smoothing_efficient aux_train_to xcomp_efficient_train dep_smoothing_and dep_in_deploy dep_smoothing_in det_framework_a amod_framework_distributed pobj_in_framework prep_framework_than det_scheme_a nn_scheme_contextdependent amod_scheme_smoothing pobj_than_scheme dep_as_such prep_scheme_as pobj_as_Kneser-Ney ' +p,6815,'We can credit DUC with the emergence of automatic methods for evaluation such as ROUGE which allow quick measurement of systems during development and enable evaluation of larger amounts of data ',Lin,'nsubj_credit_We aux_credit_can dobj_credit_DUC prep_DUC_with det_emergence_the pobj_with_emergence prep_emergence_of amod_methods_automatic pobj_of_methods prep_credit_for pobj_for_evaluation dep_as_such prep_evaluation_as nn__ROUGE pobj_as_ nsubj_allow_which rcmod__allow amod_measurement_quick dobj_allow_measurement prep_measurement_of pobj_of_systems prep_allow_during pobj_during_development cc_credit_and conj_credit_enable dobj_enable_evaluation prep_evaluation_of amod_amounts_larger pobj_of_amounts prep_amounts_of pobj_of_data ' +n,6816,'This cost can often be substantial , as with the Penn Treebank ',Marcus,'det_cost_This nsubj_substantial_cost aux_substantial_can advmod_substantial_often cop_substantial_be advmod_substantial_as dep_as_with det_Treebank_the nn_Treebank_Penn pobj_with_Treebank ' +o,6817,'Our approach to statistical machine translation differs from the model proposed in in that : We compute the joint model P -LRB- Ws , WT -RRB- from the bilanguage corpus to account for the direct mapping of the source sentence Ws into the target sentence I ? VT that is ordered according to the source language word order ',Brown,'poss_approach_Our nsubj_differs_approach prep_approach_to amod_translation_statistical nn_translation_machine pobj_to_translation prep_differs_from det_model_the pobj_from_model partmod_model_proposed prep_proposed_in pobj_in_ prep_proposed_in det_VT_that nsubj_compute_We dep_:_compute det_P_the amod_P_joint nn_P_model dobj_compute_P appos_P_Ws dep_Ws_WT prep_compute_from det_corpus_the amod_corpus_bilanguage pobj_from_corpus aux_account_to xcomp_compute_account prep_account_for det_mapping_the amod_mapping_direct pobj_for_mapping prep_mapping_of det_Ws_the nn_Ws_source nn_Ws_sentence pobj_of_Ws prep_account_into det_I_the nn_I_target nn_I_sentence pobj_into_I pobj_in_VT nsubjpass_ordered_that auxpass_ordered_is rcmod_model_ordered prep_ordered_according dep_according_to det_order_the nn_order_source nn_order_language nn_order_word pobj_to_order ' +o,6818,'For a given choice of q and f , the IIS algorithm can be used to find maximum likelihood values for the parameters ~ ',Berger,'prep_used_For det_choice_a amod_choice_given pobj_For_choice prep_choice_of pobj_of_q cc_q_and conj_q_f det__the nn__IIS nn__algorithm nsubjpass_used_ aux_used_can auxpass_used_be aux_find_to purpcl_used_find amod_values_maximum nn_values_likelihood dobj_find_values prep_find_for det_~_the nn_~_parameters pobj_for_~ ' +o,6819,'1 shows that Wikipedia can indeed be used as a sense inventory for sense disambiguation ',Mihalcea,'num__1 nsubj_shows_ complm_used_that nsubjpass_used_Wikipedia aux_used_can advmod_used_indeed auxpass_used_be ccomp_shows_used prep_used_as det_inventory_a nn_inventory_sense pobj_as_inventory prep_inventory_for nn_disambiguation_sense pobj_for_disambiguation ' +p,6820,'Preparing tagged corpora either by hand is labour-intensive and potentially error-prone , and although a semi-automatic approach can be used , it is a good thing to reduce the human involvement as much as possible ',Marcus,'csubj_labour-intensive_Preparing amod_corpora_tagged dobj_Preparing_corpora advmod_Preparing_either prep_Preparing_by pobj_by_hand cop_labour-intensive_is cc_labour-intensive_and advmod_error-prone_potentially conj_labour-intensive_error-prone cc_labour-intensive_and mark_used_although det_approach_a amod_approach_semi-automatic nsubjpass_used_approach aux_used_can auxpass_used_be advcl_thing_used dep_used_ nsubj_thing_it cop_thing_is det_thing_a amod_thing_good conj_labour-intensive_thing aux_reduce_to infmod_thing_reduce det_involvement_the amod_involvement_human dobj_reduce_involvement advmod_much_as advmod_reduce_much dep_much_as pobj_as_possible ' +o,6821,'Using dictionaries as network of lexical items or senses has been quite popular for word sense disambiguation before losing ground to statistical approaches , even though tried a revival of such methods ',Mihalcea,'dobj_Using_dictionaries mark_popular_as nsubj_popular_network prep_network_of amod_items_lexical pobj_of_items cc_items_or conj_items_senses aux_popular_has dep_popular_been advmod_popular_quite advcl_Using_popular prep_popular_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation partmod_disambiguation_ prep__before pcomp_before_losing dobj_losing_ground prep_losing_to amod_approaches_statistical pobj_to_approaches advmod_tried_even dep_tried_though nsubj_tried_ ccomp_popular_tried det_revival_a dobj_tried_revival prep_revival_of amod_methods_such pobj_of_methods ' +o,6822,'Method Number of frames Number of verbs Linguistic resources F-Score -LRB- evaluation based on a gold standard -RRB- Coverage on a corpus C Manning 19 200 POS tagger + simple finite state parser 58 T Briscoe & J Carroll 161 14 Full parser 55 A Sarkar & D Zeman 137 914 Annotated treebank 88 D Kawahara et al ',Dunning,'dobj_Method_Number prep_Number_of nn_Number_frames pobj_of_Number prep_Number_of amod_F-Score_verbs nn_F-Score_Linguistic nn_F-Score_resources pobj_of_F-Score dep_Number_evaluation partmod_evaluation_based prep_based_on det_standard_a amod_standard_gold pobj_on_standard advmod_Number_Coverage dep_Coverage_on det__a nn__corpus nn__C nn__Manning pobj_on_ number_200_19 dep_Number_200 dep_T_POS amod_T_tagger cc_tagger_+ conj_tagger_simple amod_T_finite nn_T_state nn_T_parser num_T_58 prep_200_T nn__Briscoe cc_Briscoe_& conj_Briscoe_J nn__Carroll dep_T_ number_14_161 num_Sarkar_14 nn_Sarkar_Full nn_Sarkar_parser num_Sarkar_55 nn_Sarkar_A prep__Sarkar cc_Sarkar_& nn__D nn__Zeman conj_Sarkar_ number_914_137 dep_Sarkar_914 nsubj_treebank_Annotated rcmod_914_treebank dobj_treebank_88 nn_al_D nn_al_Kawahara nn_al_et nsubj_Method_al ' +o,6823,' and in merely using binary unigram features , corresponding to the 17,744 unstemmed word or punctuation types with count 4 in the full 2000-document corpus ',Pang,'cc__and conj__ prep__in advmod_using_merely pcomp_in_using amod_features_binary nn_features_unigram dobj_using_features appos_features_corresponding prep_corresponding_to det_word_the num_word_17,744 amod_word_unstemmed pobj_to_word cc_word_or nn_types_punctuation conj_word_types prep_using_with pobj_with_count num_count_4 prep_count_in det_corpus_the amod_corpus_full amod_corpus_2000-document pobj_in_corpus ' +o,6824,'The recurrence property had been utilized to extract keywords or key-phrases from text ',Smadja,'det_property_The nn_property_recurrence nsubjpass_utilized_property aux_utilized_had auxpass_utilized_been aux_extract_to xcomp_utilized_extract dobj_extract_keywords cc_keywords_or conj_keywords_key-phrases prep_extract_from pobj_from_text ' +o,6825,'1 Introduction Hyponymy relations can play a crucial role in various NLP systems , and there have been many attempts to develop automatic methods to acquire hyponymy relations from text corpora ',Fleischman,'dep_play_1 nn_relations_Introduction nn_relations_Hyponymy nsubj_play_relations aux_play_can det_role_a amod_role_crucial dobj_play_role prep_play_in amod_systems_various nn_systems_NLP pobj_in_systems cc_play_and expl_been_there aux_been_have conj_play_been amod_attempts_many dobj_been_attempts aux_develop_to infmod_attempts_develop amod_methods_automatic dobj_develop_methods aux_acquire_to xcomp_develop_acquire amod_relations_hyponymy dobj_acquire_relations prep_acquire_from nn_corpora_text pobj_from_corpora ' +o,6826,'As a result , we can use collocation measures like point-wise mutual information or the log-likelihood ratio to predict the strong association for a given cue ',Dunning,'prep_use_As det_result_a pobj_As_result nsubj_use_we aux_use_can nn_measures_collocation dobj_use_measures prep_use_like amod__point-wise amod__mutual nn__information pobj_like_ cc__or det__the amod__log-likelihood nn__ratio conj__ aux_predict_to xcomp_use_predict det_association_the amod_association_strong dobj_predict_association prep_association_for det_cue_a amod_cue_given pobj_for_cue ' +o,6827,'Thus , a lot of alignment techniques have been suggested at ; the sentence , phrase , nomt t -RRB- hrase , word , collocation and terminology level ',Berger,'advmod_suggested_Thus det_lot_a nsubjpass_suggested_lot prep_lot_of amod_techniques_alignment pobj_of_techniques aux_suggested_have auxpass_suggested_been prep_suggested_at det__the nn__sentence pobj_at_ nn__phrase conj__ amod__nomt nn__t amod__hrase conj__ nn__word conj__ nn__collocation conj__ cc__and amod_level_terminology conj__level ' +o,6828,' s approach for English resolves three LDD types in parser output trees without traces and coindexation -LRB- Figure 2 -LRB- b -RRB- -RRB- , ie topicalisation -LRB- TOPIC -RRB- , wh-movement in relative clauses -LRB- TOPIC REL -RRB- and interrogatives -LRB- FOCUS -RRB- ',Cahill,'dep_approach_ dep_approach_s prep_approach_for amod_resolves_English pobj_for_resolves num_types_three nn_types_LDD nsubj_approach_types prep_approach_in nn_trees_parser nn_trees_output pobj_in_trees prep_trees_without pobj_without_traces cc_trees_and conj_trees_coindexation dep_coindexation_Figure num_Figure_2 dep_-LRB-_b nn_topicalisation_ie conj_coindexation_topicalisation abbrev_topicalisation_TOPIC conj_coindexation_wh-movement prep_wh-movement_in amod_clauses_relative pobj_in_clauses nn_REL_TOPIC abbrev_wh-movement_REL cc_coindexation_and conj_coindexation_interrogatives abbrev_interrogatives_FOCUS ' +o,6829,'REALM uses an HMM trained on a large corpus to help determine whether the arguments of a candidate relation are of the appropriate type ',Downey,'nsubj_uses_REALM det_HMM_an dobj_uses_HMM partmod_HMM_trained prep_trained_on det_corpus_a amod_corpus_large pobj_on_corpus aux_help_to xcomp_trained_help dep_help_determine complm_are_whether det_arguments_the nsubj_are_arguments prep_arguments_of det_relation_a nn_relation_candidate pobj_of_relation ccomp_determine_are prep_are_of det_type_the amod_type_appropriate pobj_of_type ' +o,6830,'c2007 Association for Computational Linguistics Structural Correspondence Learning for Dependency Parsing Nobuyuki Shimizu Information Technology Center University of Tokyo Tokyo , Japan shimizu @ rdlitcu-tokyoacjp Hiroshi Nakagawa Information Technology Center University of Tokyo Tokyo , Japan nakagawa @ dlitcu-tokyoacjp Abstract Following , we present an application of structural correspondence learning to non-projective dependency parsing ',Blitzer,'nn_Association_c2007 prep_Association_for nn_Learning_Computational nn_Learning_Linguistics nn_Learning_Structural nn_Learning_Correspondence pobj_for_Learning prep_Learning_for nn_University_Dependency nn_University_Parsing nn_University_Nobuyuki nn_University_Shimizu nn_University_Information nn_University_Technology nn_University_Center pobj_for_University prep_University_of nn_Tokyo_Tokyo pobj_of_Tokyo nn_shimizu_Japan appos_Tokyo_shimizu dep_University_@ amod_University_rdlitcu-tokyoacjp nn_University_Hiroshi nn_University_Nakagawa nn_University_Information nn_University_Technology nn_University_Center dep_present_University prep_University_of nn_Tokyo_Tokyo pobj_of_Tokyo nn_nakagawa_Japan appos_Tokyo_nakagawa dep_present_@ amod__dlitcu-tokyoacjp nn__Abstract nn__Following dep_@_ nsubj_present_we dep_Association_present det_application_an dobj_present_application prep_application_of amod_correspondence_structural pobj_of_correspondence partmod_correspondence_learning aux_non-projective_to xcomp_learning_non-projective amod__dependency amod__parsing dobj_non-projective_ ' +o,6831,'In shift-reduce parsing , further mistakes are often caused by previous ones , so only the first mistake in each sentence -LRB- if there is one -RRB- is easily identifiable ; 7 this is also the argument for early update in applying perceptron learning to these incremental parsing algorithms -LRB- see also Section 2 -RRB- ',Collins,'prep_caused_In amod_parsing_shift-reduce pobj_In_parsing amod_mistakes_further nsubjpass_caused_mistakes auxpass_caused_are advmod_caused_often prep_caused_by amod_ones_previous pobj_by_ones dep_caused_so advmod_mistake_only det_mistake_the amod_mistake_first nsubj_is_mistake prep_mistake_in det_sentence_each pobj_in_sentence mark_is_if expl_is_there dep_sentence_is nsubj_is_one dep_caused_is advmod_identifiable_easily acomp_is_identifiable nsubj_argument_7 dep_7_this cop_argument_is advmod_argument_also det_argument_the parataxis_caused_argument prep_argument_for amod_update_early pobj_for_update prep_update_in pcomp_in_applying dobj_applying_perceptron partmod_perceptron_learning prep_learning_to det__these amod__incremental nn__parsing nn__algorithms pobj_to_ dep_argument_see advmod_see_also dobj_see_Section num_Section_2 ' +o,6832,'We have also used ROUGE evaluation approach which is based on n-gram co-occurrences between machine summaries and ideal human summaries ',Lin,'nsubj_used_We aux_used_have advmod_used_also amod__ROUGE nn__evaluation nn__approach dobj_used_ nsubjpass_based_which auxpass_based_is rcmod__based prep_based_on amod_co-occurrences_n-gram pobj_on_co-occurrences prep_co-occurrences_between nn_summaries_machine pobj_between_summaries cc_summaries_and amod_summaries_ideal amod_summaries_human conj_summaries_summaries ' +o,6833,'Automatically Learning Entailment Rules from the Web Many algorithms for automatically learning paraphrases and entailment rules have been explored in recent years ',Barzilay,'nn_Rules_Automatically nn_Rules_Learning nn_Rules_Entailment nsubjpass_explored_Rules prep_Rules_from det_algorithms_the nn_algorithms_Web amod_algorithms_Many pobj_from_algorithms prep_algorithms_for advmod_learning_automatically pcomp_for_learning dobj_learning_paraphrases cc_paraphrases_and nn_rules_entailment conj_paraphrases_rules aux_explored_have auxpass_explored_been prep_explored_in amod_years_recent pobj_in_years ' +o,6834,'In particular , we use the name\\/instance lists described by and available on Fleischmans web page to generate features between names and nominals -LRB- this list contains a110a111a85 pairs mined from a112a73a96 GBs of news data -RRB- ',Fleischman,'prep_use_In pobj_In_particular nsubj_use_we det_lists_the amod_lists_name\\/instance dobj_use_lists partmod_lists_described prep_described_by pobj_by_ cc__and conj__available prep_described_on nn_page_Fleischmans nn_page_web pobj_on_page aux_generate_to xcomp_described_generate dobj_generate_features prep_features_between pobj_between_names cc_names_and conj_names_nominals det_list_this nsubj_contains_list parataxis_use_contains amod_pairs_a110a111a85 dobj_contains_pairs partmod_pairs_mined prep_mined_from amod_GBs_a112a73a96 pobj_from_GBs prep_GBs_of nn_data_news pobj_of_data ' +o,6835,'3 Tagging 31 Corpus To facilitate comparison with previous results , we used the UPenn Treebank corpus ',Marcus,'dep_used_Tagging num_Corpus_31 dobj_Tagging_Corpus aux_facilitate_To xcomp_Tagging_facilitate dobj_facilitate_comparison prep_facilitate_with amod_results_previous pobj_with_results nsubj_used_we dep_3_used det_corpus_the nn_corpus_UPenn nn_corpus_Treebank dobj_used_corpus ' +o,6836,'p0 -LRB- t w -RRB- is calculated by ME models as follows : p0 -LRB- t w -RRB- = 1Y -LRB- w -RRB- exp braceleftBigg Hsummationdisplay h = 1 hgh -LRB- w , t -RRB- bracerightBigg , -LRB- 20 -RRB- 709 Language Features English Prefixes of 0 up to four characters , suffixes of 0 up to four characters , 0 contains Arabic numerals , 0 contains uppercase characters , 0 contains hyphens ',Berger,'nsubjpass_calculated_p0 nn_w_t appos_p0_w auxpass_calculated_is prep_calculated_by nn_models_ME pobj_by_models mark_follows_as advcl_calculated_follows nsubj_=_ dep__p0 nn_w_t appos__w ccomp_follows_= num_exp_1Y nn_exp_w dobj_=_exp partmod_exp_braceleftBigg nn_h_Hsummationdisplay nsubj_=_h ccomp_braceleftBigg_= num_hgh_1 dobj_=_hgh appos_hgh_w dep_w_t partmod_hgh_bracerightBigg dep_bracerightBigg_20 nn_Prefixes_709 nn_Prefixes_Language nn_Prefixes_Features nn_Prefixes_English nsubj_contains_Prefixes prep_Prefixes_of pobj_of_0 advmod_contains_up dep_up_to num_characters_four pobj_to_characters appos_characters_suffixes prep_suffixes_of pobj_of_0 advmod_contains_up dep_up_to num_characters_four pobj_to_characters num_numerals_0 nn_numerals_contains nn_numerals_Arabic appos_characters_numerals appos_numerals_0 dep_calculated_contains amod_characters_uppercase dobj_contains_characters cc_contains_0 conj_contains_contains dobj_contains_hyphens ' +o,6837,'a11a29a9 thea13 thea15 a1a4a3a6a5 a11a29a9 thea13 thea15 a11a29a9 thea15 a11a29a9 thea15a1a0 a2 since a11a2a9 thea13 thea15a4a3 a11a29a9 thea15 a11a29a9 thea15 Also note that in the case of phraseness of a bigram , the equation looks similar to pointwise mutual information , but they are different ',Church,'amod_a2_a11a29a9 amod_a2_thea13 amod_a2_thea15 amod_a2_a1a4a3a6a5 amod_a2_a11a29a9 amod_a2_thea13 amod_a2_thea15 amod_a2_a11a29a9 amod_a2_thea15 amod_a2_a11a29a9 nn_a2_thea15a1a0 nsubj_note_a2 prep_a2_since amod_thea15_a11a2a9 nn_thea15_thea13 nn_thea15_thea15a4a3 nn_thea15_a11a29a9 nn_thea15_thea15 amod_thea15_a11a29a9 pobj_since_thea15 advmod_note_Also complm_looks_that prep_looks_in det_case_the pobj_in_case prep_case_of pobj_of_phraseness prep_phraseness_of det_bigram_a pobj_of_bigram det_equation_the nsubj_looks_equation ccomp_note_looks acomp_looks_similar aux_pointwise_to xcomp_similar_pointwise amod__mutual nn__information dobj_pointwise_ cc_looks_but nsubj_different_they cop_different_are conj_looks_different ' +o,6838,' proposed a summarization system based on the draft and revision proposed a system based on extraction and cut-and-paste generation Our abstractors performed the same cut-and-paste operations that Jing and McKeown noted in their work , and we think that our two-step model will be a reasonable starting point for our subsequent research ',Jing,'nsubj_proposed_ det_system_a nn_system_summarization dobj_proposed_system prep_proposed_based dep_based_on det__the nn__draft cc_draft_and conj_draft_revision pobj_on_ partmod__proposed det_system_a dobj_proposed_system prep_proposed_based dep_based_on pobj_on_extraction cc_extraction_and amod_generation_cut-and-paste conj_extraction_generation poss_abstractors_Our nsubj_performed_abstractors dep_proposed_performed det_operations_the amod_operations_same amod_operations_cut-and-paste dobj_performed_operations complm_noted_that nsubj_noted_Jing cc_Jing_and conj_Jing_McKeown ccomp_performed_noted prep_noted_in poss_work_their pobj_in_work cc_performed_and nsubj_think_we conj_performed_think complm_point_that poss_model_our amod_model_two-step nsubj_point_model aux_point_will cop_point_be det_point_a amod_point_reasonable amod_point_starting ccomp_think_point prep_point_for poss_research_our amod_research_subsequent pobj_for_research ' +p,6839,'Research in this direction was pioneered by , who developed Inversion Transduction Grammars to capture crosslingual grammar variations such as phrase reorderings ',Wu,'nsubjpass_pioneered_Research prep_Research_in det_direction_this pobj_in_direction auxpass_pioneered_was dep_developed_pioneered prep_pioneered_by pobj_by_ nsubj_developed_who nn_Grammars_Inversion nn_Grammars_Transduction nsubj_capture_Grammars aux_capture_to xcomp_developed_capture amod_variations_crosslingual nn_variations_grammar dobj_capture_variations dep_as_such prep_variations_as nn_reorderings_phrase pobj_as_reorderings ' +o,6840,'213 Correlation analysis As a correlation measure between terms , we use mutual information ',Church,'num_analysis_213 amod_analysis_Correlation prep_use_As det_measure_a nn_measure_correlation pobj_As_measure prep_measure_between pobj_between_terms nsubj_use_we dep_analysis_use amod_information_mutual dobj_use_information ' +o,6841,'The f-structure annotation algorithm used for inducing LFG resources from the Penn-II treebank for English uses configurational , categorial , function tag and trace information ',Cahill,'det_algorithm_The amod_algorithm_f-structure nn_algorithm_annotation nsubj_uses_algorithm partmod_algorithm_used prep_used_for pcomp_for_inducing nn_resources_LFG dobj_inducing_resources prep_inducing_from det_treebank_the amod_treebank_Penn-II pobj_from_treebank prep_treebank_for nn__English pobj_for_ dobj_uses_configurational conj_configurational_categorial nn_tag_function conj_configurational_tag cc_configurational_and amod_information_trace conj_configurational_information ' +n,6842,'2 Previous work on Sentiment Analysis Some prior studies on sentiment analysis focused on the document-level classification of sentiment where a document is assumed to have only a single sentiment , thus these studies are not applicable to our goal ',Turney,'num_work_2 amod_work_Previous dep_studies_work prep_work_on nn_Analysis_Sentiment pobj_on_Analysis dep_studies_Some amod_studies_prior dep_analysis_studies prep_studies_on pobj_on_sentiment partmod_analysis_focused prep_focused_on det_classification_the amod_classification_document-level pobj_on_classification prep_classification_of nn__sentiment pobj_of_ advmod_assumed_where det_document_a nsubjpass_assumed_document auxpass_assumed_is advcl_focused_assumed aux_have_to xcomp_assumed_have advmod_sentiment_only det_sentiment_a amod_sentiment_single dobj_have_sentiment advmod_assumed_thus det_studies_these nsubj_applicable_studies cop_applicable_are neg_applicable_not dep_assumed_applicable prep_applicable_to poss_goal_our pobj_to_goal ' +o,6843,'Building on a recent proposal in this direction by , we propose a generic method of this sort , and we test it on a set of unrelated tasks , reporting good performance across the board with very little task-specific tweaking ',Turney,'prep_Building_on det_proposal_a amod_proposal_recent pobj_on_proposal prep_proposal_in det_direction_this pobj_in_direction prep_Building_by pobj_by_ nsubj_propose_we dep_Building_propose det_method_a amod_method_generic dobj_propose_method prep_method_of det_sort_this pobj_of_sort cc_Building_and nsubj_test_we conj_Building_test dobj_test_it prep_test_on det_set_a pobj_on_set prep_set_of amod_tasks_unrelated pobj_of_tasks xcomp_test_reporting amod_performance_good dobj_reporting_performance prep_reporting_across det_board_the pobj_across_board prep_board_with advmod_tweaking_very advmod_tweaking_little amod_tweaking_task-specific pobj_with_tweaking ' +o,6844,'Instances of this work include information extraction , ontology induction and resource acquisition ',Biadsy,'nsubj_include_Instances prep_Instances_of det_work_this pobj_of_work nn_extraction_information dobj_include_extraction amod_induction_ontology conj_extraction_induction cc_extraction_and nn_acquisition_resource conj_extraction_acquisition ' +o,6845,'We further assume that the degree of difficulty of a phrase is directly correlated with the quality of the translation produced by the MT system , which can be approximated using an automatic evaluation metric , such as BLEU ',Papineni,'nsubj_assume_We advmod_assume_further complm_correlated_that det_degree_the nsubjpass_correlated_degree prep_degree_of pobj_of_difficulty prep_difficulty_of det_phrase_a pobj_of_phrase auxpass_correlated_is advmod_correlated_directly ccomp_assume_correlated prep_correlated_with det_quality_the pobj_with_quality prep_quality_of det_translation_the pobj_of_translation partmod_translation_produced prep_produced_by det_system_the nn_system_MT pobj_by_system nsubjpass_approximated_which aux_approximated_can auxpass_approximated_be rcmod_system_approximated purpcl_approximated_using det_metric_an amod_metric_automatic nn_metric_evaluation dobj_using_metric dep_as_such prep_system_as pobj_as_BLEU ' +p,6846,'Since its introduction to the Natural Language Processing -LRB- NLP -RRB- community , ME-based classifiers have been shown to be effective in various NLP tasks ',Berger,'prep_shown_Since poss__its nn__introduction prep_introduction_to det_Processing_the nn_Processing_Natural nn_Processing_Language pobj_to_Processing abbrev_introduction_NLP nn__community pobj_Since_ amod_classifiers_ME-based nsubjpass_shown_classifiers aux_shown_have auxpass_shown_been aux_effective_to cop_effective_be xcomp_shown_effective prep_effective_in amod_tasks_various nn_tasks_NLP pobj_in_tasks ' +o,6847,'We adopted IOB -LRB- IOB2 -RRB- labeling , where the rst word of an entity of class C is labeled B-C , the words in the entity are labeled I-C , and other words are labeled O ',Ramshaw,'nsubj_adopted_We nn__IOB nn__IOB2 nn__labeling dobj_adopted_ advmod_labeled_where det_word_the nn_word_rst nsubjpass_labeled_word prep_word_of det_entity_an pobj_of_entity prep_entity_of nn_C_class pobj_of_C auxpass_labeled_is rcmod__labeled dobj_labeled_B-C det_words_the nsubjpass_labeled_words prep_words_in det_entity_the pobj_in_entity auxpass_labeled_are ccomp_adopted_labeled dobj_labeled_I-C cc_adopted_and amod_words_other nsubjpass_labeled_words auxpass_labeled_are conj_adopted_labeled dobj_labeled_O ' +o,6848,'If human-aligned data is available , the EMD algorithm provides higher baseline alignments than GIZA + + that have led to better MT performance ',Fraser,'mark_available_If amod_data_human-aligned nsubj_available_data cop_available_is advcl_provides_available det_algorithm_the nn_algorithm_EMD nsubj_provides_algorithm amod_alignments_higher nn_alignments_baseline dobj_provides_alignments prep_provides_than nn_+_GIZA nn_+_+ pobj_than_+ nsubj_led_that aux_led_have rcmod_+_led prep_led_to amod_performance_better nn_performance_MT pobj_to_performance ' +o,6849,'2 Related Work Sentiment Classi cation Traditionally , categorization of opinion texts has been cast as a binary classication task ',Pang,'dep_cast_2 nn_Traditionally_Related nn_Traditionally_Work nn_Traditionally_Sentiment nn_Traditionally_Classi nn_Traditionally_cation nsubjpass_cast_Traditionally nsubjpass_cast_categorization prep_categorization_of nn_texts_opinion pobj_of_texts aux_cast_has auxpass_cast_been prep_cast_as det_task_a amod_task_binary nn_task_classication pobj_as_task ' +p,6850,'We employ a robust statistical parser to determine the constituent structure for each sentence , from which subjects -LRB- s -RRB- , objects -LRB- o -RRB- , and relations other than subject or object -LRB- x -RRB- are identified ',Collins,'nsubj_employ_We det__a amod__robust amod__statistical nn__parser nsubj_determine_ aux_determine_to xcomp_employ_determine det_structure_the amod_structure_constituent dobj_determine_structure prep_determine_for det_sentence_each pobj_for_sentence rel_identified_from pobj_from_which nsubjpass_identified_subjects dep_-LRB-_s conj_subjects_objects appos_objects_o cc_subjects_and conj_subjects_relations dep_than_other prep_relations_than pobj_than_subject cc_subject_or conj_subject_object appos_subjects_x auxpass_identified_are rcmod_sentence_identified ' +o,6851,' describe a method of disambiguation , where disambiguation questions are dynamically constructed on the basis of an analysis of the differences among the closest routing destination vectors ',Chu-Carroll,'nsubj_describe_ det_method_a dobj_describe_method prep_method_of pobj_of_disambiguation advmod_constructed_where nn_questions_disambiguation nsubjpass_constructed_questions auxpass_constructed_are advmod_constructed_dynamically rcmod_disambiguation_constructed prep_constructed_on det_basis_the pobj_on_basis prep_basis_of det_analysis_an pobj_of_analysis prep_analysis_of det_differences_the pobj_of_differences prep_constructed_among det_vectors_the amod_vectors_closest amod_vectors_routing nn_vectors_destination pobj_among_vectors ' +o,6852,'The unit of utterance corresponds to the unit of segment in the original BLEU and NIST studies ',Papineni,'det_unit_The nsubj_corresponds_unit prep_unit_of pobj_of_utterance prep_corresponds_to det_unit_the pobj_to_unit prep_unit_of pobj_of_segment prep_segment_in det_BLEU_the amod_BLEU_original pobj_in_BLEU cc_BLEU_and nn_studies_NIST conj_BLEU_studies ' +o,6853,'Another possible comparison could be with a version of sentiment classification method applied to Chinese ',Turney,'det_comparison_Another amod_comparison_possible nsubj_be_comparison aux_be_could prep_be_with det_version_a pobj_with_version prep_version_of amod_method_ nn_method_sentiment nn_method_classification pobj_of_method partmod_method_applied prep_applied_to pobj_to_Chinese ' +o,6854,'Then the word alignment is refined by performing growdiag-final method ',Koehn,'det_alignment_the nn_alignment_word dep_Then_alignment auxpass_refined_is rcmod_alignment_refined prep_refined_by pcomp_by_performing amod_method_growdiag-final dobj_performing_method ' +p,6855,'Unlike Choueka , identify as collocations both interrupted and uninterrupted sequences of words ',Church,'prep_identify_Unlike nn__Choueka pobj_Unlike_ nsubj_identify_ mark_collocations_as advcl_identify_collocations preconj_interrupted_both amod_sequences_interrupted cc_interrupted_and conj_interrupted_uninterrupted dobj_collocations_sequences prep_sequences_of pobj_of_words ' +o,6856,'3 24 Intonation Annotations For our intonation annotation , we have annotated the intonational phrase boundaries , using the ToBI -LRB- Tones and Break Indices -RRB- definition ',Brown,'number_24_3 num_Annotations_24 nn_Annotations_Intonation nsubj_annotated_Annotations prep_annotated_For poss_annotation_our nn_annotation_intonation pobj_For_annotation nsubj_annotated_we aux_annotated_have det_boundaries_the amod_boundaries_intonational nn_boundaries_phrase dobj_annotated_boundaries xcomp_annotated_using det__the nn__ToBI dep_ToBI_Tones cc_Tones_and nn_Indices_Break conj_Tones_Indices nn__definition dobj_using_ ' +o,6857,'For extrinsic evaluation of machine translation , we use the BLEU metric ',Papineni,'prep_use_For amod_evaluation_extrinsic pobj_For_evaluation prep_evaluation_of nn_translation_machine pobj_of_translation nsubj_use_we det_metric_the amod_metric_BLEU dobj_use_metric ' +o,6858,'Previous work in statistical synchronous grammars has been limited to forms of synchronous context-free grammar ',Wu,'amod_work_Previous nsubjpass_limited_work prep_work_in amod_grammars_statistical amod_grammars_synchronous pobj_in_grammars aux_limited_has auxpass_limited_been prep_limited_to pobj_to_forms prep_forms_of amod_grammar_synchronous amod_grammar_context-free pobj_of_grammar ' +o,6859,'In this respect it resembles Wus 264 bilingual bracketer , but ours uses a different extraction method that allows more than one lexical item in a rule , in keeping with the phrasebased philosophy ',Wu,'prep_Wus_In det_respect_this pobj_In_respect nsubj_Wus_it aux_Wus_resembles num__264 amod__bilingual nn__bracketer dobj_Wus_ cc_Wus_but nsubj_uses_ours conj_Wus_uses det_method_a amod_method_different nn_method_extraction dobj_uses_method nsubj_allows_that rcmod_method_allows dep_than_more quantmod_one_than num_item_one amod_item_lexical dobj_allows_item prep_allows_in det_rule_a pobj_in_rule prep_allows_in pcomp_in_keeping prep_keeping_with det_philosophy_the amod_philosophy_phrasebased pobj_with_philosophy ' +o,6860,'For this reason there is currently a great deal of interest in methods which incorporate syntactic information within statistical machine translation systems -LRB- eg , see -RRB- ',Berger,'prep_is_For det_reason_this pobj_For_reason expl_is_there advmod_is_currently det_deal_a amod_deal_great nsubj_is_deal prep_deal_of pobj_of_interest prep_interest_in pobj_in_methods nsubj_incorporate_which rcmod_methods_incorporate amod_information_syntactic dobj_incorporate_information prep_incorporate_within amod_systems_statistical nn_systems_machine nn_systems_translation pobj_within_systems dep_deal_eg dep_eg_see dobj_eg_ ' +p,6861,'The state-of-the-art SMT system Moses implements a distance-based reordering model and a distortion model , operating with rewrite patterns extracted from a phrase alignment table ',Koehn,'det_Moses_The amod_Moses_state-of-the-art nn_Moses_SMT nn_Moses_system nsubj_implements_Moses det__a amod__distance-based nn__reordering nn__model dobj_implements_ cc__and det_model_a nn_model_distortion conj__model xcomp_implements_operating prep_operating_with amod_patterns_rewrite pobj_with_patterns partmod_patterns_extracted prep_extracted_from det__a nn__phrase nn__alignment nn__table pobj_from_ ' +p,6862,'Maximum Entropy -LRB- MaxEnt -RRB- principle has been successfully applied in many classification and tagging tasks ',Ratnaparkhi,'nn_Entropy_Maximum appos_Entropy_MaxEnt nsubjpass_applied_principle aux_applied_has auxpass_applied_been advmod_applied_successfully dep_Entropy_applied prep_applied_in amod_classification_many pobj_in_classification cc_classification_and nn_tasks_tagging conj_classification_tasks ' +o,6863,'Using our WSD model to constrain the translation candidates given to the decoder hurts translation quality , as measured by the automated BLEU metric ',Papineni,'poss_model_our nn_model_WSD dobj_Using_model aux_constrain_to xcomp_Using_constrain det_candidates_the nn_candidates_translation dobj_constrain_candidates prep_constrain_given dep_given_to det_decoder_the pobj_to_decoder dep_Using_hurts nn_quality_translation nsubj_Using_quality mark_measured_as dep_quality_measured prep_measured_by det_metric_the amod_metric_automated nn_metric_BLEU pobj_by_metric ' +o,6864,'In the first step , the scores are initialized according to the G 2 statistic ',Dunning,'prep_initialized_In det_step_the amod_step_first pobj_In_step det_scores_the nsubjpass_initialized_scores auxpass_initialized_are prep_initialized_according dep_according_to det_statistic_the nn_statistic_G num_statistic_2 pobj_to_statistic ' +o,6865,'-LRB- levelopment of cor1 -RRB- ora with morl -RRB- ho-synta -LRB- : ti -LRB- : and syntacti -LRB- : mmotation , ',Ramshaw,'number_cor1_levelopment quantmod_cor1_of punct_morl_ora dep_morl_with dep_cor1_morl dep_cor1_ho-synta dep_ho-synta_ti cc_ti_and conj_ti_syntacti dep__mmotation dep_syntacti_ appos__ ' +o,6866,'The translation models and lexical scores were estimated on the training corpus whichwasautomaticallyalignedusingGiza + + in both directions between source and target and symmetrised using the growing heuristic ',Och,'det_models_The nn_models_translation nsubjpass_estimated_models cc_models_and nn_scores_lexical conj_models_scores auxpass_estimated_were prep_estimated_on det_whichwasautomaticallyalignedusingGiza_the nn_whichwasautomaticallyalignedusingGiza_training nn_whichwasautomaticallyalignedusingGiza_corpus pobj_on_whichwasautomaticallyalignedusingGiza amod_whichwasautomaticallyalignedusingGiza_+ cc_+_+ conj_+_ prep_estimated_in det_directions_both pobj_in_directions prep_directions_between pobj_between_source cc_source_and conj_source_target cc_estimated_and conj_estimated_symmetrised xcomp_symmetrised_using det__the amod__growing amod__heuristic dobj_using_ ' +o,6867,'These 30 questions are determined by growing a classification tree on the word vocabulary as described in ',Brown,'det_questions_These num_questions_30 nsubjpass_determined_questions auxpass_determined_are prep_determined_by pcomp_by_growing det_tree_a nn_tree_classification dobj_growing_tree prep_growing_on det_vocabulary_the nn_vocabulary_word pobj_on_vocabulary mark_described_as advcl_growing_described prt_described_in ' +o,6868,'660 2 Statistical Coreference Resolution Model Our coreference system uses a binary entity-mention model PL( je, m) (henceforth link model ) to score the action of linking a mention m to an entity e. In our implementation, the link model is computed as PL(L = 1je, m) max mprimee PL(L = 1je, mprime, m), (1) where mprime is one mention in entity e, and the basic model building block PL(L = 1je, mprime, m) is an exponential or maximum entropy model (Berger et al. , 1996): PL(Lje, mprime, m) = exp braceleftbig summationtext i igi(e, m prime, m, L)bracerightbig Z(e, mprime, m), (2) where Z(e, mprime, m) is a normalizing factor to ensure that PL( je, mprime, m) is a probability, fgi(e, mprime, m, L)g are features and fig are feature weights.',Berger,'' +o,6869,'Bitexts also play a role in less automated applications such as concordancing for bilingual lexicography , computer-assisted language learning , and tools for translators -LRB- eg ',Brown,'nsubj_play_Bitexts advmod_play_also det_role_a dobj_play_role prep_role_in amod_applications_less amod_applications_automated pobj_in_applications dep_as_such prep_applications_as pcomp_as_concordancing prep_concordancing_for amod__bilingual nn__lexicography pobj_for_ amod_language_computer-assisted nsubj_learning_language conj_play_learning cc_play_and nsubj_translators_tools advmod_translators_for conj_play_translators dobj_translators_eg ' +o,6870,'To measure the translation quality , we use the BLEU score and the NIST score ',Papineni,'aux_measure_To dep_use_measure det_quality_the nn_quality_translation dobj_measure_quality nsubj_use_we det__the amod__BLEU nn__score dobj_use_ cc__and det__the amod__NIST nn__score conj__ ' +o,6871,'Moreover , our approach integrates the abbreviation translation component into the baseline system in a natural way , and thus is able to make use of the minimum-error-rate training to automatically adjust the model parameters to reflect the change of the integrated system over the baseline system ',Och,'advmod_integrates_Moreover poss_approach_our nsubj_integrates_approach det_component_the nn_component_abbreviation nn_component_translation dobj_integrates_component prep_integrates_into det_system_the amod_system_baseline pobj_into_system prep_system_in det_way_a amod_way_natural pobj_in_way cc_integrates_and advmod_integrates_thus cop_able_is conj_integrates_able aux_make_to xcomp_able_make dobj_make_use prep_use_of det__the amod__minimum-error-rate nn__training pobj_of_ aux_adjust_to advmod_adjust_automatically xcomp_make_adjust det_parameters_the nn_parameters_model nsubj_reflect_parameters aux_reflect_to xcomp_adjust_reflect det_change_the dobj_reflect_change prep_change_of det_system_the amod_system_integrated pobj_of_system prep_system_over det_system_the nn_system_baseline pobj_over_system ' +o,6872,'4 Method-2 : Simple Chunk-based Extraction To overcome the shortcomings of the Brill tagger in identifying particles , we next look to full chunk 2Note , this is the same as the maximum span length of 5 used by , and above the maximum attested NP length of 3 from our corpus study -LRB- see Section 22 -RRB- ',Smadja,'num_Method-2_4 dep_length_Method-2 dep_length_Simple nn_Extraction_Chunk-based nsubj_overcome_Extraction aux_overcome_To xcomp_Simple_overcome det_shortcomings_the dobj_overcome_shortcomings prep_shortcomings_of det_tagger_the nn_tagger_Brill pobj_of_tagger prep_overcome_in pcomp_in_identifying nsubj_look_particles nsubj_look_we advmod_look_next ccomp_identifying_look prep_look_to amod_2Note_full nn_2Note_chunk pobj_to_2Note nsubj_length_this cop_length_is det_length_the amod_length_same prep_same_as det_span_the amod_span_maximum pobj_as_span prep_length_of pobj_of_5 partmod_length_used prep_used_by pobj_by_ cc_by_and conj_by_above det_length_the amod_length_maximum amod_length_attested nn_length_NP pobj_above_length prep_length_of pobj_of_3 prep_used_from poss_study_our nn_study_corpus pobj_from_study dep_length_see dobj_see_Section num_Section_22 ' +o,6873,'As notes , many tasks in computational linguistics are simply more difficult than the content analysis classifications addressed by Krippendorff , and according to Fleiss , kappa values between 4 and 75 indicate fair to good agreement anyhow ',Carletta,'prep_difficult_As amod_notes_ pobj_As_notes amod_tasks_many nsubj_difficult_tasks prep_tasks_in amod_linguistics_computational pobj_in_linguistics cop_difficult_are advmod_difficult_simply advmod_difficult_more mark_addressed_than det_classifications_the nn_classifications_content nn_classifications_analysis nsubj_addressed_classifications ccomp_difficult_addressed prep_addressed_by pobj_by_Krippendorff cc_difficult_and prep_indicate_according dep_according_to nn__Fleiss pobj_to_ nn_values_kappa nsubj_indicate_values prep_values_between pobj_between_4 cc_4_and conj_4_75 conj_difficult_indicate acomp_indicate_fair prep_fair_to amod_anyhow_good nn_anyhow_agreement pobj_to_anyhow ' +o,6874,'We rerank derivations with cube growing , a lazy beam search algorithm ',Huang,'nsubj_derivations_We nn_derivations_rerank prep_derivations_with pobj_with_cube amod_cube_growing det_algorithm_a amod_algorithm_lazy nn_algorithm_beam nn_algorithm_search dobj_derivations_algorithm ' +o,6875,'Two main extensions from that work that we are making use of are : 1 -RRB- proofs falling below a user defined cost threshold halt the search 2 -RRB- a simple variable typing system reduces the number of axioms written and the size of the search space ',Hobbs,'num_extensions_Two amod_extensions_main nsubj_are_extensions prep_extensions_from det_work_that pobj_from_work complm_making_that nsubj_making_we aux_making_are dep_work_making dobj_making_use prep_use_of dep_defined_1 nsubj_defined_proofs partmod_proofs_falling prep_falling_below det_user_a pobj_below_user parataxis_are_defined nn_threshold_cost nsubj_halt_threshold ccomp_defined_halt det_search_the dobj_halt_search dep_reduces_2 det_system_a amod_system_simple amod_system_variable nn_system_typing nsubj_reduces_system dep_are_reduces det_number_the dobj_reduces_number prep_number_of pobj_of_axioms partmod_axioms_written cc_axioms_and det_size_the conj_axioms_size prep_size_of det_space_the nn_space_search pobj_of_space ' +o,6876,'While calculate n-gram matches on non-labelled head-modifier sequences derived by head-extraction rules from syntactic trees , we automatically evaluate the quality of translation by calculating an f-score on labeled dependency structures produced by a LexicalFunctional Grammar -LRB- LFG -RRB- parser ',Liu,'mark_matches_While nn_n-gram_ nn_n-gram_calculate nsubj_matches_n-gram advcl_evaluate_matches prep_matches_on amod_sequences_non-labelled amod_sequences_head-modifier pobj_on_sequences partmod_sequences_derived prep_derived_by amod_rules_head-extraction pobj_by_rules prep_derived_from amod_trees_syntactic pobj_from_trees nsubj_evaluate_we advmod_evaluate_automatically det_quality_the dobj_evaluate_quality prep_quality_of pobj_of_translation prep_evaluate_by pcomp_by_calculating det_parser_an amod_parser_f-score prep_f-score_on amod_structures_labeled amod_structures_dependency pobj_on_structures partmod_structures_produced prep_produced_by det_Grammar_a nn_Grammar_LexicalFunctional pobj_by_Grammar abbrev_Grammar_LFG dobj_calculating_parser ' +o,6877,'4 Testing the Four Hypotheses The question of why self-training helps in some cases but not others has inspired various theories ',McClosky,'num_Testing_4 det_Four_the nsubj_Hypotheses_Four dep_Testing_Hypotheses det_question_The dobj_Hypotheses_question prep_question_of advmod_helps_why nsubj_helps_self-training pcomp_of_helps prep_helps_in det__some nn__cases pobj_in_ dep_not_but cc__not conj__others nsubj_inspired_ aux_inspired_has dep_Testing_inspired amod_theories_various dobj_inspired_theories ' +o,6878,'Large treebanks are available for major languages , however these are often based on a speci c text type or genre , eg nancial newspaper text -LRB- the Penn-II Treebank -RRB- ',Marcus,'amod_treebanks_Large cop_available_are rcmod_treebanks_available prep_available_for amod_languages_major pobj_for_languages advmod_based_however nsubjpass_based_these auxpass_based_are advmod_based_often dep_languages_based prep_based_on det_speci_a pobj_on_speci dep_type_c nn_type_text dep_treebanks_type cc_type_or conj_type_genre partmod_type_eg amod_text_nancial nn_text_newspaper dobj_eg_text det__the nn__Penn-II nn__Treebank appos_text_ ' +o,6879,'4 The Experiment For our experiment , we used a tree-bank grammar induced from sections 2-21 of the Penn Wall Street Journal text , with section 22 reserved for testing ',Marcus,'dep_used_4 det_Experiment_The dep_For_Experiment prep_used_For poss_experiment_our pobj_For_experiment nsubj_used_we det_grammar_a amod_grammar_tree-bank dobj_used_grammar partmod_grammar_induced prep_induced_from pobj_from_sections num_sections_2-21 prep_sections_of det__the nn__Penn nn__Wall nn__Street nn__Journal nn__text pobj_of_ prep_used_with pobj_with_section num_section_22 partmod_section_reserved prep_reserved_for pobj_for_testing ' +o,6880,'test additional resources JESS-CM -LRB- CRF\\/HMM -RRB- 9735 9740 1G-word unlabeled data 9728 9733 9715 9724 crude company name detector -LRB- sup ',Shen,'amod_resources_additional dobj_test_resources nsubj_test_JESS-CM abbrev_JESS-CM_CRF\\/HMM number_9740_9735 num_data_9740 amod_data_1G-word amod_data_unlabeled dep_JESS-CM_data partmod_data_ number_9733_9728 num__9733 dobj__ number_9724_9715 num_name_9724 amod_name_crude nn_name_company dep__name nn_sup_detector dep_name_sup ' +o,6881,'We use the GIZA + + implementation of IBM Model 4 coupled with the phrase extraction heuristics of Koehn et al ',Och,'nsubj_use_We det_implementation_the amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dobj_use_implementation prep_implementation_of nn__IBM nn__Model num__4 pobj_of_ partmod__coupled prep_coupled_with det_heuristics_the nn_heuristics_phrase nn_heuristics_extraction pobj_with_heuristics prep_heuristics_of pobj_of_Koehn cc_Koehn_et conj_Koehn_al ' +p,6882,'In the well-known so-called IBM word alignment models , re-estimating the model parameters depends on the empirical probability P -LRB- ek , fk -RRB- for each sentence pair -LRB- ek , fk -RRB- ',Brown,'prep__In det_word_the amod_word_well-known amod_word_so-called nn_word_IBM pobj_In_word nn_models_alignment nsubj__models xcomp__re-estimating det_parameters_the nn_parameters_model nsubj_depends_parameters ccomp_re-estimating_depends prep_depends_on det_P_the amod_P_empirical nn_P_probability pobj_on_P appos_P_ek dep_ek_fk prep_P_for det_pair_each nn_pair_sentence pobj_for_pair appos_pair_ek dep_ek_fk ' +p,6883,'The classification is performed with a statistical approach , built around the maximum entropy -LRB- MaxEnt -RRB- principle , that has the advantage of combining arbitrary types of information in making a classification decision ',Berger,'det_classification_The nsubjpass_performed_classification auxpass_performed_is prep_performed_with det_approach_a amod_approach_statistical pobj_with_approach partmod_approach_built prep_built_around det_entropy_the amod_entropy_maximum nn__entropy appos__MaxEnt nn__principle pobj_around_ nsubj_has_that rcmod__has det_advantage_the dobj_has_advantage prep_advantage_of pcomp_of_combining amod_types_arbitrary dobj_combining_types prep_types_of pobj_of_information prep_combining_in pcomp_in_making det_decision_a nn_decision_classification dobj_making_decision ' +o,6884,'4 5 Experiments 51 Evaluation Measures We evaluated the proposed method using four evaluation measures , BLEU , NIST , WER -LRB- word error rate -RRB- , and PER -LRB- position independent word error rate -RRB- ',Papineni,'num_Experiments_4 num_Experiments_5 nsubj_Measures_Experiments num_Measures_51 nn_Measures_Evaluation nsubj_evaluated_We rcmod_Measures_evaluated det_method_the amod_method_proposed dobj_evaluated_method partmod_method_using num_measures_four nn_measures_evaluation dobj_using_measures nn__BLEU conj_measures_ nn__NIST conj_measures_ conj_measures_WER nn_rate_word nn_rate_error dep_WER_rate cc_measures_and conj_measures_PER nn_rate_position amod_rate_independent nn_rate_word nn_rate_error dep_Measures_rate ' +o,6885,'21 Linear Models for NLP We follow the framework outlined in ',Collins,'num_Models_21 nn_Models_Linear nsubj_outlined_Models prep_Models_for pobj_for_NLP nsubj_follow_We rcmod_NLP_follow det_framework_the dobj_follow_framework prt_outlined_in ' +o,6886,'This approach has also been used by -LRB- Dagan and Itai , 1994 ; Gale et al , 1992 ; Shiitze , 1992 ; Gale et al , 1993 ; , 1995 ; Gale and Church , 1Lunar is not an unknown word in English , Yeltsin finds its translation in the 4-th candidate ',Yarowsky,'det_approach_This nsubjpass_used_approach aux_used_has advmod_used_also auxpass_used_been prep_used_by pobj_by_Dagan cc_Dagan_and conj_Dagan_Itai nsubj_finds_1994 dep_1994_Gale cc_Gale_et conj_Gale_al appos_Gale_1992 dep_Gale_Shiitze appos_Shiitze_1992 dep_Gale_Gale cc_Gale_et conj_Gale_al appos_Gale_1993 dep_Gale_ appos__1995 dep_Gale_Gale cc_Gale_and conj_Gale_Church nsubj_word_1Lunar cop_word_is neg_word_not det_word_an amod_word_unknown dep_1994_word prep_word_in pobj_in_English dep_1994_Yeltsin rcmod_Dagan_finds poss_translation_its dobj_finds_translation prep_finds_in det_candidate_the amod_candidate_4-th pobj_in_candidate ' +p,6887,'Furthermore , the BLEU score performance suggests that our model is not very powerful , but some interesting hints can be found in Table 3 when we compare our method with a 5-gram language model to a state-of-the-art system Moses based on various evaluation metrics , including BLEU score , NIST score , METEOR , TER , WER and PER ',Koehn,'advmod_suggests_Furthermore det_performance_the amod_performance_BLEU nn_performance_score nsubj_suggests_performance complm_powerful_that poss_model_our nsubj_powerful_model cop_powerful_is neg_powerful_not advmod_powerful_very ccomp_suggests_powerful cc_suggests_but det_hints_some amod_hints_interesting nsubjpass_found_hints aux_found_can auxpass_found_be conj_suggests_found prep_found_in pobj_in_Table num_Table_3 advmod_compare_when nsubj_compare_we advcl_found_compare poss_method_our dobj_compare_method prep_compare_with det_model_a amod_model_5-gram nn_model_language pobj_with_model prep_model_to det_system_a amod_system_state-of-the-art pobj_to_system nn__Moses nsubj_based_ dep_suggests_based prep_based_on amod_metrics_various nn_metrics_evaluation pobj_on_metrics prep_based_including amod_score_BLEU pobj_including_score amod__NIST nn__score conj_score_ nn__METEOR conj_score_ nn__TER conj_score_ conj_score_WER cc_score_and conj_score_PER ' +o,6888,'For phrase-based translation model training , we used the GIZA + + toolkit , and 10M bilingual sentences ',Och,'prep_used_For amod_training_phrase-based nn_training_translation nn_training_model pobj_For_training nsubj_used_we det__the nn__GIZA amod__+ cc_+_+ conj_+_toolkit dobj_used_ cc__and num_sentences_10M amod_sentences_bilingual conj__sentences ' +o,6889,'This feature , which is based on the lexical parameters of the IBM Model 1 , provides a complementary probability for each tuple in the translation table ',Brown,'det_feature_This nsubj_provides_feature nsubjpass_based_which auxpass_based_is rcmod_feature_based prep_based_on det_parameters_the amod_parameters_lexical pobj_on_parameters prep_parameters_of det__the nn__IBM nn__Model num__1 pobj_of_ det_probability_a amod_probability_complementary dobj_provides_probability prep_probability_for det_tuple_each pobj_for_tuple prep_tuple_in det_table_the nn_table_translation pobj_in_table ' +p,6890,'Disambiguation of a limited number of words is not hard , and necessary context information can be carefully collected and hand-crafted to achieve high disambiguation accuracy as shown in ',Yarowsky,'nsubj_hard_Disambiguation prep_Disambiguation_of det_number_a amod_number_limited pobj_of_number prep_number_of pobj_of_words cop_hard_is neg_hard_not cc_hard_and amod_information_necessary nn_information_context nsubjpass_collected_information aux_collected_can auxpass_collected_be advmod_collected_carefully conj_hard_collected cc_collected_and conj_collected_hand-crafted aux_achieve_to xcomp_hand-crafted_achieve amod_accuracy_high nn_accuracy_disambiguation dobj_achieve_accuracy mark_shown_as advcl_achieve_shown prep_shown_in ' +p,6891,' and Bikel and Chiang has demonstrated the applicability of the model for Czech and Chinese ',Collins,'nsubj_demonstrated_ cc__and conj__Bikel cc__and nn__Chiang conj__ aux_demonstrated_has det_applicability_the dobj_demonstrated_applicability prep_applicability_of det_model_the amod_model_ pobj_of_model prep_model_for pobj_for_Czech cc_Czech_and conj_Czech_Chinese ' +o,6892,'Some researchers have explored the use of Wikipedia information to improve the disambiguation process ',Cucerzan,'det__Some nn__researchers nsubj_explored_ aux_explored_have det_use_the dobj_explored_use prep_use_of nn_information_Wikipedia pobj_of_information aux_improve_to xcomp_explored_improve det_process_the nn_process_disambiguation dobj_improve_process ' +o,6893,'The trees may be learned directly from parallel corpora , or provided by a parser trained on hand-annotated treebanks ',Wu,'det_trees_The nsubjpass_learned_trees aux_learned_may auxpass_learned_be advmod_learned_directly prep_learned_from amod__parallel nn__corpora pobj_from_ cc_learned_or conj_learned_provided prep_provided_by det_parser_a pobj_by_parser partmod_parser_trained prep_trained_on amod__hand-annotated nn__treebanks pobj_on_ ' +o,6894,'The model is often further restricted so that each source word is assigned to exactly one target word ',Brown,'det_model_The nsubj_restricted_model cop_restricted_is advmod_restricted_often advmod_often_further advmod_assigned_so dep_assigned_that det_word_each nn_word_source nsubjpass_assigned_word auxpass_assigned_is ccomp_restricted_assigned aux_target_to advmod_target_exactly dep_exactly_one xcomp_assigned_target dobj_target_word ' +o,6895,'are the labeled parsing recall and precision , respectively , as defined in -LRB- slightly different from -RRB- ',Collins,'cop_parsing_are det_labeled_the nsubj_parsing_labeled dobj_parsing_recall cc_recall_and conj_recall_precision advmod_parsing_respectively mark_defined_as advcl_parsing_defined prep_defined_in pobj_in_ advmod_different_slightly dep__different prep_different_from pobj_from_ ' +o,6896,' and Hanks 1990 ; Smadja and McKeown 1990 -RRB- ',Church,'cc__and conj__Hanks num_Hanks_1990 dep__Smadja cc_Smadja_and conj_Smadja_McKeown num_McKeown_1990 ' +n,6897,'Since Czech is a language with relatively high degree of word-order freedom , and its sentences contain certain syntactic phenomena , such as discontinuous constituents -LRB- non-projective constructions -RRB- , which can not be straightforwardly handled using the annotation scheme of Penn Treebank , based on phrase-structure trees , we decided to adopt for the PCEDT the dependency-based annotation scheme of the Prague Dependency Treebank PDT ',Marcus,'mark_language_Since nsubj_language_Czech cop_language_is det_language_a advcl_decided_language prep_language_with advmod_high_relatively amod_degree_high pobj_with_degree prep_degree_of amod_freedom_word-order pobj_of_freedom cc_language_and poss_sentences_its nsubj_contain_sentences conj_language_contain amod_phenomena_certain amod_phenomena_syntactic dobj_contain_phenomena dep_as_such prep_phenomena_as amod_constituents_discontinuous pobj_as_constituents amod_constructions_non-projective appos_constituents_constructions nsubjpass_handled_which aux_handled_can neg_handled_not auxpass_handled_be advmod_handled_straightforwardly rcmod_constituents_handled purpcl_handled_using det_scheme_the nn_scheme_annotation dobj_using_scheme prep_scheme_of nn__Penn nn__Treebank pobj_of_ prep_handled_based dep_based_on amod_trees_phrase-structure pobj_on_trees nsubj_decided_we aux_adopt_to xcomp_decided_adopt prep_adopt_for det_PCEDT_the pobj_for_PCEDT det_scheme_the amod_scheme_dependency-based nn_scheme_annotation dobj_adopt_scheme prep_scheme_of det__the nn__Prague nn__Dependency nn__Treebank nn__PDT pobj_of_ ' +o,6898,'The sampler reasons over the infinite space of possible translation units without recourse to arbitrary restrictions -LRB- eg , constraints drawn from a wordalignment or a grammar fixed a priori -LRB- Blunsom et al , 1f and e are the input and output sentences respectively ',Zhang,'det_reasons_The nn_reasons_sampler prep_reasons_over det_space_the amod_space_infinite pobj_over_space prep_space_of amod_units_possible nn_units_translation pobj_of_units prep_units_without amod_restrictions_recourse dep_recourse_to dep_recourse_arbitrary pobj_without_restrictions dep_reasons_eg nsubj_sentences_constraints partmod_constraints_drawn prep_drawn_from det__a amod__wordalignment pobj_from_ cc__or det_grammar_a conj__grammar partmod_grammar_fixed dep_priori_a advmod_fixed_priori dobj_fixed_Blunsom cc_Blunsom_et nn_1f_al conj_Blunsom_1f cc_Blunsom_and conj_Blunsom_e cop_sentences_are det_sentences_the nn_sentences_input cc_input_and conj_input_output dep_reasons_sentences advmod_sentences_respectively ' +o,6899,'The first-sense heuristic can be thought of as striving for maximal specificity at the risk of precluding some admissible senses -LRB- reduced recall -RRB- , 7Allowing for multiple fine-grained senses to be judged as appropriate in a given context goes back at least to Sussna ; discussed more recently by , eg , ',Navigli,'det_heuristic_The amod_heuristic_first-sense nsubjpass_thought_heuristic aux_thought_can auxpass_thought_be prep_thought_of dep_of_as pcomp_as_striving prep_striving_for amod_specificity_maximal pobj_for_specificity prep_striving_at det_risk_the pobj_at_risk prep_risk_of pcomp_of_precluding det_senses_some amod_senses_admissible dobj_precluding_senses amod_recall_reduced appos_senses_recall partmod_senses_7Allowing prep_7Allowing_for amod_senses_multiple amod_senses_fine-grained pobj_for_senses aux_judged_to auxpass_judged_be xcomp_7Allowing_judged prep_judged_as amod_eg_appropriate prep_appropriate_in det_context_a amod_context_given pobj_in_context dep_context_goes advmod_goes_back dep_to_at dep_at_least dep_back_to nn__Sussna pobj_to_ partmod__discussed advmod_recently_more advmod_discussed_recently prep_discussed_by pobj_as_eg ' +p,6900,'One of the most successful metrics for judging machine-generated text is BLEU ',Papineni,'nsubj_BLEU_One prep_One_of det_metrics_the advmod_successful_most amod_metrics_successful pobj_of_metrics prep_metrics_for pcomp_for_judging amod_text_machine-generated dobj_judging_text cop_BLEU_is ' +o,6901,'5 Experimental Data The sense-tagged text and feature set used in these experiments are the same as in ',Berger,'num_Data_5 nn_Data_Experimental det_text_The amod_text_sense-tagged nsubj_are_text cc_text_and nn_set_feature conj_text_set partmod_text_used prep_used_in det_experiments_these pobj_in_experiments dep_Data_are det_same_the dep_as_same advmod_are_as prep_are_in pobj_in_ ' +p,6902,'Among them , the unsupervised algorithm using decisiontrees has achieved promising performance ',Yarowsky,'prep_achieved_Among pobj_Among_them det_algorithm_the amod_algorithm_unsupervised nsubj_achieved_algorithm partmod_algorithm_using amod__decisiontrees dobj_using_ aux_achieved_has xcomp_achieved_promising dobj_promising_performance ' +o,6903,'The information for semi-supervised sense disambiguation is usually obtained from bilingual corpora -LRB- eg parallel corpora or untagged monolingual corpora in two languages -RRB- , or sense-tagged seed examples ',Yarowsky,'det_information_The nsubjpass_obtained_information prep_information_for amod_disambiguation_semi-supervised nn_disambiguation_sense pobj_for_disambiguation auxpass_obtained_is advmod_obtained_usually prep_obtained_from amod__bilingual nn__corpora nn_corpora_eg nn_corpora_parallel dep__corpora cc_corpora_or amod_corpora_untagged nn_corpora_monolingual conj_corpora_corpora prep_corpora_in num_languages_two pobj_in_languages pobj_from_ cc__or amod_examples_sense-tagged nn_examples_seed conj__examples ' +o,6904,'Recently , specific probabilistic tree-based models have been proposed not only for machine translation , but also for This work was supported by DARPA contract F49620-001-0337 and ARDA contract MDA904-02-C-0450 ',Wu,'advmod_proposed_Recently amod_models_specific amod_models_probabilistic amod_models_tree-based nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been cc_for_not dep_not_only prep_proposed_for nn__machine nn__translation pobj_for_ dep_also_but cc_for_also conj_for_for det_work_This pobj_for_work auxpass_supported_was rcmod_work_supported prep_supported_by amod_F49620-001-0337_DARPA nn_F49620-001-0337_contract pobj_by_F49620-001-0337 cc_F49620-001-0337_and nn_MDA904-02-C-0450_ARDA nn_MDA904-02-C-0450_contract conj_F49620-001-0337_MDA904-02-C-0450 ' +p,6905,'Recent projects in semisupervised and unsupervised tagging also show significant progress ',Johnson,'amod_projects_Recent nsubj_show_projects prep_projects_in amod__semisupervised pobj_in_ cc__and amod__unsupervised conj__ partmod__tagging advmod_tagging_also amod_progress_significant dobj_show_progress ' +o,6906,'On the contrary , a string-to-tree decoder -LRB- eg , -RRB- is a parser that applies string-to-tree rules to obtain a target parse for the source string ',Shen,'prep_parser_On det_contrary_the pobj_On_contrary det_decoder_a amod_decoder_string-to-tree nsubj_parser_decoder appos_decoder_eg dep_eg_ cop_parser_is det_parser_a nsubj_applies_that rcmod_parser_applies amod_rules_string-to-tree dobj_applies_rules aux_obtain_to xcomp_applies_obtain det_parse_a nn_parse_target dobj_obtain_parse prep_obtain_for det_string_the nn_string_source pobj_for_string ' +o,6907,'Each element in vectorw gives a weight to its corresponding element in -LRB- y -RRB- , which is the count of a particular feature over the whole sentence y We calculate the vectorw value by supervised learning , using the averaged perceptron algorithm , given in Figure 1 ',Collins,'det_element_Each nsubj_gives_element prep_element_in pobj_in_vectorw det_weight_a dobj_gives_weight prep_gives_to poss_element_its amod_element_corresponding pobj_to_element prep_element_in pobj_in_y nsubj_count_which cop_count_is det_count_the rcmod_y_count prep_count_of det_feature_a amod_feature_particular pobj_of_feature prep_feature_over det_y_the amod_y_whole nn_y_sentence pobj_over_y nsubj_calculate_We dep_gives_calculate det_value_the nn_value_vectorw dobj_calculate_value prep_calculate_by amod_learning_supervised pobj_by_learning xcomp_calculate_using det__the amod__averaged nn__perceptron nn__algorithm dobj_using_ prep_calculate_given dep_given_in pobj_in_Figure num_Figure_1 ' +o,6908,' and Turney and Littman -LRB- 2002 -RRB- exploit the first two generalizations for unsupervised sentiment classification of movie reviews ',Turney,'nsubj_exploit_ cc__and conj__Turney cc_Turney_and conj_Turney_Littman appos_Turney_2002 det_generalizations_the amod_generalizations_first num_generalizations_two dobj_exploit_generalizations prep_generalizations_for amod_classification_unsupervised nn_classification_sentiment pobj_for_classification prep_classification_of nn_reviews_movie pobj_of_reviews ' +o,6909,' , who retrain the tagger and reach accuracies of 93 \% using CTB-I ',Ratnaparkhi,'nsubj_retrain_who rcmod__retrain det_tagger_the amod_tagger_ dobj_retrain_tagger cc_retrain_and conj_retrain_reach dobj_reach_accuracies prep__of num_\%_93 pobj_of_\% partmod_\%_using dobj_using_CTB-I ' +o,6910,'We chose nouns that occur a minimum of 10 times in the corpus, have no undetermined translations and at least five different translations in the six nonEnglish languages, and have the log likelihood score of at least 18; that is: LL(T T, T S ) = = 2 1 ij n* j * j*i ij n log 18 where n ij stands for the number of times T T and T S have been seen together in aligned sentences, n i* and n *j stand for the number occurrences of T T and T S, respectively, and n ** represents the total 4 We computed raw percentages only; common measures of annotator agreement such as the Kappa statistic (Carletta, 1996) proved to be inappropriate for our two-category (yesno) classification scheme.',Carletta,'' +o,6911,'It is known that ITGs do not induce the class of inside-out alignments discussed in ',Wu,'nsubjpass_known_It auxpass_known_is complm_induce_that nsubj_induce_ITGs aux_induce_do neg_induce_not ccomp_known_induce det_class_the dobj_induce_class prep_class_of amod_alignments_inside-out pobj_of_alignments partmod_alignments_discussed prep_discussed_in ' +o,6912,'1 Introduction and Previous Research It is by now commonplace knowledge that accurate syntactic parsing is not possible given only a context-free grammar with standard Penn Treebank labels -LRB- eg , S , NP , etc -RRB- ',Marcus,'dep_Introduction_1 cc_Introduction_and amod_Research_Previous conj_Introduction_Research nsubj_is_It dep_Introduction_is prep_is_by advmod_commonplace_now amod_knowledge_commonplace pobj_by_knowledge complm_possible_that amod_parsing_accurate dep_accurate_syntactic nsubj_possible_parsing cop_possible_is neg_possible_not ccomp_knowledge_possible partmod_knowledge_given advmod_grammar_only det_grammar_a amod_grammar_context-free dobj_given_grammar prep_given_with amod_labels_standard nn_labels_Penn nn_labels_Treebank nn_labels_ pobj_with_labels dep_given_eg appos_eg_S appos_eg_NP dep_eg_etc ' +o,6913,'Their weights are optimized wrt BLEU score using the algorithm described in ',Och,'poss_weights_Their nsubjpass_optimized_weights auxpass_optimized_are nn_score_wrt nn_score_BLEU dobj_optimized_score partmod_score_using det_algorithm_the dobj_using_algorithm partmod_algorithm_described prep_described_in ' +n,6914,'13 give an informal example , but do not elaborate on it ',Huang,'num__13 nsubj_give_ det_example_an amod_example_informal dobj_give_example cc_give_but aux_elaborate_do neg_elaborate_not conj_give_elaborate prep_elaborate_on pobj_on_it ' +p,6915,'1 Introduction The emergence of phrase-based statistical machine translation -LRB- PSMT -RRB- has been one of the major developments in statistical approaches to translation ',Koehn,'nsubj_been_1 partmod_1_Introduction det_emergence_The dobj_Introduction_emergence prep_emergence_of amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_PSMT advmod_Introduction_ aux_been_has dobj_been_one prep_one_of det_developments_the amod_developments_major pobj_of_developments prep_developments_in amod_approaches_statistical pobj_in_approaches prep_been_to pobj_to_translation ' +o,6916,' describe how to learn hundreds of millions of treetransformation rules from a parsed , aligned Chinese\\/English corpus , and et al ',Galley,'nsubj_describe_ advmod_learn_how aux_learn_to ccomp_describe_learn dep_millions_hundreds quantmod_millions_of dobj_learn_millions prep_millions_of nn_rules_treetransformation pobj_of_rules prep_rules_from det_corpus_a amod_corpus_parsed amod_corpus_aligned nn_corpus_Chinese\\/English pobj_from_corpus cc_millions_and conj_millions_ cc__et conj__al ' +o,6917,' found that such smoothing during training gives almost identical results on translation metrics ',Och,'nsubj_found_ complm_gives_that amod_smoothing_such nsubj_gives_smoothing prep_smoothing_during pobj_during_training ccomp_found_gives advmod_identical_almost amod_results_identical dobj_gives_results prep_results_on nn_metrics_translation pobj_on_metrics ' +o,6918,'One is how to learn a statistical model to estimate the conditional probability , and the other is how to generate confusion set C of a given query q 4.1 Maximum Entropy Model for Query Spelling Correction We take a feature-based approach to model the posterior probability . Specifically we use the maximum entropy model (Berger et al. , 1996) for this task: = exp , =1 exp( (, ) =1 ) (2) where exp( (, ) =1 ) is the normalization factor; , is a feature function defined over query q and correction candidate c, while is the corresponding feature weight.',Berger,'' +o,6919,'We can mentionhere only part of this work : for monolingualextraction , and -LRB- Kupiec , 1993 ; Wu ,1994 ; Smadjaetal ',Smadja,'nsubj_mentionhere_We aux_mentionhere_can advmod_part_only dobj_mentionhere_part prep_part_of det_work_this pobj_of_work dep_part_ prep__for pobj_for_monolingualextraction cc_part_and nn_1993_Kupiec conj_part_1993 dep_part_Wu num_Wu_,1994 dep_part_Smadjaetal ' +o,6920,'Previous workonsentimentanalysishascoveredawiderange of tasks , including polarity classification , opinion extraction , and opinion source assignment ',Pang,'amod_workonsentimentanalysishascoveredawiderange_Previous prep_workonsentimentanalysishascoveredawiderange_of pobj_of_tasks prep_tasks_including amod__polarity nn__classification pobj_including_ nn__opinion nn__extraction conj__ cc__and nn__opinion nn__source nn__assignment conj__ ' +o,6921,'The second one is heuristic and tries to use a wordaligned corpus ',Koehn,'det_one_The amod_one_second nsubj_heuristic_one cop_heuristic_is cc_heuristic_and conj_heuristic_tries aux_use_to xcomp_tries_use det_corpus_a amod_corpus_wordaligned dobj_use_corpus ' +o,6922,'Statistical data about these various cooccurrence relations is employed for a variety of applications , such as speech recognition , language generation , lexicography , machine translation , information retrieval and various disambiguation tasks ',Church,'nn_data_Statistical nsubjpass_employed_data prep_data_about det_relations_these amod_relations_various amod_relations_cooccurrence pobj_about_relations auxpass_employed_is prep_employed_for det_variety_a pobj_for_variety prep_variety_of pobj_of_applications dep_as_such prep_variety_as nn__speech nn__recognition pobj_as_ nn__language nn__generation conj__ amod__lexicography conj__ nn__machine nn__translation conj__ nn__information nn__retrieval conj__ cc__and amod_tasks_various nn_tasks_disambiguation conj__tasks advmod_employed_ ' +o,6923,'Collins and Koo introduced an improved reranking model for parsing which includes a hidden layer of semantic features ',Koo,'nn__Collins cc_Collins_and conj_Collins_Koo nsubj_introduced_ det_model_an amod_model_improved nn_model_reranking dobj_introduced_model prep_model_for pobj_for_parsing nsubj_includes_which rcmod_model_includes det_layer_a amod_layer_hidden dobj_includes_layer prep_layer_of amod_features_semantic pobj_of_features ' +o,6924,'1 Introduction Many different statistical tests have been proposed to measure the strength of word similarity or word association in natural language texts ',Church,'num_Introduction_1 amod_tests_Many amod_tests_different amod_tests_statistical nsubjpass_proposed_tests aux_proposed_have auxpass_proposed_been dep_Introduction_proposed aux_measure_to xcomp_proposed_measure det_strength_the dobj_measure_strength prep_strength_of nn_similarity_word pobj_of_similarity cc_similarity_or nn_association_word conj_similarity_association prep_similarity_in amod_texts_natural nn_texts_language pobj_in_texts ' +o,6925,'As in most other statistical parsing systems we therefore use the pruning technique described in Goodman and which assigns a score to each item in the chart equal to the product of the inside probability of the item and its prior probability ',Collins,'prep_use_As dep_As_in advmod_other_most amod_systems_other amod_systems_statistical nn_systems_parsing pobj_in_systems nsubj_use_we advmod_use_therefore det_technique_the nn_technique_pruning dobj_use_technique partmod_technique_described prep_described_in nn__Goodman pobj_in_ cc__and conj__ nsubj_assigns_which rcmod_technique_assigns det_score_a dobj_assigns_score prep_assigns_to det_item_each pobj_to_item prep_item_in det_chart_the pobj_in_chart amod_chart_equal prep_equal_to det_product_the pobj_to_product prep_product_of det_probability_the amod_probability_inside pobj_of_probability prep_probability_of det_item_the pobj_of_item cc_item_and poss_probability_its amod_probability_prior conj_item_probability ' +o,6926,'Some work has been done on adding new terms and relations to WordNet and FACTOTUM ',Snow,'det_work_Some nsubjpass_done_work aux_done_has auxpass_done_been prep_done_on pcomp_on_adding amod_terms_new dobj_adding_terms cc_terms_and conj_terms_relations prep_adding_to nn__WordNet pobj_to_ cc__and nn__FACTOTUM conj__ ' +o,6927,'The computation mechanism of GP and LP bears a resemblance to the EM algorithm , which iteratively computes maximum likelihood estimates from incomplete data ',Brown,'det_mechanism_The nn_mechanism_computation nsubj_bears_mechanism prep_mechanism_of pobj_of_GP cc_GP_and conj_GP_LP det_resemblance_a dobj_bears_resemblance prep_bears_to det__the nn__EM nn__algorithm pobj_to_ nsubj_computes_which advmod_computes_iteratively rcmod__computes amod_estimates_maximum nn_estimates_likelihood dobj_computes_estimates prep_computes_from amod_data_incomplete pobj_from_data ' +p,6928,'Their idea has proven effective for estimating the statistics of unknown words in previous studies ',Ratnaparkhi,'poss_idea_Their nsubj_proven_idea aux_proven_has acomp_proven_effective prep_effective_for pcomp_for_estimating det_statistics_the dobj_estimating_statistics prep_statistics_of amod_words_unknown pobj_of_words prep_words_in amod_studies_previous pobj_in_studies ' +o,6929,'Using BLEU as a metric , our method achieves an absolute improvement of 006 -LRB- 2213 \% relative -RRB- as compared with the standard model trained with 5,000 L f - L e sentence pairs for French-Spanish translation ',Papineni,'dep_achieves_Using nn__BLEU dobj_Using_ prep__as det_metric_a pobj_as_metric poss_method_our nsubj_achieves_method det_improvement_an amod_improvement_absolute dobj_achieves_improvement prep_improvement_of pobj_of_006 num_\%_2213 measure_relative_\% dep_006_relative prep_achieves_as prep_achieves_compared dep_compared_with det_model_the amod_model_standard pobj_with_model partmod_model_trained prep_trained_with num_L_5,000 pobj_with_L dep_L_f nn_pairs_L nn_pairs_e nn_pairs_sentence dep_model_pairs prep_pairs_for amod_translation_French-Spanish pobj_for_translation ' +o,6930,'Due to its popularity for unsupervised POS induction research and its often-used tagset , for our initial research , we use the Wall Street Journal -LRB- WSJ -RRB- portion of the Penn Treebank , with 36 tags -LRB- plus 9 punctuation tags -RRB- , and we use sections 00-18 , leaving held-out data for future experiments4 Defining frequent frames as those occurring at 4Even if we wanted child-directed speech , the CHILDES database uses coarse POS tags ',Marcus,'dep_to_Due prep_use_to poss_popularity_its pobj_to_popularity prep_popularity_for amod__unsupervised dep__POS nn__induction nn__research pobj_for_ cc_popularity_and poss_tagset_its amod_tagset_often-used conj_popularity_tagset prep_tagset_for poss_research_our amod_research_initial pobj_for_research nsubj_use_we det_portion_the nn_Journal_Wall nn_Journal_Street nn_portion_Journal abbrev_Journal_WSJ dobj_use_portion prep_portion_of det__the nn__Penn nn__Treebank pobj_of_ prep_portion_with num_tags_36 pobj_with_tags dep_tags_plus num_tags_9 nn_tags_punctuation dep_plus_tags cc_use_and nsubj_use_we conj_use_use dobj_use_sections tmod_use_00-18 xcomp_use_leaving amod_data_held-out dobj_leaving_data prep_leaving_for amod_experiments4_future pobj_for_experiments4 partmod_experiments4_Defining amod_frames_frequent dobj_Defining_frames prep_Defining_as pobj_as_those partmod_those_occurring prep_occurring_at pobj_at_4Even mark_wanted_if nsubj_wanted_we advcl_uses_wanted amod_speech_child-directed dobj_wanted_speech det__the nn__CHILDES nn__database nsubj_uses_ dep_use_uses amod_tags_coarse dep_tags_POS dobj_uses_tags ' +o,6931,'We also tested the flat syntactic feature set proposed in s work ',Luo,'nsubj_tested_We advmod_tested_also det_set_the amod_set_flat amod_set_syntactic nn_set_feature dobj_tested_set partmod_set_proposed prep_proposed_in amod_work_ nn_work_s pobj_in_work ' +o,6932,'BLEU , NIST ',Papineni,'nn__BLEU nn__NIST appos__ ' +n,6933,'Like WASP1 , the phrase extraction algorithm of PHARAOH is based on the output of a word alignment model such as GIZA + + , which performs poorly when applied directly to MRLs -LRB- Section 32 -RRB- ',Koehn,'prep_based_Like pobj_Like_WASP1 det_algorithm_the nn_algorithm_phrase nn_algorithm_extraction nsubjpass_based_algorithm prep_algorithm_of pobj_of_PHARAOH auxpass_based_is prep_based_on det_output_the pobj_on_output prep_output_of det_model_a nn_model_word nn_model_alignment pobj_of_model dep_as_such prep_model_as pobj_as_GIZA amod_GIZA_+ cc_+_+ conj_+_ nsubj_performs_which rcmod_GIZA_performs advmod_performs_poorly advmod_applied_when advcl_performs_applied advmod_applied_directly prep_applied_to pobj_to_MRLs dep_applied_Section num_Section_32 ' +p,6934,'We use the popular online learning algorithm of structured perceptron with parameter averaging ',Collins,'nsubj_use_We det_algorithm_the amod_algorithm_popular amod_algorithm_online nn_algorithm_learning dobj_use_algorithm prep_algorithm_of amod_perceptron_structured pobj_of_perceptron prep_use_with nn_averaging_parameter pobj_with_averaging ' +o,6935,'However , as Categorial Grammar formalisms do not usually change the lexical entries of words to deal with movement , but use further rules , the lexicons learned here will be valid over corpora with movement ',Marcus,'advmod_learned_However mark_change_as nn_formalisms_Categorial nn_formalisms_Grammar nsubj_change_formalisms aux_change_do neg_change_not advmod_change_usually dep_learned_change det_entries_the amod_entries_lexical dobj_change_entries prep_entries_of pobj_of_words aux_deal_to xcomp_change_deal prep_deal_with pobj_with_movement cc_change_but conj_change_use amod_rules_further dobj_use_rules advmod_use_ det_lexicons_the nsubj_learned_lexicons nsubj_valid_here aux_valid_will cop_valid_be ccomp_learned_valid prep_valid_over pobj_over_corpora prep_valid_with pobj_with_movement ' +o,6936,'For this we aligned 170,863 pairs of Arabic\\/English newswire sentences from LDC , trained a state-of-the-art syntax-based statistical machine translation system on these sentences and alignments , and measured BLEU scores on a separate set of 1298 newswire test sentences ',Papineni,'dep_we_this pobj_For_we dep_we_aligned num_pairs_170,863 dobj_aligned_pairs prep_pairs_of nn_sentences_Arabic\\/English nn_sentences_newswire pobj_of_sentences prep_aligned_from pobj_from_LDC conj_aligned_trained det__a amod__state-of-the-art amod__syntax-based amod__statistical nn__machine nn__translation nn__system dobj_trained_ prep__on det_sentences_these pobj_on_sentences cc_sentences_and conj_sentences_alignments cc_aligned_and conj_aligned_measured num_scores_BLEU measure__scores acomp_measured_ prep_measured_on det_set_a amod_set_separate pobj_on_set prep_set_of num_sentences_1298 nn_sentences_newswire nn_sentences_test pobj_of_sentences ' +o,6937,'The left-to-right parser would likely improve if we were to use a left-corner transform ',Collins,'det_parser_The amod_parser_left-to-right nsubj_improve_parser aux_improve_would advmod_improve_likely mark_were_if nsubj_were_we advcl_improve_were aux_use_to xcomp_were_use det_left-corner_a nsubj_transform_left-corner ccomp_use_transform ' +o,6938,'Goldwater and Griffiths evaluated against the reduced tag set of 17 tags developed by Smith and Eisner , while evaluated against the full Penn Treebank tag set ',Johnson,'nn__Goldwater cc_Goldwater_and conj_Goldwater_Griffiths nsubj_evaluated_ prep_evaluated_against det_set_the amod_set_reduced nn_set_tag pobj_against_set prep_set_of num_tags_17 pobj_of_tags partmod_tags_developed prep_developed_by nn__Smith cc_Smith_and conj_Smith_Eisner pobj_by_ mark_evaluated_while nn__ nsubj_evaluated_ advcl_developed_evaluated prep_evaluated_against det_tag_the amod_tag_full nn_tag_Penn nn_tag_Treebank pobj_against_tag partmod_tag_set ' +p,6939,'This algorithm is referred to as GHKM and is widely used in SSMT systems ',Galley,'det_algorithm_This nsubjpass_referred_algorithm auxpass_referred_is prep_referred_to quantmod__as number__GHKM pobj_to_ cc_referred_and auxpass_used_is advmod_used_widely conj_referred_used prep_used_in amod_systems_SSMT pobj_in_systems ' +o,6940,'The unknown word tokens are with respect to Training I. Data set Sect\'ns Token Unknown Training I 26-270, 600-931 213986 Training II 600-931, 500-527, 1001-1039 204701 Training III 001-270, 301-527, 590-593, 600-1039, 1043-1151 485321 Devset 23839 2849 XH 001-025 7844 381 HKSAR 500-527 8202 1168 SM 590-593, 1001-1002 7793 1300 Test set 23522 2957 XH 271-300 8008 358 HKSAR 528-554 7153 1020 SM 594-596, 1040-1042 8361 1579 5.2 The model Our model builds on research into loglinear models by Ng and Low (2004), Toutanova et al. , (2003) and Ratnaparkhi (1996).',Ratnaparkhi,'' +o,6941,'1 Introduction A ` pain in the neck \' for NLP in languages of the Indo-Aryan family -LRB- eg Hindi-Urdu , Bangla and Kashmiri -RRB- is the fact that most verbs -LRB- nearly half of all instances in Hindi -RRB- occur as complex predicates multi-word complexes which function as a single verbal unit in terms of argument and event structure ',Marcus,'num_Introduction_1 det_pain_A nsubj_fact_pain prep_pain_in det_neck_the poss__neck pobj_in_ prep__for pobj_for_NLP prep_NLP_in pobj_in_languages prep_languages_of det_family_the amod_family_Indo-Aryan pobj_of_family nn_Hindi-Urdu_eg appos_family_Hindi-Urdu dep_Hindi-Urdu_Bangla cc_Bangla_and conj_Bangla_Kashmiri cop_fact_is det_fact_the dep_Introduction_fact complm_occur_that amod_verbs_most nsubj_occur_verbs quantmod_half_nearly dep_verbs_half prep_half_of det_instances_all pobj_of_instances prep_instances_in pobj_in_Hindi ccomp_fact_occur prep_occur_as amod_predicates_complex pobj_as_predicates amod_complexes_multi-word dobj_occur_complexes nsubj_function_which rcmod_complexes_function prep_function_as det_unit_a amod_unit_single amod_unit_verbal pobj_as_unit prep_unit_in pobj_in_terms prep_terms_of nn__argument cc_argument_and conj_argument_event nn__structure pobj_of_ ' +o,6942,'It generates a vector of 5 numeric values for each phrase pair: phrase translation probability: ( f|e) = count( f, e) count(e),(e| f) = count( f, e) count( f) 2http://www.phramer.org/ Java-based open-source phrase based SMT system 3http://www.isi.edu/licensed-sw/carmel/ 4http://www.speech.sri.com/projects/srilm/ 5http://www.iccs.inf.ed.ac.uk/pkoehn/training.tgz 150 lexical weighting (Koehn et al. , 2003): lex( f|e,a) = nproductdisplay i=1 1 |{j|(i, j) a}| summationdisplay (i,j)a w(fi|ej) lex(e|f,a) = mproductdisplay j=1 1 |{i|(i, j) a}| summationdisplay (i,j)a w(ej|fi) phrase penalty: ( f|e) = e; log(( f|e)) = 1 2.2 Decoding We used the Pharaoh decoder for both the Minimum Error Rate Training (Och, 2003) and test dataset decoding.',Och,'' +o,6943,'We can confirm that changing the dimensionality parameter h has rather little effect -LRB- Table 4 -RRB- , which is in line with previous findings ',Blitzer,'nsubj_confirm_We aux_confirm_can complm_has_that csubj_has_changing det_h_the amod_h_dimensionality nn_h_parameter dobj_changing_h ccomp_confirm_has advmod_little_rather amod_effect_little dobj_has_effect number_4_Table dep_effect_4 nsubj_is_which rcmod_effect_is prep_is_in pobj_in_line prep_line_with amod_findings_previous pobj_with_findings ' +o,6944,'This includes both the parsers that attach probabilities to parser moves , but also those of the lexicalized PCFG variety ',Collins,'nsubj_includes_This predet_parsers_both det_parsers_the dobj_includes_parsers nsubj_attach_that rcmod_parsers_attach dobj_attach_probabilities prep_attach_to nn__parser nn__moves pobj_to_ dep_also_but cc_parsers_also conj_parsers_those prep_those_of det_variety_the amod_variety_lexicalized nn_variety_PCFG pobj_of_variety ' +o,6945,'We also test our language model using leave-one-out cross-validation on the Penn Treebank -LRB- WSJ -RRB- , giving us 8674 \% accuracy -LRB- see Table 1 -RRB- ',Marcus,'nsubj_test_We advmod_test_also poss_model_our nn_model_language dobj_test_model dep_model_using amod_cross-validation_leave-one-out dobj_using_cross-validation prep_using_on det__the nn__Penn nn__Treebank pobj_on_ abbrev__WSJ xcomp_using_giving iobj_giving_us num_\%_8674 dobj_giving_\% dobj_giving_accuracy parataxis_using_see dobj_see_Table num_Table_1 ' +o,6946,'Classes can be induced directly from the corpus or taken from a manually crafted taxonomy ',Brown,'nsubjpass_induced_Classes aux_induced_can auxpass_induced_be advmod_induced_directly prep_induced_from det__the nn__corpus pobj_from_ cc_induced_or conj_induced_taken prep_taken_from det_manually_a pobj_from_manually partmod_manually_crafted advmod__taxonomy dep_crafted_ ' +o,6947,'32 System Combination Scheme In our work , we use a sentence-level system combination model to select best translation hypothesis from the candidate pool -LRB- -RRB- This method can also be viewed to be a hypotheses reranking model since we only use the existing translations instead of performing decoding over a confusion network as done in the word-level combination method ',Rosti,'num_Scheme_32 nn_Scheme_System nn_Scheme_Combination prep_use_In poss_work_our pobj_In_work nsubj_use_we dep_Scheme_use det_model_a amod_model_sentence-level nn_model_system nn_model_combination nsubj_select_model aux_select_to xcomp_use_select prt_select_best nn_hypothesis_translation dobj_select_hypothesis prep_select_from det_-LRB-_the nn_-LRB-_candidate nn_-LRB-_pool det_method_This nsubjpass_viewed_method aux_viewed_can advmod_viewed_also auxpass_viewed_be dep_Scheme_viewed aux_model_to cop_model_be det_model_a amod_model_hypotheses amod_model_reranking purpcl_viewed_model mark_use_since nsubj_use_we advmod_use_only advcl_model_use det_translations_the amod_translations_existing dobj_use_translations dep_of_instead prep_use_of pcomp_of_performing xcomp_performing_decoding prt_decoding_over det_network_a nn_network_confusion dobj_decoding_network advmod_done_as amod_network_done prep_decoding_in det_method_the amod_method_word-level nn_method_combination pobj_in_method ' +o,6948,'Methods have been proposed for automatic evaluation in MT -LRB- eg , BLEU -RRB- ',Papineni,'nsubjpass_proposed_Methods aux_proposed_have auxpass_proposed_been prep_proposed_for amod_evaluation_automatic pobj_for_evaluation prep_evaluation_in pobj_in_MT appos_MT_eg nn__BLEU dep_eg_ ' +o,6949,'Some work identifies inflammatory texts -LRB- eg , -RRB- or classifies reviews as positive or negative -LRB- -RRB- ',Pang,'det_work_Some nsubj_identifies_work amod_texts_inflammatory dobj_identifies_texts appos_texts_eg dep_eg_ cc_identifies_or conj_identifies_classifies dobj_classifies_reviews advmod_positive_as amod_reviews_positive cc_positive_or conj_positive_negative appos_reviews_ ' +o,6950,' put forward and discussed n-gram models based on classes of words ',Brown,'nsubj_put_ advmod_put_forward cc_put_and conj_put_discussed amod_models_n-gram dobj_discussed_models prep_discussed_based dep_based_on pobj_on_classes prep_classes_of pobj_of_words ' +o,6951,' develop a bottom-up decoder for BTG that uses only phrase pairs ',Wu,'advmod_develop_ det_decoder_a amod_decoder_bottom-up dobj_develop_decoder prep_decoder_for nn__BTG pobj_for_ nsubj_uses_that rcmod__uses amod_pairs_only nn_pairs_phrase dobj_uses_pairs ' +o,6952,'We use maximum marginal decoding , which reports performs better than Viterbi decoding ',Johnson,'nsubj_use_We amod_decoding_maximum amod_decoding_marginal dobj_use_decoding rel_performs_which amod_reports_ nsubj_performs_reports rcmod_decoding_performs advmod_performs_better prep_performs_than nn_decoding_Viterbi pobj_than_decoding ' +p,6953,'Stochastic models have been widely used in POS tagging for simplicity and language independence of the models ',Cutting,'amod__Stochastic nn__models nsubjpass_used_ aux_used_have auxpass_used_been advmod_used_widely prep_used_in pobj_in_POS xcomp_used_tagging prep_tagging_for nn_independence_simplicity cc_simplicity_and conj_simplicity_language pobj_for_independence prep_independence_of det_models_the pobj_of_models ' +o,6954,'On the machine-learning side , it would be interesting to generalize the ideas of large-margin classi cation to sequence models , strengthening the results of and leading to new optimal training algorithms with stronger guarantees against over tting ',Collins,'prep_interesting_On det_side_the amod_side_machine-learning pobj_On_side nsubj_interesting_it aux_interesting_would cop_interesting_be aux_generalize_to xcomp_interesting_generalize det_ideas_the dobj_generalize_ideas prep_ideas_of amod_cation_large-margin nn_cation_classi pobj_of_cation prep_generalize_to nn_models_sequence pobj_to_models dep_generalize_strengthening det_results_the dobj_strengthening_results prep_results_of pobj_of_ cc_strengthening_and conj_strengthening_leading prep_leading_to amod_algorithms_new amod_algorithms_optimal nn_algorithms_training pobj_to_algorithms prep_algorithms_with amod_guarantees_stronger pobj_with_guarantees dep_over_against prep_leading_over pobj_over_tting ' +o,6955,'Therefore , domain adaptation methods have recently been proposed in several NLP areas , eg , word sense disambiguation , statistical parsing , and lexicalized-grammar parsing ',Blitzer,'advmod_proposed_Therefore nn_methods_domain nn_methods_adaptation nsubjpass_proposed_methods aux_proposed_have advmod_proposed_recently auxpass_proposed_been prep_proposed_in amod_areas_several nn_areas_NLP pobj_in_areas nn__eg dep__word nn__sense nn__disambiguation appos_areas_ amod__statistical amod__parsing conj__ cc__and amod__lexicalized-grammar amod__parsing conj__ ' +o,6956,'Baseline Pharaoh with phrases extracted from IBM Model 4 training with maximum phrase length 7 and extraction method diag-growthfinal Lex Phrase-decoder simulation : using only the initial lexical rules from the phrase table , all with LHS X , the Glue rule , and a binary reordering rule with its own reordering-feature XCat All nonterminals merged into a single X nonterminal : simulation of the system Hiero ',Koehn,'nn_Pharaoh_Baseline prep_Pharaoh_with pobj_with_phrases partmod_phrases_extracted prep_extracted_from nn_Model_IBM pobj_from_Model num_Model_4 dep_Pharaoh_training prep_training_with amod_length_maximum nn_length_phrase pobj_with_length num_length_7 cc_length_and nn_simulation_extraction nn_simulation_method nn_simulation_diag-growthfinal num_simulation_ nn_simulation_Lex nn_simulation_Phrase-decoder conj_length_simulation parataxis_training_using advmod_rules_only det_rules_the amod_rules_initial amod_rules_lexical dobj_using_rules prep_rules_from det_table_the nn_table_phrase pobj_from_table advmod_merged_all dep_all_with nn_X_LHS pobj_with_X det_rule_the nn_rule_Glue conj_X_rule cc_X_and det_rule_a amod_rule_binary nn_rule_reordering conj_X_rule prep_rule_with poss_nonterminals_its amod_nonterminals_own amod_nonterminals_reordering-feature nn_nonterminals_XCat nn_nonterminals_All pobj_with_nonterminals ccomp_training_merged prep_merged_into det_nonterminal_a amod_nonterminal_single nn_nonterminal_X pobj_into_nonterminal parataxis_training_simulation prep_simulation_of det__the nn__system nn__Hiero pobj_of_ ' +o,6957,'We believe the benefit to limiting the size of n is connected to observation that as n increases , the accuracy of an n-gram model increases , but the reliability of our parameter estimates , drawn as they must be from a limited training text , decreases ',Brown,'nsubj_believe_We det_benefit_the nsubjpass_connected_benefit prep_benefit_to pcomp_to_limiting det_size_the dobj_limiting_size prep_size_of pobj_of_n auxpass_connected_is ccomp_believe_connected prep_connected_to amod_observation_ pobj_to_observation complm_increases_that prep_increases_as nn_increases_n pobj_as_increases det_accuracy_the nsubj_increases_accuracy prep_accuracy_of det_model_an amod_model_n-gram pobj_of_model ccomp_connected_increases cc_increases_but det_reliability_the nsubj_decreases_reliability prep_reliability_of poss_estimates_our nn_estimates_parameter pobj_of_estimates partmod_reliability_drawn mark_be_as nsubj_be_they aux_be_must advcl_drawn_be prep_be_from det_text_a amod_text_limited nn_text_training pobj_from_text conj_increases_decreases ' +o,6958,'Several artificial techniques have been used so that classifiers can be developed and tested without having to invest in manually tagging the data : and Sch\\/itze -LRB- 1995 -RRB- have acquired training and testing materials by creating pseudowords from existing nonhomographic forms ',Yarowsky,'amod_techniques_Several amod_techniques_artificial nsubjpass_used_techniques aux_used_have auxpass_used_been advmod_developed_so dep_developed_that nsubjpass_developed_classifiers aux_developed_can auxpass_developed_be ccomp_used_developed cc_developed_and conj_developed_tested prep_developed_without pcomp_without_having aux_invest_to xcomp_having_invest prep_invest_in advmod_tagging_manually pcomp_in_tagging det_data_the dobj_tagging_data nsubj_acquired_ cc__and conj__Sch\\/itze appos__1995 aux_acquired_have parataxis_used_acquired nn_materials_training cc_training_and conj_training_testing dobj_acquired_materials prep_acquired_by pcomp_by_creating dobj_creating_pseudowords prep_creating_from amod_forms_existing amod_forms_nonhomographic pobj_from_forms ' +o,6959,' employ Multiple Sequence Alignment to align strings extracted from closely related news articles ',Barzilay,'nsubj_employ_ amod__Multiple nn__Sequence nn__Alignment dobj_employ_ prep_employ_to nn_strings_align pobj_to_strings partmod_strings_extracted prep_extracted_from advmod_related_closely amod_articles_related nn_articles_news pobj_from_articles ' +o,6960,'In contrast , the latter computes four definite probabilities which are included as features within a machine-learning classifier from the Web in an attempt to overcome data sparseness problem ',Bean,'prep_computes_In pobj_In_contrast det_latter_the nsubj_computes_latter num_probabilities_four amod_probabilities_definite dobj_computes_probabilities nsubjpass_included_which auxpass_included_are rcmod_probabilities_included prep_included_as pobj_as_features prep_features_within det_classifier_a amod_classifier_machine-learning pobj_within_classifier prep_included_from det_problem_the amod_problem_Web prep_Web_in det_attempt_an pobj_in_attempt aux_overcome_to infmod_attempt_overcome amod_data_ dobj_overcome_data nn_problem_sparseness pobj_from_problem ' +o,6961,'216 The Maximum Entropy Principle is to nd a model p = argmax pC H -LRB- p -RRB- , which means a probability model p -LRB- y x -RRB- that maximizes entropy H -LRB- p -RRB- ',Berger,'det__The nn__Maximum nn__Entropy nn__Principle nsubj_is_ ccomp_216_is aux_nd_to xcomp_is_nd det_p_a nn_p_model nsubj_pC_p dep_pC_= nn_pC_argmax ccomp_nd_pC nsubj_216_H appos_H_p nsubj_means_which rcmod_H_means det_p_a nn_p_probability nn_p_model dobj_means_p number_x_y dep_p_x nsubj_maximizes_that rcmod_p_maximizes amod_H_entropy dobj_maximizes_H appos_H_p ' +o,6962,'A similar use of the term phrase exists in machine translation , where phrases are often pairs of word sequences consistent with word-based alignments ',Koehn,'det_use_A amod_use_similar nsubj_exists_use prep_use_of det_phrase_the nn_phrase_term pobj_of_phrase prep_exists_in nn_translation_machine pobj_in_translation advmod_consistent_where nsubj_consistent_phrases cop_consistent_are advmod_consistent_often dep_often_pairs prep_pairs_of nn_sequences_word pobj_of_sequences rcmod_translation_consistent prep_consistent_with amod_alignments_word-based pobj_with_alignments ' +o,6963,'Accurate measurement of semantic similarity between lexical units such as words or phrases is important for numerous tasks in natural language processing such as word sense disambiguation , synonym extraction , and automatic thesauri generation ',Lin,'amod_measurement_Accurate nsubj_important_measurement prep_measurement_of amod_similarity_semantic pobj_of_similarity prep_similarity_between amod_units_lexical pobj_between_units dep_as_such prep_units_as pobj_as_words cc_words_or conj_words_phrases cop_important_is prep_important_for amod_tasks_numerous pobj_for_tasks prep_tasks_in amod_processing_natural nn_processing_language pobj_in_processing dep_as_such prep_processing_as nn__word nn__sense nn__disambiguation pobj_as_ nn__synonym nn__extraction conj__ cc__and amod__automatic nn__thesauri nn__generation conj__ ' +o,6964,' -RRB- , better language-specific preprocessing and restructuring , additional feature functions such as word class language models , and minimum error rate training to optimize parameters ',Och,'amod__better amod__language-specific nn__preprocessing dep_models_ cc__and nn__restructuring conj__ amod_functions_additional nn_functions_feature appos__functions dep_as_such prep_functions_as nn_class_word pobj_as_class nn_models_language dobj__models cc_models_and amod__minimum nn__error nn__rate nn__training conj_models_ prep__to amod_parameters_optimize pobj_to_parameters ' +o,6965,'Most current transliteration systems use a generative model for transliteration such as freely available GIZA + +1 , an implementation of the IBM alignment models ',Brown,'advmod_systems_Most amod_systems_current nn_systems_transliteration nsubj_use_systems det_model_a amod_model_generative dobj_use_model prep_model_for pobj_for_transliteration dep_as_such prep_transliteration_as advmod_available_freely amod__available nn__GIZA nn__+ num__+1 pobj_as_ det_implementation_an appos_transliteration_implementation prep_implementation_of det_models_the nn_models_IBM nn_models_alignment pobj_of_models ' +o,6966,'In addition to portability experiments with the parsing model of , provided a comprehensive analysis of parser portability ',Collins,'prep_provided_In dep_In_addition dep_In_to amod_experiments_portability pobj_In_experiments prep_experiments_with det_model_the amod_model_parsing pobj_with_model prep_model_of pobj_of_ nsubj_provided_ det_analysis_a amod_analysis_comprehensive dobj_provided_analysis prep_analysis_of nn_portability_parser pobj_of_portability ' +n,6967,'This restriction is necessary because the problem of optimizing many-to-many alignments 5 Our preliminary experiments with n-gram-based overlap measures , such as BLEU and ROUGE , show that these metrics do not correlate with human judgments on the fusion task , when tested against two reference outputs ',Papineni,'det_restriction_This nsubj_necessary_restriction cop_necessary_is mark_alignments_because det_problem_the nsubj_alignments_problem prep_problem_of amod_many-to-many_optimizing pobj_of_many-to-many advcl_necessary_alignments dobj_alignments_5 poss_experiments_Our amod_experiments_preliminary nsubj_show_experiments prep_experiments_with amod_measures_n-gram-based amod_measures_overlap pobj_with_measures dep_as_such prep_experiments_as nn__BLEU pobj_as_ cc__and nn__ROUGE conj__ dep_necessary_show complm_correlate_that det_metrics_these nsubj_correlate_metrics aux_correlate_do neg_correlate_not ccomp_show_correlate prep_correlate_with amod_judgments_human pobj_with_judgments prep_correlate_on det_task_the nn_task_fusion pobj_on_task advmod_tested_when advcl_correlate_tested prep_tested_against num_outputs_two nn_outputs_reference pobj_against_outputs ' +o,6968,'Since parsing is just an initial stage of natural language understanding , the project was focused not just on obtaining syntactic trees alone -LRB- as is done in many other parsed corpora , for example , Penn TreeBank or Tiger -RRB- ',Marcus,'mark_stage_Since nsubj_stage_parsing cop_stage_is advmod_stage_just det_stage_an amod_stage_initial advcl_focused_stage prep_stage_of amod_understanding_natural nn_understanding_language pobj_of_understanding det_project_the nsubjpass_focused_project auxpass_focused_was dep_on_not dep_on_just prep_focused_on pcomp_on_obtaining amod_trees_syntactic dobj_obtaining_trees advmod_obtaining_alone mark_done_as auxpass_done_is advcl_obtaining_done prep_done_in amod_corpora_many amod_corpora_other amod_corpora_parsed pobj_in_corpora prep_done_for pobj_for_example nn__Penn nn__TreeBank conj_example_ cc_example_or nn__Tiger conj_example_ ' +o,6969,'There exists a variety of different metrics , eg , word error rate , position-independent word error rate , BLEU score , NIST score , METEOR , GTM ',Papineni,'expl_exists_There det_variety_a dobj_exists_variety prep_variety_of amod_metrics_different pobj_of_metrics appos_metrics_eg nn_rate_word nn_rate_error appos_metrics_rate amod_rate_position-independent nn_rate_word nn_rate_error appos_metrics_rate amod__BLEU nn__score appos_metrics_ amod__NIST nn__score appos_metrics_ nn__METEOR appos_metrics_ nn__GTM appos_metrics_ ' +n,6970,'Although the authors of stated that they would discuss the search problem in a follow-up arti cle , so far there have no publications devoted to the decoding issue for statistical machine translation ',Brown,'mark_stated_Although det_authors_the nsubj_stated_authors prep_authors_of pobj_of_ complm_discuss_that nsubj_discuss_they aux_discuss_would ccomp_stated_discuss det_problem_the nn_problem_search dobj_discuss_problem prep_discuss_in det_cle_a amod_cle_follow-up nn_cle_arti pobj_in_cle advmod_far_so advmod_discuss_far expl_have_there dep_far_have det_publications_no dobj_have_publications partmod_publications_devoted prep_devoted_to det_issue_the amod_issue_decoding pobj_to_issue prep_issue_for amod_translation_statistical nn_translation_machine pobj_for_translation ' +o,6971,'In this work , model fit is reported in terms of the likelihood ratio statistic , G 2 , and its significance ',Dunning,'prep_reported_In det_work_this pobj_In_work nn_fit_model nsubjpass_reported_fit auxpass_reported_is prep_reported_in pobj_in_terms prep_terms_of det_statistic_the nn_statistic_likelihood nn_statistic_ratio pobj_of_statistic conj_statistic_G num_G_2 cc_statistic_and poss_significance_its conj_statistic_significance ' +p,6972,' solved relational similarity problems using the Web as a corpus ',Nakov,'nsubj_solved_ amod_problems_relational nn_problems_similarity dobj_solved_problems xcomp_solved_using det_Web_the dobj_using_Web prep_using_as det_corpus_a pobj_as_corpus ' +o,6973,'Two main approaches have generally been considered : rule-based probabilistic ',Cutting,'num_approaches_Two amod_approaches_main aux_considered_have advmod_considered_generally auxpass_considered_been dep_approaches_considered amod__rule-based dep_approaches_ amod__probabilistic ' +n,6974,'This strategy is commonly used in MT evaluation , because of BLEUs well-known problems with documents of small size ',Papineni,'det_strategy_This nsubjpass_used_strategy auxpass_used_is advmod_used_commonly prep_used_in nn_evaluation_MT pobj_in_evaluation dep_in_because dep_in_of nn_problems_BLEUs amod_problems_well-known pobj_of_problems prep_used_with pobj_with_documents prep_documents_of amod_size_small pobj_of_size ' +o,6975,' -RRB- , and distributional methods -LRB- eg , et al ',Bergsma,'nn_distributional_and dep__distributional nn_eg_methods dep_distributional_eg dobj__ dep__et advmod__al ' +o,6976,'33 Corpora Our labeled data comes from the Penn Treebank and consists of about 40,000 sentences from Wall Street Journal -LRB- WSJ -RRB- articles 153 annotated with syntactic information ',Marcus,'nn_Our_Corpora nsubj_labeled_Our rcmod_33_labeled nsubj_comes_data ccomp_labeled_comes prep_comes_from det__the nn__Penn nn__Treebank pobj_from_ cc_comes_and conj_comes_consists prep_33_of quantmod_40,000_about num_sentences_40,000 pobj_of_sentences prep_sentences_from nn_articles_Wall nn_articles_Street nn_articles_Journal abbrev_articles_WSJ pobj_from_articles advmod_annotated_153 partmod_articles_annotated prep_annotated_with amod_information_syntactic pobj_with_information ' +o,6977,'2 System Description 21 Data Representation In this paper , we change the representation of the original data as follows : Bracketed representation of roles is converted into IOB2 representation Word tokens are collapsed into base phrase -LRB- BP -RRB- tokens ',Ramshaw,'num_Description_2 nn_Description_System num_Description_21 nn_Representation_Data dep_In_Representation prep_change_In det_paper_this pobj_In_paper nsubj_change_we dep_Description_change det_representation_the dobj_change_representation prep_representation_of det_data_the amod_data_original pobj_of_data mark_follows_as advcl_change_follows nn_representation_Bracketed nsubjpass_converted_representation prep_representation_of pobj_of_roles auxpass_converted_is parataxis_change_converted prep_converted_into nn_representation_IOB2 pobj_into_representation nn_tokens_ nn_tokens_Word nsubjpass_collapsed_tokens auxpass_collapsed_are dep_Description_collapsed prep_collapsed_into amod_tokens_base nn_tokens_phrase nn_tokens_BP pobj_into_tokens ' +n,6978,'Our system improves over the latent named-entity tagging in , from 61 \% to 87 \% ',Haghighi,'poss_system_Our nsubj_improves_system prep_improves_over det_tagging_the amod_tagging_latent amod_tagging_named-entity pobj_over_tagging prep_tagging_in pobj_in_ prep_improves_from number_\%_61 dep_\%_\% dep_\%_to number_\%_87 pobj_from_\% ' +o,6979,'Models that support non-monotonic decoding generally include a distortion cost , such as aibi11 where ai is the starting position of the foreign phrasefi andbi1 is the ending position of phrase fi1 ',Koehn,'nsubj_position_Models nsubj_support_that rcmod_Models_support amod_decoding_non-monotonic nsubj_include_decoding advmod_include_generally ccomp_support_include det_cost_a nn_cost_distortion dobj_include_cost dep_as_such prep_cost_as pobj_as_aibi11 advmod_position_where nsubj_position_ai cop_position_is det_position_the amod_position_starting rcmod_aibi11_position prep_position_of det_andbi1_the amod_andbi1_foreign nn_andbi1_phrasefi pobj_of_andbi1 cop_position_is det_position_the nn_position_ending prep_position_of nn_fi1_phrase pobj_of_fi1 ' +o,6980,'The transcription probabilities can then be easily learned from the alignments induced by GIZA + + , using a scoring function ',Koehn,'det_probabilities_The nn_probabilities_transcription nsubjpass_learned_probabilities aux_learned_can advmod_learned_then auxpass_learned_be advmod_learned_easily prep_learned_from det_alignments_the pobj_from_alignments partmod_alignments_induced prep_induced_by nn_+_GIZA nn_+_+ pobj_by_+ xcomp_induced_using det_function_a amod_function_scoring dobj_using_function ' +o,6981,'Further , we can learn the channel probabilities in an unsupervised manner using a variant of the EM algorithm similar to machine translation , and statistical language understanding ',Brown,'advmod_learn_Further nsubj_learn_we aux_learn_can det_probabilities_the nn_probabilities_channel dobj_learn_probabilities prep_learn_in det_manner_an amod_manner_unsupervised pobj_in_manner partmod_manner_using det_variant_a nsubj_similar_variant prep_variant_of det_algorithm_the nn_algorithm_EM pobj_of_algorithm xcomp_using_similar prep_similar_to nn__machine nn__translation pobj_to_ cc__and amod__statistical nn__language nn__understanding conj__ ' +p,6982,'Albeit simple , the algorithm has proven to be very efficient and accurate for the task of parse selection ',Zettlemoyer,'amod_simple_Albeit nsubj_proven_simple det_algorithm_the appos_simple_algorithm aux_proven_has aux_efficient_to cop_efficient_be advmod_efficient_very xcomp_proven_efficient cc_efficient_and conj_efficient_accurate prep_efficient_for det_task_the pobj_for_task prep_task_of amod_selection_parse pobj_of_selection ' +o,6983,' and Uryupina -LRB- 2003 -RRB- have already employed a definite probability measure in a similar way , although the way the ratio is computed is slightly different ',Bean,'nsubj_measure_ cc__and conj__Uryupina appos__2003 aux_measure_have advmod_measure_already dep_measure_employed det_measure_a amod_measure_definite nn_measure_probability prep_measure_in det_way_a amod_way_similar pobj_in_way mark_different_although det_way_the nsubj_different_way det_ratio_the nsubjpass_computed_ratio auxpass_computed_is rcmod_way_computed cop_different_is advmod_different_slightly advcl_measure_different ' +o,6984,'Partitioning 2 : Medium and low frequency words As noted in , log-likelihood statistics are able to capture word bi-gram regularities ',Dunning,'dep_Medium_Partitioning dobj_Partitioning_2 cc_Medium_and amod_words_low nn_words_frequency conj_Medium_words mark_noted_As advcl_able_noted prep_noted_in pobj_in_ amod_statistics_log-likelihood nsubj_able_statistics cop_able_are dep_Medium_able aux_capture_to xcomp_able_capture nn_regularities_word nn_regularities_bi-gram dobj_capture_regularities ' +o,6985,'Perhaps this was not observed earlier since studied only base NPs , most of which are short ',Ramshaw,'advmod_observed_Perhaps nsubjpass_observed_this auxpass_observed_was neg_observed_not advmod_observed_earlier dep_earlier_since nsubj_studied_ dep_observed_studied amod_NPs_only nn_NPs_base dobj_studied_NPs nsubj_short_most prep_most_of pobj_of_which cop_short_are rcmod_NPs_short ' +o,6986,'Previous work from showed improvements in perplexity-oriented measures using mixture-based translation lexicon ',Brown,'amod_work_Previous nsubj_showed_work prep_work_from pobj_from_ dobj_showed_improvements prep_showed_in amod_measures_perplexity-oriented pobj_in_measures xcomp_showed_using amod_lexicon_mixture-based nn_lexicon_translation dobj_using_lexicon ' +o,6987,'2 Learning algorithm The translation model is a standard linear model , which we train using MIRA , following Watanabe et al ',Och,'num_algorithm_2 nn_algorithm_Learning det_model_The nn_model_translation nsubj__model cop__is det__a amod__standard amod__linear nn__model dep_algorithm_ dobj_train_which nsubj_train_we rcmod__train xcomp_train_using nn__MIRA dobj_using_ prep__following pobj_following_Watanabe cc_Watanabe_et conj_Watanabe_al ' +o,6988,'However , in the Grammar Association context , when developing -LRB- using Bayes decomposition -RRB- the basic equations of the system presented in , it is said that the reverse model for a28 a13a37a3a38a5a39a32a21a0a35a7 does not seem to admit a simple factorization which is also correct and convenient , so crude heuristics were adopted in the mathematical development of the expression to be maximized ',Brown,'advmod_said_However prep_said_in det_context_the nn_context_Grammar nn_context_Association pobj_in_context advmod_developing_when dep_said_developing parataxis_developing_using nn_decomposition_Bayes dobj_using_decomposition det_equations_the amod_equations_basic nsubj_presented_equations prep_equations_of det_system_the pobj_of_system ccomp_developing_presented prep_presented_in pobj_in_ nsubjpass_said_it auxpass_said_is complm_seem_that det_model_the amod_model_reverse nsubj_seem_model prep_model_for amod_a13a37a3a38a5a39a32a21a0a35a7_a28 pobj_for_a13a37a3a38a5a39a32a21a0a35a7 aux_seem_does neg_seem_not ccomp_said_seem aux_admit_to xcomp_seem_admit det_factorization_a amod_factorization_simple dobj_admit_factorization nsubj_correct_which cop_correct_is advmod_correct_also rcmod_factorization_correct cc_correct_and conj_correct_convenient dep_seem_so amod_heuristics_crude nsubjpass_adopted_heuristics auxpass_adopted_were dep_seem_adopted prep_adopted_in det_development_the amod_development_mathematical pobj_in_development prep_development_of det_expression_the pobj_of_expression aux_maximized_to auxpass_maximized_be xcomp_adopted_maximized ' +o,6989,'The translation and reference files are analyzed by a treebank-based , probabilistic Lexical-Functional Grammar -LRB- LFG -RRB- parser , which produces a set of dependency triples for each input ',Cahill,'det_files_The nn_files_translation cc_translation_and conj_translation_reference nsubjpass_analyzed_files auxpass_analyzed_are prep_analyzed_by det__a amod__treebank-based dep_treebank-based_probabilistic nn__Lexical-Functional nn__Grammar abbrev__LFG nn__parser pobj_by_ nsubj_produces_which rcmod__produces det_set_a dobj_produces_set prep_set_of amod_triples_dependency pobj_of_triples prep_triples_for det_input_each pobj_for_input ' +o,6990,'The use of such relations -LRB- mainly relations between verbs or nouns and their arguments and modifiers -RRB- for various purposes has received growing attention in recent research ',Smadja,'det_use_The nsubj_received_use prep_use_of amod_relations_such pobj_of_relations advmod_relations_mainly dep_relations_relations prep_relations_between pobj_between_verbs cc_verbs_or conj_verbs_nouns cc_verbs_and poss_arguments_their conj_verbs_arguments cc_arguments_and conj_arguments_modifiers prep_relations_for amod_purposes_various pobj_for_purposes aux_received_has amod_attention_growing dobj_received_attention prep_attention_in amod_research_recent pobj_in_research ' +p,6991,'Two popular techniques that incorporate the error criterion are Minimum Error Rate Training -LRB- MERT -RRB- and Minimum BayesRisk -LRB- MBR -RRB- decoding ',Och,'num_techniques_Two amod_techniques_popular nsubj__techniques nsubj_incorporate_that rcmod_techniques_incorporate det_criterion_the nn_criterion_error dobj_incorporate_criterion cop__are nn__Minimum nn__Error nn__Rate nn__Training abbrev__MERT cc__and nn_BayesRisk_Minimum conj__BayesRisk abbrev_BayesRisk_MBR amod__decoding dep__ ' +o,6992,'This amounts to performing binary text categorization under categories Objective and Subjective ; 2 ',Pang,'nsubj_amounts_This prep_amounts_to pcomp_to_performing amod_categorization_binary nn_categorization_text dobj_performing_categorization prep_performing_under nn_Objective_categories pobj_under_Objective cc_Objective_and nn__Subjective conj_Objective_ dep_Objective_2 ' +o,6993,'In contrast , semi-supervised domain adaptation is the scenario in which , in addition to the labeled source data , we only have unlabeled and no labeled target domain data ',Blitzer,'prep_scenario_In pobj_In_contrast amod__semi-supervised nn__domain nn__adaptation nsubj_scenario_ cop_scenario_is det_scenario_the rel_labeled_in dep_in_which pobj_in_in pobj_in_addition prep_addition_to det_data_the amod_data_labeled nn_data_source pobj_to_data nsubj_labeled_we advmod_labeled_only aux_labeled_have advmod_labeled_unlabeled cc_unlabeled_and conj_unlabeled_no rcmod_scenario_labeled nn_data_target nn_data_domain dobj_labeled_data ' +o,6994,'With the success of collaborative sites like Amazons Mechanical Turk 1 , one 1http : \\/ \\/ wwwmturkcom \\/ 59 can provide the task of annotation to multiple oracles on the internet ',Snow,'prep_provide_With det_success_the pobj_With_success prep_success_of amod_sites_collaborative pobj_of_sites prep_sites_like nn_Turk_Amazons nn_Turk_Mechanical pobj_like_Turk num_Turk_1 num_1http_one nsubj_provide_1http amod_wwwmturkcom_\\/ amod_wwwmturkcom_\\/ dep_1http_wwwmturkcom partmod_wwwmturkcom_\\/ dobj_\\/_59 aux_provide_can det_task_the dobj_provide_task prep_task_of pobj_of_annotation prep_provide_to amod_oracles_multiple pobj_to_oracles prep_provide_on det_internet_the pobj_on_internet ' +o,6995,'Still , a confidence range for BLEU can be estimated by bootstrapping ',Och,'advmod_estimated_Still det_range_a nn_range_confidence nsubjpass_estimated_range prep_range_for pobj_for_BLEU aux_estimated_can auxpass_estimated_be prep_estimated_by pcomp_by_bootstrapping ' +o,6996,'Some authors have already designed similar matching techniques , such as the ones described in and ',Snow,'det_authors_Some nsubj_designed_authors aux_designed_have advmod_designed_already amod_techniques_similar amod_techniques_matching dobj_designed_techniques dep_as_such prep_techniques_as det_ones_the pobj_as_ones partmod_ones_described prep_described_in nn_and_ pobj_in_and ' +n,6997,'The generalized perceptron proposed by is closely related to CRFs , but the best CRF training methods seem to have a slight edge over the generalized perceptron ',Collins,'det_perceptron_The amod_perceptron_generalized nsubjpass_related_perceptron partmod_perceptron_proposed prep_proposed_by pobj_by_ auxpass_related_is advmod_related_closely prep_related_to pobj_to_CRFs cc_related_but det_methods_the amod_methods_best nn_methods_CRF nn_methods_training nsubj_seem_methods conj_related_seem aux_have_to xcomp_seem_have det_edge_a amod_edge_slight dobj_have_edge prep_edge_over det_perceptron_the amod_perceptron_generalized pobj_over_perceptron ' +o,6998,'The feature templates in that were left out were the ones that look at the previous word , the word two positions before the current , and the word two positions after the current ',Ratnaparkhi,'det_templates_The nn_templates_feature nsubj_ones_templates prep_templates_in pobj_in_ nsubjpass_left_that auxpass_left_were rcmod__left prt_left_out cop_ones_were det_ones_the nsubj_look_that rcmod_ones_look prep_look_at det_word_the amod_word_previous pobj_at_word det_word_the appos_word_word num_positions_two dobj_look_positions prep_positions_before det_current_the pobj_before_current cc_positions_and det_word_the conj_positions_word num_positions_two dep_word_positions prep_look_after det_current_the pobj_after_current ' +o,6999,'In our case , we computed a likelihood ratio score for all pairs of English tokens and Inuktitut substrings of length ranging from 3 to 10 characters ',Dunning,'prep_computed_In poss_case_our pobj_In_case nsubj_computed_we det__a nn__likelihood nn__ratio nn__score dobj_computed_ prep_computed_for det_pairs_all pobj_for_pairs prep_pairs_of amod_tokens_English pobj_of_tokens cc_pairs_and amod_substrings_Inuktitut conj_pairs_substrings prep_substrings_of pobj_of_length xcomp_computed_ranging prep_ranging_from number_10_3 dep_10_to num_characters_10 pobj_from_characters ' +o,7000,'More recently , Haffari and Sarkar have extended the work of and given a better mathematical understanding of self-training algorithms ',Abney,'advmod_recently_More advmod_extended_recently nn__Haffari cc_Haffari_and conj_Haffari_Sarkar nsubj_extended_ aux_extended_have det_work_the dobj_extended_work prep_work_of pobj_of_ cc_extended_and conj_extended_given det_understanding_a amod_understanding_better amod_understanding_mathematical dobj_given_understanding prep_understanding_of amod_algorithms_self-training pobj_of_algorithms ' +o,7001,'Word alignments are provided by GIZA + + with grow-diag-final combination , with infrastructure for alignment combination and phrase extraction provided by the shared task ',Och,'amod_alignments_Word nsubjpass_provided_alignments auxpass_provided_are prep_provided_by pobj_by_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_provided_with amod_combination_grow-diag-final pobj_with_combination prep_provided_with pobj_with_infrastructure prep_infrastructure_for amod_combination_alignment pobj_for_combination cc_combination_and nn_extraction_phrase conj_combination_extraction partmod_combination_provided prep_provided_by det_task_the amod_task_shared pobj_by_task ' +o,7002,'As a result , they are being used in a variety of applications , such as question answering , speech recognition , language modeling , language generation and , most notably , machine translation ',Galley,'prep_used_As det_result_a pobj_As_result nsubjpass_used_they aux_used_are auxpass_used_being prep_used_in det_variety_a pobj_in_variety prep_variety_of pobj_of_applications dep_as_such prep_variety_as nn__question nn__answering pobj_as_ nn__speech nn__recognition conj__ nn__language nn__modeling conj__ nn__language nn__generation conj__ cc__and advmod_notably_most dep__notably nn_translation_machine conj__translation ' +o,7003,'By labeling Treeb ~ n ~ nodes with Gr ~ ramar rule names , and not with phrasal and clausal n ~ raes , as in other -LRB- non-gr ~ rarnar-based -RRB- treebanks \' , we gain access to all information provided by the Grammar regarding each ~ reebank node ',Marcus,'prep_gain_By pcomp_By_labeling nn_nodes_Treeb nn_nodes_~ nn_nodes_n nn_nodes_~ dobj_labeling_nodes prep_labeling_with nn_names_Gr nn_names_~ nn_names_ramar nn_names_rule pobj_with_names cc_with_and conj_with_not dep_with_with nn_raes_phrasal cc_phrasal_and conj_phrasal_clausal nn_raes_n nn_raes_~ pobj_with_raes advmod_gain_as dep_as_in amod_treebanks_other amod_~_non-gr dep_rarnar-based_~ dep_other_rarnar-based poss__treebanks pobj_in_ nsubj_gain_we dobj_gain_access prep_gain_to det_information_all pobj_to_information partmod_information_provided prep_provided_by det_Grammar_the pobj_by_Grammar prep_Grammar_regarding det_node_each amod_node_~ nn_node_reebank pobj_regarding_node ' +o,7004,'The usefulness of likelihood ratios for collocation detection has been made explicit by and has been confirmed by an evaluation of various collocation detection methods carried out by Evert and Krenn ',Dunning,'det_usefulness_The nsubjpass_made_usefulness prep_usefulness_of nn_ratios_likelihood pobj_of_ratios prep_ratios_for nn_detection_collocation pobj_for_detection aux_made_has auxpass_made_been dep_made_explicit prep_made_by pobj_by_ cc_made_and aux_confirmed_has auxpass_confirmed_been conj_made_confirmed prep_confirmed_by det_evaluation_an pobj_by_evaluation prep_evaluation_of amod_methods_various nn_methods_collocation nn_methods_detection pobj_of_methods partmod_methods_carried prt_carried_out prep_carried_by nn__Evert cc_Evert_and conj_Evert_Krenn pobj_by_ ' +o,7005,'We use maximum entropy modeling to directly model the conditional probability a17a19a18a20a2a21a15a23a22a24a26a25 , where each a27a5a15 in a24a29a28a30a18a31a27a32a4a33a6a7a8a9a8a9a8a9a6a23a27a34a11a14a25 is an observation associated with the corresponding speaker a2 a15 a27 a15 is represented here by only one variable for notational ease , but it possibly represents several lexical , durational , structural , and acoustic observations ',Berger,'nsubj_use_We amod__maximum amod__entropy nn__modeling nsubj_model_ aux_model_to advmod_model_directly xcomp_use_model det_a17a19a18a20a2a21a15a23a22a24a26a25_the amod_a17a19a18a20a2a21a15a23a22a24a26a25_conditional nn_a17a19a18a20a2a21a15a23a22a24a26a25_probability dobj_model_a17a19a18a20a2a21a15a23a22a24a26a25 advmod_observation_where det_a27a5a15_each nsubj_observation_a27a5a15 prep_a27a5a15_in pobj_in_a24a29a28a30a18a31a27a32a4a33a6a7a8a9a8a9a8a9a6a23a27a34a11a14a25 cop_observation_is det_observation_an advcl_use_observation partmod_observation_associated prep_associated_with det_speaker_the amod_speaker_corresponding pobj_with_speaker amod_a15_a2 amod_a15_a15 nn_a15_a27 nsubjpass_represented_a15 auxpass_represented_is dep_use_represented advmod_represented_here prep_represented_by advmod_variable_only num_variable_one pobj_by_variable prep_variable_for amod_ease_notational pobj_for_ease cc_represented_but nsubj_represents_it advmod_represents_possibly conj_represented_represents amod_lexical_several dobj_represents_lexical amod_lexical_durational dep_durational_structural cc_lexical_and amod_observations_acoustic conj_lexical_observations ' +p,7006,'Besides relative frequencies , lexical weights are widely used to estimate how well the words in f translate the words in e To do this , one needs first to estimate a lexical translation probability distribution w -LRB- e f -RRB- by relative frequency from the same word alignments in the training corpus : w -LRB- e f -RRB- = count -LRB- f , e -RRB- summationtext e count -LRB- f , e -RRB- -LRB- 3 -RRB- Note that a special source NULL token is added to each source sentence and aligned to each unaligned target word ',Koehn,'prep_used_Besides amod_frequencies_relative pobj_Besides_frequencies amod__lexical nn__weights nsubjpass_used_ auxpass_used_are advmod_used_widely aux_estimate_to xcomp_used_estimate advmod_translate_how advmod_how_well det_words_the nsubj_translate_words prep_words_in dep_in_f ccomp_estimate_translate det_words_the dobj_translate_words prep_words_in dep_in_e aux_do_To xcomp_translate_do dobj_do_this num_needs_one measure_first_needs amod_this_first aux_estimate_to infmod_this_estimate det_w_a amod_w_lexical nn_w_translation nn_w_probability nn_w_distribution dobj_estimate_w dep_f_e dep_-LRB-_f prep_w_by amod_frequency_relative pobj_by_frequency prep_w_from det_alignments_the amod_alignments_same nn_alignments_word pobj_from_alignments prep_alignments_in det_corpus_the nn_corpus_training pobj_in_corpus nsubj_=_w dep_f_e dep_-LRB-_f rcmod_w_= nsubj_summationtext_count dep_count_f appos_f_e xcomp_=_summationtext dep_Note_e nsubj_Note_count dep_count_f appos_f_e appos_count_3 ccomp_summationtext_Note complm_added_that det_token_a amod_token_special nn_token_source nn_token_NULL nsubjpass_added_token auxpass_added_is ccomp_Note_added prep_added_to det_sentence_each nn_sentence_source pobj_to_sentence cc_added_and conj_added_aligned prep_aligned_to det_word_each amod_word_unaligned nn_word_target pobj_to_word ' +o,7007,'Although we have argued -LRB- section 2 -RRB- that this is unlikely to succeed , to our knowledge , we are the first to investigate the matter empirically11 The best-known MT aligner is undoubtedly GIZA + + , which contains implementations of various IBM models , as well as the HMM model of Vogel et al ',Brown,'mark_argued_Although nsubj_argued_we aux_argued_have advcl_first_argued dep_argued_section num_section_2 complm_unlikely_that nsubj_unlikely_this cop_unlikely_is ccomp_argued_unlikely aux_succeed_to xcomp_unlikely_succeed prep_unlikely_to poss_knowledge_our pobj_to_knowledge nsubj_first_we cop_first_are det_first_the aux_investigate_to dep_first_investigate det_empirically11_the nn_empirically11_matter dobj_investigate_empirically11 det_aligner_The amod_aligner_best-known nn_aligner_MT nsubj_GIZA_aligner cop_GIZA_is advmod_GIZA_undoubtedly dep_first_GIZA amod_GIZA_+ cc_+_+ conj_+_ nsubj_contains_which rcmod_GIZA_contains dobj_contains_implementations prep_implementations_of amod__various nn__IBM nn__models pobj_of_ dep_well_as cc_implementations_well dep_well_as det_model_the nn_model_HMM conj_implementations_model prep_model_of pobj_of_Vogel cc_Vogel_et conj_Vogel_al ' +o,7008,'Since text planners can not generate either the requisite syntactic variation or quantity of text , developed an evaluation strategy for HALOGEN employing a substitute : sentence parses from the Penn TreeBank , a corpus that includes texts from newspapers such as the Wall Street Journal , and which have been hand-annotated for syntax by linguists ',Marcus,'mark_generate_Since nn_planners_text nsubj_generate_planners aux_generate_can neg_generate_not advcl_developed_generate advmod_generate_either det_variation_the amod_variation_requisite amod_variation_syntactic dobj_generate_variation cc_variation_or conj_variation_quantity prep_quantity_of pobj_of_text nsubj_developed_ det_strategy_an nn_strategy_evaluation dobj_developed_strategy prep_developed_for pobj_for_HALOGEN xcomp_developed_employing det_substitute_a dobj_employing_substitute nn_parses_sentence dep_substitute_parses prep_parses_from det__the nn__Penn nn__TreeBank pobj_from_ det_corpus_a appos__corpus nsubj_includes_that rcmod_corpus_includes dobj_includes_texts prep_includes_from pobj_from_newspapers dep_as_such prep_newspapers_as det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_as_Journal cc_substitute_and nsubjpass_hand-annotated_which aux_hand-annotated_have auxpass_hand-annotated_been conj_substitute_hand-annotated prep_hand-annotated_for pobj_for_syntax prep_hand-annotated_by pobj_by_linguists ' +o,7009,'Indeed , only few earlier works reported inter-judge agreement level , and those that did reported rather low Kappa values , such as 054 and 055 063 ',Barzilay,'advmod_reported_Indeed advmod_reported_only amod_works_few amod_works_earlier nsubj_reported_works amod_level_inter-judge nn_level_agreement dobj_reported_level cc_reported_and nsubj_reported_those nsubj_did_that rcmod_those_did conj_reported_reported advmod_low_rather amod_values_low nn_values_Kappa dobj_reported_values dep_as_such prep_values_as number_063_054 number_063_ cc_063_and conj_063_055 num__063 pobj_as_ ' +p,7010,'Perhaps more importantly , discriminative models have been shown to offer competitive performance on a variety of sequential and structured learning tasks in NLP that are traditionally tackled via generative models , such as letter-to-phoneme conversion , semantic role labeling , syntactic parsing , language modeling , and machine translation ',Liang,'advmod_shown_Perhaps advmod_importantly_more advmod_shown_importantly amod_models_discriminative nsubjpass_shown_models aux_shown_have auxpass_shown_been aux_offer_to xcomp_shown_offer amod_performance_competitive dobj_offer_performance prep_offer_on det_variety_a pobj_on_variety prep_variety_of amod_tasks_sequential cc_sequential_and conj_sequential_structured nn_tasks_learning pobj_of_tasks prep_offer_in pobj_in_NLP nsubjpass_tackled_that auxpass_tackled_are advmod_tackled_traditionally rcmod_NLP_tackled prep_tackled_via amod_models_generative pobj_via_models dep_as_such prep_NLP_as amod__letter-to-phoneme nn__conversion pobj_as_ amod__semantic nn__role nn__labeling conj__ amod__syntactic nn__parsing conj__ nn__language nn__modeling conj__ cc__and nn_translation_machine conj__translation ' +o,7011,'The bidirectional word alignmentisusedtoobtainlexicalphrasetranslationpairs using heuristics presented in and ',Och,'det_word_The amod_word_bidirectional nsubj_alignmentisusedtoobtainlexicalphrasetranslationpairs_word xcomp_alignmentisusedtoobtainlexicalphrasetranslationpairs_using dobj_using_heuristics partmod_heuristics_presented prep_presented_in pobj_in_ cc__and conj__ ' +o,7012,'An alternative to linear models is the log-linear models suggested by ',Och,'det_alternative_An nsubj_models_alternative prep_alternative_to amod_models_linear pobj_to_models cop_models_is det_models_the amod_models_log-linear partmod_models_suggested prep_suggested_by ' +o,7013,'Later taggers have managed to improve Brills figures a little bit , to just above 97 \% on the Wall Street Journal corpus using Hidden Markov Models , HMM and Conditional Random Fields , CRF ; eg , and Toutanova et al ',Collins,'advmod_taggers_Later nsubj_managed_taggers aux_managed_have aux_improve_to xcomp_managed_improve nn_figures_Brills dobj_improve_figures det_bit_a amod_bit_little dep_figures_bit prep_managed_to advmod_above_just quantmod_97_above num_\%_97 pobj_to_\% prep_\%_on det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_on_corpus partmod_corpus_using nn_Models_Hidden nn_Models_Markov dobj_using_Models conj_Models_HMM cc_Models_and nn_CRF_Conditional nn_CRF_Random nn_CRF_Fields conj_Models_CRF dep_Models_eg conj_Models_ cc_Models_and conj_Models_Toutanova cc_Toutanova_et conj_Toutanova_al ' +o,7014,'The novel algorithm differs computationally from earlier work in discriminative training algorithms for SMT as follows : a90 No computationally expensive a57 - best lists are generated during training : for each input sentence a single block sequence is generated on each iteration over the training data ',Och,'det_algorithm_The amod_algorithm_novel nsubj_differs_algorithm advmod_differs_computationally prep_differs_from amod_work_earlier pobj_from_work prep_work_in amod_algorithms_discriminative nn_algorithms_training pobj_in_algorithms prep_differs_for nn__SMT pobj_for_ mark_follows_as dep__follows dobj_follows_a90 det_a57_No advmod_expensive_computationally amod_a57_expensive dep_a90_a57 amod_lists_best nsubjpass_generated_lists auxpass_generated_are dep_a90_generated prep_generated_during pobj_during_training prep_differs_for det_sentence_each nn_sentence_input pobj_for_sentence det_sequence_a amod_sequence_single nn_sequence_block nsubjpass_generated_sequence auxpass_generated_is dep_differs_generated prep_generated_on det_iteration_each pobj_on_iteration prep_iteration_over det_data_the nn_data_training pobj_over_data ' +o,7015,'Introduction Since Eric Brill first introduced the method of Transformation-Based Learning -LRB- TBL -RRB- it has been used to learn rules for many natural language processing tasks , such as part-of-speech tagging , PPattachment disambiguation , text chunking , spelling correction , dialogue act tagging and ellipsis resolution ',Ramshaw,'mark_introduced_Since nn_Brill_Eric nsubj_introduced_Brill advmod_introduced_first advcl_act_introduced det_method_the dobj_introduced_method prep_method_of nn_Learning_Transformation-Based pobj_of_Learning abbrev_Learning_TBL nsubjpass_used_it aux_used_has auxpass_used_been rcmod_method_used aux_learn_to xcomp_used_learn dobj_learn_rules prep_rules_for amod_tasks_many amod_tasks_natural nn_tasks_language nn_tasks_processing pobj_for_tasks dep_as_such prep_rules_as amod__part-of-speech amod__tagging nn__ dep__PPattachment nn__disambiguation pobj_as_ nn__text amod__chunking appos__ xcomp_introduced_spelling nn__correction dobj_spelling_ nsubj_act_dialogue dep_Introduction_act amod__tagging dobj_act_ cc__and nn_resolution_ellipsis conj__resolution amod_resolution_ ' +o,7016,'For nonprojective parsing , the analogy to the inside algorithm is the O -LRB- n3 -RRB- matrix-tree algorithm , which is dominated asymptotically by a matrix determinant ',Koo,'prep_algorithm_For amod_parsing_nonprojective pobj_For_parsing det_analogy_the nsubj_algorithm_analogy prep_analogy_to det_algorithm_the amod_algorithm_inside pobj_to_algorithm cop_algorithm_is det_algorithm_the nn_algorithm_O appos_algorithm_n3 amod_algorithm_matrix-tree nsubjpass_dominated_which auxpass_dominated_is rcmod_algorithm_dominated advmod_dominated_asymptotically prep_dominated_by det_determinant_a nn_determinant_matrix pobj_by_determinant ' +o,7017,'GIZA + + and the heuristics grow-diag-final-and are used to generate m-ton word alignments ',Och,'nn__GIZA nn__+ nn__+ nsubjpass_used_ cc__and det_grow-diag-final-and_the amod_grow-diag-final-and_heuristics conj__grow-diag-final-and auxpass_used_are aux_generate_to xcomp_used_generate amod_alignments_m-ton nn_alignments_word dobj_generate_alignments ' +o,7018,'The translation probability can also be discriminatively trained such as in ',Tillmann,'det_probability_The nn_probability_translation nsubjpass_trained_probability aux_trained_can advmod_trained_also auxpass_trained_be advmod_trained_discriminatively dep_as_such prep_trained_as dep_as_in ' +o,7019,'To help our model learn that it is desirable to copy answer words into the question , we add to each corpus a list of identical dictionary word pairs w iw i For each corpus , we use GIZA , a publicly available SMT package that implements the IBM models , to train a QA noisy-channel model that maps flattened answer parse trees , obtained using the cut procedure described in Section 31 , into questions ',Brown,'aux_help_To dep_add_help poss_model_our nsubj_learn_model ccomp_help_learn complm_desirable_that nsubj_desirable_it cop_desirable_is ccomp_learn_desirable aux_copy_to xcomp_desirable_copy nn_words_answer dobj_copy_words prep_copy_into det_question_the pobj_into_question nsubj_add_we prep_add_to det_corpus_each pobj_to_corpus det_list_a dobj_add_list prep_list_of amod_pairs_identical nn_pairs_dictionary nn_pairs_word pobj_of_pairs nn_iw_w nsubj_i_iw rcmod_list_i prep_use_For det_corpus_each pobj_For_corpus nsubj_use_we ccomp_i_use nn__GIZA nsubj_train_ det_package_a advmod_available_publicly amod_package_available nn_package_SMT appos__package nsubj_implements_that rcmod_package_implements det__the nn__IBM nn__models dobj_implements_ aux_train_to xcomp_use_train det_model_a nn_model_QA amod_model_noisy-channel dobj_train_model complm_flattened_that nsubj_flattened_maps ccomp_train_flattened nn_trees_answer nn_trees_parse dobj_flattened_trees partmod_trees_obtained xcomp_obtained_using det_procedure_the nn_procedure_cut dobj_using_procedure partmod_procedure_described prep_described_in pobj_in_Section num_Section_31 prep_flattened_into pobj_into_questions ' +o,7020,'4 Data Collection We evaluated out method by running RASP over Brown Corpus and Wall Street Journal , as contained in the Penn Treebank ',Marcus,'num_Collection_4 nn_Collection_Data nsubj_evaluated_We rcmod_Collection_evaluated prt_evaluated_out dobj_evaluated_method prep_evaluated_by pcomp_by_running dobj_running_RASP prep_RASP_over nn_Corpus_Brown pobj_over_Corpus cc_Corpus_and nn_Journal_Wall nn_Journal_Street conj_Corpus_Journal mark_contained_as advcl_evaluated_contained prep_contained_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +o,7021,'translation systems and use Moses to search for the best target sentence ',Och,'nn_systems_translation nsubj__systems cc__and conj__use nn__Moses dobj__ prep__to pobj_to_search prep__for det_sentence_the amod_sentence_best nn_sentence_target pobj_for_sentence ' +o,7022,'51 Comparison to self-training For completeness , we also compared our results to the self-learning algorithm , which has commonly been referred to as bootstrapping in natural language processing and originally popularized by the work of Yarowsky in word sense disambiguation ',Abney,'num_Comparison_51 prep_Comparison_to pcomp_to_self-training prep_compared_For pobj_For_completeness nsubj_compared_we advmod_compared_also ccomp_self-training_compared poss_results_our dobj_compared_results prep_compared_to det_algorithm_the amod_algorithm_self-learning pobj_to_algorithm nsubjpass_referred_which aux_referred_has advmod_referred_commonly auxpass_referred_been rcmod_algorithm_referred aux_bootstrapping_to advmod_bootstrapping_as xcomp_referred_bootstrapping prep_bootstrapping_in amod_processing_natural nn_processing_language pobj_in_processing cc_referred_and advmod_popularized_originally conj_referred_popularized prep_popularized_by det_work_the pobj_by_work prep_work_of pobj_of_Yarowsky prep_popularized_in nn_disambiguation_word nn_disambiguation_sense pobj_in_disambiguation ' +o,7023,'The acquisition of clues is a key technology in these research efforts , as seen in learning methods for document-level SA and for phraselevel SA ',Kanayama,'det_acquisition_The nsubj_technology_acquisition prep_acquisition_of pobj_of_clues cop_technology_is det_technology_a amod_technology_key prep_technology_in det_efforts_these nn_efforts_research pobj_in_efforts mark_seen_as advcl_technology_seen prep_seen_in pcomp_in_learning dobj_learning_methods prep_learning_for amod__document-level nn__SA pobj_for_ cc_for_and conj_for_for amod_SA_phraselevel pobj_for_SA ' +o,7024,'Re-decoding based regeneration re-decodes the source sentence using original LM as well as new trans105 lation and reordering models that are trained on the source-to-target N-best translations generated in the first pass ',Rosti,'amod__Re-decoding partmod__based nn_re-decodes_regeneration dobj_based_re-decodes det_sentence_the nn_sentence_source nsubj_using_sentence dep__using amod_LM_original dobj_using_LM dep_well_as cc_LM_well dep_well_as amod_lation_new nn_lation_trans105 conj_LM_lation cc_lation_and nn_models_reordering conj_lation_models nsubjpass_trained_that auxpass_trained_are rcmod_lation_trained prep_trained_on det_translations_the amod_translations_source-to-target amod_translations_N-best pobj_on_translations partmod_translations_generated prep_generated_in det_pass_the amod_pass_first pobj_in_pass ' +o,7025,'Following , the slot labels are drawn from a set of classes constructed by extending each label by three additional symbols , Beginning\\/Inside\\/Outside -LRB- B\\/I\\/O -RRB- ',Ramshaw,'prep_drawn_Following pobj_Following_ det_labels_the nn_labels_slot nsubjpass_drawn_labels auxpass_drawn_are prep_drawn_from det_set_a pobj_from_set prep_set_of pobj_of_classes partmod_classes_constructed prep_constructed_by pcomp_by_extending det_label_each dobj_extending_label prep_extending_by num_symbols_three amod_symbols_additional pobj_by_symbols appos_symbols_Beginning\\/Inside\\/Outside appos_Beginning\\/Inside\\/Outside_B\\/I\\/O ' +p,7026,'While error-driven training techniques are commonly used to improve the performance of phrasebased translation systems , this paper presents a novel block sequence translation approach to SMT that is similar to sequential natural language annotation problems 727 such as part-of-speech tagging or shallow parsing , both in modeling and parameter training ',Och,'mark_used_While amod_techniques_error-driven nn_techniques_training nsubjpass_used_techniques auxpass_used_are advmod_used_commonly advcl_presents_used aux_improve_to xcomp_used_improve det_performance_the dobj_improve_performance prep_performance_of amod__phrasebased nn__translation nn__systems pobj_of_ det_paper_this nsubj_presents_paper det_approach_a amod_approach_novel nn_approach_block nn_approach_sequence nn_approach_translation dobj_presents_approach prep_presents_to pobj_to_SMT nsubj_similar_that cop_similar_is rcmod_SMT_similar aux_sequential_to xcomp_similar_sequential amod_problems_natural nn_problems_language nn_problems_annotation nsubj_727_problems xcomp_sequential_727 dep_as_such prep_727_as amod_tagging_part-of-speech pobj_as_tagging cc_tagging_or nn_parsing_shallow conj_tagging_parsing dep_to_both conj_to_in nn_training_modeling cc_modeling_and conj_modeling_parameter pobj_in_training ' +o,7027,'The proxy slot denotes a semantic individual which serves the role of an event instance in a partially Davidsonian scheme , as in or ',Hobbs,'det_slot_The amod_slot_proxy nsubj_denotes_slot det_individual_a amod_individual_semantic dobj_denotes_individual nsubj_serves_which rcmod_individual_serves det_role_the dobj_serves_role prep_role_of det_instance_an nn_instance_event pobj_of_instance prep_serves_in det_scheme_a nn_scheme_partially nn_scheme_Davidsonian pobj_in_scheme dep_in_as dep_in_in pobj_in_ cc__or conj__ ' +o,7028,'42 Base Model II Using the translation model II , where alignments are dependent on word\\/entity positions and word\\/entity sequence lengths , we have p -LRB- w e -RRB- = mproductdisplay j = 1 lsummationdisplay i = 0 p -LRB- aj = i j , m , l -RRB- p -LRB- wj ei -RRB- -LRB- 2 -RRB- where aj = i means that wj is aligned with ei ',Brown,'nsubj_means_42 nn_II_Base nn_II_Model nsubj_=_II partmod_II_Using det_model_the nn_model_translation dobj_Using_model nn__II tmod_Using_ advmod_dependent_where nsubj_dependent_alignments cop_dependent_are dep__dependent prep_dependent_on amod_positions_word\\/entity pobj_on_positions cc_positions_and amod_lengths_word\\/entity nn_lengths_sequence conj_positions_lengths nsubj_p_we aux_p_have ccomp_dependent_p number_e_w nsubj_j_e dep_j_= amod_j_mproductdisplay ccomp_p_j rcmod_42_= num_i_1 amod_i_lsummationdisplay dobj_=_i dep_p_= num_p_0 rcmod_i_p dep_i_aj nn_j_= nn_j_i dobj_aj_j appos_j_m tmod_aj_l dep_i_p nn_ei_wj appos_p_ei appos_p_2 advmod_i_where nsubj_i_aj dep_i_= rcmod_p_i complm_aligned_that nsubjpass_aligned_wj auxpass_aligned_is ccomp_means_aligned prep_aligned_with pobj_with_ei ' +o,7029,'These were combined using the Grow Diag Final And symmetrization heuristic ',Koehn,'nsubjpass_combined_These auxpass_combined_were xcomp_combined_using dep_Final_the nn_Final_Grow nn_Final_Diag nsubj_heuristic_Final cc_Final_And conj_Final_symmetrization xcomp_using_heuristic ' +o,7030,'If the target CFG is purely binary branching , then the previous theoretical and linguistic analyses suggest that much of the requisite constituent and word order transposition may be accommodated without change to the mirrored ITG ',Wu,'mark_branching_If det_CFG_the nn_CFG_target nsubj_branching_CFG cop_branching_is advmod_branching_purely advmod_branching_binary advcl_suggest_branching advmod_analyses_then det_analyses_the amod_analyses_previous amod_analyses_theoretical cc_theoretical_and conj_theoretical_linguistic nsubj_suggest_analyses amod_analyses_ complm_accommodated_that nsubjpass_accommodated_much prep_much_of det_constituent_the amod_constituent_requisite pobj_of_constituent cc_constituent_and nn_transposition_word nn_transposition_order conj_constituent_transposition aux_accommodated_may auxpass_accommodated_be ccomp_suggest_accommodated prep_accommodated_without pobj_without_change prep_accommodated_to det_ITG_the amod_ITG_mirrored pobj_to_ITG ' +o,7031,'First , we considered single sentences as documents , and tokens as sentences -LRB- we define a token as a sequence of characters delimited by 1In our case , the score we seek to globally maximize by dynamic programming is not only taking into account the length criteria described in but also a cognate-based one similar to ',Dunning,'advmod_considered_First nsubj_considered_we amod_sentences_single dobj_considered_sentences prep_considered_as pobj_as_documents cc_considered_and conj_considered_tokens mark_taking_as nsubj_taking_sentences nsubj_define_we rcmod_sentences_define det_token_a dobj_define_token prep_define_as det_sequence_a pobj_as_sequence prep_sequence_of pobj_of_characters partmod_characters_delimited prep_delimited_by nn_case_1In poss_case_our pobj_by_case det_score_the appos_case_score nsubj_seek_we rcmod_score_seek aux_maximize_to advmod_maximize_globally xcomp_seek_maximize prep_maximize_by amod_programming_dynamic pobj_by_programming aux_taking_is neg_taking_not advmod_taking_only advcl_tokens_taking prep_taking_into pobj_into_account det_criteria_the nn_criteria_length dobj_taking_criteria partmod_criteria_described prep_described_in pobj_in_ dep_also_but cc_criteria_also det_one_a amod_one_cognate-based conj_criteria_one amod_one_similar dep_similar_to dep_similar_ ' +o,7032,'For the maximum entropy classifier , we estimate the weights by maximizing the likelihood of a heldout set , using the standard IIS algorithm ',Berger,'prep_estimate_For det_classifier_the amod_classifier_maximum nn_classifier_entropy pobj_For_classifier nsubj_estimate_we det_weights_the dobj_estimate_weights prep_estimate_by pcomp_by_maximizing det_likelihood_the dobj_maximizing_likelihood prep_likelihood_of det_set_a amod_set_heldout pobj_of_set xcomp_estimate_using det_algorithm_the amod_algorithm_standard nn_algorithm_IIS dobj_using_algorithm ' +o,7033,'22 The Crossing Constraint According to , crossing constraint can be defined in the following ',Wu,'det_Constraint_The nn_Constraint_Crossing nsubjpass_defined_Constraint prep_Constraint_According dep__to pobj_According_ partmod__crossing dobj_crossing_constraint aux_defined_can auxpass_defined_be dep_22_defined prep_defined_in det_following_the pobj_in_following ' +o,7034,' attain 982 \% coverage and a BLEU score of 06652 on the standard WSJ test set -LRB- Section 23 -RRB- ',Cahill,'advmod_attain_ num_\%_982 dobj_attain_\% dep_\%_coverage cc_\%_and det_score_a amod_score_BLEU conj_\%_score prep_score_of pobj_of_06652 prep_attain_on det_set_the amod_set_standard nn_set_WSJ nn_set_test pobj_on_set appos_set_Section num_Section_23 ' +o,7035,'Previous publications on Meteor have described the details underlying the metric and have extensively compared its performance with Bleu and several other MT evaluation metrics ',Banerjee,'amod_publications_Previous nsubj_described_publications prep_publications_on nn__Meteor pobj_on_ aux_described_have det_details_the dobj_described_details xcomp_described_underlying det_metric_the dobj_underlying_metric cc_described_and aux_compared_have advmod_compared_extensively conj_described_compared poss_performance_its dobj_compared_performance prep_compared_with pobj_with_Bleu cc_Bleu_and amod_metrics_several amod_metrics_other nn_metrics_MT nn_metrics_evaluation conj_Bleu_metrics ' +o,7036,'1 Word associations -LRB- co-occurrences , or joint frequencies -RRB- have a wide range of applications including : speech recognition , optical character recognition , and information retrieval -LRB- IR -RRB- ',Dunning,'num_associations_1 nn_associations_Word nsubj_have_co-occurrences cc_co-occurrences_or amod_frequencies_joint conj_co-occurrences_frequencies rcmod_associations_have det_range_a amod_range_wide dobj_have_range prep_range_of pobj_of_applications prep_have_including nn_recognition_speech pobj_including_recognition amod_recognition_optical nn_recognition_character conj_recognition_recognition cc_recognition_and conj_recognition_information dep_associations_retrieval abbrev_retrieval_IR ' +o,7037,'(Snow et al., 2006; Nakov & Hearst, 2008).',Nakov,'' +o,7038,'Our study is also different from these previous ones in that measuring the agreement among annotators became an issue ',Carletta,'poss_study_Our nsubj_different_study cop_different_is advmod_different_also prep_different_from det_ones_these amod_ones_previous pobj_from_ones rel_issue_in pobj_in_that csubj_issue_measuring det_agreement_the dobj_measuring_agreement prep_measuring_among pobj_among_annotators cop_issue_became det_issue_an rcmod_ones_issue ' +o,7039,'Our approach differs from the corpus-based surface generation approaches of and ',Berger,'poss_approach_Our nsubj_differs_approach prep_differs_from det_approaches_the amod_approaches_corpus-based nn_approaches_surface nn_approaches_generation pobj_from_approaches prep_approaches_of nn_and_ pobj_of_and ' +o,7040,'In our experiments we use a grammar with a start symbol S , a single preterminal C , and two nonterminals A and B used to ensure that only one parse can generate any given word-level alignment -LRB- ignoring insertions and deletions -RRB- ',Wu,'prep_use_In poss_experiments_our pobj_In_experiments nsubj_use_we det_grammar_a dobj_use_grammar prep_grammar_with det_symbol_a nn_symbol_start pobj_with_symbol nsubj_used_S det_C_a amod_C_single amod_C_preterminal conj_S_C cc_S_and num_nonterminals_two conj_S_nonterminals dep_nonterminals_A cc_A_and conj_A_B dep_use_used aux_ensure_to xcomp_used_ensure complm_generate_that advmod_parse_only num_parse_one nsubj_generate_parse aux_generate_can ccomp_ensure_generate det_alignment_any amod_alignment_given amod_alignment_word-level dobj_generate_alignment dep_alignment_ignoring dobj_ignoring_insertions cc_insertions_and conj_insertions_deletions ' +o,7041,'There are good reasons for using such a hand-crafted , genre-specific verb lexicon instead of a general resource such as WordNet or Levins classes : Many verbs used in the domain of scientific argumentation have assumed a specialized meaning , which our lexicon readily encodes ',Dunning,'expl_are_There amod_reasons_good nsubj_are_reasons prep_reasons_for pcomp_for_using prep_using_such det_lexicon_a amod_lexicon_hand-crafted amod_lexicon_genre-specific nn_lexicon_verb dep_such_lexicon dep_of_instead prep_lexicon_of det_resource_a amod_resource_general pobj_of_resource dep_as_such prep_resource_as nn_classes_WordNet cc_WordNet_or conj_WordNet_Levins nn_classes_ pobj_as_classes amod_verbs_Many nsubj_assumed_verbs partmod_verbs_used prep_used_in det_domain_the pobj_in_domain prep_domain_of amod_argumentation_scientific pobj_of_argumentation aux_assumed_have parataxis_are_assumed det_meaning_a amod_meaning_specialized dobj_assumed_meaning dobj_encodes_which poss_lexicon_our nsubj_encodes_lexicon advmod_encodes_readily dep_assumed_encodes ' +o,7042,'Here , we compare two similarity measures : the familiar BLEU score and a score based on string kernels ',Papineni,'advmod_compare_Here nsubj_compare_we num_measures_two nn_measures_similarity dobj_compare_measures det__the amod__familiar dep_familiar_BLEU nn__score dep_measures_ cc__and det_score_a conj__score partmod_score_based prep_based_on nn_kernels_string pobj_on_kernels ' +o,7043,'Although this approach can give inaccurate estimates , the counts given to the incorrect senses will disperse randomly throughout the hierarchy as noise , and by accumulating counts up the hierarchy we will tend to gather counts from the correct senses of related words ',Dunning,'mark_give_Although det_approach_this nsubj_give_approach aux_give_can advcl_disperse_give amod_estimates_inaccurate dobj_give_estimates det_counts_the nsubj_disperse_counts prep_counts_given dep_given_to det_senses_the amod_senses_incorrect pobj_to_senses aux_disperse_will advmod_disperse_randomly prep_disperse_throughout det_hierarchy_the pobj_throughout_hierarchy prep_hierarchy_as pobj_as_noise cc_as_and conj_as_by pcomp_by_accumulating dobj_accumulating_counts advmod_accumulating_up det_hierarchy_the dep_up_hierarchy nsubj_tend_we aux_tend_will dep_disperse_tend aux_gather_to xcomp_tend_gather dobj_gather_counts prep_gather_from det_senses_the amod_senses_correct pobj_from_senses prep_senses_of amod__related nn__words pobj_of_ ' +o,7044,'Most probabilistic parsing research including , for example , work by by , and Charniak -LRB- 1997 -RRB- is based on branching process models ',Collins,'advmod_probabilistic_Most amod_research_probabilistic amod_research_parsing nsubjpass_based_research prep_research_including dep_including_for pobj_for_example pobj_including_work prep_work_by dep_by_by pobj_by_ cc_work_and conj_work_Charniak appos_Charniak_1997 auxpass_based_is prep_based_on pcomp_on_branching nn_models_process dobj_branching_models advmod_branching_ ' +o,7045,'2 Baseline Coreference Resolution System Our baseline coreference system implements the standard machine learning approach to coreference resolution -LRB- see Ng and Cardie , , Yang and Su -LRB- 2007 -RRB- , for instance -RRB- , which consists of probabilistic classification and clustering , as described below ',Ponzetto,'num_system_2 nn_system_Baseline nn_system_Coreference nn_system_Resolution nn_system_System nn_system_Our nn_system_baseline nn_system_coreference nsubj_implements_system det_machine_the amod_machine_standard dobj_implements_machine partmod_machine_learning dobj_learning_approach aux_coreference_to xcomp_learning_coreference dobj_coreference_resolution dep_machine_see dobj_see_Ng cc_Ng_and nn__Cardie conj_Ng_ conj__ conj__Yang cc__and conj__Su appos__2007 conj_Ng_for pobj_for_instance nsubj_consists_which rcmod_machine_consists prep_consists_of amod_classification_probabilistic pobj_of_classification cc_classification_and conj_classification_clustering mark_described_as advcl_consists_described advmod_described_below ' +o,7046,'The training and decoding system of our SMT used the publicly available Pharaoh 2 ',Koehn,'det_system_The nn_system_training cc_training_and conj_training_decoding nsubj_used_system prep_system_of poss_SMT_our pobj_of_SMT det_Pharaoh_the advmod_available_publicly amod_Pharaoh_available dobj_used_Pharaoh amod_Pharaoh_ tmod__2 ' +o,7047,' propose factored translation models that combine feature functions to handle syntactic , morphological , and other linguistic information in a log-linear model ',Koehn,'advmod_propose_ amod_models_factored nn_models_translation dobj_propose_models nsubj_combine_that rcmod_models_combine nn_functions_feature nsubj_handle_functions aux_handle_to xcomp_combine_handle amod_information_syntactic conj_syntactic_morphological cc_syntactic_and conj_syntactic_other dep_syntactic_linguistic dobj_handle_information prep_handle_in det_model_a amod_model_log-linear pobj_in_model ' +o,7048,'This therefore suggests that better parameters are likely to be learned in the 2 generative coreference model mirrors this in the posterior distribution which it assigns to mention types given their salience -LRB- see their Table 1 -RRB- ',Haghighi,'nsubj_suggests_This advmod_suggests_therefore complm_likely_that amod_parameters_better nsubj_likely_parameters cop_likely_are ccomp_suggests_likely aux_learned_to auxpass_learned_be xcomp_likely_learned prep_learned_in det_generative_the num_generative_2 amod_generative_ pobj_in_generative amod_model_coreference nsubj_mirrors_model dep_suggests_mirrors dobj_mirrors_this prep_this_in det_distribution_the nn_distribution_posterior pobj_in_distribution rel_assigns_which nsubj_assigns_it rcmod_this_assigns aux_mention_to xcomp_assigns_mention dobj_mention_types prep_mention_given poss_salience_their dep_given_salience dep_salience_see poss_Table_their dobj_see_Table dep_Table_1 ' +p,7049,'5 Bidirectional Sequence Classification Bidirectional POS tagging , the current state of the art for English , has some properties that make it appropriate for Icelandic ',Shen,'num__5 nn__Bidirectional nn__Sequence nn__Classification nn__Bidirectional dep__POS nn__tagging nsubj_has_ det_state_the amod_state_current appos__state prep_state_of det_art_the pobj_of_art prep_art_for pobj_for_English det_properties_some dobj_has_properties nsubj_make_that rcmod_properties_make nsubj_appropriate_it xcomp_make_appropriate prep_appropriate_for pobj_for_Icelandic ' +o,7050,'1 Introduction Over the last few years , several automatic metrics for machine translation -LRB- MT -RRB- evaluation have been introduced , largely to reduce the human cost of iterative system evaluation during the development cycle ',Papineni,'num_Introduction_1 dep_Introduction_Over det_years_the amod_years_last amod_years_few pobj_Over_years amod_metrics_several amod_metrics_automatic appos_years_metrics dep_metrics_for nn_translation_machine pobj_for_translation abbrev_translation_MT nsubjpass_introduced_evaluation aux_introduced_have auxpass_introduced_been rcmod_translation_introduced advmod_reduce_largely aux_reduce_to dep_for_reduce det_cost_the amod_cost_human dobj_reduce_cost prep_cost_of amod_evaluation_iterative nn_evaluation_system pobj_of_evaluation prep_reduce_during det_cycle_the nn_cycle_development pobj_during_cycle ' +o,7051,'Most of the annotation approaches tackling these issues , however , are aimed at performing classifications at either the document level , or the sentence or word level ',Pang,'nsubj_approaches_Most prep_Most_of det_annotation_the pobj_of_annotation xcomp_approaches_tackling det_issues_these nsubjpass_aimed_issues advmod_aimed_however auxpass_aimed_are ccomp_tackling_aimed prep_aimed_at pcomp_at_performing dobj_performing_classifications prep_performing_at dep__either det__the nn__document nn__level dep_either_ cc_either_or det_sentence_the conj_either_sentence cc_sentence_or conj_sentence_word nn__level pobj_at_ ' +o,7052,'This logistic regression is also called Maxent as it finds the distribution with maximum entropy that properly estimates the average of each feature over the training data ',Berger,'det_regression_This amod_regression_logistic nsubjpass_called_regression auxpass_called_is advmod_called_also dobj_called_Maxent mark_finds_as nsubj_finds_it advcl_called_finds det_distribution_the dobj_finds_distribution prep_finds_with amod_entropy_maximum pobj_with_entropy nsubj_estimates_that advmod_estimates_properly rcmod_entropy_estimates det_average_the dobj_estimates_average prep_average_of det_feature_each pobj_of_feature prep_feature_over det_data_the nn_data_training pobj_over_data ' +o,7053,'The idea of threading EEs to their antecedents in a stochastic parser was proposed by , following the GPSG tradition ',Collins,'det_idea_The nsubjpass_proposed_idea prep_idea_of pcomp_of_threading dobj_threading_EEs prep_threading_to poss_antecedents_their pobj_to_antecedents prep_antecedents_in det_parser_a amod_parser_stochastic pobj_in_parser auxpass_proposed_was prep_proposed_by pobj_by_ prep_proposed_following det__the nn__GPSG nn__tradition pobj_following_ ' +o,7054,'The mutual information clustering algorithm were used for this ',Brown,'det_information_The amod_information_mutual nsubjpass_used_information partmod_information_clustering nn__algorithm dobj_clustering_ auxpass_used_were prep_used_for pobj_for_this ' +o,7055,'1993 ; Chang et al , 1992 ; Collins and Brooks , 1995 ; Fujisaki , 1989 ; Hindle and Rooth , 1991 ; Hindle and Rooth , 1993 ; Jelinek et al , 1990 ; Magerman and , 1991 ; Magerman , 1995 ; Ratnaparkhi et al , 1994 ; Resnik , 1993 ; Su and Chang , 1988 -RRB- ',Marcus,'dep_1993_Chang cc_Chang_et conj_Chang_al appos_Chang_1992 dep_Chang_Collins cc_Collins_and conj_Collins_Brooks appos_Collins_1995 dep_Chang_Fujisaki appos_Fujisaki_1989 dep_Chang_Hindle cc_Hindle_and conj_Hindle_Rooth appos_Hindle_1991 dep_Chang_Hindle cc_Hindle_and conj_Hindle_Rooth appos_Hindle_1993 dep_Chang_Jelinek cc_Jelinek_et conj_Jelinek_al appos_Jelinek_1990 dep_Chang_Magerman cc_Magerman_and conj_Magerman_ appos_Magerman_1991 dep_Chang_Magerman appos_Magerman_1995 dep_Chang_Ratnaparkhi cc_Ratnaparkhi_et conj_Ratnaparkhi_al appos_Ratnaparkhi_1994 dep_Chang_Resnik appos_Resnik_1993 dep_Chang_Su cc_Su_and conj_Su_Chang appos_Chang_1988 ' +o,7056,'Building on the annotations from the Wall Street Journal (WSJ) portion of the Penn Treebank (Marcus et al., 1993), the project added several new layers of semantic annotations, such as coreference information, word senses, etc. In its first release (LDC2007T21) through the Linguistic Data Consortium (LDC), the project manually sense-tagged more than 40,000 examples belonging to hundreds of noun and verb types with an ITA of 90\%, based on a coarse-grained sense inventory, where each word has an average of only 3.2 senses.',Marcus,'' +o,7057,'Using a variant of the voted perceptron , we discriminatively trained our parser in an on-line fashion ',Collins,'dep_discriminatively_Using det_variant_a dobj_Using_variant prep_variant_of det__the amod__voted nn__perceptron pobj_of_ nsubj_discriminatively_we dep_discriminatively_trained poss_parser_our dobj_trained_parser prep_trained_in det_fashion_an amod_fashion_on-line pobj_in_fashion ' +o,7058,'The approach combines statistical and knowledge-based methods , but unlike many recent corpus-based approaches to sense disambiguation , it takes as its starting point the assumption that senseannotated training text is not available ',Marcus,'det_approach_The nsubj_combines_approach amod_methods_statistical cc_statistical_and conj_statistical_knowledge-based dobj_combines_methods cc_combines_but prep_takes_unlike amod_approaches_many amod_approaches_recent amod_approaches_corpus-based pobj_unlike_approaches prep_approaches_to nn__sense nn__disambiguation pobj_to_ nsubj_takes_it conj_combines_takes prep_takes_as poss_point_its amod_point_starting pobj_as_point det_assumption_the nsubj_available_assumption nsubj_senseannotated_that rcmod_assumption_senseannotated nn_text_training dobj_senseannotated_text cop_available_is neg_available_not dep_combines_available ' +o,7059,'Traditionally , maximum-likelihood estimation from relative frequencies is used to obtain conditional probabilities , eg , p -LRB- s t -RRB- = c -LRB- s , t -RRB- \\/ summationtexts c -LRB- s , t -RRB- -LRB- since the estimation problems for p -LRB- s t -RRB- and p -LRB- t s -RRB- are symmetrical , we will usually refer only to p -LRB- s t -RRB- for brevity -RRB- ',Koehn,'advmod_used_Traditionally amod_estimation_maximum-likelihood nsubjpass_used_estimation prep_estimation_from amod_frequencies_relative pobj_from_frequencies auxpass_used_is aux_obtain_to xcomp_used_obtain amod_probabilities_conditional nsubj__probabilities ccomp_obtain_ nsubj_=_eg appos_eg_p amod_t_s appos_p_t ccomp__= dobj_=_c appos_c_s dep_s_t nsubj_summationtexts_\\/ rcmod_c_summationtexts dobj_summationtexts_c appos_c_s dep_s_t mark_symmetrical_since det_problems_the nn_problems_estimation nsubj_symmetrical_problems prep_problems_for pobj_for_p amod_t_s appos_p_t cc_p_and conj_p_p nn_s_t appos_p_s cop_symmetrical_are dep_c_symmetrical nsubj_refer_we aux_refer_will advmod_refer_usually dep_symmetrical_refer advmod_refer_only prep_refer_to pobj_to_p amod_t_s appos_p_t prep_p_for pobj_for_brevity ' +p,7060,'We chose the perceptron for the training algorithm because it has shown good performance on other NLP tasks ; in particular , reported good performance for a perceptron tagger compared to a Maximum Entropy tagger ',Collins,'nsubj_chose_We det_perceptron_the dobj_chose_perceptron prep_chose_for det_algorithm_the nn_algorithm_training pobj_for_algorithm mark_shown_because nsubj_shown_it aux_shown_has advcl_chose_shown amod_performance_good dobj_shown_performance prep_shown_on amod_tasks_other nn_tasks_NLP pobj_on_tasks prep_reported_in pobj_in_particular nsubj_reported_ parataxis_chose_reported amod_performance_good dobj_reported_performance prep_reported_for det_tagger_a nn_tagger_perceptron pobj_for_tagger prep_reported_compared dep_compared_to det_tagger_a nn_tagger_Maximum nn_tagger_Entropy pobj_to_tagger ' +o,7061,' and Lee -RRB- can be generally divided into three types : discounting , class-based smoothing , and distance-weighted averaging ',Brown,'nn__ cc__and conj__Lee aux_divided_can auxpass_divided_be advmod_divided_generally dep__divided prep_divided_into num_types_three pobj_into_types nn__discounting dep__ amod__class-based nn__smoothing conj__ cc__and amod__distance-weighted nn__averaging conj__ ' +o,7062,'Many existing systems for statistical machine translation implement models presented by : The correspondence between the words in the source and the target strings is described by alignments that assign target word positions to each source word position ',Brown,'amod_systems_Many amod_systems_existing prep_systems_for amod__statistical nn__machine nn__translation pobj_for_ dep_systems_implement dobj_implement_models partmod_models_presented prep_presented_by pobj_by_ det_correspondence_The dep_models_correspondence prep_correspondence_between det_words_the pobj_between_words prep_words_in det_source_the pobj_in_source cc_implement_and det_strings_the nn_strings_target nsubjpass_described_strings auxpass_described_is conj_implement_described prep_described_by pobj_by_alignments nsubj_assign_that rcmod_alignments_assign nn_positions_target nn_positions_word dobj_assign_positions prep_assign_to det_position_each nn_position_source nn_position_word pobj_to_position ' +o,7063,'However , they do not elaborate on how the comparisons are done , or on how effective the program is describes a heuristic approach to forming unlabeled clusters of closely related senses in an MRD ',Dolan,'advmod_elaborate_However nsubj_elaborate_they aux_elaborate_do neg_elaborate_not prep_elaborate_on advmod_done_how det_comparisons_the nsubjpass_done_comparisons auxpass_done_are pcomp_on_done cc_on_or conj_on_on advmod__how dep_how_effective det_program_the nsubj__program cop__is pcomp_on_ ccomp__describes det_approach_a amod_approach_heuristic dobj_describes_approach prep_approach_to pcomp_to_forming amod_clusters_unlabeled dobj_forming_clusters prep_clusters_of advmod_related_closely amod_senses_related pobj_of_senses prep_senses_in det_MRD_an pobj_in_MRD ' +o,7064,' , various classification models and linguistic features have been proposed to improve the classification performance ',Pang,'amod_models_ amod_models_various nn_models_classification nsubjpass_proposed_models cc_models_and amod_features_linguistic conj_models_features aux_proposed_have auxpass_proposed_been aux_improve_to xcomp_proposed_improve det_performance_the nn_performance_classification dobj_improve_performance ' +p,7065,'The state-of-the art taggers are using feature sets discribed in the corresponding articles -LRB- , , and -RRB- , Morce supervised and Morce semi-supervised are using feature set desribed in section 4 ',Collins,'det_taggers_The amod_taggers_state-of-the nn_taggers_art nsubj_using_taggers aux_using_are nn_sets_feature dobj_using_sets partmod_sets_discribed prep_discribed_in det_articles_the amod_articles_corresponding pobj_in_articles appos_articles_ dep__ dep__ cc__and conj__ nsubj_supervised_Morce conj_using_supervised cc_using_and nsubj_using_Morce amod_Morce_semi-supervised aux_using_are conj_using_using nn_set_feature dobj_using_set partmod_set_desribed prep_desribed_in pobj_in_section num_section_4 ' +o,7066,'One of the simplest models in the context of lexical triggers is the IBM model 1 which captures lexical dependencies between source and target words ',Brown,'nsubj__One prep_One_of det_models_the amod_models_simplest pobj_of_models prep_models_in det_context_the pobj_in_context prep_context_of amod_triggers_lexical pobj_of_triggers cop__is det__the nn__IBM nn__model num__1 nsubj_captures_which rcmod__captures amod_dependencies_lexical dobj_captures_dependencies prep_dependencies_between nn_words_source cc_source_and conj_source_target pobj_between_words ' +o,7067,'grow-diagfinal -RRB- ',Koehn,'nsubj__grow-diagfinal ' +o,7068,' , -RRB- and the exploitation of advanced techniques that involve joint learning -LRB- eg , Daume III and Marcu -RRB- and joint inference -LRB- eg , Denis and Baldridge -RRB- for coreference resolution and a related extraction task ',Ponzetto,'amod__ cc__and det_exploitation_the conj__exploitation prep_exploitation_of amod_techniques_advanced pobj_of_techniques nsubj_involve_that rcmod_exploitation_involve amod_learning_joint dobj_involve_learning dep_learning_eg nn_III_Daume conj_eg_III cc_eg_and nn__Marcu conj_eg_ cc_learning_and amod_inference_joint conj_learning_inference dep__eg conj_eg_Denis cc_eg_and nn__Baldridge conj_eg_ prep__for amod_resolution_coreference pobj_for_resolution cc_resolution_and det_task_a amod_task_related nn_task_extraction conj_resolution_task ' +o,7069,'This permits us to make exact comparisons with the parser of Yamada and Matsumoto , but also the parsers of and Charniak -LRB- 2000 -RRB- , which are evaluated on the same data set in Yamada and Matsumoto ',Collins,'nsubj_permits_This nsubj_make_us aux_make_to xcomp_permits_make amod_comparisons_exact dobj_make_comparisons prep_make_with det_parser_the pobj_with_parser prep_parser_of nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto pobj_of_ dep_also_but cc_parser_also det_parsers_the conj_parser_parsers prep_parsers_of pobj_of_ cc__and conj__Charniak appos__2000 nsubjpass_evaluated_which auxpass_evaluated_are rcmod__evaluated prep_evaluated_on det_data_the amod_data_same pobj_on_data partmod_data_set prep_set_in nn__Yamada cc_Yamada_and conj_Yamada_Matsumoto pobj_in_ ' +o,7070,'In fact , the largest source of English dependency trees is automatically generated from the Penn Treebank and is by convention exclusively projective ',Marcus,'prep_generated_In pobj_In_fact det_source_the amod_source_largest nsubjpass_generated_source prep_source_of amod_trees_English amod_trees_dependency pobj_of_trees auxpass_generated_is advmod_generated_automatically prep_generated_from det__the nn__Penn nn__Treebank pobj_from_ cc_generated_and conj_generated_is prep_is_by pobj_by_convention advmod_projective_exclusively amod_convention_projective ' +o,7071,'Normally , :8 is considered a good agreement ',Carletta,'advmod_considered_Normally nsubjpass_considered_:8 auxpass_considered_is det_agreement_a amod_agreement_good dobj_considered_agreement ' +o,7072,'Recent work on this task explored a variety of methodologies to address this issue ',Johnson,'amod__Recent nn__work nsubj_explored_ prep__on det_task_this pobj_on_task det_variety_a dobj_explored_variety prep_variety_of pobj_of_methodologies aux_address_to xcomp_explored_address det_issue_this dobj_address_issue ' +o,7073,'The maximum entropy models used here are similar in form to those in ',Ratnaparkhi,'det_models_The amod_models_maximum amod_models_entropy nsubj_similar_models partmod_models_used advmod_used_here cop_similar_are prep_similar_in pobj_in_form prep_form_to pobj_to_those prep_similar_in ' +o,7074,'Our baseline is the phrase-based MT system of ',Koehn,'poss_baseline_Our nsubj_system_baseline cop_system_is det_system_the amod_system_phrase-based nn_system_MT prep_system_of ' +o,7075,'We will provide a more detailed and systematic comparison between MAXIMUM ENTROPY MODELING and MEMORY BASED LEARNING for morpho-syntactic disambiguation and we investigate whether earlier observed differences in tagging accuracy can be attributed to algorithm bias , information source issues or both ',Berger,'nsubj_provide_We aux_provide_will det_comparison_a advmod_detailed_more amod_comparison_detailed cc_detailed_and conj_detailed_systematic dobj_provide_comparison prep_comparison_between nn__MAXIMUM nn__ENTROPY nn__MODELING pobj_between_ cc__and nn__MEMORY nn__BASED nn__LEARNING conj__ prep_provide_for amod_disambiguation_morpho-syntactic pobj_for_disambiguation cc_disambiguation_and nsubj_investigate_we conj_disambiguation_investigate complm_attributed_whether advmod_differences_earlier amod_differences_observed nsubjpass_attributed_differences prep_differences_in amod_accuracy_tagging pobj_in_accuracy aux_attributed_can auxpass_attributed_be ccomp_investigate_attributed prep_attributed_to nn_bias_algorithm pobj_to_bias nn_issues_information nn_issues_source conj_bias_issues cc_bias_or conj_bias_both ' +o,7076,'So we propose forest reranking , a technique inspired by forest rescoring that approximately reranks the packed forest of exponentially many parses ',Huang,'cc_propose_So nsubj_propose_we nn_reranking_forest dobj_propose_reranking det_technique_a appos_reranking_technique partmod_technique_inspired prep_inspired_by nn__forest nn__rescoring pobj_by_ complm_reranks_that nsubj_reranks_approximately ccomp_inspired_reranks det_forest_the amod_forest_packed dobj_reranks_forest prep_forest_of amod_parses_exponentially amod_parses_many pobj_of_parses ' +o,7077,'We use the standard NIST MTEval data sets for the years 2003 , 2004 and 2005 -LRB- henceforth MT03 , MT04 and MT05 , respectively -RRB- 6 We report results in terms of case-insensitive 4gram BLEU scores ',Papineni,'nsubj_use_We det_data_the amod_data_standard nn_data_NIST nn_data_MTEval nsubj_sets_data ccomp_use_sets prep_sets_for dep_years_the pobj_for_years tmod_years_2003 num_2003_2004 dep_2003_and dep_2003_2005 amod_MT03_henceforth dep_years_MT03 conj_MT03_MT04 cc_MT03_and conj_MT03_MT05 advmod_MT03_respectively dobj_sets_6 nsubj_report_We rcmod_6_report dobj_report_results prep_results_in pobj_in_terms prep_terms_of amod_scores_case-insensitive amod_scores_4gram nn_scores_BLEU nn_scores_ pobj_of_scores ' +o,7078,'We adopted the chunk representation proposed by and used four different tags : B-NUC and B-SAT for nucleus and satellite-initial tokens , and I-NUC and I-SAT for non-initial tokens , ie , tokens inside a nucleus and satellite span ',Ramshaw,'nsubj_adopted_We det_representation_the nn_representation_chunk dobj_adopted_representation dep_representation_proposed prep_proposed_by pobj_by_ cc_proposed_and conj_proposed_used num_tags_four amod_tags_different dobj_used_tags dep_representation_B-NUC cc_B-NUC_and conj_B-NUC_B-SAT prep_B-NUC_for nn_tokens_nucleus cc_nucleus_and conj_nucleus_satellite-initial pobj_for_tokens cc_representation_and conj_representation_I-NUC cc_I-NUC_and conj_I-NUC_I-SAT prep_I-NUC_for amod_tokens_non-initial pobj_for_tokens appos_tokens_ie appos_ie_tokens prep_tokens_inside det_span_a nn_span_nucleus cc_nucleus_and conj_nucleus_satellite pobj_inside_span ' +o,7079,'Figures 1 and 2 present best results in the learning experiments for the complete set of patterns used in the collocation approach , over two of our evaluation corpora11 Type Positions Tags\\/Words Features Accuracy Precision Recall GIS 1 W 1254 097 096 098 IIS 1 T 136 095 096 094 NB 1 T 136 088 097 084 9 see Rish , 2001 , Ratnaparkhi , 1997 and et al , 1996 for a formal description of these algorithms ',Berger,'num_Figures_1 cc_Figures_and conj_Figures_2 amod_results_present dep_present_best nsubj_used_results prep_results_in det_experiments_the nn_experiments_learning pobj_in_experiments prep_experiments_for det_set_the amod_set_complete pobj_for_set prep_set_of pobj_of_patterns dep_Figures_used prep_used_in det_approach_the nn_approach_collocation pobj_in_approach prep_used_over pobj_over_two prep_two_of poss_Features_our nn_Features_evaluation nn_Features_corpora11 nn_Features_Type nn_Features_Positions nn_Features_Tags\\/Words pobj_of_Features nn_T_Accuracy nn_T_Precision nn_T_Recall nn_T_GIS num_T_1 nn_T_W num_T_1254 num_T_097 num_T_096 num_T_098 nn_T_IIS num_T_1 nn_T_T num_T_136 num_T_095 num_T_096 num_T_094 nn_T_NB num_T_1 nsubj_136_T dep_Figures_136 num_097_088 iobj_136_097 number_9_084 dobj_136_9 dep_9_see dobj_see_Rish conj_Rish_2001 conj_Rish_Ratnaparkhi conj_Rish_1997 cc_Rish_and conj_Rish_ cc__et conj__al appos_9_1996 prep_1996_for det_description_a amod_description_formal pobj_for_description prep_description_of det_algorithms_these pobj_of_algorithms ' +o,7080,'Tag test data using the POS-tagger described in ',Ratnaparkhi,'nn_data_test dobj_Tag_data partmod_data_using det_POS-tagger_the dobj_using_POS-tagger partmod_POS-tagger_described prep_described_in ' +o,7081,'In , the significance of an association -LRB- x , y -RRB- is measured by the mutual information I -LRB- x , y -RRB- , ie the probability of observing x and y together , compared with the probability of observing x and y independently ',Church,'prep_measured_In pcomp_In_ dep__ det_significance_the dobj__significance prep_significance_of det_association_an pobj_of_association dep_association_x appos_x_y auxpass_measured_is prep_measured_by det_information_the amod_information_mutual pobj_by_information nsubjpass_measured_I dep_I_x appos_x_y partmod_I_ie det_probability_the dobj_ie_probability prep_probability_of pcomp_of_observing dobj_observing_x cc_x_and conj_x_y advmod_observing_together prep_observing_compared dep_compared_with det_probability_the pobj_with_probability prep_probability_of pcomp_of_observing dobj_observing_x cc_x_and conj_x_y advmod_observing_independently ' +o,7082,'The precision rate using the lexical statistics approach can reach around 60 \% if both word bi-gram extraction and n-gram extractions are taking into account ',Smadja,'det_rate_The nn_rate_precision nsubj_\%_rate partmod_rate_using det_approach_the amod_approach_lexical nn_approach_statistics dobj_using_approach aux_\%_can dep_\%_reach prt_\%_around num_\%_60 mark_taking_if det_extraction_both nn_extraction_word nn_extraction_bi-gram nsubj_taking_extraction cc_extraction_and amod_extractions_n-gram conj_extraction_extractions aux_taking_are advcl_\%_taking prep_taking_into pobj_into_account ' +o,7083,'Many methods exist for clustering , eg , ',Brown,'amod_methods_Many nsubj_exist_methods prep_exist_for pobj_for_clustering appos_clustering_eg ' +o,7084,'The weights for these models are determined using the method described in ',Och,'det_weights_The nsubjpass_determined_weights prep_weights_for det_models_these pobj_for_models auxpass_determined_are xcomp_determined_using det_method_the dobj_using_method partmod_method_described prep_described_in ' +o,7085,'TopSense is tested on 20 words extensively investigated in recent WSD literature ',Yarowsky,'nsubjpass_tested_TopSense auxpass_tested_is prep_tested_on num_words_20 pobj_on_words advmod_investigated_extensively partmod_words_investigated prep_investigated_in amod_literature_recent nn_literature_WSD pobj_in_literature ' +o,7086,'But there is also extensive research focused on including linguistic knowledge in metrics among others ',Owczarzak,'cc_is_But expl_is_there advmod_is_also amod_research_extensive nsubj_is_research partmod_research_focused prt_focused_on prep_focused_including amod_knowledge_linguistic pobj_including_knowledge prep_knowledge_in pobj_in_metrics partmod_metrics_ prep__among pobj_among_others ' +o,7087,' studied a method for word sense disambiguation using unlabeled data ',Yarowsky,'nsubj_studied_ det_method_a dobj_studied_method prep_method_for nn_disambiguation_word nn_disambiguation_sense pobj_for_disambiguation partmod_disambiguation_using amod_data_unlabeled dobj_using_data ' +o,7088,'Obtaining a word-aligned corpus usually involves training a word-based translation models in each directions and combining the resulting alignments ',Brown,'det_corpus_a amod_corpus_word-aligned dobj_Obtaining_corpus advmod_involves_usually dep_Obtaining_involves dobj_involves_training det_models_a amod_models_word-based nn_models_translation nsubj_Obtaining_models dep_models_ prep__in det_directions_each pobj_in_directions cc__and conj__combining det_alignments_the amod_alignments_resulting dobj_combining_alignments ' +o,7089,'The other utilizes a sort of parallel texts , such as multiple translation of the same text , corresponding articles from multiple news sources , and bilingual corpus ',Barzilay,'det_utilizes_The amod_utilizes_other det_sort_a dep_utilizes_sort prep_sort_of amod_texts_parallel pobj_of_texts dep_as_such prep_sort_as amod_translation_multiple pobj_as_translation prep_translation_of det__the amod__same nn__text pobj_of_ amod_articles_corresponding conj__articles prep_articles_from amod__multiple nn__news nn__sources pobj_from_ cc__and amod__bilingual nn__corpus conj__ ' +o,7090,'For ROUGE-S and ROUGE-SU , we use three variations following : the maximum skip distances are 4 , 9 and infinity 7 ',Lin,'prep_use_For pobj_For_ROUGE-S cc_ROUGE-S_and conj_ROUGE-S_ROUGE-SU nsubj_use_we num_variations_three dobj_use_variations prep_variations_following pobj_following_ det_distances_the amod_distances_maximum amod_distances_skip nsubj_4_distances cop_4_are parataxis_use_4 conj_4_9 cc_4_and conj_4_infinity num_infinity_7 ' +p,7091,' saw a LFMS improvement of 08 \% over their baseline discriminative parser after adding punctuation features , one of which encoded the sentence-final punctuation ',Collins,'nsubj_saw_ det_improvement_a nn_improvement_LFMS dobj_saw_improvement prep_improvement_of num_\%_08 pobj_of_\% prep_\%_over poss_parser_their amod_parser_baseline nn_parser_discriminative pobj_over_parser prep_saw_after pcomp_after_adding nn_features_punctuation dobj_adding_features nsubj_encoded_one prep_one_of pobj_of_which rcmod_features_encoded det_punctuation_the amod_punctuation_sentence-final dobj_encoded_punctuation ' +o,7092,'Finally , other approaches rely on reviews with numeric ratings from websites and train -LRB- semi - -RRB- supervised learning algorithms to classify reviews as positive or negative , or in more fine-grained scales ',Pang,'advmod_rely_Finally amod_approaches_other nsubj_rely_approaches prep_rely_on pobj_on_reviews prep_reviews_with amod_ratings_numeric pobj_with_ratings prep_ratings_from pobj_from_websites dobj_rely_ cc__and conj__train dep_-LRB-_semi partmod__supervised xcomp_supervised_learning dobj_learning_algorithms aux_classify_to xcomp_learning_classify nsubj_positive_reviews advmod_positive_as xcomp_classify_positive cc_positive_or conj_positive_negative cc__or conj__in amod_scales_more amod_scales_fine-grained pobj_in_scales ' +o,7093,'The first one makes use of the advances in the parsing technology or on the availability of large parsed corpora -LRB- eg Trcebank -RRB- to produce algorithms inspired by Hobbs \' baseline method ',Marcus,'det_one_The amod_one_first nsubj_makes_one dobj_makes_use prep_use_of det_advances_the pobj_of_advances prep_makes_in det_technology_the amod_technology_parsing pobj_in_technology cc_in_or conj_in_on det_availability_the pobj_on_availability prep_availability_of amod_corpora_large amod_corpora_parsed pobj_of_corpora nn__eg nn__Trcebank appos_corpora_ aux_produce_to infmod_availability_produce dobj_produce_algorithms partmod_algorithms_inspired prep_inspired_by poss__Hobbs amod__baseline nn__method pobj_by_ ' +o,7094,'The probabilities are ordered according to , at least my , intuition with pronoun being the most likely , followed by proper nouns , followed by common nouns , a fact also noted by ',Haghighi,'det_probabilities_The nsubjpass_ordered_probabilities auxpass_ordered_are prep_ordered_according dep_according_to dep_my_at dep_at_least dep_intuition_my pobj_to_intuition prep_ordered_with pobj_with_pronoun cop__being det__the advmod_likely_most amod__likely partmod_pronoun_ partmod__followed prep_followed_by amod__proper nn__nouns pobj_by_ partmod__followed prep_followed_by amod_nouns_common pobj_by_nouns advmod_noted_ det_fact_a nsubj_noted_fact advmod_noted_also dep_ordered_noted prep_noted_by ' +o,7095,'Therefore , structure divergence and parse errors are two of the major issues that may largely compromise the performance of syntax-based SMT ',Zhang,'advmod_two_Therefore nn_divergence_structure nsubj_two_divergence cc_divergence_and nn_errors_parse conj_divergence_errors cop_two_are prep_two_of det_issues_the amod_issues_major pobj_of_issues nsubj_compromise_that aux_compromise_may advmod_compromise_largely rcmod_issues_compromise det_performance_the dobj_compromise_performance prep_performance_of amod_SMT_syntax-based pobj_of_SMT ' +n,7096,'2 Previous Work It is helpful to compare this approach with recent efforts in statistical MT Phrase-based models are good at learning local translations that are pairs of -LRB- consecutive -RRB- sub-strings , but often insufficient in modeling the reorderings of phrases themselves , especially between language pairs with very different word-order ',Koehn,'num_Work_2 amod_Work_Previous nsubj_helpful_It cop_helpful_is dep_Work_helpful aux_compare_to xcomp_helpful_compare det_approach_this dobj_compare_approach prep_compare_with amod_efforts_recent pobj_with_efforts prep_efforts_in amod_models_statistical nn_models_MT nn_models_Phrase-based pobj_in_models nsubj_good_ cop_good_are dep_Work_good prep_good_at pcomp_at_learning amod_translations_local dobj_learning_translations nsubj_pairs_that cop_pairs_are rcmod_translations_pairs prep_pairs_of dep_sub-strings_consecutive pobj_of_sub-strings cc_good_but advmod_good_often dep_good_insufficient prep_insufficient_in pcomp_in_modeling det_reorderings_the dobj_modeling_reorderings prep_reorderings_of pobj_of_phrases advmod_phrases_themselves advmod_reorderings_especially dep_reorderings_between nn_pairs_language dep_reorderings_pairs prep_pairs_with advmod_different_very amod_word-order_different pobj_with_word-order ' +o,7097,'Work focusses on analyzing subjective features of text or speech , such as sentiment , opinion , emotion or point of view ',Pang,'dobj_Work_focusses prep_Work_on pcomp_on_analyzing amod_features_subjective dobj_analyzing_features prep_features_of pobj_of_text cc_text_or conj_text_speech dep_as_such prep_features_as pobj_as_sentiment conj_sentiment_opinion conj_sentiment_emotion cc_sentiment_or conj_sentiment_point prep_sentiment_of pobj_of_view ' +o,7098,'The NIST BLEU-4 is a variant of BLEU and is computed as a49a51a50 a2a16a52a53a6 a0a9a8a10a0a12a11a54a13a55a15 a26a57a56a33a58a60a59 a43 a61a63a62 a64 a65a67a66a69a68 a28a71a70a46a72a74a73 a65 a6 a0a9a8a10a0a3a11a54a13a19a75a77a76 a6 a0a9a8a10a0a3a11a54a13 -LRB- 2 -RRB- where a73 a65 a6 a0a78a8a10a0a3a11a54a13 is the precision of a79 - grams in the hypothesis a0 given the reference a0 a11 and a76 a6 a0a78a8a10a0a3a11a54a13a81a80 a43 is a brevity penalty ',Papineni,'det_BLEU-4_The nn_BLEU-4_NIST nsubj_variant_BLEU-4 cop_variant_is det_variant_a prep_variant_of amod_a0a9a8a10a0a3a11a54a13_BLEU ccomp_BLEU_ cc__and auxpass_computed_is conj__computed prep_computed_as amod_a2a16a52a53a6_a49a51a50 pobj_as_a2a16a52a53a6 amod_a0a9a8a10a0a3a11a54a13_a0a9a8a10a0a12a11a54a13a55a15 amod_a0a9a8a10a0a3a11a54a13_a26a57a56a33a58a60a59 amod_a0a9a8a10a0a3a11a54a13_a43 amod_a0a9a8a10a0a3a11a54a13_a61a63a62 amod_a0a9a8a10a0a3a11a54a13_a64 amod_a0a9a8a10a0a3a11a54a13_a65a67a66a69a68 amod_a0a9a8a10a0a3a11a54a13_a28a71a70a46a72a74a73 amod_a0a9a8a10a0a3a11a54a13_a65 amod_a0a9a8a10a0a3a11a54a13_a6 amod_a0a9a8a10a0a3a11a54a13_a0a9a8a10a0a3a11a54a13a19a75a77a76 amod_a0a9a8a10a0a3a11a54a13_a6 pobj_of_a0a9a8a10a0a3a11a54a13 appos_a0a9a8a10a0a3a11a54a13_2 advmod_precision_where amod_a0a78a8a10a0a3a11a54a13_a73 amod_a0a78a8a10a0a3a11a54a13_a65 nn_a0a78a8a10a0a3a11a54a13_a6 nsubj_precision_a0a78a8a10a0a3a11a54a13 cop_precision_is det_precision_the ccomp_a0a9a8a10a0a3a11a54a13_precision prep_precision_of pobj_of_a79 nsubj_penalty_grams prep_grams_in det_a0_the nn_a0_hypothesis pobj_in_a0 partmod_a0_given det_reference_the iobj_given_reference dobj_given_a0 dep_a0_a11 cc_a0_and amod_a43_a76 amod_a43_a6 nn_a43_a0a78a8a10a0a3a11a54a13a81a80 conj_a0_a43 cop_penalty_is det_penalty_a amod_penalty_brevity parataxis_variant_penalty ' +o,7099,'1 Introduction Inversion transduction grammar -LRB- ITG -RRB- constraints provide coherent structural constraints on the relationship between a sentence and its translation ',Wu,'num_grammar_1 nn_grammar_Introduction nn_grammar_Inversion nn_grammar_transduction nn_constraints_grammar abbrev_constraints_ITG nsubj__constraints xcomp__provide amod_constraints_coherent amod_constraints_structural dobj_provide_constraints prep_provide_on det_relationship_the pobj_on_relationship prep_relationship_between det_sentence_a pobj_between_sentence cc_sentence_and poss_translation_its conj_sentence_translation ' +o,7100,'Examples of this are bilexical grammars -- such as Eisner and Satta , Charniak , -- where the lexical heads of each constituent are annotated on both the rightand left-hand sides of the context-free rules , under the constraint that every constituent inherits the lexical head from exactly one of its children , and the lexical head of a POS is its terminal item ',Collins,'nsubj_grammars_Examples prep_Examples_of pobj_of_this cop_grammars_are amod_grammars_bilexical dep_as_such dep__as pobj_as_Eisner cc_Eisner_and nn__Satta conj_Eisner_ nn__Charniak dep_grammars_ nn__ appos__ advmod_annotated_where det_heads_the amod_heads_lexical nsubjpass_annotated_heads prep_heads_of det_constituent_each pobj_of_constituent auxpass_annotated_are rcmod_grammars_annotated prep_annotated_on predet_sides_both det_sides_the amod_sides_rightand amod_sides_left-hand pobj_on_sides prep_sides_of det_rules_the amod_rules_context-free pobj_of_rules prep_annotated_under det_constraint_the pobj_under_constraint complm_inherits_that det_constituent_every nsubj_inherits_constituent dep_constraint_inherits det_head_the amod_head_lexical dobj_inherits_head prep_inherits_from advmod_one_exactly pobj_from_one prep_one_of poss_children_its pobj_of_children cc_grammars_and det_head_the amod_head_lexical nsubj_item_head prep_head_of det_POS_a pobj_of_POS cop_item_is poss_item_its amod_item_terminal conj_grammars_item ' +o,7101,'For a second set of parsing experiments , we used the WSJ portion of the Penn Tree Bank and Helmut Schmids enrichment program tmod ',Marcus,'prep_used_For det_set_a amod_set_second pobj_For_set prep_set_of amod_experiments_parsing pobj_of_experiments nsubj_used_we det_portion_the nn_portion_WSJ dobj_used_portion prep_portion_of det__the nn__Penn nn__Tree nn__Bank pobj_of_ cc__and nn__Helmut nn__Schmids nn__enrichment nn__program amod__tmod conj__ ' +p,7102,'The current state of the art is represented by the so-called phrase-based translation approach ',Koehn,'det_state_The amod_state_current nsubjpass_represented_state prep_state_of det_art_the pobj_of_art auxpass_represented_is prep_represented_by det_approach_the amod_approach_so-called amod_approach_phrase-based nn_approach_translation pobj_by_approach ' +n,7103,'When compared to other kernel methods , our approach performs better than those based on the Tree kernel , and is only 02 \% worse than the best results achieved by a kernel method for parsing ',Collins,'advmod_performs_When prep_performs_compared dep_compared_to amod_methods_other nn_methods_kernel pobj_to_methods poss_approach_our nsubj_performs_approach advmod_performs_better mark_based_than nsubj_based_those dep_better_based prep_based_on det__the nn__Tree nn__kernel pobj_on_ cc_performs_and cop_worse_is quantmod_02_only num_\%_02 measure_worse_\% conj_performs_worse prep_worse_than det_results_the amod_results_best pobj_than_results partmod_results_achieved prep_achieved_by det_method_a nn_method_kernel pobj_by_method prep_method_for amod__parsing pobj_for_ ' +o,7104,'To this end we follow the method introduced by , ie by sliding a window of a given size over some texts ',Church,'prep_follow_To det_end_this pobj_To_end nsubj_follow_we det_method_the nsubj_introduced_method ccomp_follow_introduced prep_introduced_by pobj_by_ appos__ie prep_introduced_by pcomp_by_sliding det_window_a dobj_sliding_window prep_window_of det_size_a amod_size_given pobj_of_size prep_size_over det_texts_some pobj_over_texts ' +o,7105,'The MBT POS tagger is used to provide POS information ',Berger,'det__The nn__MBT dep__POS nn__tagger nsubjpass_used_ auxpass_used_is aux_provide_to xcomp_used_provide dep_information_POS dobj_provide_information ' +o,7106,'In this paper , we modify the method in Albrecht and Hwa to only prepare human reference translations for the training examples , and then evaluate the translations produced by the subject systems against the references using BLEU score ',Papineni,'prep_modify_In det_paper_this pobj_In_paper nsubj_modify_we det_method_the dobj_modify_method prep_modify_in nn__Albrecht cc_Albrecht_and conj_Albrecht_Hwa pobj_in_ aux_prepare_to advmod_prepare_only xcomp_modify_prepare amod_translations_human nn_translations_reference dobj_prepare_translations prep_translations_for det_examples_the nn_examples_training pobj_for_examples cc_prepare_and advmod_prepare_then conj_prepare_evaluate det_translations_the dobj_evaluate_translations partmod_translations_produced prep_produced_by det_systems_the nn_systems_subject pobj_by_systems prep_systems_against det_references_the pobj_against_references xcomp_produced_using amod_score_BLEU dobj_using_score ' +o,7107,'The model weights are trained using the standard ranking perceptron ',Collins,'det_weights_The nn_weights_model nsubjpass_trained_weights auxpass_trained_are xcomp_trained_using det_perceptron_the amod_perceptron_standard amod_perceptron_ranking dobj_using_perceptron ' +o,7108,'Carletta suggests that content analysis researchers consider K -RRB- 8 as good reliability , with67 -LRB- \\/ ~ \' -LRB- 8 allowing tentative conclusions to be drawn ',Carletta,'nsubj_suggests_Carletta complm_consider_that nn_researchers_content nn_researchers_analysis nsubj_consider_researchers ccomp_suggests_consider dep_8_K dobj_consider_8 prep_8_as amod_reliability_good pobj_as_reliability amod_~_with67 amod_~_\\/ appos_8_~ number_allowing_8 dep_~_allowing amod_conclusions_tentative dep_allowing_conclusions aux_drawn_to auxpass_drawn_be xcomp_consider_drawn ' +p,7109,'In addition to the classical window-based technique , some studies investigated the use of lexico-syntactic patterns -LRB- eg , X or Y -RRB- to get more accurate co-occurrence statistics ',Bollegala,'prep_investigated_In pobj_In_addition prep_addition_to det_technique_the amod_technique_classical amod_technique_window-based pobj_to_technique det_studies_some nsubj_investigated_studies det_use_the dobj_investigated_use prep_use_of amod_patterns_lexico-syntactic pobj_of_patterns appos_patterns_eg dep_eg_X cc_X_or conj_X_Y aux_get_to xcomp_investigated_get advmod_accurate_more amod_statistics_accurate amod_statistics_co-occurrence dobj_get_statistics ' +o,7110,'-LRB- 1 -RRB- provides evidence that should be chosen by optimizing an objective function basd on the evaluation metric of interest , rather than likelihood ',Och,'dep_provides_1 nsubj_provides_ dobj_provides_evidence nsubjpass_chosen_that aux_chosen_should auxpass_chosen_be dep_provides_chosen prep_chosen_by pcomp_by_optimizing det_function_an amod_function_objective dobj_optimizing_function partmod_function_basd prep_basd_on det_metric_the nn_metric_evaluation pobj_on_metric prep_metric_of pobj_of_interest cc_function_rather dep_rather_than conj_function_likelihood ' +o,7111,'2 Related Work Given its potential usefulness in coreference resolution , anaphoricity determination has been studied fairly extensively in the literature and can be classified into three categories : heuristic rule-based , statistics-based and learning-based ',Bean,'num_Work_2 amod_Work_Related prep_studied_Given poss_usefulness_its amod_usefulness_potential dep_Given_usefulness prep_usefulness_in amod_resolution_coreference pobj_in_resolution amod_determination_anaphoricity nsubjpass_studied_determination aux_studied_has auxpass_studied_been dep_Work_studied advmod_studied_fairly dep_in_extensively prep_studied_in det_literature_the pobj_in_literature cc_studied_and aux_classified_can auxpass_classified_be conj_studied_classified prep_classified_into num_categories_three pobj_into_categories amod__heuristic amod__rule-based dep_categories_ amod__statistics-based conj__ cc__and amod__learning-based conj__ ' +o,7112,'Proceedings of the 40th Annual Meeting of the Association for cently , semantic resources have also been used in collocation discovery , smoothing and model estimation and text classi cation ',Brown,'nsubjpass_used_Proceedings prep_Proceedings_of det_Meeting_the amod_Meeting_40th amod_Meeting_Annual pobj_of_Meeting prep_Meeting_of det_Association_the pobj_of_Association prep_Association_for amod_resources_cently amod_resources_semantic pobj_for_resources aux_used_have advmod_used_also auxpass_used_been prep_used_in amod__collocation nn__discovery nn__ conj__smoothing cc__and conj__model nn__estimation pobj_in_ cc__and nn__text nn__classi nn__cation conj__ ' +o,7113,'3TheData For our experiments we used a version of the British National Corpus parsed with the statistical parser of ',Collins,'dep_3TheData_For poss_experiments_our pobj_For_experiments nsubj_used_we rcmod_experiments_used det_version_a dobj_used_version prep_version_of det_Corpus_the nn_Corpus_British nn_Corpus_National pobj_of_Corpus partmod_Corpus_parsed prep_parsed_with det_parser_the amod_parser_statistical pobj_with_parser prep_parser_of ' +o,7114,'1 Introduction During the last decade , statistical machine translation -LRB- SMT -RRB- systems have evolved from the original word-based approach into phrase-based translation systems ',Koehn,'num_Introduction_1 dep_Introduction_During det_decade_the amod_decade_last pobj_During_decade amod_translation_statistical nn_translation_machine appos_decade_translation abbrev_translation_SMT nsubj_evolved_systems aux_evolved_have rcmod_translation_evolved prep_evolved_from det__the amod__original amod__word-based nn__approach pobj_from_ prep_evolved_into amod_systems_phrase-based nn_systems_translation pobj_into_systems ' +o,7115,'In terms of relative performance , Naive Bayes tends to do the worst and SVMs tend to do the best , although the 12http : \\/ \\/ wwwenglishbhamacuk\\/stafi\\/oliver \\/ software\\/tagger\\/indexhtm 13 unsupervised algorithm uses bigrams containing an adjective or an adverb ',Turney,'prep_tends_In pobj_In_terms prep_terms_of amod_performance_relative pobj_of_performance nn_Bayes_Naive nsubj_tends_Bayes aux_do_to xcomp_tends_do det_worst_the nsubj_tend_worst cc_worst_and conj_worst_SVMs ccomp_do_tend aux_do_to xcomp_tend_do det_best_the dobj_do_best mark_unsupervised_although det_software\\/tagger\\/indexhtm_the nn_software\\/tagger\\/indexhtm_12http nn_software\\/tagger\\/indexhtm_\\/ amod_software\\/tagger\\/indexhtm_\\/ nn_software\\/tagger\\/indexhtm_wwwenglishbhamacuk\\/stafi\\/oliver nn_software\\/tagger\\/indexhtm_\\/ nsubj_unsupervised_software\\/tagger\\/indexhtm num__13 dep_software\\/tagger\\/indexhtm_ advcl_tends_unsupervised nn_uses_algorithm nsubj_bigrams_uses ccomp_unsupervised_bigrams xcomp_bigrams_containing det_adjective_an dobj_containing_adjective cc_adjective_or det_adverb_an conj_adjective_adverb ' +o,7116,'Thus , we used the five taggers , MBL , MXPOST , fnTBL , TnT , and IceTagger3 , in the same manner as described in , but with the following minor changes ',Ratnaparkhi,'advmod_used_Thus nsubj_used_we det_taggers_the num_taggers_five dobj_used_taggers nn__MBL conj_taggers_ nn__MXPOST conj_taggers_ amod__fnTBL conj_taggers_ conj_taggers_TnT cc_taggers_and conj_taggers_IceTagger3 conj_used_in det_manner_the amod_manner_same pobj_in_manner advmod_described_as amod_manner_described prep_manner_in pobj_in_ cc_used_but conj_used_with det_changes_the amod_changes_following amod_changes_minor pobj_with_changes ' +p,7117,'Among the various knowledge-based and data-driven word sense disambiguation methods that have been proposed to date , supervised systems have been constantly observed as leading to the highest performance ',Yarowsky,'prep_observed_Among det__the amod__various amod__knowledge-based pobj_Among_ cc__and amod_methods_data-driven nn_methods_ nn_methods_word nn_methods_sense nn_methods_disambiguation conj__methods nsubjpass_proposed_that aux_proposed_have auxpass_proposed_been rcmod_methods_proposed prep_proposed_to pobj_to_date amod_systems_supervised nsubjpass_observed_systems aux_observed_have auxpass_observed_been advmod_observed_constantly prep_observed_as pcomp_as_leading prep_leading_to det_performance_the amod_performance_highest pobj_to_performance ' +o,7118,'In this case , one is often required to find the translation -LRB- s -RRB- in the hypergraph that are most similar to the desired translations , with similarity computed via some automatic metric such as BLEU ',Papineni,'prep_required_In det_case_this pobj_In_case nsubjpass_required_one auxpass_required_is advmod_required_often aux_find_to xcomp_required_find det_translation_the dobj_find_translation dep_-LRB-_s prep_find_in det_hypergraph_the pobj_in_hypergraph nsubj_similar_that cop_similar_are advmod_similar_most rcmod_hypergraph_similar prep_similar_to det_translations_the amod_translations_desired pobj_to_translations prep_required_with pobj_with_similarity partmod_similarity_computed prep_computed_via det_metric_some amod_metric_automatic pobj_via_metric dep_as_such prep_metric_as pobj_as_BLEU ' +p,7119,'Discriminative taggers and chunkers have been the state-of-the-art for more than a decade ',Ratnaparkhi,'amod_taggers_Discriminative nsubj_been_taggers cc_taggers_and conj_taggers_chunkers aux_been_have det_state-of-the-art_the dobj_been_state-of-the-art prep_state-of-the-art_for dep_than_more quantmod_a_than num_decade_a pobj_for_decade ' +o,7120,'4 Maxilnum Entropy The model used here for sentence-boundary detection is based on the maximum entropy model used for POS tagging in ',Ratnaparkhi,'num_Entropy_4 nn_Entropy_Maxilnum det_model_The nsubjpass_based_model partmod_model_used advmod_used_here prep_used_for amod_detection_sentence-boundary pobj_for_detection auxpass_based_is dep_Entropy_based prep_based_on det_model_the amod_model_maximum nn_model_entropy pobj_on_model partmod_model_used prep_used_for pobj_for_POS partmod_POS_tagging prt_tagging_in ' +o,7121,'Using techniques described in , Church and Hanks -LRB- 1990 -RRB- , and Hindle and Rooth -LRB- 1991 -RRB- , Figure 4 shows some examples of the most frequent V-O pairs from the AP corpus ',Church,'dep_shows_Using dobj_Using_techniques partmod_techniques_described prep_described_in pobj_in_ conj__Church cc__and conj__Hanks appos_techniques_1990 cc_techniques_and conj_techniques_Hindle cc_Hindle_and conj_Hindle_Rooth appos_Hindle_1991 nsubj_shows_Figure num_Figure_4 det_examples_some dobj_shows_examples prep_examples_of det_pairs_the advmod_frequent_most amod_pairs_frequent amod_pairs_V-O pobj_of_pairs prep_pairs_from det_corpus_the nn_corpus_AP pobj_from_corpus ' +o,7122,'As mentioned earlier , both of these methods are based on Collinss averaged-perceptron algorithm for sequence labeling ',Collins,'mark_mentioned_As advcl_based_mentioned advmod_mentioned_earlier nsubjpass_based_both prep_both_of det_methods_these pobj_of_methods auxpass_based_are prep_based_on nn_algorithm_Collinss amod_algorithm_averaged-perceptron pobj_on_algorithm prep_based_for nn_labeling_sequence pobj_for_labeling ' +o,7123,'32 F-Structure Based NLD Recovery presented a NLD recovery algorithm operating at LFG f-structure for treebankbased LFG approximations ',Cahill,'num_F-Structure_32 dep_F-Structure_Based nn__NLD nn__Recovery dep_Based_ partmod__presented det_operating_a nn_operating_NLD nn_operating_recovery nn_operating_algorithm dobj_presented_operating prep_presented_at nn_f-structure_LFG pobj_at_f-structure prep_f-structure_for amod_approximations_treebankbased nn_approximations_LFG pobj_for_approximations ' +o,7124,'5 SMT Experiments 51 Experimental Setup We used publicly available resources for all our tests : for decoding we used Moses and our parallel data was taken from the Spanish-English section of Europarl ',Koehn,'num_SMT_5 nsubj_Experiments_SMT iobj_Experiments_51 nn_Setup_Experimental dobj_Experiments_Setup nsubj_used_We rcmod_Setup_used advmod_available_publicly amod_resources_available dobj_used_resources prep_used_for predet_tests_all poss_tests_our pobj_for_tests prep_Experiments_for pcomp_for_decoding nsubj_used_we ccomp_decoding_used nn__Moses nsubjpass_taken_ cc__and poss_data_our amod_data_parallel conj__data auxpass_taken_was ccomp_used_taken prep_taken_from det_section_the amod_section_Spanish-English pobj_from_section prep_section_of pobj_of_Europarl ' +o,7125,' and used comparable news articles to obtain sentence level paraphrases ',Barzilay,'nsubj_used_ cc__and conj__ amod_articles_comparable nn_articles_news dobj_used_articles aux_obtain_to xcomp_used_obtain nn_paraphrases_sentence nn_paraphrases_level dobj_obtain_paraphrases ' +o,7126,'Several techniques and results have been reported on learning subcategorization frames -LRB- SFs -RRB- from text corpora ',Dunning,'amod_techniques_Several nsubjpass_reported_techniques cc_techniques_and conj_techniques_results aux_reported_have auxpass_reported_been prep_reported_on pcomp_on_learning nn_frames_subcategorization dobj_learning_frames abbrev_frames_SFs prep_learning_from nn__text nn__corpora pobj_from_ ' +o,7127,'To find these pairs automatically , wetrainedanon-sequentiallog-linearmodel that achieves a 902 accuracy ',Lin,'aux_find_To dep__find det_pairs_these dobj_find_pairs advmod_find_automatically nsubj__wetrainedanon-sequentiallog-linearmodel nsubj_achieves_that rcmod_wetrainedanon-sequentiallog-linearmodel_achieves det_accuracy_a num_accuracy_902 dobj_achieves_accuracy ' +o,7128,'As with the graph-based parser , we use the discriminative perceptron to train the transition-based model -LRB- see Figure 5 -RRB- ',Collins,'prep_use_As dep_As_with det_parser_the amod_parser_graph-based pobj_with_parser nsubj_use_we det_perceptron_the amod_perceptron_discriminative dobj_use_perceptron partmod_perceptron_ aux_train_to xcomp__train det_model_the amod_model_transition-based dobj_train_model parataxis_use_see dobj_see_Figure num_Figure_5 ' +o,7129,'Consequently , semi-supervised learning , which combines both labeled and unlabeled data , has been applied to some NLP tasks such as word sense disambiguation , classification , clustering , named entity classification , and parsing ',Yarowsky,'advmod_applied_Consequently amod_learning_semi-supervised nsubjpass_applied_learning nsubj_combines_which rcmod_learning_combines preconj_labeled_both amod_data_labeled cc_labeled_and conj_labeled_unlabeled dobj_combines_data aux_applied_has auxpass_applied_been prep_applied_to det_tasks_some nn_tasks_NLP pobj_to_tasks dep_as_such prep_tasks_as nn__word nn__sense nn__disambiguation pobj_as_ nn__classification appos__ amod__clustering dep_clustering_ conj__named nn__entity nn__classification dobj_named_ cc__and amod__parsing conj__ ' +o,7130,'1 Introduction B (Papineni et al., 2002) was one of the first automatic evaluation metrics for machine translation (MT), and despite being challenged by a number of alternative metrics (Melamed et al., 2003; Banerjee and Lavie, 2005; Snover et al., 2006; Chan and Ng, 2008), it remains the standard in the statistical MTliterature.Callison-Burchetal.(2006)havesubjected B to a searching criticism, with two realworld case studies of significant failures of correlation between B and human adequacy/fluency judgments.Bothcasesinvolvecomparisonsbetween statistical MT systems and other translation methods (human post-editing and a rule-based MT system), and they recommend that the use of B be restrictedtocomparisonsbetweenrelatedsystemsor different versions of the same systems.',Banerjee,'' +o,7131,'The supertagger uses a log-linear model to define a distribution over the lexical category set for each word and the previous two categories and the forward backward algorithm efficiently sums over all histories to give a distribution for each word ',Ratnaparkhi,'det_supertagger_The nsubj_uses_supertagger det_model_a amod_model_log-linear dobj_uses_model aux_define_to infmod_model_define det_distribution_a dobj_define_distribution prep_define_over det_category_the amod_category_lexical pobj_over_category partmod_category_set prep_set_for det_word_each pobj_for_word cc_word_and det_categories_the amod_categories_previous num_categories_two conj_word_categories nsubj_sums_ cc__and det_algorithm_the advmod_algorithm_forward amod_algorithm_backward conj__algorithm advmod_sums_efficiently dep_uses_sums prep_sums_over det_histories_all pobj_over_histories aux_give_to xcomp_sums_give det_distribution_a dobj_give_distribution prep_distribution_for det_word_each pobj_for_word ' +o,7132,'The reader is referred to and for details of MI clustering , but we will first briefly summarize the MI clustering and then describe our hierarchical clustering algorithm ',Brown,'det_reader_The nsubjpass_referred_reader auxpass_referred_is prep_referred_to pobj_to_ cc__and conj__ prep_referred_for pobj_for_details prep_details_of nn_clustering_MI pobj_of_clustering cc_referred_but nsubj_summarize_we aux_summarize_will advmod_summarize_first dep_first_briefly conj_referred_summarize det_clustering_the nn_clustering_MI dobj_summarize_clustering cc_referred_and nsubj_describe_then conj_referred_describe poss_algorithm_our amod_algorithm_hierarchical nn_algorithm_clustering dobj_describe_algorithm ' +o,7133,'However , they can be usefully employed during system development , for example , for quickly assessing modeling ideas or for comparing across different system configurations ',Papineni,'advmod_employed_However nsubjpass_employed_they aux_employed_can auxpass_employed_be advmod_employed_usefully prep_employed_during nn_development_system pobj_during_development prep_employed_for pobj_for_example conj_for_for advmod_assessing_quickly pcomp_for_assessing nn_ideas_modeling dobj_assessing_ideas cc_for_or conj_for_for pcomp_for_comparing prep_comparing_across amod_configurations_different nn_configurations_system pobj_across_configurations ' +o,7134,'We follow the method used by , which encodes the matching with a gazetteer entity using IOB tags , with the modication for Japanese ',Kazama,'nsubj_follow_We det_method_the dobj_follow_method partmod_method_used prep_used_by pobj_by_ nsubj_encodes_which rcmod__encodes det_matching_the dobj_encodes_matching prep_encodes_with det_entity_a nn_entity_gazetteer pobj_with_entity partmod_entity_using nn_tags_IOB dobj_using_tags prep_using_with det_modication_the pobj_with_modication prep_modication_for pobj_for_Japanese ' +p,7135,'To alleviate this effort , various semi-supervised learning algorithms such as self-training , cotraining , transductive SVM and many others have been proposed and successfully applied under different assumptions and settings ',Yarowsky,'aux_alleviate_To dep_proposed_alleviate det_effort_this dobj_alleviate_effort amod_algorithms_various amod_algorithms_semi-supervised nn_algorithms_learning nsubjpass_proposed_algorithms dep_as_such prep_algorithms_as amod__self-training pobj_as_ amod__cotraining conj__ amod__transductive nn__SVM conj__ cc__and amod_others_many conj__others aux_proposed_have auxpass_proposed_been cc_proposed_and advmod_applied_successfully conj_proposed_applied prep_applied_under amod_assumptions_different pobj_under_assumptions cc_assumptions_and conj_assumptions_settings ' +n,7136,'Lexical relationships under the standard IBM models do not account for many-to-many mappings , and phrase extraction relies heavily on the accuracy of the IBM word-toword alignment ',Brown,'amod_relationships_Lexical nsubj_account_relationships prep_relationships_under det__the amod__standard nn__IBM nn__models pobj_under_ aux_account_do neg_account_not prep_account_for amod_mappings_many-to-many pobj_for_mappings cc_account_and nn_extraction_phrase nsubj_relies_extraction conj_account_relies advmod_relies_heavily prep_relies_on det_accuracy_the pobj_on_accuracy prep_accuracy_of det_alignment_the nn_alignment_IBM amod_alignment_word-toword pobj_of_alignment ' +o,7137,'We tuned the parameters of these features with Minimum Error Rate Training -LRB- MERT -RRB- on the NIST MT03 Evaluation data set -LRB- 919 sentences -RRB- , and then test the MT performance on NIST MT03 and MT05 Evaluation data ',Och,'nsubj_tuned_We det_parameters_the dobj_tuned_parameters prep_parameters_of det_features_these pobj_of_features prep_tuned_with nn__Minimum nn__Error nn__Rate nn__Training abbrev__MERT pobj_with_ prep__on det_set_the nn_set_NIST nn_set_MT03 nn_set_Evaluation nn_set_data pobj_on_set num_sentences_919 appos_set_sentences cc_tuned_and advmod_test_then conj_tuned_test det_performance_the nn_performance_MT dobj_test_performance prep_performance_on amod_MT03_NIST pobj_on_MT03 cc_MT03_and nn__MT05 nn__Evaluation nn__data conj_MT03_ ' +o,7138,'This source is very important for repairs that do not have initial retracing , and is the mainstay of the ` parser-first \' approach -- keep trying alternative corrections until one of them parses ',Marcus,'det_source_This nsubj_important_source cop_important_is advmod_important_very prep_important_for pobj_for_repairs nsubj_have_that aux_have_do neg_have_not rcmod_repairs_have amod_retracing_initial dobj_have_retracing cc_important_and cop_mainstay_is det_mainstay_the conj_important_mainstay prep_mainstay_of det_approach_the amod_approach_parser-first pobj_of_approach partmod_approach_ parataxis__keep xcomp_keep_trying amod_corrections_alternative dobj_trying_corrections mark_parses_until nsubj_parses_one prep_one_of pobj_of_them advcl__parses ' +o,7139,'glish nouns first appeared in ',Hindle,'amod_nouns_glish nsubj_appeared_nouns advmod_appeared_first prt_appeared_in ' +o,7140,'For the English experiments , we use the now-standard training and test sets that were introduced in 2 ',Ramshaw,'prep_use_For det_experiments_the amod_experiments_English pobj_For_experiments nsubj_use_we det_training_the amod_training_now-standard dobj_use_training cc_training_and nn_sets_test conj_training_sets nsubjpass_introduced_that auxpass_introduced_were rcmod_training_introduced prep_introduced_in pobj_in_ num__2 ' +o,7141,'Barzilay and Lee learned paraphrasing patterns as pairs of word lattices , which are then used to produce sentence level paraphrases ',Barzilay,'nn__Barzilay cc_Barzilay_and conj_Barzilay_Lee nsubj_learned_ amod_patterns_paraphrasing dobj_learned_patterns prep_learned_as pobj_as_pairs prep_pairs_of nn_lattices_word pobj_of_lattices nsubjpass_used_which auxpass_used_are advmod_used_then rcmod_lattices_used aux_produce_to xcomp_used_produce nn_paraphrases_sentence nn_paraphrases_level dobj_produce_paraphrases ' +o,7142,'Current work has been spurred by two papers , and ',Yarowsky,'amod_work_Current nsubjpass_spurred_work aux_spurred_has auxpass_spurred_been prep_spurred_by num_papers_two pobj_by_papers conj_papers_ cc_papers_and conj_papers_ ' +o,7143,'Given two sentences X and Y , the WLCS score of X and Y can be computed using the similar dynamic programming procedure as stated in ',Lin,'num_sentences_two dep_Given_sentences nsubjpass_computed_X cc_X_and conj_X_Y det_score_the nn_score_WLCS appos_X_score prep_score_of pobj_of_X cc_X_and conj_X_Y aux_computed_can auxpass_computed_be rcmod_sentences_computed purpcl_computed_using det_procedure_the amod_procedure_similar amod_procedure_dynamic nn_procedure_programming dobj_using_procedure mark_stated_as advcl_using_stated prt_stated_in ' +o,7144,'For all non-LEAF systems , we take the best performing of the union , refined and intersection symmetrization heuristics to combine the 1-to-N and M-to-1 directions resulting in a M-to-N alignment ',Och,'prep_take_For det_systems_all amod_systems_non-LEAF pobj_For_systems nsubj_take_we dep_performing_the dep_performing_best nsubj_heuristics_performing prep_performing_of det_symmetrization_the nn_symmetrization_union amod_union_refined cc_union_and conj_union_intersection pobj_of_symmetrization ccomp_take_heuristics acomp_heuristics_ aux_combine_to xcomp__combine det_directions_the amod_directions_1-to-N cc_1-to-N_and conj_1-to-N_M-to-1 dobj_combine_directions partmod_directions_resulting prep_resulting_in det_alignment_a amod_alignment_M-to-N pobj_in_alignment ' +o,7145,'A number of part-of-speech taggers are readily available and widely used , all trained and retrainable on text corpora ',Marcus,'det_number_A nsubj_available_number prep_number_of amod_taggers_part-of-speech pobj_of_taggers cop_available_are advmod_available_readily cc_available_and advmod_used_widely conj_available_used advmod_trained_all conj_available_trained cc_available_and conj_available_retrainable prep_retrainable_on nn__text nn__corpora pobj_on_ ' +o,7146,'The Xerox experiments correspond to something between D1 and D2 , and between TO and T1 , in that there is some initial biasing of the probabilities ',Cutting,'det_experiments_The nn_experiments_Xerox nsubj__experiments dep__correspond prep_correspond_to pobj_to_something prep_something_between pobj_between_D1 cc_D1_and conj_D1_D2 cc_to_and conj_to_between pobj_between_TO cc_TO_and conj_TO_T1 rel_is_in pobj_in_that expl_is_there rcmod_TO_is det_biasing_some amod_biasing_initial nsubj_is_biasing prep_biasing_of det_probabilities_the pobj_of_probabilities ' +o,7147,'The traditional framework presented in assumes a generative process where the source sentence is passed through a noisy stochastic process to produce the target sentence ',Brown,'det_framework_The amod_framework_traditional nsubj_assumes_framework partmod_framework_presented prep_presented_in pobj_in_ det_process_a amod_process_generative dobj_assumes_process advmod_passed_where det_sentence_the nn_sentence_source nsubjpass_passed_sentence auxpass_passed_is rcmod_process_passed prep_passed_through det_process_a advmod_stochastic_noisy amod_process_stochastic pobj_through_process aux_produce_to xcomp_passed_produce det_sentence_the nn_sentence_target dobj_produce_sentence ' +o,7148,'Experimental results were only reported for the METEOR metric ',Banerjee,'amod_results_Experimental nsubjpass_reported_results auxpass_reported_were advmod_reported_only prep_reported_for det_metric_the amod_metric_METEOR pobj_for_metric ' +o,7149,'1 Introduction Since their appearance , BLEU and NIST have been the standard tools used for evaluating the quality of machine translation ',Papineni,'num_Introduction_1 prep_been_Since poss_appearance_their pobj_Since_appearance nn__BLEU nsubj_been_ cc__and nn__NIST conj__ aux_been_have dep_Introduction_been det_tools_the amod_tools_standard dobj_been_tools partmod_tools_used prep_used_for pcomp_for_evaluating det_quality_the dobj_evaluating_quality prep_quality_of nn_translation_machine pobj_of_translation ' +o,7150,'Expansion of the equivalent sentence set can be applied to automatic evaluation of machine translation quality , for example ',Papineni,'nsubjpass_applied_Expansion prep_Expansion_of det_set_the nn_set_equivalent nn_set_sentence pobj_of_set aux_applied_can auxpass_applied_be prep_applied_to amod_evaluation_automatic pobj_to_evaluation prep_evaluation_of nn__machine nn__translation nn__quality pobj_of_ prep_applied_for pobj_for_example ' +o,7151,'The kappa statistic has become the de facto standard to assess inter-annotator agreement ',Carletta,'det__The nn__kappa nn__statistic nsubj_standard_ aux_standard_has cop_standard_become det_standard_the amod_standard_de dep_de_facto aux_assess_to xcomp_standard_assess amod_agreement_inter-annotator dobj_assess_agreement ' +o,7152,'8 An alternative formula for G 2 is given in , but the two are equivalent ',Dunning,'det_formula_An amod_formula_alternative nsubjpass_given_formula prep_formula_for pobj_for_G num_G_2 auxpass_given_is dep_8_given prep_given_in pobj_in_ cc_given_but dep_two_the nsubj_equivalent_two cop_equivalent_are conj_given_equivalent ' +p,7153,'Then the same system weights are applied to both IncHMM and Joint Decoding - based approaches , and the feature weights of them are trained using the max-BLEU training method proposed by Och and refined by ',Moore,'advmod_applied_Then det_weights_the amod_weights_same nn_weights_system nsubjpass_applied_weights auxpass_applied_are prep_applied_to preconj_Decoding_both nn_Decoding_IncHMM cc_IncHMM_and conj_IncHMM_Joint pobj_to_Decoding dep_to_based dep_based_approaches cc_applied_and det_weights_the nn_weights_feature nsubjpass_trained_weights prep_weights_of pobj_of_them auxpass_trained_are conj_applied_trained xcomp_trained_using det_method_the amod_method_max-BLEU nn_method_training dobj_using_method dep_method_proposed prep_proposed_by nn__Och pobj_by_ cc_proposed_and conj_proposed_refined prep_refined_by ' +o,7154,' 00: the current input token and the previous one have the same parent 90: one ancestor of the current input token and the previous input token have the same parent 09: the current input token and one ancestor of the previous input token have the same parent 99 one ancestor of the current input token and one ancestor of the previous input token have the same parent Compared with the B-Chunk and I-Chunk used in Ramshaw and Marcus(1995)~, structural relations 99 and 90 correspond to B-Chunk which represents the first word of the chunk, and structural relations 00 and 09 correspond to I-Chunk which represents each other in the chunk while 90 also means the beginning of the sentence and 09 means the end of the sentence.',Ramshaw,'' +o,7155,'In a set of transformational rules is used for modifying the classification of words ',Ramshaw,'pcomp_In_ det_set_a nsubjpass_used_set prep_set_of amod_rules_transformational pobj_of_rules auxpass_used_is ccomp__used prep_used_for pcomp_for_modifying det_classification_the dobj_modifying_classification prep_classification_of pobj_of_words ' +o,7156,'A problem mentioned in is that the algorithm that computes the compressed representation might need to retain the entire database in memory ; in their paper , they design strategies to work around this problem ',Talbot,'det_problem_A nsubj_is_problem partmod_problem_mentioned prep_mentioned_in pobj_in_ complm_need_that det_algorithm_the nsubj_need_algorithm nsubj_computes_that rcmod_algorithm_computes det_representation_the amod_representation_compressed dobj_computes_representation aux_need_might ccomp_is_need aux_retain_to xcomp_need_retain det_database_the amod_database_entire dobj_retain_database prep_retain_in pobj_in_memory prep_design_in poss_paper_their pobj_in_paper nsubj_design_they parataxis_is_design nsubj_work_strategies aux_work_to xcomp_design_work prt_work_around det_problem_this dobj_work_problem ' +o,7157,'Decoding with an SCFG -LRB- eg , translating from Chinese to English using the above grammar -RRB- can be cast as a parsing problem -LRB- see Section 3 for details -RRB- , in which case we need to binarize a synchronous rule with more than two nonterminals to achieve polynomial time algorithms ',Zhang,'nsubjpass_cast_Decoding prep_Decoding_with det_SCFG_an pobj_with_SCFG dep_SCFG_eg xcomp_eg_translating prep_translating_from pobj_from_Chinese prep_translating_to pobj_to_English partmod_English_using det_grammar_the amod_grammar_above dobj_using_grammar aux_cast_can auxpass_cast_be prep_cast_as det_problem_a amod_problem_parsing pobj_as_problem dep_problem_see dobj_see_Section num_Section_3 prep_Section_for pobj_for_details rel_need_in det_case_which pobj_in_case nsubj_need_we rcmod_problem_need aux_binarize_to xcomp_need_binarize det_rule_a amod_rule_synchronous dobj_binarize_rule prep_rule_with dep_than_more quantmod_two_than num_nonterminals_two pobj_with_nonterminals aux_achieve_to xcomp_binarize_achieve amod_algorithms_polynomial nn_algorithms_time dobj_achieve_algorithms ' +o,7158,'6 Related Work A pioneering antecedent for our work is , who trained a Collins-style generative parser over a syntactic structure augmented with the template entity and template relations annotations for the MUC-7 shared task ',Collins,'num_Work_6 amod_Work_Related det_antecedent_A amod_antecedent_pioneering nsubj__antecedent prep_antecedent_for poss_work_our pobj_for_work cop__is dep_Work_ nsubj_trained_who rcmod__trained det__a amod__Collins-style amod__generative nn__parser nsubj_augmented_ prep__over det_structure_a amod_structure_syntactic pobj_over_structure dep_Work_augmented prep_augmented_with det_entity_the amod_entity_template pobj_with_entity cc_entity_and nn_annotations_template nn_annotations_relations conj_entity_annotations prep_augmented_for det_task_the nn_task_MUC-7 amod_task_shared pobj_for_task ' +p,7159,'Aside from purely linguistic interest , bracket structure has been empirically shown to be highly effective at constraining subsequent training of , for example , stochastic context-free grammars ',Brown,'dep_from_Aside advcl_shown_from advmod_linguistic_purely amod_interest_linguistic pobj_from_interest nn_structure_bracket nsubjpass_shown_structure aux_shown_has auxpass_shown_been advmod_shown_empirically aux_effective_to cop_effective_be advmod_effective_highly xcomp_shown_effective prep_effective_at pcomp_at_constraining amod_training_subsequent dobj_constraining_training prep_training_of dep_of_for pobj_for_example amod__stochastic amod__context-free nn__grammars pobj_of_ ' +o,7160,'440 respondence learning -LRB- SCL -RRB- domain adaptation algorithm for use in sentiment classification ',Blitzer,'amod_learning_respondence nsubj_440_learning abbrev_learning_SCL nn_algorithm_domain nn_algorithm_adaptation dep_learning_algorithm partmod_algorithm_ prep__for pobj_for_use prep_use_in nn_classification_sentiment pobj_in_classification ' +o,7161,'A single translation is then selected by finding the candidate that yields the best overall score or by cotraining ',Och,'det_translation_A amod_translation_single nsubjpass_selected_translation auxpass_selected_is advmod_selected_then prep_selected_by pcomp_by_finding det_candidate_the dobj_finding_candidate nsubj_yields_that rcmod_candidate_yields det_score_the amod_score_best amod_score_overall dobj_yields_score advmod_yields_ cc_by_or conj_by_by amod__cotraining pobj_by_ ' +o,7162,'The tools used are the Moses toolkit for decoding and training , GIZA + + for word alignment , and SRILM for language models ',Och,'det_tools_The nsubj__tools partmod_tools_used cop__are det__the nn__Moses nn__toolkit prep__for pobj_for_decoding cc_decoding_and conj_decoding_training nn_+_GIZA nn_+_+ conj__+ prep_+_for nn__word nn__alignment pobj_for_ cc__and nn__SRILM conj__ prep__for nn_models_language pobj_for_models ' +o,7163,'611 Nugget-Based Pyramid Evaluation For our first approach we used a nugget-based evaluation methodology ',Lin,'num_Evaluation_611 amod_Evaluation_Nugget-Based nn_Evaluation_Pyramid dep_For_Evaluation prep_used_For poss_approach_our amod_approach_first pobj_For_approach nsubj_used_we det_methodology_a amod_methodology_nugget-based nn_methodology_evaluation dobj_used_methodology ' +o,7164,'Many methods for calculating the similarity have been proposed ',Lin,'amod_methods_Many nsubjpass_proposed_methods prep_methods_for pcomp_for_calculating det_similarity_the dobj_calculating_similarity aux_proposed_have auxpass_proposed_been ' +o,7165,'It has been shown that one sense per discourse property can improve the performance of bootstrapping algorithm ',Yarowsky,'nsubjpass_shown_It aux_shown_has auxpass_shown_been complm_improve_that num_sense_one nsubj_improve_sense prep_sense_per nn_property_discourse pobj_per_property aux_improve_can ccomp_shown_improve det_performance_the dobj_improve_performance prep_performance_of amod_algorithm_bootstrapping pobj_of_algorithm ' +o,7166,'Post-editing of automatic annotation has been pursued in various projects ',Marcus,'nsubjpass_pursued_Post-editing prep_Post-editing_of amod_annotation_automatic pobj_of_annotation aux_pursued_has auxpass_pursued_been prep_pursued_in amod_projects_various pobj_in_projects ' +o,7167,'It is a fundamental and often a necessary step before linguistic knowledge acquisitions , such as training a phrase translation table in phrasal machine translation -LRB- MT -RRB- system , or extracting hierarchial phrase rules or synchronized grammars in syntax-based translation framework ',Koehn,'nsubj_fundamental_It cop_fundamental_is det_fundamental_a cc_fundamental_and conj_fundamental_often det_step_a amod_step_necessary dep_often_step dep_often_before amod_acquisitions_linguistic nn_acquisitions_knowledge pobj_before_acquisitions dep_as_such prep_often_as pcomp_as_training det_table_a nn_table_phrase nn_table_translation dobj_training_table prep_training_in amod_translation_phrasal nn_translation_machine pobj_in_translation abbrev_translation_MT dep_translation_system num_system_ cc_system_or conj_system_extracting amod_rules_hierarchial nn_rules_phrase dobj_extracting_rules cc_rules_or nn_grammars_synchronized conj_rules_grammars prep_extracting_in amod_framework_syntax-based nn_framework_translation pobj_in_framework ' +p,7168,'Word segmentation and POS tagging in a joint process have received much attention in recent research and have shown improvements over a pipelined fashion ',Zhang,'amod_segmentation_Word nsubj_received_segmentation cc_segmentation_and conj_segmentation_POS partmod_segmentation_tagging prep_tagging_in det_process_a amod_process_joint pobj_in_process aux_received_have amod_attention_much dobj_received_attention prep_received_in amod_research_recent pobj_in_research cc_received_and aux_shown_have conj_received_shown dobj_shown_improvements prep_improvements_over det_fashion_a amod_fashion_pipelined pobj_over_fashion ' +o,7169,'Some researchers then tried to automatically extract paraphrase rules , which facilitates the rule-based PG methods ',Barzilay,'det_researchers_Some nsubj_tried_researchers advmod_tried_then aux_extract_to advmod_extract_automatically xcomp_tried_extract amod__paraphrase nn__rules dobj_extract_ nsubj_facilitates_which rcmod__facilitates det_methods_the amod_methods_rule-based nn_methods_PG dobj_facilitates_methods ' +o,7170,'For such cases , unsupervised approaches have been developed for predicting relations , by using sentences containing discourse connectives as training data ',Marcu,'prep_developed_For amod_cases_such pobj_For_cases amod_approaches_unsupervised nsubjpass_developed_approaches aux_developed_have auxpass_developed_been prep_developed_for pcomp_for_predicting dobj_predicting_relations prep_developed_by pcomp_by_using dobj_using_sentences xcomp_using_containing nn_connectives_discourse dobj_containing_connectives prep_containing_as nn_data_training pobj_as_data ' +o,7171,'The model scaling factors 1 , ,5 and the word and phrase penalties are optimized with respect to some evaluation criterion , eg BLEU score ',Och,'det_factors_The amod_factors_model nn_factors_scaling nsubjpass_optimized_1 conj_1_,5 cc_1_and det_penalties_the nn_penalties_word cc_word_and conj_word_phrase conj_1_penalties auxpass_optimized_are dep_factors_optimized prep_optimized_with pobj_with_respect prep_optimized_to det__some nn__evaluation nn__criterion pobj_to_ nn_score_eg nn_score_BLEU appos__score ' +o,7172,'84 52 Machine translation on Europarl corpus We further tested our WDHMM on a phrase-based machine translation system to see whether our improvement on word alignment can also improve MT accuracy measured by BLEU score ',Papineni,'number_52_84 num_translation_52 nn_translation_Machine nsubj_corpus_translation prep_translation_on pobj_on_Europarl nsubj_tested_We advmod_tested_further ccomp_corpus_tested poss_WDHMM_our dobj_tested_WDHMM prep_tested_on det_system_a amod_system_phrase-based nn_system_machine nn_system_translation pobj_on_system aux_see_to xcomp_tested_see complm_improve_whether poss_improvement_our nsubj_improve_improvement prep_improvement_on nn_alignment_word pobj_on_alignment aux_improve_can advmod_improve_also ccomp_see_improve nn_accuracy_MT dobj_improve_accuracy partmod_accuracy_measured prep_measured_by amod_score_BLEU pobj_by_score ' +o,7173,'MT output was evaluated using the standard evaluation metric BLEU 2 The parameters of the MT System were optimized for BLEU metric on NIST MTEval2002 test sets using minimum error rate training , and the systems were tested on NIST MTEval2003 test sets for both languages ',Och,'nn_output_MT nsubjpass_evaluated_output auxpass_evaluated_was xcomp_evaluated_using det_evaluation_the amod_evaluation_standard dobj_using_evaluation amod_BLEU_metric nsubj__BLEU dep_evaluated_ dep_optimized_2 det_parameters_The nsubjpass_optimized_parameters prep_parameters_of det_System_the nn_System_MT pobj_of_System auxpass_optimized_were dep__optimized prep_optimized_for amod_metric_BLEU pobj_for_metric prep_metric_on amod_sets_NIST nn_sets_MTEval2002 nn_sets_test pobj_on_sets xcomp_optimized_using amod__minimum nn__error nn__rate nn__training dobj_using_ cc_optimized_and det_systems_the nsubjpass_tested_systems auxpass_tested_were conj_optimized_tested prep_tested_on amod_sets_NIST nn_sets_MTEval2003 nn_sets_test pobj_on_sets prep_sets_for det_languages_both pobj_for_languages ' +o,7174,'Someworkwithintheframework of synchronous grammars , while others create a generative story that includes a parse tree provided for one of the sentences ',Wu,'nsubj__Someworkwithintheframework prep_Someworkwithintheframework_of amod_grammars_synchronous pobj_of_grammars mark_create_while nsubj_create_others advcl__create det_story_a amod_story_generative dobj_create_story nsubj_includes_that rcmod_story_includes det_tree_a amod_tree_parse dobj_includes_tree prep_includes_provided dep_provided_for pobj_for_one prep_one_of det__the nn__sentences pobj_of_ ' +o,7175,'1 Introduction Most of the current work in statistical machine translation builds on word replacement models developed at IBM in the early 1990s ',Brown,'num_Most_1 nn_Most_Introduction nsubj_builds_Most prep_Most_of det_work_the amod_work_current pobj_of_work prep_work_in amod_translation_statistical nn_translation_machine pobj_in_translation prep_builds_on nn_models_word nn_models_replacement pobj_on_models partmod_models_developed prep_developed_at pobj_at_IBM prep_developed_in det_1990s_the amod_1990s_early pobj_in_1990s ' +o,7176,'1 Introduction Nowadays , statistical machine translation is mainly based on phrases ',Koehn,'num_translation_1 nn_translation_Introduction nn_translation_Nowadays amod_translation_statistical nn_translation_machine nsubjpass_based_translation auxpass_based_is advmod_based_mainly prep_based_on pobj_on_phrases ' +o,7177,' present a knowledge-lean algorithm that uses multiple-sequence alignment to 177 learn generate sentence-level paraphrases essentially from unannotated corpus data alone ',Barzilay,'nsubj_present_ det_algorithm_a amod_algorithm_knowledge-lean nsubj_learn_algorithm nsubj_uses_that rcmod_algorithm_uses amod_alignment_multiple-sequence dobj_uses_alignment prep_uses_to pobj_to_177 ccomp_present_learn xcomp_learn_generate amod_paraphrases_sentence-level dobj_generate_paraphrases advmod_generate_essentially prep_generate_from amod_data_unannotated nn_data_corpus pobj_from_data advmod_generate_alone ' +o,7178,'6 Related Work In machine translation , the concept of packed forest is first used by to characterize the search space of decoding with language models ',Huang,'num_Work_6 amod_Work_Related prep_used_In nn_translation_machine pobj_In_translation det_concept_the nsubjpass_used_concept prep_concept_of amod_forest_packed pobj_of_forest auxpass_used_is advmod_used_first rcmod_Work_used prep_used_by pobj_by_ aux_characterize_to xcomp_used_characterize det_space_the nn_space_search dobj_characterize_space prep_Work_of pcomp_of_decoding prep_decoding_with nn_models_language pobj_with_models ' +o,7179,'Statistical data about these various cooccurrence relations is employed for a variety of applications , such as speech recognition , language generation , lexicography , machine translation , information retrieval and various disambiguation tasks ',Hindle,'nn_data_Statistical nsubjpass_employed_data prep_data_about det_relations_these amod_relations_various amod_relations_cooccurrence pobj_about_relations auxpass_employed_is prep_employed_for det_variety_a pobj_for_variety prep_variety_of pobj_of_applications dep_as_such prep_variety_as nn__speech nn__recognition pobj_as_ nn__language nn__generation conj__ amod__lexicography conj__ nn__machine nn__translation conj__ nn__information nn__retrieval conj__ cc__and amod_tasks_various nn_tasks_disambiguation conj__tasks ' +o,7180,'\' -LRB- ` here are three main approaches in tagging problem : rule-based approach , statistical approach and connectionist approach ',Marcus,'advmod_approaches_here cop_approaches_are num_approaches_three amod_approaches_main prep_approaches_in pcomp_in_tagging dobj_tagging_problem amod__rule-based nn__approach dep_problem_ amod__statistical nn__approach conj__ cc__and amod__connectionist nn__approach conj__ ' +o,7181,' frame the problem of detecting subjective sentences as finding the minimum cut in a graph representation of the sentences ',Pang,'advmod_frame_ det_problem_the dobj_frame_problem prep_problem_of pcomp_of_detecting amod_sentences_subjective dobj_detecting_sentences prep_detecting_as pcomp_as_finding det_cut_the amod_cut_minimum dobj_finding_cut prep_finding_in det_representation_a nn_representation_graph pobj_in_representation prep_representation_of det_sentences_the pobj_of_sentences ' +o,7182,'Machine learning methods should be interchangeable : Transformation-based learning -LRB- TBL -RRB- and Memory-based learning -LRB- MBL -RRB- have been applied to many different problems , so a single interchangeable component should be used to represent each method ',Marcus,'csubj_interchangeable_learning dobj_learning_methods aux_interchangeable_should cop_interchangeable_be dep_Machine_interchangeable amod_learning_Transformation-based attr_interchangeable_learning abbrev_learning_TBL amod_learning_ cc__and conj__Memory-based partmod_learning_learning abbrev_learning_MBL nsubjpass_applied_ aux_applied_have auxpass_applied_been dep_Machine_applied prep_applied_to amod_problems_many amod_problems_different pobj_to_problems dep_applied_so det_component_a amod_component_single dep_single_interchangeable nsubjpass_used_component aux_used_should auxpass_used_be dep_applied_used aux_represent_to purpcl_used_represent det_method_each dobj_represent_method ' +o,7183,' , , Kim and Hovy -LRB- 2004 -RRB- and others -RRB- , however , the research described in this paper uses the information retrieval -LRB- IR -RRB- paradigm which has also been used by some researchers ',Turney,'nsubj_uses_ dep__ dep__ conj__Kim cc_Kim_and conj_Kim_Hovy appos_Kim_2004 cc__and conj__others advmod_uses_however det_research_the nsubj_uses_research partmod_research_described prep_described_in det_paper_this pobj_in_paper det_paradigm_the nn_paradigm_information nn_paradigm_retrieval abbrev_paradigm_IR dobj_uses_paradigm nsubjpass_used_which aux_used_has advmod_used_also auxpass_used_been rcmod_paradigm_used prep_used_by det_researchers_some pobj_by_researchers ' +o,7184,' and Koo Discriminative Reranking for NLP Della Pietra 1996 ; Della Pietra , Della Pietra , and Lafferty 1997 -RRB- , or conjugate gradient methods ',Collins,'dep__and nn_Reranking_Koo nn_Reranking_Discriminative dep__Reranking prep_Reranking_for nn_Pietra_NLP nn_Pietra_Della pobj_for_Pietra num_Pietra_1996 nn_Pietra_Della dep__Pietra nn_Pietra_Della appos_Pietra_Pietra cc_Pietra_and conj_Pietra_Lafferty num_Lafferty_1997 cc_Pietra_or amod__conjugate amod__gradient nn__methods conj_Pietra_ ' +o,7185,' ; -RRB- ',Liu,'dep__ ' +o,7186,'One of our goals was to use for our study only information that could be annotated reliably , as we believe this will make our results easier to replicate ',Carletta,'nsubj_was_One prep_One_of poss_goals_our pobj_of_goals aux_use_to xcomp_was_use prep_use_for poss_study_our pobj_for_study amod_information_only dobj_use_information nsubjpass_annotated_that aux_annotated_could auxpass_annotated_be rcmod_information_annotated advmod_annotated_reliably advmod_annotated_ mark_believe_as nsubj_believe_we advcl_annotated_believe nsubj_make_this aux_make_will ccomp_believe_make poss_results_our nsubj_easier_results xcomp_make_easier aux_replicate_to xcomp_easier_replicate ' +o,7187,'A related example would be a version of synchronous CFG that allows only one pair of linked nonterminals and any number of unlinked nonterminals , which could be bitextparsed in O -LRB- n5 -RRB- time , whereas inversion transduction grammar takes O -LRB- n6 -RRB- ',Wu,'det_example_A amod_example_related nsubj_version_example aux_version_would cop_version_be det_version_a prep_version_of amod_CFG_synchronous pobj_of_CFG nsubj_allows_that rcmod_version_allows advmod_pair_only num_pair_one dobj_allows_pair prep_pair_of amod_nonterminals_linked pobj_of_nonterminals cc_nonterminals_and det_number_any conj_nonterminals_number prep_number_of amod_nonterminals_unlinked pobj_of_nonterminals nsubjpass_bitextparsed_which aux_bitextparsed_could auxpass_bitextparsed_be rcmod_nonterminals_bitextparsed prep_bitextparsed_in nn_time_O appos_time_n5 pobj_in_time mark_takes_whereas amod__inversion nn__transduction nn__grammar nsubj_takes_ advcl_allows_takes dobj_takes_O appos_O_n6 ' +o,7188,'In fact , the WtoP model is a segmental Hidden Markov Model , in which states emit observation sequences ',Berger,'prep__In pobj_In_fact det_model_the nn_model_WtoP nsubj__model cop__is det__a amod__segmental nn__Hidden nn__Markov nn__Model rel_emit_in pobj_in_which nsubj_emit_states rcmod__emit nn_sequences_observation dobj_emit_sequences ' +o,7189,'We use GIZA + + for 5 http://iit-itinrc-cnrcgcca/projects-projets/portage_ehtml 176 word alignment , and the Pharaoh system suite to build the phrase table and decode ',Och,'nsubj_use_We dobj_use_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_GIZA_for num_alignment_5 amod_alignment_http://iit-itinrc-cnrcgcca/projects-projets/portage_ehtml nn_alignment_176 nn_alignment_word pobj_for_alignment cc_GIZA_and det_suite_the nn_suite_Pharaoh nn_suite_system conj_GIZA_suite aux_build_to xcomp_use_build det_table_the nn_table_phrase dobj_build_table cc_table_and nn__decode conj_table_ ' +o,7190,'2 Bidirectional Dependency Networks When building probabilistic models for tag sequences , we often decompose the global probability of sequences using a directed graphical model -LRB- eg , an HMM or a conditional Markov model -LRB- CMM -RRB- -RRB- ',Ratnaparkhi,'nsubj_Networks_2 nn_Networks_Bidirectional nn_Networks_Dependency advmod_decompose_When dep_decompose_building amod_models_probabilistic dobj_building_models prep_building_for nn_sequences_tag pobj_for_sequences nsubj_decompose_we advmod_decompose_often dep_Networks_decompose det_probability_the amod_probability_global dobj_decompose_probability prep_probability_of pobj_of_sequences xcomp_decompose_using det_model_a amod_model_directed amod_model_graphical dobj_using_model dep__eg det__an nn__HMM conj_eg_ cc_eg_or det_model_a amod_model_conditional nn_model_Markov conj_eg_model abbrev_model_CMM parataxis_Networks_ ' +o,7191,'Previous research in this area includes several models which incorporate hidden variables ',Titov,'amod_research_Previous nsubj_includes_research prep_research_in det_area_this pobj_in_area amod_models_several dobj_includes_models nsubj_incorporate_which rcmod_models_incorporate amod_variables_hidden dobj_incorporate_variables ' +o,7192,'22 Maximum Entropy Model The maximum entropy model estimates a probability distribution from training data ',Berger,'number_Maximum_22 num_Model_Maximum nn_Model_Entropy det__The amod__maximum amod__entropy nn__model nsubj_estimates_ dep_Model_estimates det_distribution_a nn_distribution_probability dobj_estimates_distribution prep_estimates_from nn_data_training pobj_from_data ' +p,7193,'35 Anaphoricity Determination Finally , several coreference systems have successfully incorporated anaphoricity determination 660 modules -LRB- eg Ng and Cardie and -RRB- ',Bean,'nn_Determination_Anaphoricity nsubj_incorporated_Determination advmod_incorporated_Finally amod_systems_several nn_systems_coreference nsubj_incorporated_systems aux_incorporated_have advmod_incorporated_successfully rcmod_35_incorporated amod_determination_anaphoricity dobj_incorporated_determination num_modules_660 dep_35_modules nn_Ng_eg dep_modules_Ng cc_Ng_and nn__Cardie conj_Ng_ cc__and conj__ ' +o,7194,'This is known as cost-based abduction ',Hobbs,'nsubjpass_known_This auxpass_known_is prep_known_as amod_abduction_cost-based pobj_as_abduction ' +o,7195,'In other words , learning with L1 regularization naturally has an intrinsic effect of feature selection , which results in an 97 efficient and interpretable inference with almost the same performance as L2 regularization ',Gao,'prep_has_In amod_words_other pobj_In_words partmod_words_learning prep_learning_with nn_regularization_L1 pobj_with_regularization advmod_has_naturally det_effect_an amod_effect_intrinsic dobj_has_effect prep_effect_of pobj_of_feature nsubj_has_selection nsubj_results_which rcmod_selection_results prep_results_in det_inference_an num_inference_97 amod_inference_efficient cc_efficient_and conj_efficient_interpretable pobj_in_inference prep_inference_with advmod_performance_almost det_performance_the amod_performance_same pobj_with_performance prep_performance_as nn_regularization_L2 pobj_as_regularization ' +o,7196,'There are several other approaches such as Ji and Ploux and the already mentioned ',Rapp,'expl_are_There amod_approaches_several amod_approaches_other nsubj_are_approaches dep_as_such prep_approaches_as pobj_as_Ji cc_Ji_and conj_Ji_Ploux amod_approaches_ cc__and conj__the advmod_mentioned_already partmod_approaches_mentioned ' +o,7197,'Examples of this work include a system by Liu et al , and experiments by Hindle and Rooth , and Resnik and Hearst 2 These efforts had mixed success , suggesting that while multi-level preference scores are problematic , integrating some corpus data does not solve the problems ',Marcus,'nsubj_include_Examples prep_Examples_of det_work_this pobj_of_work det_system_a dobj_include_system prep_system_by nn__Liu cc_Liu_et conj_Liu_al pobj_by_ cc_system_and conj_system_experiments prep_experiments_by nn__Hindle cc_Hindle_and conj_Hindle_Rooth pobj_by_ cc_system_and conj_system_Resnik cc_Resnik_and nn__Hearst conj_Resnik_ dep_mixed_2 det_efforts_These nsubj_mixed_efforts aux_mixed_had dep_include_mixed dobj_mixed_success xcomp_mixed_suggesting complm_solve_that mark_problematic_while amod_scores_multi-level nn_scores_preference nsubj_problematic_scores cop_problematic_are advcl_solve_problematic csubj_solve_integrating det_data_some nn_data_corpus dobj_integrating_data aux_solve_does neg_solve_not ccomp_suggesting_solve det_problems_the dobj_solve_problems ' +o,7198,'First , researchers are divided between a general method -LRB- that attempts to apply WSD to all the content words of texts , the option taken in this paper -RRB- and one that is applied only to a small trial selection of texts words -LRB- for example -RRB- ',Yarowsky,'advmod_divided_First nsubjpass_divided_researchers auxpass_divided_are prep_divided_between det_method_a amod_method_general pobj_between_method nsubj_attempts_that rcmod_method_attempts aux_apply_to xcomp_attempts_apply dobj_apply_WSD prep_apply_to predet_words_all det_words_the nn_words_content pobj_to_words prep_words_of pobj_of_texts det_option_the appos_method_option partmod_option_taken prep_taken_in det_paper_this pobj_in_paper cc_method_and conj_method_one nsubjpass_applied_that auxpass_applied_is rcmod_one_applied advmod_applied_only prep_applied_to det_selection_a amod_selection_small nn_selection_trial pobj_to_selection prep_selection_of nn_words_texts pobj_of_words dep_divided_for nn__example nn__ pobj_for_ ' +o,7199,'Deeper syntax , eg phrase or dependency structures , has been shown useful in generative models , heuristic-based models and even for syntactically motivated models such as ITG ',Wu,'amod_syntax_Deeper nsubjpass_shown_syntax nn_phrase_eg appos_syntax_phrase cc_phrase_or amod_structures_dependency conj_phrase_structures aux_shown_has auxpass_shown_been amod__useful prep_useful_in amod__generative nn__models pobj_in_ amod_models_heuristic-based appos__models dobj_shown_ cc__and advmod__even prep_shown_for advmod_motivated_syntactically amod_models_motivated pobj_for_models dep_as_such prep_models_as pobj_as_ITG ' +o,7200,'1 Introduction Bilingual data -LRB- including bilingual sentences and bilingual terms -RRB- are critical resources for building many applications , such as machine translation and cross language information retrieval ',Brown,'num_data_1 nn_data_Introduction nn_data_Bilingual nsubj_resources_data dep_data_including amod_sentences_bilingual pobj_including_sentences cc_sentences_and amod_terms_bilingual conj_sentences_terms cop_resources_are amod_resources_critical prep_resources_for pcomp_for_building amod_applications_many dobj_building_applications dep_as_such prep_resources_as nn__machine nn__translation pobj_as_ cc__and amod__cross nn__language nn__information nn__retrieval conj__ ' +o,7201,'In the BBN model , as with Model 2 of , modifying nonterminals are generated conditioning both on the parent P and its head child H Unlike Model 2 of , they are also generated conditioning on the previously generated modifying nonterminal , L \\/ -1 or Pq-1 , and there is no subcat frame or distance feature ',Collins,'dep_generated_In det_model_the nn_model_BBN pobj_In_model mark_generated_as prep_generated_with pobj_with_Model num_Model_2 prep_Model_of pobj_of_ amod_nonterminals_modifying nsubjpass_generated_nonterminals auxpass_generated_are dep_generated_generated dobj_generated_conditioning dep_on_both prep_generated_on det_P_the nn_P_parent pobj_on_P cc_generated_and poss_child_its nn_child_head dep_H_child conj_generated_H dep_generated_Unlike pobj_Unlike_Model num_Model_2 dep_generated_of pobj_of_ nsubjpass_generated_they auxpass_generated_are advmod_generated_also dobj_generated_conditioning prep_generated_on det_nonterminal_the advmod_generated_previously amod_nonterminal_generated amod_nonterminal_modifying pobj_on_nonterminal nn_-1_L nn_-1_\\/ appos_nonterminal_-1 cc_-1_or conj_-1_Pq-1 cc_generated_and expl_is_there conj_generated_is det_frame_no amod_frame_subcat nsubj_is_frame cc_frame_or nn_feature_distance conj_frame_feature ' +o,7202,'For example , the sentence My father is \\* work in the laboratory is parsed as : -LRB- S -LRB- NP My father -RRB- -LRB- VP is -LRB- NP work -RRB- -RRB- -LRB- PP in the laboratory -RRB- -RRB- 2The abbreviations s -LRB- is or has -RRB- and d -LRB- would or had -RRB- compound the ambiguities ',Collins,'prep_is_For pobj_For_example det_sentence_the nsubj_is_sentence poss_father_My dep_sentence_father dep_parsed_\\* nsubjpass_parsed_work prep_work_in det_laboratory_the pobj_in_laboratory auxpass_parsed_is dep_is_parsed dep_as_ advmod_parsed_as dep_parsed_S parataxis_S_NP poss_father_My dobj_NP_father nsubj_is_VP parataxis_S_is nn_work_NP dep_is_work parataxis_S_PP prep_PP_in det_laboratory_the pobj_in_laboratory amod_abbreviations_2The nsubj_s_abbreviations dep_is_s parataxis_s_is cc_is_or conj_is_has cc_s_and nsubj_compound_d dep_d_would cc_would_or conj_would_had conj_s_compound det_ambiguities_the dobj_compound_ambiguities ' +o,7203,'For this reason , paraphrase poses a great challenge for many Natural Language Processing -LRB- NLP -RRB- tasks , just as ambiguity does , notably in text summarization and NL generation ',Barzilay,'prep_poses_For det_reason_this pobj_For_reason nsubj_poses_paraphrase det_challenge_a amod_challenge_great dobj_poses_challenge prep_challenge_for amod_tasks_many nn_tasks_Natural nn_tasks_Language nn_tasks_Processing abbrev_tasks_NLP pobj_for_tasks advmod_does_just dep_does_as nsubj_does_ambiguity dep_poses_does advmod_does_notably prep_does_in nn_summarization_text pobj_in_summarization cc_summarization_and nn_generation_NL conj_summarization_generation ' +o,7204,'The Chinese text was tagged using the MXPOST maximum-entropy part of speech tagging tool trained on the Penn Chinese Treebank 51 ; the English text was tagged using the TnT part of speech tagger trained on the Wall Street Journal portion of the English Penn treebank ',Ratnaparkhi,'det_text_The amod_text_Chinese nsubjpass_tagged_text auxpass_tagged_was xcomp_tagged_using det_part_the amod_part_MXPOST amod_part_maximum-entropy dobj_using_part prep_part_of nn_tool_speech amod_tool_tagging pobj_of_tool advmod_trained_ partmod_tool_trained prep_trained_on det_Treebank_the nn_Treebank_Penn nn_Treebank_Chinese pobj_on_Treebank num_Treebank_51 det_text_the nn_text_English nsubjpass_tagged_text auxpass_tagged_was parataxis_tagged_tagged xcomp_tagged_using det_part_the nn_part_TnT dobj_using_part prep_part_of nn__speech nn__tagger pobj_of_ partmod__trained prep_trained_on det_portion_the nn_portion_Wall nn_portion_Street nn_portion_Journal pobj_on_portion prep_portion_of det_treebank_the nn_treebank_English nn_treebank_Penn pobj_of_treebank ' +o,7205,'Thus , the Penn Treebank of American English has been used to train and evaluate the best available parsers of unrestricted English text ',Marcus,'advmod_used_Thus det_Treebank_the nn_Treebank_Penn nsubjpass_used_Treebank prep_Treebank_of nn__American nn__English pobj_of_ aux_used_has auxpass_used_been aux_train_to xcomp_used_train cc_train_and conj_train_evaluate det_parsers_the dep_available_best amod_parsers_available dobj_train_parsers prep_parsers_of amod__unrestricted amod__English nn__text pobj_of_ ' +o,7206,'Feature comparison measures : to convert two feature sets into a scalar value , several measures have been proposed , such as cosine , Lins measure , Kullback-Leibler -LRB- KL -RRB- divergence and its variants ',Lin,'amod_measures_Feature nn_measures_comparison aux_convert_to infmod_measures_convert num_sets_two nn_sets_feature dobj_convert_sets prep_convert_into det_value_a nn_value_scalar pobj_into_value amod_measures_several nsubjpass_proposed_measures aux_proposed_have auxpass_proposed_been dep_value_proposed dep_as_such prep_value_as pobj_as_cosine nn__Lins nn__measure conj_cosine_ amod_divergence_Kullback-Leibler nn_divergence_KL conj_cosine_divergence cc_cosine_and poss_variants_its conj_cosine_variants ' +o,7207,'In , a search stops if there is no hypothesis compatible with the gold standard in the queue of candidates ',Collins,'prep_stops_In pobj_In_ det_search_a nsubj_stops_search mark_is_if expl_is_there advcl_stops_is det_hypothesis_no nsubj_is_hypothesis amod_hypothesis_compatible prep_compatible_with det_standard_the amod_standard_gold pobj_with_standard prep_standard_in det_queue_the pobj_in_queue prep_queue_of pobj_of_candidates ' +o,7208,'We combine different parametrization of -LRB- smoothed -RRB- BLEU , NIST , and TER , to give a total of roughly 100 features ',Papineni,'nsubj_combine_We amod_parametrization_different dobj_combine_parametrization prep_parametrization_of dep_of_smoothed nn__BLEU pobj_of_ nn__NIST conj__ cc__and nn__TER conj__ aux_give_to xcomp_combine_give det_total_a dobj_give_total prep_total_of quantmod_100_roughly num_features_100 pobj_of_features ' +n,7209,'Despite ME theory and its related training algorithm do not set restrictions on the range of feature functions1 , popular NLP text books and research papers seem to limit them to binary features ',Berger,'prep_set_Despite nn_theory_ME pobj_Despite_theory cc_theory_and poss_algorithm_its amod_algorithm_related nn_algorithm_training conj_theory_algorithm advmod_set_ aux_set_do neg_set_not dobj_set_restrictions prep_set_on det_range_the pobj_on_range prep_range_of pobj_of_feature nsubj_set_functions1 amod_books_popular nn_books_NLP nn_books_text appos_functions1_books nn_papers_ cc__and conj__research nsubj_seem_papers advmod_seem_ rcmod_books_seem aux_limit_to xcomp_seem_limit dobj_limit_them prep_limit_to amod_features_binary pobj_to_features ' +o,7210,'5 Data-driven Dependency Parsing Models for data-driven dependency parsing can be roughly divided into two paradigms : Graph-based and transition-based models ',McDonald,'num_Models_5 amod_Models_Data-driven nn_Models_Dependency nn_Models_Parsing nsubjpass_divided_Models prep_Models_for amod_parsing_data-driven amod_parsing_dependency pobj_for_parsing aux_divided_can auxpass_divided_be advmod_divided_roughly prep_divided_into num_paradigms_two pobj_into_paradigms amod_models_Graph-based cc_Graph-based_and conj_Graph-based_transition-based dep_paradigms_models ' +o,7211,'In this framework , the source language , let-s say English , is assumed to be generated by a noisy probabilistic source1 Most of the current statistical MT systems treat this source as a sequence of words ',Brown,'prep_assumed_In det_framework_this pobj_In_framework det_language_the nn_language_source nsubjpass_assumed_language appos_language_let-s rcmod_let-s_say dobj_say_English auxpass_assumed_is aux_generated_to auxpass_generated_be xcomp_assumed_generated prep_generated_by det_source1_a advmod_probabilistic_noisy amod_source1_probabilistic pobj_by_source1 nsubj_treat_Most prep_Most_of det_systems_the amod_systems_current amod_systems_statistical nn_systems_MT pobj_of_systems rcmod_source1_treat det_source_this dobj_treat_source prep_treat_as det_sequence_a pobj_as_sequence prep_sequence_of pobj_of_words ' +o,7212,'Restricting phrases to syntactic constituents has been shown to harm performance , so we tighten our definition of a violation to disregard cases where the only point of overlap is obscured by our phrasal resolution ',Koehn,'amod_phrases_Restricting nsubjpass_shown_phrases prep_phrases_to amod_constituents_syntactic pobj_to_constituents aux_shown_has auxpass_shown_been aux_harm_to xcomp_shown_harm nn__performance dobj_harm_ mark_tighten_so nsubj_tighten_we advcl_shown_tighten poss_definition_our dobj_tighten_definition prep_definition_of det_violation_a pobj_of_violation aux_cases_to dep_cases_disregard infmod_violation_cases advmod_obscured_where det_point_the amod_point_only nsubjpass_obscured_point prep_point_of pobj_of_overlap auxpass_obscured_is advcl_cases_obscured prep_obscured_by poss_resolution_our amod_resolution_phrasal pobj_by_resolution ' +o,7213,'They may rely only on this information -LRB- eg , -RRB- , or they may combine it with additional information as well -LRB- eg , -RRB- ',Turney,'nsubj_rely_They aux_rely_may advmod_rely_only prep_rely_on det_information_this pobj_on_information appos_information_eg dep_eg_ cc_rely_or nsubj_combine_they aux_combine_may conj_rely_combine dobj_combine_it prep_combine_with amod_information_additional pobj_with_information advmod_well_as advmod_combine_well dep_well_eg dep_eg_ ' +o,7214,'Like the models of Goodman , the additional features in our model are generated probabilistically , whereas in the parser of distance measures are assumed to be a function of the already generated structure and are not generated explicitly ',Collins,'prep_generated_Like det_models_the pobj_Like_models prep_models_of nn__Goodman pobj_of_ det_features_the amod_features_additional nsubjpass_generated_features prep_features_in poss_model_our pobj_in_model auxpass_generated_are advmod_generated_probabilistically dep_generated_whereas prep_assumed_in det_parser_the pobj_in_parser prep_parser_of nn__ pobj_of_ nn_measures_distance nsubjpass_assumed_measures auxpass_assumed_are dep_generated_assumed aux_function_to cop_function_be det_function_a xcomp_assumed_function prep_function_of det_structure_the advmod_generated_already amod_structure_generated pobj_of_structure cc_assumed_and auxpass_generated_are neg_generated_not conj_assumed_generated advmod_generated_explicitly ' +o,7215,'Their weights are optimized wrt BLEU score using the algorithm described in ',Och,'poss_weights_Their nsubjpass_optimized_weights auxpass_optimized_are nn_score_wrt nn_score_BLEU dobj_optimized_score partmod_score_using det_algorithm_the dobj_using_algorithm partmod_algorithm_described prep_described_in ' +o,7216,'1 Introduction In statistical machine translation , output translations are evaluated by their similarity to human reference translations , where similarity is most often measured by BLEU ',Papineni,'num_Introduction_1 prep_evaluated_In amod_translation_statistical nn_translation_machine pobj_In_translation nn_translations_output nsubjpass_evaluated_translations auxpass_evaluated_are dep_Introduction_evaluated prep_evaluated_by poss_similarity_their pobj_by_similarity prep_evaluated_to amod_translations_human nn_translations_reference pobj_to_translations advmod_measured_where nsubjpass_measured_similarity auxpass_measured_is advmod_often_most advmod_measured_often rcmod_translations_measured prep_measured_by pobj_by_BLEU ' +o,7217,'As a baseline , we use an IBM Model 4 system3 with a greedy decoder4 ',Brown,'prep_use_As det_baseline_a pobj_As_baseline nsubj_use_we det_system3_an nn_system3_IBM nn_system3_Model num_system3_4 nn_system3_ dobj_use_system3 prep_system3_with det__a amod__greedy amod__decoder4 pobj_with_ ' +o,7218,'For a detailed description of each algorithm , readers are referred to for the boosting algorithm , Collins -LRB- 2002 -RRB- for perceptron learning , and Gao et al ',Collins,'prep_referred_For det_description_a amod_description_detailed pobj_For_description prep_description_of det_algorithm_each pobj_of_algorithm nsubjpass_referred_readers auxpass_referred_are aux__to xcomp_referred_ prep__for det_algorithm_the amod_algorithm_boosting pobj_for_algorithm conj_algorithm_Collins appos_Collins_2002 prep_Collins_for nn_learning_perceptron pobj_for_learning cc_algorithm_and conj_algorithm_Gao cc_Gao_et conj_Gao_al ' +o,7219,'Of these , only feature weights can be trained , for which we used minimum error rate training with version 104 of IBM-style BLEU in case-insensitive mode ',Papineni,'prep_trained_Of pobj_Of_these amod_weights_only nn_weights_feature nsubjpass_trained_weights aux_trained_can auxpass_trained_be prep_trained_for dep_used_which nsubj_used_we pcomp_for_used amod_training_minimum nn_training_error nn_training_rate dobj_used_training prep_used_with pobj_with_version num_version_104 prep_version_of nn__IBM-style nn__BLEU pobj_of_ prep_used_in amod_mode_case-insensitive pobj_in_mode ' +o,7220,'Word Senses Sample Size Feedback Size \% Correct \% Correct per Sense Total drug narcotic 65 100 923 905 medicine 83 65 891 sentence judgment 23 327 1000 925 grammar 4 42 500 suit court 212 1,461 986 948 garment 21 81 550 player performer 48 230 875 923 participant 44 1,552 977 the feedback sets -RRB- consisted of a few dozen examples , in comparison to thousands of examples needed in other corpus-based methods ',Yarowsky,'nn_Feedback_Word nn_Feedback_Senses nn_Feedback_Sample nn_Feedback_Size dep_Correct_Feedback num_\%_Size dep_Feedback_\% num_\%_Correct dep_Correct_\% dep_medicine_Correct prep_Correct_per nn_drug_Sense amod_drug_Total pobj_per_drug amod_drug_narcotic number_100_65 num_905_100 number_905_923 dep_narcotic_905 dep_consisted_medicine num_medicine_83 number_891_65 num_consisted_891 nn_consisted_sentence nn_consisted_judgment num_consisted_23 number_500_327 cc_500_1000 conj_500_925 cc_500_grammar conj_500_4 number_500_42 num_court_500 nn_court_suit measure_212_court amod_consisted_212 number_977_1,461 cc_977_986 conj_977_948 cc_977_garment conj_977_21 number_977_81 number_977_550 cc_977_player conj_977_performer number_977_48 number_977_230 number_977_875 number_977_923 cc_977_participant conj_977_44 number_977_1,552 dep_212_977 det_feedback_the dep_977_feedback dep_977_sets prep_consisted_of quantmod_dozen_a quantmod_dozen_few num_examples_dozen pobj_of_examples prep_examples_in pobj_in_comparison prep_comparison_to pobj_to_thousands prep_thousands_of pobj_of_examples partmod_examples_needed prep_needed_in amod_methods_other amod_methods_corpus-based pobj_in_methods ' +p,7221,'1 Introduction Phrase-based statistical machine translation models have achieved significant improvements in translation accuracy over the original IBM word-based model ',Koehn,'num__1 nn__Introduction nn__Phrase-based amod__statistical nn__machine nn__translation nn__models nsubj_achieved_ aux_achieved_have amod_improvements_significant dobj_achieved_improvements prep_improvements_in nn_accuracy_translation pobj_in_accuracy prep_achieved_over det_model_the amod_model_original nn_model_IBM amod_model_word-based pobj_over_model ' +o,7222,'4 Method-2 : Simple Chunk-based Extraction To overcome the shortcomings of the Brill tagger in identifying particles , we next look to full chunk 2Note , this is the same as the maximum span length of 5 used by Smadja , and above the maximum attested NP length of 3 from our corpus study -LRB- see Section 22 -RRB- ',Dunning,'num_Method-2_4 dep_length_Method-2 dep_length_Simple nn_Extraction_Chunk-based nsubj_overcome_Extraction aux_overcome_To xcomp_Simple_overcome det_shortcomings_the dobj_overcome_shortcomings prep_shortcomings_of det_tagger_the nn_tagger_Brill pobj_of_tagger prep_overcome_in pcomp_in_identifying nsubj_look_particles nsubj_look_we advmod_look_next ccomp_identifying_look prep_look_to amod_2Note_full nn_2Note_chunk pobj_to_2Note nsubj_length_this cop_length_is det_length_the amod_length_same prep_same_as det_span_the amod_span_maximum pobj_as_span prep_length_of pobj_of_5 partmod_length_used prep_used_by nn__Smadja pobj_by_ cc_by_and conj_by_above det_length_the amod_length_maximum amod_length_attested nn_length_NP pobj_above_length prep_length_of pobj_of_3 prep_used_from poss_study_our nn_study_corpus pobj_from_study dep_length_see dobj_see_Section num_Section_22 ' +o,7223,'2 Previous work on Sentiment Analysis Some prior studies on sentiment analysis focused on the document-level classification of sentiment where a document is assumed to have only a single sentiment , thus these studies are not applicable to our goal ',Pang,'num_work_2 amod_work_Previous dep_studies_work prep_work_on nn_Analysis_Sentiment pobj_on_Analysis dep_studies_Some amod_studies_prior dep_analysis_studies prep_studies_on pobj_on_sentiment partmod_analysis_focused prep_focused_on det_classification_the amod_classification_document-level pobj_on_classification prep_classification_of nn__sentiment pobj_of_ advmod_assumed_where det_document_a nsubjpass_assumed_document auxpass_assumed_is advcl_focused_assumed aux_have_to xcomp_assumed_have advmod_sentiment_only det_sentiment_a amod_sentiment_single dobj_have_sentiment advmod_assumed_thus det_studies_these nsubj_applicable_studies cop_applicable_are neg_applicable_not dep_assumed_applicable prep_applicable_to poss_goal_our pobj_to_goal ' +o,7224,'tactic parser ',Collins,'nn_parser_tactic ' +o,7225,'The real-valued features include the following: a block translation score derived from phrase occurrence statistics a4a9a113a77a11, a trigram language model to predict target words a4a179a112a229 a78a204a11, a lexical weighting score for the block internal words a4a127a202a204a11, a distortion model a4a0a207a229 a218a147a11 as well as the negative target phrase length a4a60a36a87a11 . The transition cost is computed as a19 a4a20a6 a23 a6 a39 a11a224a15 a27 a28 a30a89a32 a4a7a6 a83 a6a20a39a34a11, where a27 a199a230a227 a228 is a weight vector that sums up to a113a89a35a116 : a228 a13a26a17 a10 a27 a13a217a15a231a113a25a35a116 . The weights are trained using a procedure similar to (Och, 2003) on held-out test data.',Och,'' +o,7226,'The first adaptation includes theswap-operation , whilethesecondadaptationincludesphoneticsegment distances , which are generated by applying an iterative pointwise mutual information -LRB- PMI -RRB- procedur ',Church,'det_adaptation_The amod_adaptation_first nsubj_includes_adaptation amod__theswap-operation dobj_includes_ amod_distances_whilethesecondadaptationincludesphoneticsegment appos__distances nsubjpass_generated_which auxpass_generated_are rcmod__generated prep_generated_by pcomp_by_applying det_procedur_an amod_procedur_iterative dep_iterative_pointwise amod_procedur_mutual nn_procedur_information abbrev_procedur_PMI dobj_applying_procedur ' +o,7227,'In comparison we introduce 28 several metrics coefficients reported in Albrecht and Hwa including smoothed BLEU , METEOR , HWCM , and the metric proposed in Albrecht and Hwa using the full feature set ',Banerjee,'prep_reported_In pobj_In_comparison nsubj_introduce_we rcmod_comparison_introduce num_coefficients_28 amod_coefficients_several nn_coefficients_metrics dobj_introduce_coefficients prep_reported_in pobj_in_Albrecht cc_Albrecht_and conj_Albrecht_Hwa nsubj_reported_ prep__including nn__smoothed nn__BLEU pobj_including_ nn__METEOR conj__ nn__HWCM conj__ cc__and det_metric_the conj__metric amod_metric_proposed prep_metric_in pobj_in_Albrecht cc__and nn__Hwa conj__ partmod__using det_set_the amod_set_full nn_set_feature dobj_using_set ' +o,7228,'(14), where i is the parameter to be estimated and f i (a, b) is a feature function corresponding to i (Berger et al., 1996; Ratnaparkhi, 1997): P(E P |E G ) productdisplay i P(ep i |ep i1 ik ,eg i+k ik ) (11) P(C P |E G ,E P ) (12) productdisplay i P(cp i |cp i1 ik ,eg, ep i+k ik ) P(C G |E G ,E P ,C P ) (13) productdisplay i P(cg i |cg i1 ik ,eg, ep, cp i+k ik ) P(b|a)= exp( summationtext i i f i (a, b)) summationtext b prime exp( summationtext i i f i (a, b prime )) (14) f i (a, b) is a binary function returning TRUE or FALSE based on context a and output b. If f i (a, b)=1, its corresponding model parameter i contributes toward conditional probability P(b|a) (Berger et al., 1996; Ratnaparkhi, 1997).',Berger,'' +o,7229,'32 Compound Noun Interpretation The task of interpreting the semantics of noun compounds is one which has recently received considerable attention ',Turney,'num_Interpretation_32 amod_Interpretation_Compound nn_Interpretation_Noun det_task_The nsubj_one_task prep_task_of pcomp_of_interpreting det_semantics_the dobj_interpreting_semantics prep_semantics_of nn_compounds_noun pobj_of_compounds cop_one_is dep_Interpretation_one nsubj_received_which aux_received_has advmod_received_recently rcmod_one_received amod_attention_considerable dobj_received_attention ' +o,7230,'1 Introduction Recently , extracting questions , contexts and answers from post discussions of online forums incurs increasing academic attention ',Ding,'nsubj_incurs_1 partmod_1_Introduction advmod_Introduction_Recently xcomp_Introduction_extracting dobj_extracting_questions conj_questions_contexts cc_questions_and conj_questions_answers prep_extracting_from nn_discussions_post pobj_from_discussions prep_discussions_of amod_forums_online pobj_of_forums xcomp_incurs_increasing amod_attention_academic dobj_increasing_attention ' +o,7231,'Second , we will discuss the work done by who use clustering of paraphrases to induce rewriting rules ',Barzilay,'advmod_discuss_Second nsubj_discuss_we aux_discuss_will det_work_the dobj_discuss_work partmod_work_done prep_done_by pobj_by_ nsubj_use_who rcmod_work_use nsubj_induce_clustering prep_clustering_of pobj_of_paraphrases aux_induce_to xcomp_use_induce amod_rules_rewriting dobj_induce_rules ' +o,7232,'8http : \\/ \\/ svmlightjoachimsorg 9Our replication of the Wan et al model is approximate , because we used different preprocessing tools : MXPOST for POS tagging , MSTParser for parsing , and Dan Bikels interface -LRB- http://wwwcisupennedu/dbikel/ softwarehtml # wn -RRB- to WordNet for lemmatization information ',Ratnaparkhi,'nn_\\/_\\/ nsubj_approximate_\\/ partmod_\\/_svmlightjoachimsorg amod_replication_9Our dobj_svmlightjoachimsorg_replication prep_replication_of det_model_the nn_model_Wan cc_Wan_et conj_Wan_al pobj_of_model aux_approximate_is dep_8http_approximate mark_used_because nsubj_used_we advcl_approximate_used amod_tools_different amod_tools_preprocessing dobj_used_tools dep_tools_MXPOST prep_MXPOST_for pobj_for_POS xcomp_used_tagging nn_MSTParser_ dobj_tagging_MSTParser prep_MSTParser_for amod__parsing pobj_for_ cc_used_and nn_Bikels_Dan dep_interface_Bikels conj_used_interface dep_softwarehtml_http://wwwcisupennedu/dbikel/ dep_interface_softwarehtml dep_wn_# dep_softwarehtml_wn prep_interface_to nn__WordNet pobj_to_ prep__for nn_information_lemmatization pobj_for_information ' +o,7233,'To derive the joint counts c -LRB- s , t -RRB- from which p -LRB- s t -RRB- and p -LRB- t s -RRB- are estimated , we use the phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Koehn,'aux_derive_To dep_use_derive det_counts_the amod_counts_joint dobj_derive_counts dep_counts_c appos_counts_s dep_s_t rel_estimated_from pobj_from_which nsubjpass_estimated_p amod_t_s appos_p_t cc_p_and conj_p_p nn_s_t appos_p_s auxpass_estimated_are rcmod_counts_estimated nsubj_use_we dep_generated_use det_algorithm_the nn_algorithm_phrase nn_algorithm_induction nsubj_described_algorithm ccomp_use_described prep_described_in pobj_in_ prep_described_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments xcomp_generated_using nn_model_IBM dobj_using_model num__2 nsubj_generated_ ' +p,7234,'1 Introduction Todays statistical machine translation systems rely on high quality phrase translation pairs to acquire state-of-the-art performance , see ',Koehn,'num_Todays_1 nn_Todays_Introduction amod_systems_statistical nn_systems_machine nn_systems_translation nsubj_rely_systems dep_Todays_rely prt_rely_on amod_pairs_high nn_pairs_quality nn_pairs_phrase nn_pairs_translation dobj_rely_pairs aux_acquire_to xcomp_rely_acquire amod_performance_state-of-the-art dobj_acquire_performance dep_rely_see ' +o,7235,'The mutual information Ml -LRB- x , y -RRB- is defined as the following formula ',Church,'det_Ml_The amod_Ml_mutual nn_Ml_information nsubjpass_defined_Ml dep_Ml_x appos_x_y auxpass_defined_is prep_defined_as det_formula_the amod_formula_following pobj_as_formula ' +o,7236,'In order to be able to compare the edit distance with the other metrics, we have used the following formula(Wen et al., 2002)whichnormalisesthe minimum edit distance by the length of the longest questionand transformsit into a similaritymetric: normalisededitdistance = 1 edit dist(q1,q2)max(| q 1 |,| q2 |) Word Ngram Overlap This metric compares the word n-gramsin both questions: ngramoverlap = 1N Nsummationdisplay n=1 | Gn(q1) Gn(q2) | min(| Gn(q1) |,| Gn(q2) |) where Gn(q) is the set of n-grams of length n in question q and N usually equals 4 (Barzilay and Lee, 2003;Cordeiroet al., 2007).',Barzilay,'' +o,7237,'24 Reordering Reordering features take many forms in MT In phrase-based systems , reordering is accomplished both within phrase pairs -LRB- local reordering -RRB- as well as through distance-based distortion models and lexicalized reordering models ',Koehn,'num_features_24 amod_features_Reordering nn_features_Reordering nsubj_take_features amod_forms_many dobj_take_forms prep_take_in pobj_in_MT prep_accomplished_In amod_systems_phrase-based pobj_In_systems nsubjpass_accomplished_reordering auxpass_accomplished_is dep_take_accomplished dep_within_both prep_accomplished_within amod_pairs_phrase pobj_within_pairs amod_reordering_local appos_pairs_reordering dep_well_as cc_within_well dep_well_as conj_within_through amod__distance-based nn__distortion nn__models pobj_through_ cc__and amod_models_lexicalized nn_models_reordering conj__models ' +o,7238,'In this work , we use the GIZA + + implementation of IBM Model 5 ',Brown,'prep_use_In det_work_this pobj_In_work nsubj_use_we det__the amod__GIZA cc_GIZA_+ conj_GIZA_+ nn__implementation dobj_use_ prep__of nn_Model_IBM pobj_of_Model num_Model_5 ' +n,7239,'4 Conclusions Compared with other word alignment algorithms , word_align does not require sentence alignment as input , and was shown to produce useful alignments for small and noisy corpora ',Brown,'num_Conclusions_4 nsubjpass__Conclusions prep_Conclusions_Compared dep_Compared_with amod_algorithms_other nn_algorithms_word nn_algorithms_alignment pobj_with_algorithms nsubj_require_word_align aux_require_does neg_require_not parataxis__require nn_alignment_sentence dobj_require_alignment prep_require_as pobj_as_input cc__and auxpass_shown_was conj__shown aux_produce_to xcomp_shown_produce amod_alignments_useful dobj_produce_alignments prep_produce_for amod_corpora_small cc_small_and conj_small_noisy pobj_for_corpora ' +o,7240,' Merkel , Nilsson , & Ahrenberg -LRB- 1994 -RRB- have constructed a system that uses frequency of recurrent segments to determine long phrases ',Church,'nn_Merkel_ nsubj_constructed_Merkel appos_Merkel_Nilsson cc_Merkel_& conj_Merkel_Ahrenberg appos_Merkel_1994 aux_constructed_have det_system_a dobj_constructed_system nsubj_uses_that rcmod_system_uses nsubj_determine_frequency prep_frequency_of amod_segments_recurrent pobj_of_segments aux_determine_to xcomp_uses_determine amod_phrases_long dobj_determine_phrases ' +o,7241,'Candidate translations are scored by a linear combination of models , weighted according to Minimum Error Rate Training or MERT ',Och,'amod_translations_Candidate nsubjpass_scored_translations auxpass_scored_are prep_scored_by det_combination_a amod_combination_linear pobj_by_combination prep_combination_of pobj_of_models partmod_models_weighted prep_weighted_according dep_according_to nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_to_Training cc_Training_or conj_Training_MERT ' +o,7242,'For 1-best search , we use the cube pruning technique which approximately intersects the translation forest with the LM ',Huang,'prep_use_For amod_search_1-best pobj_For_search nsubj_use_we det__the nn__cube nn__pruning nn__technique dobj_use_ nsubj_intersects_which advmod_intersects_approximately rcmod__intersects det_forest_the nn_forest_translation dobj_intersects_forest prep_intersects_with det_LM_the pobj_with_LM ' +o,7243,'213 Proceedings of EACL \'99 Table 2: The result of disambiguation experiment(two senses) (6) [__ 122 \'-~cause~ e~\'ect ~ require a-~ \'-Telose, open, ~ rrect(~ \'-\'(fall, decline, win} ] 278 \'-~feel, think, sense T T 280 {hit, attack, strike} I 250 {leave, remain, go} [ 183 gcty t ~Ol accomplish, operate\'}-216 --{occur, happen, ~ --{order, request, arrange-\'~\'~ 240 \'-~ass, adopt, ~ 274 -\'~roduce, create, gro\'~~\'--\'2~ --~ush, attack, pull~ -~s~ve, 223 \'-{ship, put, send} {stop, end, move} {add, append, total} {keep, maintain, protect} Total 215(77.3 181(72.4 160(87.4 349(92.3) ~-~ Correct(\%)] 83(77.0) 113(86.2) I 169(87.5) J Yarowsky used an unsupervised learning procedure to perform noun WSD (Yarowsky, 1995).',Yarowsky,'' +p,7244,'There is usually not a considerable difference between the two methods in terms of the accuracy of the resulting model , but L1 regularization has a significant advantage in practice ',Gao,'expl_is_There advmod_is_usually neg_is_not det_difference_a amod_difference_considerable nsubj_is_difference prep_is_between det_methods_the num_methods_two pobj_between_methods prep_methods_in pobj_in_terms prep_terms_of det_accuracy_the pobj_of_accuracy prep_accuracy_of det__the amod__resulting nn__model pobj_of_ cc_is_but nn_regularization_L1 nsubj_has_regularization conj_is_has det_advantage_a amod_advantage_significant dobj_has_advantage prep_advantage_in pobj_in_practice ' +p,7245,'We have chosen the Maximum Entropy tagger for a comparison with our universal tagger , since it achieved -LRB- by a small margin -RRB- the best overall result on Slovene as reported there of taggers available to us -LRB- MBT , the best overall , was not freely available to us at the time of writing -RRB- ',Ratnaparkhi,'nsubj_chosen_We aux_chosen_have det__the nn__Maximum nn__Entropy nn__tagger dobj_chosen_ prep__for det_comparison_a pobj_for_comparison prep_chosen_with poss_tagger_our amod_tagger_universal pobj_with_tagger mark_achieved_since nsubj_achieved_it advcl_chosen_achieved dep_reported_by det_margin_a amod_margin_small pobj_by_margin det_result_the amod_result_best amod_result_overall nsubj_reported_result prep_result_on pobj_on_Slovene advmod_Slovene_as ccomp_achieved_reported expl_taggers_there measure_of_ advmod_taggers_of ccomp_reported_taggers acomp_taggers_available prep_available_to pobj_to_us dep_taggers_MBT det_overall_the advmod_overall_best appos_MBT_overall cop_available_was neg_available_not advmod_available_freely dep_MBT_available prep_available_to pobj_to_us prep_available_at det_time_the pobj_at_time prep_time_of pcomp_of_writing ' +p,7246,'The pioneering work of introduced NP chunking as a machine-learning problem , with standard datasets and evaluation metrics ',Ramshaw,'det_work_The amod_work_pioneering nsubj_introduced_work prep_work_of pobj_of_ nn_chunking_NP dobj_introduced_chunking prep_introduced_as det_problem_a amod_problem_machine-learning pobj_as_problem prep_introduced_with amod_datasets_standard pobj_with_datasets cc_datasets_and nn_metrics_evaluation conj_datasets_metrics ' +o,7247,' ` Word Association Norms , Mutual Information , and Lexicography , \' Computational Linguistics Vol ',Church,'nn_Norms_Word nn_Norms_Association dobj__Norms nn_Information_Mutual conj_Norms_Information cc_Norms_and conj_Norms_Lexicography nn_Vol_Computational nn_Vol_Linguistics nsubj__Vol ' +o,7248,'Furthermore , these systems have tackled the problem at different levels of granularity , from the document level , sentence level , phrase level , as well as the speaker level in debates ',Pang,'advmod_tackled_Furthermore det_systems_these nsubj_tackled_systems aux_tackled_have det_problem_the dobj_tackled_problem prep_tackled_at amod_levels_different pobj_at_levels prep_levels_of pobj_of_granularity prep_tackled_from det__the nn__document nn__level nn__ nn__sentence nn__level appos__ nn__phrase nn__level pobj_from_ dep_well_as cc__well dep_well_as det_level_the nn_level_speaker conj__level prep_level_in pobj_in_debates partmod_debates_ ' +o,7249,'Practical Model 4 systems therefore make substantial search approximations ',Brown,'nn_Model_Practical num_systems_4 nsubj_make_systems advmod_make_therefore dep_Model_make amod_approximations_substantial nn_approximations_search dobj_make_approximations ' +o,7250,'The thesaurus was produced using the metric described by with input from the grammatical relation data extracted using the 90 million words of written English from the British National Corpus -LRB- BNC -RRB- using the RASP parser ',Lin,'det_thesaurus_The nsubjpass_produced_thesaurus auxpass_produced_was xcomp_produced_using det__the amod__metric amod__described prep_described_by pcomp_by_ prep__with pobj_with_input prep__from det_data_the amod_data_grammatical nn_data_relation pobj_from_data partmod_data_extracted xcomp_extracted_using det_words_the number_million_90 num_words_million dobj_using_words prep_words_of amod_English_written pobj_of_English prep_using_from det__the amod__British nn__National nn__Corpus abbrev__BNC pobj_from_ partmod__using det_parser_the nn_parser_RASP dobj_using_parser dobj_using_ ' +o,7251,'We use three different kinds of metrics : DR-STM Semantic Tree Matching , a la , but over DRS instead of over constituency trees ',Liu,'nsubj_use_We num_kinds_three amod_kinds_different dobj_use_kinds prep_kinds_of pobj_of_metrics nn_Matching_DR-STM nn_Matching_Semantic nn_Matching_Tree dep_kinds_Matching det__a nn__la appos_Matching_ cc_use_but conj_use_over pobj_over_DRS cc_over_instead dep_instead_of conj_over_over nn_trees_constituency pobj_over_trees ' +o,7252,'1 Introduction The overall goal of the Penn Discourse Treebank -LRB- PDTB -RRB- is to annotate the million word WSJ corpus in the Penn TreeBank with a layer of discourse annotations ',Marcus,'num_Introduction_1 det_goal_The amod_goal_overall nsubj_is_goal prep_goal_of det_Treebank_the nn_Treebank_Penn nn_Treebank_Discourse pobj_of_Treebank abbrev_Treebank_PDTB dep_Introduction_is aux_annotate_to xcomp_is_annotate det_corpus_the number_word_million num_corpus_word nn_corpus_WSJ dobj_annotate_corpus prep_corpus_in det__the nn__Penn nn__TreeBank pobj_in_ prep_annotate_with det_layer_a pobj_with_layer prep_layer_of nn_annotations_discourse pobj_of_annotations ' +o,7253,' et al 1993 -RRB- ',Brown,'nn_1993_ nn_1993_et nn_1993_al ' +o,7254,'One possible strategy is to exploit a widecoverage realizer that aims for applicability in multiple application domains ',Cahill,'num_strategy_One amod_strategy_possible nsubj_is_strategy aux_exploit_to xcomp_is_exploit det_realizer_a amod_realizer_widecoverage dobj_exploit_realizer nsubj_aims_that rcmod_realizer_aims prep_aims_for pobj_for_applicability prep_applicability_in amod_domains_multiple nn_domains_application pobj_in_domains ' +o,7255,' introduce IBM Models 1-5 for alignment modelling ; Vogel et al ',Brown,'advmod_introduce_ nn_1-5_IBM nn_1-5_Models dobj_introduce_1-5 prep_1-5_for amod_modelling_alignment pobj_for_modelling dep_1-5_Vogel cc_Vogel_et conj_Vogel_al ' +p,7256,'Furthermore , good results have been produced in other areas of NLP research using maximum entropy techniques ',Berger,'advmod_produced_Furthermore amod_results_good nsubjpass_produced_results aux_produced_have auxpass_produced_been prep_produced_in amod_areas_other pobj_in_areas prep_areas_of nn_research_NLP pobj_of_research xcomp_produced_using amod_techniques_maximum nn_techniques_entropy dobj_using_techniques ' +o,7257,'4 The Corpus We used two corpora for our analysis : hospital discharge summaries from 1991 to 1997 from the Columbia-Presbyterian Medical Center , and the January 1996 part of the Wall Street Journal corpus from the Penn TreeBank ',Marcus,'num_Corpus_4 nn_Corpus_The nsubj_used_We dep_Corpus_used num_corpora_two dobj_used_corpora prep_corpora_for poss_analysis_our pobj_for_analysis nn_summaries_hospital nn_summaries_discharge dep_corpora_summaries prep_summaries_from num_1997_1991 dep_1997_to pobj_from_1997 prep_1997_from det_Center_the nn_Center_Columbia-Presbyterian nn_Center_Medical pobj_from_Center cc_corpora_and det_part_the nn_part_January num_part_1996 conj_corpora_part prep_part_of det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal pobj_of_corpus prep_used_from det_TreeBank_the nn_TreeBank_Penn pobj_from_TreeBank ' +o,7258,'Much of this work has been fueled by the availability of large corpora annotated with syntactic structures , especially the Penn Treebank ',Marcus,'nsubjpass_fueled_Much prep_Much_of det_work_this pobj_of_work aux_fueled_has auxpass_fueled_been prep_fueled_by det_availability_the pobj_by_availability prep_availability_of amod_corpora_large pobj_of_corpora partmod_corpora_annotated prep_annotated_with amod_structures_syntactic pobj_with_structures advmod_structures_especially det_Treebank_the nn_Treebank_Penn dep_structures_Treebank ' +o,7259,'The words we want to aggregate for text analysis are not rigorous synonyms , but the role is the same , so we have to consider the syntactic relation based on the assumptions that words with the same role tend to modify or be modified by similar words ',Hindle,'det_words_The nsubj_want_we dep_words_want aux_aggregate_to xcomp_want_aggregate mark_synonyms_for nn_analysis_text nsubj_synonyms_analysis cop_synonyms_are neg_synonyms_not amod_synonyms_rigorous advcl_aggregate_synonyms cc_want_but det_role_the nsubj_same_role cop_same_is det_same_the conj_want_same dep_want_so nsubj_have_we dep_want_have aux_consider_to xcomp_have_consider det_relation_the amod_relation_syntactic dobj_consider_relation prep_consider_based dep_based_on det_assumptions_the pobj_on_assumptions complm_tend_that nsubj_tend_words prep_words_with det_role_the amod_role_same pobj_with_role ccomp_consider_tend aux_modify_to xcomp_tend_modify cc_modify_or auxpass_modified_be conj_modify_modified prep_modified_by amod_words_similar pobj_by_words ' +o,7260,'Iegar -LRB- ling l ; his l ; ypu of -LRB- : olloeation , the approaches till ilOW could be divi -LRB- led inl ; o t ; wo groups : those thai ; do uo -LRB- , refer to s ` ttbstrings of colloco , l , ions as a l -RRB- arti -LRB- : ular problem , -LRB- Church and llanks , t99 -LRB- -RRB- ; Kim and Cho , 1993 ; Nagao and Mori , 1994 -RRB- , and those that ; do ',Smadja,'amod_l_ling pobj_Iegar_l poss_l_his nsubj_do_l dep_l_ypu prep_ypu_of parataxis_be_olloeation det_approaches_the appos_olloeation_approaches dep_olloeation_till pobj_till_ilOW aux_be_could dep_of_be nsubj_do_divi partmod_divi_led dobj_led_inl dep_inl_o dep_o_t aux_groups_wo rcmod_inl_groups det_thai_those dep_groups_thai ccomp_be_do dobj_do_uo dep_to_refer dep_s_to dep_uo_s dep_s_ttbstrings prep_ttbstrings_of pobj_of_colloco dep_uo_l dep_be_ions advmod_l_as det_l_a tmod_ions_l pobj_of_arti dep_arti_ular dep_ular_problem appos_ular_Church cc_Church_and conj_Church_llanks num_-LRB-_t99 dep_ular_Kim cc_Kim_and conj_Kim_Cho appos_ular_1993 dep_1993_Nagao cc_ular_and conj_ular_Mori appos_Mori_1994 cc_arti_and conj_arti_those dep_those_that rcmod_l_do ' +o,7261,'52 Translation In order to test the translation performance of the grammars induced by our model and the GHKM method6 we report BLEU scores on sentences of up to twenty words in length from the MT03 NIST evaluation ',Papineni,'num_Translation_52 mark_test_In dep_test_order aux_test_to dep_Translation_test det_performance_the nn_performance_translation dobj_test_performance prep_performance_of det_grammars_the pobj_of_grammars partmod_grammars_induced prep_induced_by poss_model_our pobj_by_model cc_model_and det_method6_the nn_method6_GHKM conj_model_method6 nsubj_report_we rcmod_method6_report amod_scores_BLEU amod_scores_ dobj_report_scores prep_report_on pobj_on_sentences prep_sentences_of amod_evaluation_up prep_up_to num_words_twenty pobj_to_words prep_words_in pobj_in_length prep_up_from det_NIST_the nn_NIST_MT03 pobj_from_NIST pobj_of_evaluation ' +o,7262,' study the shortest hyperpath problem and Nielsen et al ',Marcus,'amod_study_ det_problem_the amod_problem_shortest nn_problem_hyperpath dep_study_problem cc_problem_and conj_problem_Nielsen cc_Nielsen_et conj_Nielsen_al ' +o,7263,'The observation probabilities for a given state , representing a certain word class , are determined by the relative frequencies of words belonging to that class -LRB- as determined by the algorithm of -RRB- ; the probabilities of other words are set to a small initial value ',Brown,'det_probabilities_The nn_probabilities_observation nsubjpass_determined_probabilities prep_probabilities_for det_state_a amod_state_given pobj_for_state partmod_determined_representing det_class_a amod_class_certain nn_class_word dobj_representing_class auxpass_determined_are prep_determined_by det_frequencies_the amod_frequencies_relative pobj_by_frequencies prep_frequencies_of pobj_of_words partmod_words_belonging prep_belonging_to det_class_that pobj_to_class mark_determined_as dep_words_determined prep_determined_by det_algorithm_the pobj_by_algorithm prep_algorithm_of pobj_of_ det_probabilities_the nsubjpass_set_probabilities prep_probabilities_of amod_words_other pobj_of_words auxpass_set_are parataxis_determined_set prep_set_to det_value_a amod_value_small amod_value_initial pobj_to_value ' +o,7264,'Applications of word clustering include language modeling , text classification , thesaurus construction and so on ',Brown,'nsubj_include_Applications prep_Applications_of nn_clustering_word pobj_of_clustering nn__language nn__modeling dobj_include_ nn__text nn__classification conj__ nn__thesaurus nn__construction conj__ cc__and advmod_on_so conj__on ' +p,7265,'3 Experiments We evaluated the effect of random feature mixing on four popular learning methods : Perceptron , MIRA , SVM and Maximum entropy ; with 4 NLP datasets : 20 Newsgroups1 , Reuters , Sentiment and Spam ',Blitzer,'num_Experiments_3 nsubj_evaluated_We rcmod_Experiments_evaluated det_effect_the dobj_evaluated_effect prep_effect_of amod_feature_random pobj_of_feature partmod_feature_mixing prep_mixing_on num_methods_four amod_methods_popular nn_methods_learning pobj_on_methods dep_methods_Perceptron nn__MIRA appos_Perceptron_ appos_Perceptron_SVM cc_SVM_and amod_entropy_Maximum conj_SVM_entropy prep_Perceptron_with num_datasets_4 nn_datasets_NLP pobj_with_datasets num_Newsgroups1_20 dep_methods_Newsgroups1 nn__Reuters conj_methods_ nn__Sentiment conj_methods_ cc_methods_and nn__Spam conj_methods_ ' +o,7266,'Following , we computed the skip bi-gram score using both the sentence pool and the query pool ',Lin,'prep_computed_Following pobj_Following_ nsubj_computed_we det_score_the amod_score_skip amod_score_bi-gram dobj_computed_score partmod_score_using preconj_pool_both det_pool_the nn_pool_sentence dobj_using_pool cc_pool_and det_pool_the amod_pool_query conj_pool_pool ' +o,7267,'Cohen \'s Kappa ~ ',Carletta,'poss_~_Cohen possessive_Cohen_\'s nn_~_Kappa ' +o,7268,'Computational linguistics research generally attaches great value to high kappa measures , which indicate high human agreement on a particular task ',Carletta,'amod_research_linguistics dobj_Computational_research advmod_attaches_generally dep_Computational_attaches amod_value_great dobj_attaches_value prep_attaches_to amod_measures_high nn_measures_kappa pobj_to_measures nsubj_Computational_ nsubj_indicate_which rcmod__indicate amod_agreement_high amod_agreement_human dobj_indicate_agreement prep_agreement_on det_task_a amod_task_particular pobj_on_task ' +p,7269,'Currently , the best-performing English NP interpretation methods in computational linguistics focus mostly on two consecutive noun instances -LRB- noun compounds -RRB- and are either -LRB- weakly -RRB- supervised , knowledge-intensive , , , , , , , , or use statistical models on large collections of unlabeled data , , , ',Snow,'advmod_focus_Currently det_methods_the amod_methods_best-performing nn_methods_English nn_methods_NP nn_methods_interpretation nsubj_focus_methods prep_methods_in amod_linguistics_computational pobj_in_linguistics ccomp_knowledge-intensive_focus advmod_focus_mostly prep_focus_on num_instances_two amod_instances_consecutive nn_instances_noun pobj_on_instances nn_compounds_noun appos_instances_compounds cc_focus_and cop_supervised_are dep_supervised_either dep_supervised_weakly conj_focus_supervised nn__ appos__ appos__ appos__ appos__ appos__ appos__ dobj_knowledge-intensive_ cc__or conj__use amod_models_statistical nsubj_knowledge-intensive_models prep_models_on amod_collections_large pobj_on_collections prep_collections_of amod_data_unlabeled pobj_of_data amod_data_ dep__ dep__ dep__ ' +o,7270,'available -RRB- : SCISSOR , an integrated syntactic-semantic parser ; KRISP , an SVM-based parser using string kernels ; WASP , a system based on synchronous grammars ; Z&C 3 , a probabilistic parser based on relaxed CCG grammars ; and LU , a generative model with discriminative reranking ',Zettlemoyer,'nn__SCISSOR dobj_-RRB-_ det_parser_an amod_parser_integrated amod_parser_syntactic-semantic appos__parser nn__KRISP dep__ det_parser_an amod_parser_SVM-based appos__parser partmod_parser_using nn_kernels_string dobj_using_kernels nn__WASP dep__ det_system_a appos__system partmod_system_based prep_based_on amod_grammars_synchronous pobj_on_grammars dep__Z&C num_Z&C_ num_Z&C_3 det_parser_a amod_parser_probabilistic appos_Z&C_parser partmod_parser_based prep_based_on amod_grammars_relaxed nn_grammars_CCG pobj_on_grammars cc__and nn__LU conj__ det_model_a amod_model_generative appos__model prep_model_with amod_reranking_discriminative pobj_with_reranking ' +o,7271,'Our decoder is a phrase-based multi-stack imple5 mentation of the log-linear model similar to Pharaoh ',Koehn,'poss_decoder_Our nsubj_mentation_decoder cop_mentation_is det_mentation_a amod_mentation_phrase-based amod_mentation_multi-stack nn_mentation_imple5 prep_mentation_of det_model_the amod_model_log-linear pobj_of_model amod_model_similar prep_similar_to pobj_to_Pharaoh ' +o,7272,' 9436 -LRB- +006 -RRB- Table 8 : The HySOL performance with the F-score optimization technique on Chunking experiments from unlabeled data appear different from each other ',Suzuki,'number_9436_ appos_9436_+006 nsubj_different_Table num_Table_8 det_performance_The nn_performance_HySOL dep_Table_performance prep_performance_with det_technique_the amod_technique_F-score nn_technique_optimization pobj_with_technique prep_technique_on pcomp_on_Chunking amod_experiments_ dobj_Chunking_experiments prep_Chunking_from amod_data_unlabeled pobj_from_data cop_different_appear dep_9436_different prep_different_from det_other_each pobj_from_other ' +o,7273,'Co-training is related to self-training , in that an algorithm is trained on its own predictions ',Ramshaw,'amod__Co-training nsubjpass_related_ auxpass_related_is prep_related_to pobj_to_self-training rel_trained_in pobj_in_that det_algorithm_an nsubjpass_trained_algorithm auxpass_trained_is rcmod_self-training_trained prep_trained_on poss_predictions_its amod_predictions_own pobj_on_predictions ' +o,7274,'4 , we see strong parallels between TransType and ITU : language model enumerating word sequences vs 4 Initially statistical MT used a noisy-channel approach ; but recently have introduced a more general framework based on the maximum-entropy principle , which shows nice prospects in terms of flexibility and learnability ',Brown,'nsubj_see_4 nsubj_see_we amod_parallels_strong dobj_see_parallels prep_parallels_between pobj_between_TransType cc_TransType_and conj_TransType_ITU nn_model_language dep_parallels_model partmod_model_enumerating nn_sequences_word dobj_enumerating_sequences cc_model_vs num_MT_4 amod_MT_Initially amod_MT_statistical conj_model_MT partmod_MT_used det__a amod__noisy-channel nn__approach dobj_used_ cc_introduced_but advmod_introduced_recently nsubj_introduced_ aux_introduced_have parataxis_see_introduced det_framework_a advmod_general_more amod_framework_general dobj_introduced_framework prep_introduced_based dep_based_on det_principle_the amod_principle_maximum-entropy pobj_on_principle nsubj_shows_which rcmod_principle_shows amod_prospects_nice dobj_shows_prospects prep_shows_in pobj_in_terms prep_terms_of pobj_of_flexibility cc_flexibility_and conj_flexibility_learnability ' +o,7275,'GIZA + + consists of a set of statistical translation models of different complexity , namely the IBM ones ',Brown,'nn_+_GIZA nn_+_+ nsubj_consists_+ prep_consists_of det_set_a pobj_of_set prep_set_of amod_models_statistical nn_models_translation pobj_of_models prep_models_of amod_complexity_different pobj_of_complexity advmod_complexity_namely det_ones_the nn_ones_IBM dep_complexity_ones ' +o,7276,'For example , paraphrase references to make them closer to the system translation in order to obtain more reliable results when using automatic evaluation metrics like BLEU ',Papineni,'prep__For pobj_For_example amod_references_paraphrase nsubj_make_references aux_make_to xcomp__make dobj_make_them advmod_make_closer dep_closer_to det_translation_the nn_translation_system pobj_to_translation mark_obtain_in dep_obtain_order aux_obtain_to purpcl_make_obtain advmod_reliable_more amod_results_reliable dobj_obtain_results advmod_using_when advcl_obtain_using amod_metrics_automatic nn_metrics_evaluation dobj_using_metrics prep_using_like pobj_like_BLEU ' +o,7277,'This system uses all featuresof conventionalphrase-basedSMT as in ',Koehn,'det_system_This nsubj_uses_system det_conventionalphrase-basedSMT_all nn_conventionalphrase-basedSMT_featuresof dobj_uses_conventionalphrase-basedSMT prep_uses_as dep_as_in ' +o,7278,'The most obvious comparison takes on the form of a keyword analysis , which looks for the words that are significantly more frequent in the one corpus as compared to the other ',Dunning,'det_comparison_The advmod_obvious_most amod_comparison_obvious nsubj_takes_comparison prep_takes_on det_form_the pobj_on_form prep_form_of det_analysis_a amod_analysis_keyword pobj_of_analysis nsubj_looks_which rcmod_analysis_looks prep_looks_for det_words_the pobj_for_words nsubj_frequent_that cop_frequent_are advmod_frequent_significantly advmod_frequent_more rcmod_words_frequent prep_frequent_in det_corpus_the num_corpus_one pobj_in_corpus prep_frequent_as prep_frequent_compared dep_compared_to det_other_the pobj_to_other ' +o,7279,'But if one limits the information used for disambiguation of the PPattachment to include only the verb , the noun representing its object , the preposition and the main noun in the PP , the accuracy for human decision degrades from 932 \% to 882 \% on a dataset extracted from Penn Treebank ',Marcus,'cc__But mark_limits_if nsubj_limits_one dep__limits det_information_the dobj_limits_information partmod_information_used prep_used_for pobj_for_disambiguation prep_disambiguation_of det_PPattachment_the pobj_of_PPattachment aux_include_to xcomp_used_include advmod_verb_only det_verb_the dobj_include_verb det_noun_the nsubj__noun partmod_noun_representing poss_object_its dobj_representing_object det_preposition_the conj_object_preposition cc_object_and det_noun_the amod_noun_main conj_object_noun prep_representing_in det_PP_the pobj_in_PP det_accuracy_the appos_PP_accuracy prep_accuracy_for amod_degrades_human nn_degrades_decision pobj_for_degrades prep_representing_from number_\%_932 dep_\%_\% dep_\%_to number_\%_882 pobj_from_\% prep__on det_dataset_a pobj_on_dataset partmod_dataset_extracted prep_extracted_from nn_Treebank_Penn pobj_from_Treebank ' +o,7280,'With automatic refinement it is harder to guarantee improved performance than with manual refinements or with refinements based on direct lexicalization -LRB- Magerman , , Charniak -LRB- 1997 -RRB- , etc -RRB- ',Collins,'prep_harder_With amod_refinement_automatic pobj_With_refinement nsubj_harder_it cop_harder_is aux_guarantee_to xcomp_harder_guarantee amod_performance_improved dobj_guarantee_performance prep_guarantee_than dep_than_with amod__manual nn__refinements pobj_with_ cc_than_or conj_than_with pobj_with_refinements partmod_refinements_based prep_based_on amod_lexicalization_direct pobj_on_lexicalization nn__Magerman dep_guarantee_ appos__ nn_1997_Charniak appos__1997 dep__etc ' +o,7281,'Baseline We use the Moses MT system as a baseline and closely follow the example training procedure given for the WMT-07 and WMT-08 shared tasks4 In particular , we perform word alignment in each direction using GIZA + + , apply the grow-diag-finaland heuristic for symmetrization and use a maximum phrase length of 7 ',Koehn,'advmod_use_Baseline nsubj_use_We det__the nn__Moses nn__MT nn__system dobj_use_ prep__as det_baseline_a pobj_as_baseline cc_use_and advmod_use_closely conj_use_follow det_procedure_the nn_procedure_example nn_procedure_training dobj_follow_procedure prep_follow_given dep_given_for det_tasks4_the amod_tasks4_WMT-07 cc_WMT-07_and conj_WMT-07_WMT-08 amod_tasks4_shared pobj_for_tasks4 prep_perform_In pobj_In_particular nsubj_perform_we dep_use_perform nn_alignment_word dobj_perform_alignment prep_perform_in det_direction_each pobj_in_direction partmod_direction_using dobj_using_GIZA amod_GIZA_+ cc_+_+ conj_+_ conj_perform_apply det_heuristic_the amod_heuristic_grow-diag-finaland dobj_apply_heuristic prep_heuristic_for pobj_for_symmetrization cc_perform_and conj_perform_use det_length_a amod_length_maximum nn_length_phrase dobj_use_length prep_length_of pobj_of_7 ' +o,7282,'For instance , we may find metrics based on full constituent parsing , and on dependency parsing ',Liu,'prep_find_For pobj_For_instance nsubj_find_we aux_find_may dobj_find_metrics prep_find_based dep_based_on amod__full amod__constituent amod__parsing pobj_on_ cc_on_and conj_on_on pobj_on_dependency amod_dependency_parsing ' +o,7283,'High values of fall into the minimal entropy trap , while low values ofhave no effect on the model -LRB- see for an example -RRB- ',Jiao,'amod_values_High prep_values_of pobj_of_fall prep_fall_into det_trap_the amod_trap_minimal nn_trap_entropy pobj_into_trap mark_ofhave_while amod_values_low nsubj_ofhave_values dep_trap_ofhave det_effect_no dobj_ofhave_effect prep_effect_on det_model_the pobj_on_model dep_values_see dobj_see_ prep__for det_example_an pobj_for_example ' +o,7284,'Recently , various approaches to word sense division have been used in WSD research ',Dolan,'advmod_used_Recently amod_approaches_various nsubjpass_used_approaches partmod_approaches_ prep__to nn_division_word nn_division_sense pobj_to_division aux_used_have auxpass_used_been prep_used_in nn_research_WSD pobj_in_research ' +o,7285,'Our system is actually designed as a hybrid of the classic phrase-based SMT model and the kernel regression model as follows : First , for each source sentence a small relevant set of sentence pairs are retrieved from the large-scale parallel corpus ',Koehn,'poss_system_Our nsubjpass_designed_system auxpass_designed_is advmod_designed_actually prep_designed_as det_hybrid_a pobj_as_hybrid prep_hybrid_of det__the amod__classic amod__phrase-based nn__SMT nn__model pobj_of_ cc__and det_model_the nn_model_kernel nn_model_regression conj__model mark_follows_as advcl_designed_follows advmod_designed_First prep_designed_for det_sentence_each nn_sentence_source pobj_for_sentence det_set_a amod_set_small amod_set_relevant nsubjpass_retrieved_set prep_set_of nn_pairs_sentence pobj_of_pairs auxpass_retrieved_are dep_designed_retrieved prep_retrieved_from det_corpus_the amod_corpus_large-scale amod_corpus_parallel pobj_from_corpus ' +o,7286,'Typically , frequency information for rare words in the training data is used to estimate parameters for unknown words -LRB- and when these rare or unknown words are encountered during parsing , additional information may be obtained from a POS-tagger -RRB- ',Collins,'advmod_used_Typically nn_information_frequency nsubjpass_used_information prep_information_for amod_words_rare pobj_for_words prep_words_in det_data_the nn_data_training pobj_in_data auxpass_used_is aux_estimate_to xcomp_used_estimate dobj_estimate_parameters prep_parameters_for amod_words_unknown pobj_for_words cc_words_and advmod_encountered_when det_words_these amod_words_rare cc_rare_or conj_rare_unknown nsubjpass_encountered_words auxpass_encountered_are dep_information_encountered prep_encountered_during pcomp_during_parsing amod_information_additional conj_words_information aux_obtained_may auxpass_obtained_be rcmod_information_obtained prep_obtained_from det__a amod__POS-tagger pobj_from_ ' +o,7287,'1 Introduction The Inversion Transduction Grammar -LRB- ITG -RRB- of is a syntactically motivated algorithm for producing word-level alignments of pairs of translationally equivalent sentences in two languages ',Wu,'num_Introduction_1 det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction nsubj_algorithm_Grammar abbrev_Grammar_ITG prep_Grammar_of pobj_of_ cop_algorithm_is det_algorithm_a advmod_motivated_syntactically amod_algorithm_motivated dep_Introduction_algorithm prep_algorithm_for pcomp_for_producing amod_alignments_word-level dobj_producing_alignments prep_alignments_of pobj_of_pairs prep_pairs_of advmod_equivalent_translationally amod_sentences_equivalent pobj_of_sentences prep_sentences_in num_languages_two pobj_in_languages ' +o,7288,'Consider the following example : This film should be brilliant ',Pang,'det__the amod__following nn__example dobj_Consider_ det_film_This nsubj_brilliant_film aux_brilliant_should cop_brilliant_be parataxis_Consider_brilliant ' +p,7289,'Empirically the BLEU score has a high correlation with human evaluation when N = 4 for English translation evaluations ',Papineni,'det_score_the amod_score_BLEU dep_Empirically_score rcmod_score_has det_correlation_a amod_correlation_high dobj_has_correlation prep_correlation_with amod_evaluation_human pobj_with_evaluation advmod_4_when nsubj_4_N dep_4_= dep_evaluation_4 prep_correlation_for amod_evaluations_English nn_evaluations_translation pobj_for_evaluations ' +o,7290,'This contrasts with alternative alignment models such as those of Melamed and , which impose a one-to-one constraint on alignments ',Wu,'nsubj_contrasts_This prep_contrasts_with amod_models_alternative nn_models_alignment pobj_with_models dep_as_such prep_models_as pobj_as_those prep_those_of nn__Melamed pobj_of_ cc__and conj__ nsubj_impose_which rcmod_models_impose det_constraint_a amod_constraint_one-to-one dobj_impose_constraint prep_impose_on pobj_on_alignments ' +p,7291,'Brute-force methods -LRB- ie those that exploit the massive raw computing power currently available cheaply -RRB- may well produce some useful results ',Brown,'amod_methods_Brute-force nsubj_produce_methods dep_methods_ie dobj_ie_those nsubj_exploit_that rcmod_those_exploit det_power_the amod_power_massive amod_power_raw nn_power_computing nsubj_available_power advmod_available_currently xcomp_exploit_available advmod_available_cheaply aux_produce_may advmod_produce_well det_results_some amod_results_useful dobj_produce_results ' +o,7292,'However , pointed out that for the purpose of corpus statistics , where the sparseness of data is an important issue , it is better to use the log-likelihood ratio ',Dunning,'advmod_pointed_However nsubj_pointed_ prt_pointed_out complm_better_that prep_better_for det_purpose_the pobj_for_purpose prep_purpose_of nn_statistics_corpus pobj_of_statistics advmod_issue_where det_sparseness_the nsubj_issue_sparseness prep_sparseness_of pobj_of_data cop_issue_is det_issue_an amod_issue_important rcmod_statistics_issue nsubj_better_it cop_better_is ccomp_pointed_better aux_use_to xcomp_better_use det_ratio_the amod_ratio_log-likelihood dobj_use_ratio ' +p,7293,'Recent work has applied Bayesian non-parametric models to anaphora resolution , lexical acquisition and language modeling with good results ',Haghighi,'amod_work_Recent nsubj_applied_work aux_applied_has amod_models_Bayesian amod_models_non-parametric dobj_applied_models prep_applied_to nn__anaphora nn__resolution pobj_to_ amod__lexical nn__acquisition conj__ cc__and nn__language nn__modeling conj__ prep_applied_with amod_results_good pobj_with_results ' +o,7294,'In our experiments , we will use 4 different kinds of feature combinations : a157 Baseline : The 6 baseline features used in , such as cost of word penalty , cost of aligned template penalty ',Och,'prep_use_In poss_experiments_our pobj_In_experiments nsubj_use_we aux_use_will num_kinds_4 amod_kinds_different dobj_use_kinds prep_kinds_of nn_combinations_feature pobj_of_combinations amod_Baseline_a157 dep_kinds_Baseline det_features_The num_features_6 nn_features_baseline dep_Baseline_features partmod_features_used prep_used_in amod_cost_ dep_as_such dep_cost_as pobj_as_cost prep_cost_of nn_penalty_word pobj_of_penalty pobj_in_cost prep_cost_of amod_penalty_aligned nn_penalty_template pobj_of_penalty ' +p,7295,'However , the learning curve for Negra -LRB- see Figure 1 -RRB- indicates that the performance of the model is stable , even for small training sets ',Collins,'advmod_indicates_However det_curve_the nn_curve_learning nsubj_indicates_curve prep_curve_for pobj_for_Negra dep_curve_see dobj_see_Figure num_Figure_1 complm_stable_that det_performance_the nsubj_stable_performance prep_performance_of det_model_the amod_model_ pobj_of_model cop_stable_is ccomp_indicates_stable advmod_for_even dep_stable_for amod_sets_small nn_sets_training pobj_for_sets ' +p,7296,'This approach took inspiration from the pioneering work by , but it is also fundamentally different , because instead of grouping similar senses together , the CoreLex approach groups together words according to all of their senses ',Dolan,'det_approach_This nsubj_took_approach dobj_took_inspiration prep_took_from det_work_the amod_work_pioneering pobj_from_work prep_took_by pobj_by_ cc_took_but nsubj_different_it cop_different_is advmod_different_also advmod_different_fundamentally conj_took_different mark_approach_because dep_of_instead advcl_approach_of pcomp_of_grouping amod_senses_similar dobj_grouping_senses advmod_grouping_together det_CoreLex_the nsubj_approach_CoreLex advcl_different_approach dobj_approach_groups advmod_approach_together dep_took_words dep_words_according dep_according_to pobj_to_all prep_all_of poss_senses_their pobj_of_senses ' +o,7297,'? ? word proximity : For the web searches , uses the NEAR operator and considers only those documents that contain the adjectives within a specific proximity ',Turney,'dep_uses_word advmod_word_proximity prep_uses_For det_searches_the nn_searches_web pobj_For_searches nsubj_uses_ det_operator_the amod_operator_NEAR dobj_uses_operator cc_uses_and conj_uses_considers advmod_documents_only det_documents_those dobj_considers_documents nsubj_contain_that rcmod_documents_contain det_adjectives_the dobj_contain_adjectives prep_contain_within det_proximity_a amod_proximity_specific pobj_within_proximity ' +n,7298,'1 Introduction Statistical phrase-based systems have consistently delivered state-of-the-art performance in recent machine translation evaluations , yet these systems remain weak at handling word order changes ',Och,'num__1 nn__Introduction nn__Statistical amod__phrase-based nn__systems nsubj_delivered_ aux_delivered_have advmod_delivered_consistently amod_performance_state-of-the-art dobj_delivered_performance prep_delivered_in amod_evaluations_recent nn_evaluations_machine nn_evaluations_translation pobj_in_evaluations advmod_delivered_yet det_systems_these nsubj_weak_systems cop_weak_remain dep_delivered_weak prep_weak_at pcomp_at_handling nn_changes_word nn_changes_order dobj_handling_changes ' +o,7299,'We follow the approach by , who note that the semantic orientation of an adjective depends on the noun that it modifies and suggest using adjective-noun or adverb-verb pairs to extract semantic orientation ',Turney,'nsubj_follow_We det_approach_the dobj_follow_approach prep_follow_by pobj_by_ nsubj_note_who rcmod__note complm_depends_that det_orientation_the amod_orientation_semantic nsubj_depends_orientation prep_orientation_of det_adjective_an pobj_of_adjective ccomp_note_depends prep_depends_on det_noun_the pobj_on_noun complm_modifies_that nsubj_modifies_it ccomp_depends_modifies cc_modifies_and conj_modifies_suggest xcomp_suggest_using amod_pairs_adjective-noun cc_adjective-noun_or conj_adjective-noun_adverb-verb dobj_using_pairs aux_extract_to xcomp_using_extract amod_orientation_semantic dobj_extract_orientation ' +o,7300,'The reader is referred to for detailed information about phrase-based statistical machine translation ',Koehn,'det_reader_The nsubjpass_referred_reader auxpass_referred_is aux__to xcomp_referred_ prep__for amod_information_detailed pobj_for_information prep_information_about amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_about_translation ' +o,7301,' where the same idea of improving generalization and accuracy by looking at word classes instead of individual words is used ',Brown,'advmod_used_where det_idea_the amod_idea_same nsubjpass_used_idea prep_idea_of pcomp_of_improving dobj_improving_generalization cc_generalization_and conj_generalization_accuracy prep_improving_by pcomp_by_looking prep_looking_at nn_classes_word pobj_at_classes dep_of_instead prep_classes_of amod_words_individual pobj_of_words auxpass_used_is dep__used ' +o,7302,'The performance of PB-SMT system is measured with BLEU score ',Papineni,'det_performance_The nsubjpass_measured_performance prep_performance_of nn_system_PB-SMT pobj_of_system auxpass_measured_is prep_measured_with amod_score_BLEU pobj_with_score ' +o,7303,'This fact , along with the observation that machine translation quality improves as the amount of monolingual training material increases , has lead to the introduction of randomised techniques for representing large LMs in small space ',Talbot,'det_fact_This nsubj_lead_fact dep_improves_along dep_along_with det_observation_the pobj_with_observation dep_improves_that nn_quality_machine nn_quality_translation nsubj_improves_quality dep_fact_improves prep_improves_as det_amount_the pobj_as_amount prep_amount_of amod_increases_monolingual nn_increases_training nn_increases_material pobj_of_increases aux_lead_has prep_lead_to det_introduction_the pobj_to_introduction prep_introduction_of amod_techniques_randomised pobj_of_techniques prep_lead_for pcomp_for_representing amod_LMs_large dobj_representing_LMs prep_representing_in amod_space_small pobj_in_space ' +o,7304,'In , different word orderings are taken into account by training alignment models by considering all hypothesis pairs as a parallel corpus using GIZA + + ',Och,'prep_taken_In pobj_In_ amod_orderings_different nn_orderings_word nsubjpass_taken_orderings auxpass_taken_are prep_taken_into pobj_into_account prep_taken_by pcomp_by_training amod_models_alignment dobj_training_models prep_training_by pcomp_by_considering det_pairs_all nn_pairs_hypothesis dobj_considering_pairs prep_considering_as det_corpus_a amod_corpus_parallel pobj_as_corpus partmod_corpus_using acomp_using_GIZA cc_GIZA_+ conj_GIZA_+ ' +o,7305,'One way around this dif culty is to stipulate that all rules must be binary from the outset , as in inversion-transduction grammar -LRB- ITG -RRB- and the binary synchronous context-free grammar -LRB- SCFG -RRB- employed by the Hiero system to model the hierarchical phrases ',Wu,'num_way_One nsubj_is_way prep_way_around det_culty_this nn_culty_dif pobj_around_culty aux_stipulate_to xcomp_is_stipulate complm_binary_that det_rules_all nsubj_binary_rules aux_binary_must cop_binary_be ccomp_stipulate_binary prep_binary_from det_outset_the pobj_from_outset advmod_employed_as dep_as_in amod__inversion-transduction nn__grammar abbrev__ITG pobj_in_ cc__and det_grammar_the advmod_synchronous_binary amod_grammar_synchronous amod_grammar_context-free conj__grammar abbrev_grammar_SCFG partmod_outset_employed prep_employed_by det__the nn__Hiero nn__system pobj_by_ aux_model_to xcomp_employed_model det_phrases_the amod_phrases_hierarchical dobj_model_phrases ' +o,7306,'As has been pointed out by , the calculation of log assumes a binomial distribution ',Dunning,'mark_pointed_As aux_pointed_has auxpass_pointed_been advcl_assumes_pointed prt_pointed_out prep_pointed_by pobj_by_ det_calculation_the nsubj_assumes_calculation prep_calculation_of pobj_of_log det_distribution_a amod_distribution_binomial dobj_assumes_distribution ' +o,7307,'The commonly used phrase extraction approach based on word alignment heuristics -LRB- referred as ViterbiExtract algorithm for comparison in this paper -RRB- as described in is a special case of the algorithm , where candidate phrase pairs are restricted to those that respect word alignment boundaries ',Koehn,'det_approach_The advmod_used_commonly amod_approach_used nn_approach_phrase nn_approach_extraction prep_approach_based dep_based_on nn_heuristics_word nn_heuristics_alignment pobj_on_heuristics dep_heuristics_referred prep_referred_as nn_algorithm_ViterbiExtract pobj_as_algorithm prep_referred_for pobj_for_comparison prep_comparison_in det_paper_this pobj_in_paper mark_case_as csubj_case_described prep_described_in pobj_in_ cop_case_is det_case_a amod_case_special dep_heuristics_case prep_case_of det_algorithm_the pobj_of_algorithm advmod_restricted_where nn_pairs_candidate nn_pairs_phrase nsubjpass_restricted_pairs auxpass_restricted_are rcmod_algorithm_restricted prep_restricted_to pobj_to_those nsubj_respect_that rcmod_those_respect nn_boundaries_word nn_boundaries_alignment dobj_respect_boundaries ' +o,7308,' -RRB- , while exploring word-to-expression -LRB- inter-expression -RRB- relations has connections to techniques that employ more of a global-view of corpus statistics -LRB- eg , -RRB- 1 While most previousresearch exploits only one or the other type of relation , we propose a unified method that can exploit both types of semantic relation , while adapting a general purpose polarity lexicon into a domain specific one ',Kanayama,'csubj_has_ dep_exploring_while dobj__exploring amod_exploring_word-to-expression dep_word-to-expression_inter-expression advmod__relations dobj_has_connections prep_connections_to pobj_to_techniques nsubj_employ_that rcmod_techniques_employ dobj_employ_more prep_more_of det_global-view_a pobj_of_global-view prep_global-view_of nn_statistics_corpus pobj_of_statistics appos_statistics_eg dep_eg_ dep_propose_1 mark_previousresearch_While nsubj_previousresearch_most dep_propose_previousresearch dobj_previousresearch_exploits quantmod_one_only dep_exploits_one cc_exploits_or det_type_the amod_type_other conj_exploits_type prep_type_of pobj_of_relation nsubj_propose_we dep_has_propose det_method_a amod_method_unified dobj_propose_method nsubj_exploit_that aux_exploit_can rcmod_method_exploit det_types_both dobj_exploit_types prep_types_of amod_relation_semantic pobj_of_relation dep_adapting_while parataxis_propose_adapting det_lexicon_a amod_lexicon_general nn_lexicon_purpose nn_lexicon_polarity dobj_adapting_lexicon prep_adapting_into det_domain_a pobj_into_domain amod_domain_specific tmod_specific_one ' +o,7309,'We used an implementation of McDonald forcomparisonofresult ',Clarke,'nsubj_used_We det_implementation_an dobj_used_implementation prep_implementation_of nn_forcomparisonofresult_McDonald nn_forcomparisonofresult_ pobj_of_forcomparisonofresult ' +o,7310,'302 was nlanually annotated with subjeciivity chlssifications ',Marcus,'num__302 nsubjpass_annotated_ auxpass_annotated_was advmod_annotated_nlanually prep_annotated_with amod_chlssifications_subjeciivity pobj_with_chlssifications ' +o,7311,'The translation problem can be statistically formulated as in ',Brown,'det_problem_The nn_problem_translation nsubj_formulated_problem aux_formulated_can cop_formulated_be advmod_formulated_statistically prep_formulated_as dep_as_in ' +p,7312,'They have been successfully applied to accent restoration , word \' sense disambiguation 209 and homograph disambiguation ',Yarowsky,'nsubjpass_applied_They aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_to nn_restoration_accent pobj_to_restoration poss_disambiguation_word nn_disambiguation_sense conj_restoration_disambiguation dep_disambiguation_209 cc_restoration_and amod_disambiguation_homograph conj_restoration_disambiguation ' +o,7313,'Since this relation can often be determined automatically for a given text , we can readily use it to improve rank prediction ',Marcu,'mark_determined_Since det_relation_this nsubjpass_determined_relation aux_determined_can advmod_determined_often auxpass_determined_be advcl_use_determined advmod_determined_automatically prep_determined_for det__a amod__given nn__text pobj_for_ nsubj_use_we aux_use_can advmod_use_readily nsubj_improve_it aux_improve_to xcomp_use_improve amod_prediction_rank dobj_improve_prediction ' +p,7314,'Generative word alignment models , initially developed at IBM , and then augmented by an HMM-based model , have provided powerful modeling capability for word alignment ',Brown,'amod_models_Generative nn_models_word nn_models_alignment nsubj_provided_models advmod_developed_initially dep_models_developed prep_developed_at nn__IBM pobj_at_ cc_developed_and advmod_augmented_then conj_developed_augmented prep_augmented_by det__an amod__HMM-based nn__model pobj_by_ aux_provided_have amod_capability_powerful nn_capability_modeling dobj_provided_capability prep_capability_for nn_alignment_word pobj_for_alignment ' +o,7315,'Then they adapted statistical translation Model 2 to work with this model of cooccurrence ',Brown,'advmod_adapted_Then nsubj_adapted_they amod_translation_ amod_translation_statistical nsubj_Model_translation dep_adapted_Model dobj_Model_2 aux_work_to xcomp_Model_work prep_work_with det_model_this pobj_with_model prep_model_of pobj_of_cooccurrence ' +p,7316,'The limitations of the generative approach to sequence tagging , i e Hidden Markov Models , have been overcome by discriminative approaches proposed in recent years ',Collins,'det_limitations_The nsubjpass_overcome_limitations prep_limitations_of det_approach_the amod_approach_generative pobj_of_approach prep_approach_to pobj_to_sequence partmod_sequence_tagging advmod_tagging_i dep_tagging_e nn_Models_Hidden nn_Models_Markov dep_e_Models aux_overcome_have auxpass_overcome_been prep_overcome_by amod_approaches_discriminative pobj_by_approaches partmod_approaches_proposed prep_proposed_in amod_years_recent pobj_in_years ' +p,7317,'GIZA + + is a very popular system within SMT for creating word alignment from parallel corpus , in fact , the Moses training scripts uses it ',Och,'nsubj_system_GIZA amod_GIZA_+ cc_+_+ conj_+_ cop_system_is det_system_a advmod_popular_very amod_system_popular prep_system_within pobj_within_SMT prep_uses_for pcomp_for_creating nn_alignment_word dobj_creating_alignment prep_creating_from amod_corpus_parallel pobj_from_corpus prep_creating_in pobj_in_fact det_scripts_the nn_scripts_Moses nn_scripts_training nsubj_uses_scripts dep_system_uses dobj_uses_it ' +o,7318,'There are other approaches to statistical machine translation where translation is achieved through transduction of source language structure to target language structure ',Wu,'expl_are_There amod_approaches_other nsubj_are_approaches prep_approaches_to amod_translation_statistical nn_translation_machine pobj_to_translation advmod_achieved_where nsubjpass_achieved_translation auxpass_achieved_is rcmod_translation_achieved prep_achieved_through pobj_through_transduction prep_transduction_of nn_structure_source nn_structure_language pobj_of_structure aux_target_to xcomp_achieved_target nn_structure_language dobj_target_structure ' +o,7319,'2 Background : MaxEnt Models Maximum Entropy -LRB- MaxEnt -RRB- models are widely used in Natural Language Processing ',Berger,'num_Background_2 dep_used_Background nn_Entropy_MaxEnt nn_Entropy_Models nn_Entropy_Maximum nn_models_Entropy appos_models_MaxEnt nsubjpass_used_models auxpass_used_are advmod_used_widely prep_used_in nn_Processing_Natural nn_Processing_Language pobj_in_Processing ' +o,7320,'With this model , we can provide not only qualitative textual summarization such as good food and bad service , but also a numerical scoring of sentiment , ie , how good the food is and how bad the service is 2 Related Work There have been many studies on sentiment classification and opinion summarization ',Pang,'prep_provide_With det_model_this pobj_With_model nsubj_provide_we aux_provide_can cc_summarization_not dep_not_only amod_summarization_qualitative amod_summarization_textual dobj_provide_summarization dep_as_such prep_summarization_as amod_food_good pobj_as_food cc_food_and amod_service_bad conj_food_service dep_also_but cc_summarization_also det_scoring_a amod_scoring_numerical conj_summarization_scoring prep_scoring_of pobj_of_sentiment appos_sentiment_ie advmod_good_how dep_is_good det_food_the nsubj_is_food dep_scoring_is cc_is_and advmod_bad_how dep_Work_bad det_service_the nsubj_Work_service cop_Work_is num_Work_2 amod_Work_Related conj_is_Work expl_been_There aux_been_have rcmod_Work_been amod_studies_many dobj_been_studies prep_studies_on nn_classification_sentiment pobj_on_classification cc_classification_and nn_summarization_opinion conj_classification_summarization ' +o,7321,'This is confirmed by a comparison between our baseline result -LRB- F = 1 = 554 \% -RRB- and some baseline results of English base-NP chunking task -LRB- eg precision = 819 \% , recall = 782 \% , F = 1 = 800 \% -RRB- ',Ramshaw,'nsubjpass_confirmed_This auxpass_confirmed_is prep_confirmed_by det_comparison_a pobj_by_comparison prep_comparison_between poss_result_our amod_result_baseline pobj_between_result nsubj_=_F dep_result_= dobj_=_1 dep_\%_= num_\%_554 dep_1_\% cc_comparison_and det_results_some amod_results_baseline conj_comparison_results prep_results_of amod_task_English nn_task_base-NP amod_task_chunking pobj_of_task nn_precision_eg dep_task_precision dep_\%_= num_\%_819 rcmod_precision_\% appos_precision_recall dep_\%_= num_\%_782 rcmod_recall_\% appos_precision_F dep_F_= num_=_1 iobj_=_= dep_\%_800 amod__\% dobj_=_ ' +o,7322,'The first system is the Pharaoh decoder provided by for the shared data task ',Koehn,'det_system_The amod_system_first nsubj_decoder_system cop_decoder_is det_decoder_the nn_decoder_Pharaoh partmod_decoder_provided prep_provided_by pobj_by_ prep__for det_task_the amod_task_shared nn_task_data pobj_for_task ' +p,7323,'For English , self-training contributes 083 \% absolute improvement to the PCFG-LA parser , which is comparable to the improvement obtained from using semi-supervised training with the twostage parser in ',McClosky,'prep_contributes_For pobj_For_English nsubj_contributes_self-training num_\%_083 dobj_contributes_\% amod_improvement_absolute tmod_contributes_improvement prep_improvement_to det_parser_the amod_parser_PCFG-LA pobj_to_parser nsubj_comparable_which cop_comparable_is rcmod_parser_comparable prep_comparable_to det_improvement_the pobj_to_improvement partmod_improvement_obtained prep_obtained_from pcomp_from_using amod_training_semi-supervised dobj_using_training prep_using_with det_parser_the amod_parser_twostage pobj_with_parser prep_using_in ' +o,7324,'The supervised component is Collins parser , trained on the Wall Street Journal ',Collins,'det_component_The amod_component_supervised nsubj__component cop__is nn__Collins nn__parser partmod__trained prep_trained_on det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_on_Journal ' +o,7325,'The recent work of and were also sources of inspiration ',Haghighi,'det_work_The amod_work_recent nsubj_sources_work prep_work_of pobj_of_ cc__and conj__ cop_sources_were advmod_sources_also prep_sources_of pobj_of_inspiration ' +o,7326,'The features used by the decoder were the English language model log probability , logf -LRB- e f -RRB- , the lexical translation log probabilities in both directions , and a word count feature ',Koehn,'det_features_The nsubj_probability_features partmod_features_used prep_used_by det_decoder_the pobj_by_decoder cop_probability_were det_probability_the amod_probability_English nn_probability_language nn_probability_model nn_probability_log appos_probability_logf dep_f_e dep_-LRB-_f det_translation_the amod_translation_lexical nsubj_log_translation ccomp_probability_log dobj_log_probabilities prep_probabilities_in det__both nn__directions pobj_in_ cc_probability_and det_count_a nn_count_word nsubj_feature_count conj_probability_feature ' +o,7327,'In Englishto-German , this result produces results very comparable to a phrasal SMT system trained on the same data ',Koehn,'prep_produces_In pobj_In_Englishto-German det_result_this nsubj_produces_result nsubj_comparable_results advmod_comparable_very xcomp_produces_comparable prep_comparable_to det__a amod__phrasal nn__SMT nn__system pobj_to_ partmod__trained prep_trained_on det_data_the amod_data_same pobj_on_data ' +o,7328,'On smaller data sets the joint model shows performance comparable to the standard model , however the joint model does not reach the level of performance of the stan156 EN-ES ES-EN Joint 3-gram , dl4 2051 2664 5-gram , dl6 2634 2717 + lex ',Koehn,'prep_sets_On amod_data_smaller pobj_On_data dobj_sets_ det_lex_the amod_model_joint nn_shows_model nn_lex_shows nn_lex_performance amod_lex_comparable prep_comparable_to det_model_the amod_model_standard pobj_to_model advmod_reach_however det_model_the amod_model_joint nsubj_reach_model aux_reach_does neg_reach_not dep_model_reach det_level_the dobj_reach_level prep_level_of pobj_of_performance prep_performance_of det_ES-EN_the amod_ES-EN_stan156 nn_ES-EN_EN-ES pobj_of_ES-EN nn_lex_Joint amod_lex_3-gram amod_lex_dl4 number_2664_2051 num_lex_2664 amod_lex_5-gram amod_lex_dl6 number_2717_2634 num_lex_2717 amod_lex_+ dep__lex ' +o,7329,'We envision the use of a clever datastructure would reduce the complexity , but leave this to future work , as the experiments -LRB- Table 8 -RRB- show that 5Our definition implies that we only consider faithful spans to be contiguous ',Galley,'nsubj_envision_We det_use_the nsubj_reduce_use prep_use_of det_datastructure_a amod_datastructure_clever pobj_of_datastructure aux_reduce_would ccomp_envision_reduce det_complexity_the dobj_reduce_complexity cc_reduce_but conj_reduce_leave dobj_leave_this prep_leave_to amod_work_future pobj_to_work mark_show_as det_experiments_the nsubj_show_experiments number_8_Table dep_experiments_8 advcl_leave_show complm_implies_that amod_definition_5Our nsubj_implies_definition ccomp_show_implies complm_consider_that nsubj_consider_we advmod_consider_only ccomp_implies_consider amod_spans_faithful nsubj_contiguous_spans aux_contiguous_to cop_contiguous_be xcomp_consider_contiguous ' +o,7330,'For example , many statistical part-of-speech -LRB- POS -RRB- taggers have been developed and they use corpora as the training data to obtain statistical information or rules ',Ratnaparkhi,'prep_developed_For pobj_For_example amod_part-of-speech_many amod_part-of-speech_statistical nn_taggers_part-of-speech appos_taggers_POS nsubjpass_developed_taggers aux_developed_have auxpass_developed_been cc_developed_and nsubj_use_they conj_developed_use dobj_use_corpora prep_use_as det_data_the nn_data_training pobj_as_data aux_obtain_to xcomp_use_obtain amod_information_statistical dobj_obtain_information cc_information_or conj_information_rules ' +p,7331,'They are central to many parsing models , and despite their simplicity n-gram models have been very successful ',Collins,'nsubj_central_They cop_central_are prep_central_to amod_models_many amod_models_parsing pobj_to_models partmod_models_ cc_central_and prep_successful_despite poss_simplicity_their pobj_despite_simplicity amod_models_n-gram nsubj_successful_models aux_successful_have dep_successful_been advmod_successful_very conj_central_successful ' +o,7332,'Most of this prior work deals with supervised transfer learning , and thus requires labeled source domain data , though there are examples of unsupervised , semi-supervised , and transductive approaches ',Blitzer,'nsubj_deals_Most prep_Most_of det_work_this amod_work_prior pobj_of_work prep_deals_with amod_learning_supervised nn_learning_transfer pobj_with_learning cc_deals_and advmod_deals_thus conj_deals_requires dep_data_labeled nn_data_source nn_data_domain dep_requires_data mark_are_though expl_are_there advcl_data_are nsubj_are_examples prep_examples_of amod__unsupervised pobj_of_ amod__semi-supervised conj__ cc__and amod_approaches_transductive conj__approaches advmod_are_ ' +o,7333,'Word alignments were generated using Model 4 using the multi-threaded implementation of GIZA + + ',Brown,'amod_alignments_Word nsubjpass_generated_alignments auxpass_generated_were xcomp_generated_using nsubj_using_Model dep__4 dep_using_ xcomp_using_using det_implementation_the amod_implementation_multi-threaded nsubj_+_implementation prep_implementation_of pobj_of_GIZA xcomp_using_+ cc_+_+ conj_+_ ' +o,7334,'We used the NP data prepared by , hereafter RM95 ',Ramshaw,'nsubj_used_We det_NP_the dobj_used_NP tmod_used_data partmod_data_prepared prep_prepared_by pobj_by_ advmod_data_hereafter dep_data_RM95 ' +o,7335,'have been used in statistical machine translation , terminology research and translation aids , bilingual lexicography , word-sense disambiguation and information retrieval in a multilingual environment ',Brown,'aux_used_have auxpass_used_been prep_used_in amod_research_statistical nn_research_machine nn_research_translation nn_research_ dep_research_terminology pobj_in_research cc_research_and nn_aids_translation conj_research_aids amod__ amod__bilingual nn__lexicography nsubjpass_used_ amod__word-sense nn__disambiguation appos__ cc__and nn_retrieval_information conj__retrieval prep__in det__a amod__multilingual nn__environment pobj_in_ ' +o,7336,'For mutual information -LRB- MI -RRB- , we use two different equations : one for two-element compound nouns and the other for three-element compound nouns ',Church,'prep_use_For amod_information_mutual pobj_For_information abbrev_information_MI nsubj_use_we num_equations_two amod_equations_different dobj_use_equations dep_equations_one prep_one_for amod__two-element amod__compound nn__nouns pobj_for_ cc_equations_and det__the amod__other prep_other_for amod_nouns_three-element nn_nouns_compound pobj_for_nouns conj_equations_ ' +o,7337,'Until now , translation models have been evaluated either subjectively or using relative metrics , such as perplexity with respect to other models ',Dunning,'prep_evaluated_Until pobj_Until_now nn_models_translation nsubjpass_evaluated_models aux_evaluated_have auxpass_evaluated_been advmod_evaluated_either advmod__subjectively acomp_evaluated_ cc__or conj__using amod_metrics_relative dobj_using_metrics dep_as_such prep_metrics_as pobj_as_perplexity prep_perplexity_with pobj_with_respect prep_using_to amod__other nn__models pobj_to_ ' +o,7338,'The features used are : the length of t ; a single-parameter distortion penalty on phrase reordering in a , as described in ; phrase translation model probabilities ; and 4-gram language model probabilities logp -LRB- t -RRB- , using Kneser-Ney smoothing as implemented in the SRILM toolkit ',Koehn,'det_features_The nsubj_length_features partmod_features_used cop_length_are det_length_the prep_length_of pobj_of_t det_penalty_a amod_penalty_single-parameter nn_penalty_distortion dep_length_penalty prep_penalty_on nn_reordering_phrase pobj_on_reordering prep_reordering_in det_logp_a advmod_described_as amod_logp_described prep_described_in pobj_in_ nn_probabilities_phrase nn_probabilities_translation nn_probabilities_model dep__probabilities cc__and amod_probabilities_4-gram nn_probabilities_language nn_probabilities_model conj__probabilities pobj_in_logp appos_logp_t partmod_reordering_using dobj_using_Kneser-Ney partmod_Kneser-Ney_smoothing mark_implemented_as advcl_smoothing_implemented prep_implemented_in det__the nn__SRILM nn__toolkit pobj_in_ ' +o,7339,'The performance of tl , e presented tagger is measured and compared to that of two other taggers ',Cutting,'det_performance_The prep_measured_of pobj_of_tl dep_measured_e amod_tagger_presented nsubjpass_measured_tagger auxpass_measured_is dep_performance_measured cc_measured_and conj_measured_compared prep_measured_to pobj_to_that prep_that_of num_taggers_two amod_taggers_other pobj_of_taggers ' +o,7340,'The differences between a k-best and a beam-search parser -LRB- not to mention the use of dynamic programming -RRB- make a running time difference unsur17 Our score of 858 average labeled precision and recall for sentences less than or equal to 100 on Section 23 compares to : 867 in Charniak , 869 in Ratnaparkhi , 882 in , 896 in Charniak -LRB- 2000 -RRB- , and 8975 in Collins -LRB- 2000 -RRB- ',Collins,'det_differences_The nsubj_make_differences prep_differences_between det_k-best_a pobj_between_k-best cc_k-best_and det_parser_a amod_parser_beam-search conj_k-best_parser neg_mention_not aux_mention_to parataxis_make_mention det_use_the dobj_mention_use prep_use_of amod_programming_dynamic pobj_of_programming det_unsur17_a amod_unsur17_running nn_unsur17_time nn_unsur17_difference dobj_make_unsur17 poss_score_Our nsubj_compares_score prep_score_of num_average_858 pobj_of_average dep_average_labeled dobj_labeled_precision cc_precision_and conj_precision_recall prep_precision_for pobj_for_sentences advmod_labeled_less dep_less_than cc_labeled_or conj_labeled_equal prep_equal_to pobj_to_100 prep_equal_on pobj_on_Section num_Section_23 dep_make_compares prep_compares_to pobj_to_867 prep_867_in pobj_in_Charniak num_Charniak_ num_Charniak_869 prep_867_in pobj_in_Ratnaparkhi num_Ratnaparkhi_ num_Ratnaparkhi_882 prep_867_in pobj_in_ appos_867_896 prep_896_in pobj_in_Charniak appos_Charniak_2000 cc_Charniak_and conj_Charniak_8975 prep_8975_in pobj_in_Collins appos_Collins_2000 ' +o,7341,'To identify these terms,weusethelog-likelihoodstatisticsuggested by Dunning and first used in summarization by Lin and Hovy ',Dunning,'aux_identify_To det_terms,weusethelog-likelihoodstatisticsuggested_these dobj_identify_terms,weusethelog-likelihoodstatisticsuggested prep_identify_by nn__Dunning pobj_by_ cc_identify_and advmod_identify_first conj_identify_used prep_used_in pobj_in_summarization prep_used_by nn__Lin cc_Lin_and conj_Lin_Hovy pobj_by_ ' +p,7342,'1 Introduction In recent years , statistical machine translation have experienced a quantum leap in quality thanks to automatic evaluation and errorbased optimization ',Papineni,'num_Introduction_1 prep_experienced_In amod_years_recent pobj_In_years amod_translation_statistical nn_translation_machine nsubj_experienced_translation aux_experienced_have dep_Introduction_experienced det_leap_a nn_leap_quantum dobj_experienced_leap prep_leap_in nn_thanks_quality pobj_in_thanks prep_experienced_to amod__automatic nn__evaluation pobj_to_ cc__and amod__errorbased nn__optimization conj__ ' +o,7343,'Looking rst at learning times, it is obvious that learning time depends primarily on the number of training instances, which is why we can observe a difference of several orders of magnitude in learning time between the biggest training set (Czech) and the smallest training set (Slovene) 14 This is shown by Nivre and Scholz (2004) in comparison to the iterative, arc-standard algorithm of Yamada and Matsumoto (2003) and by McDonald and Nivre (2007) in comparison to the spanning tree algorithm of McDonald, Lerman, and Pereira (2006).',McDonald,'' +o,7344,'Inspired by s methodology which was originally designed for English and Penn-II treebank , our approach to Chinese non-local dependency recovery is based on Lexical-Functional Grammar -LRB- LFG -RRB- , a formalism that involves both phrase structure trees and predicate-argument structures ',Cahill,'prep_Inspired_by pobj_by_ dep_methodology_s dep_Inspired_methodology nsubjpass_designed_which auxpass_designed_was advmod_designed_originally rcmod_methodology_designed prep_designed_for pobj_for_English cc_English_and conj_English_Penn-II nsubj_Inspired_treebank poss_approach_our appos_treebank_approach prep_approach_to amod_recovery_Chinese amod_recovery_non-local nn_recovery_dependency pobj_to_recovery auxpass_based_is rcmod_recovery_based prep_based_on nn_Grammar_Lexical-Functional pobj_on_Grammar abbrev_Grammar_LFG det_formalism_a appos_Grammar_formalism nsubj_involves_that rcmod_formalism_involves preconj_trees_both nn_trees_phrase nn_trees_structure dobj_involves_trees cc_trees_and amod_structures_predicate-argument conj_trees_structures ' +o,7345,'LW was originally used to validate the quality of a phrase translation pair in MT ',Koehn,'nsubjpass_used_LW auxpass_used_was advmod_used_originally aux_validate_to xcomp_used_validate det_quality_the dobj_validate_quality prep_quality_of det_pair_a nn_pair_phrase nn_pair_translation pobj_of_pair prep_validate_in pobj_in_MT ' +o,7346,'It is sometimes assumed that estimates of entropy -LRB- eg , Shannon \'s estimate that English is 75 \% redundant , upper bound of 175 bits per character for printed English -RRB- are directly 3There are some cases where words are deliberately misspelled in order to get better output from the synthesizer , such as coyote spelled kiote ',Brown,'nsubjpass_assumed_It auxpass_assumed_is advmod_assumed_sometimes complm_3There_that nsubj_3There_estimates prep_estimates_of pobj_of_entropy dep_estimates_eg poss_estimate_Shannon possessive_Shannon_\'s appos_eg_estimate complm_redundant_that nsubj_redundant_English cop_redundant_is dep_\%_75 amod_redundant_\% dep_estimate_redundant xcomp_redundant_ amod_bound_upper dobj__bound prep_bound_of num_bits_175 pobj_of_bits prep_bits_per pobj_per_character prep_character_for nn_English_printed pobj_for_English cop_3There_are advmod_3There_directly ccomp_assumed_3There cop_cases_are det_cases_some rcmod_3There_cases advmod_misspelled_where nsubjpass_misspelled_words auxpass_misspelled_are advmod_misspelled_deliberately advcl_3There_misspelled prep_misspelled_in pobj_in_order aux_get_to xcomp_misspelled_get amod_output_better dobj_get_output prep_get_from det_synthesizer_the pobj_from_synthesizer dep_as_such prep_3There_as pobj_as_coyote partmod_coyote_spelled dep_spelled_kiote ' +o,7347,'Other representative collocation research can be found in Church and Hanks and ',Smadja,'amod_research_Other nn_research_representative nn_research_collocation nsubjpass_found_research aux_found_can auxpass_found_be prep_found_in pobj_in_Church cc_Church_and nn_and_Hanks nn_and_ conj_Church_and ' +o,7348,'In a second top-down pass similar to , we can recalculate psyn -LRB- d -RRB- for alternative derivations in the hypergraph ; potentially correcting search errors made in the first pass ',Huang,'prep_recalculate_In det_pass_a amod_pass_second amod_pass_top-down pobj_In_pass amod_pass_similar dep_similar_to dep_similar_ nsubj_recalculate_we aux_recalculate_can dobj_recalculate_psyn dep_psyn_d prep_psyn_for amod_derivations_alternative pobj_for_derivations prep_recalculate_in det_hypergraph_the pobj_in_hypergraph advmod_correcting_potentially parataxis_recalculate_correcting nn_errors_search dobj_correcting_errors partmod_errors_made prep_made_in det_pass_the amod_pass_first pobj_in_pass ' +o,7349,'298 within LFG includes the XLE ,3 , Hogan et al ',Cahill,'prep_298_within pobj_within_LFG dep_298_includes det__the nn__XLE num__,3 nsubj_298_ appos__Hogan cc_Hogan_et conj_Hogan_al ' +o,7350,'We borrow the idea of classifying definites occurring in the first sentence as chain starting from ',Bean,'nsubj_borrow_We det_idea_the dobj_borrow_idea prep_idea_of amod_definites_classifying pobj_of_definites partmod_definites_occurring prep_occurring_in det_sentence_the amod_sentence_first pobj_in_sentence prep_sentence_as pobj_as_chain prep_occurring_starting dep_starting_from ' +p,7351,'First , such a system makes use of lexical information when modeling reordering , which has previously been shown to be useful in German-to-English translation ',Lopez,'advmod_makes_First predet_system_such det_system_a nsubj_makes_system dobj_makes_use prep_use_of amod_information_lexical pobj_of_information advmod__when nn_reordering_modeling nsubj__reordering dep_use_ nsubjpass_shown_which aux_shown_has advmod_shown_previously auxpass_shown_been rcmod_use_shown aux_useful_to cop_useful_be xcomp_shown_useful prep_useful_in amod__German-to-English nn__translation pobj_in_ ' +n,7352,'Our approach not only outperformed a notoriously difficult baseline but also achieved similar performance to the approach of , without requiring their third-party data resources ',Svore,'poss_approach_Our nsubj_outperformed_approach cc_outperformed_not dep_not_only det_baseline_a advmod_difficult_notoriously amod_baseline_difficult dobj_outperformed_baseline cc_outperformed_but advmod_achieved_also conj_outperformed_achieved amod_performance_similar dobj_achieved_performance prep_achieved_to det_approach_the pobj_to_approach prep_approach_of pobj_of_ prep_achieved_without pcomp_without_requiring poss_resources_their amod_resources_third-party nn_resources_data dobj_requiring_resources ' +o,7353,' , extracts uninterrupted as well as interrupted collocations -LRB- predicative relations , rigid noun phrases and phrasal templates -RRB- ',Smadja,'dep__extracts amod_extracts_uninterrupted dep_well_as cc_uninterrupted_well dep_well_as conj_uninterrupted_interrupted dep_interrupted_collocations amod_relations_predicative dep_interrupted_relations amod_phrases_rigid nn_phrases_noun conj_relations_phrases cc_relations_and amod_templates_phrasal conj_relations_templates ' +o,7354,'Similarlyto , we define the strength of a pattern p in a category y as the precision of p in the set of documents labeled with category y , estimated using Laplace smoothing : strength -LRB- p , y -RRB- = count -LRB- p , y -RRB- + epsilon1count -LRB- p -RRB- + kepsilon1 -LRB- 3 -RRB- where count -LRB- p , y -RRB- is the number of documents labeled y containing pattern p , count -LRB- p -RRB- is the overall number of labeled documents containing p , and k is the number of domains ',Yarowsky,'nn__Similarlyto nsubj_estimated_ nsubj_define_we parataxis_estimated_define det_strength_the dobj_define_strength prep_strength_of det_p_a nn_p_pattern pobj_of_p prep_p_in det_y_a nn_y_category pobj_in_y prep_define_as det_precision_the pobj_as_precision prep_precision_of pobj_of_p prep_p_in det_set_the pobj_in_set prep_set_of pobj_of_documents partmod_documents_labeled prep_labeled_with nn_y_category pobj_with_y xcomp_estimated_using dobj_using_Laplace partmod_Laplace_smoothing nsubj_=_strength appos_strength_p dep_p_y rcmod_Laplace_= nn_kepsilon1_count appos_kepsilon1_p dep_p_y nn_kepsilon1_+ nn_kepsilon1_epsilon1count appos_kepsilon1_p amod_kepsilon1_+ dobj_=_kepsilon1 appos_kepsilon1_3 advmod_number_where nsubj_number_count appos_count_p dep_p_y cop_number_is det_number_the rcmod_kepsilon1_number prep_number_of pobj_of_documents partmod_documents_labeled dobj_labeled_y xcomp_labeled_containing nn_p_pattern dobj_containing_p nsubj_number_count appos_count_p cop_number_is det_number_the amod_number_overall ccomp_estimated_number prep_number_of amod_documents_labeled pobj_of_documents partmod_documents_containing dobj_containing_p cc_estimated_and nsubj_number_k cop_number_is det_number_the conj_estimated_number prep_number_of pobj_of_domains ' +o,7355,'There are rules , though rare , that can not be binarized synchronously at all , but can be incorporated in two-stage decoding with asynchronous binarization ',Wu,'expl_are_There nsubj_are_rules dep_rare_though amod_rules_rare nsubjpass_binarized_that aux_binarized_can neg_binarized_not auxpass_binarized_be rcmod_rules_binarized advmod_binarized_synchronously prep_binarized_at det__all pobj_at_ cc_are_but aux_incorporated_can auxpass_incorporated_be conj_are_incorporated prep_incorporated_in amod_decoding_two-stage pobj_in_decoding prep_incorporated_with amod_binarization_asynchronous pobj_with_binarization ' +o,7356,'82 Chen and Chang Topical Clustering maintains the position that intersense relations are mostly idiosyncratical , thereby making it difficult to characterize them in a general way so as to identify them ',Dolan,'nn_Chen_82 nsubj_position_Chen cc_Chen_and nn__Chang nn__Topical nn__Clustering conj_Chen_ dep_position_maintains det_position_the complm_idiosyncratical_that amod_relations_intersense nsubj_idiosyncratical_relations cop_idiosyncratical_are advmod_idiosyncratical_mostly ccomp_position_idiosyncratical advmod_making_thereby xcomp_idiosyncratical_making nsubj_difficult_it xcomp_making_difficult aux_characterize_to xcomp_difficult_characterize dobj_characterize_them prep_characterize_in det_way_a amod_way_general pobj_in_way advmod_as_so advmod_characterize_as aux_identify_to dep_as_identify dobj_identify_them ' +o,7357,'3 Formulation Following Klein and Manning , we use weighted directed hypergraphs as an abstraction of the probabilistic parsing problem ',Marcus,'nn_Formulation_3 prep_use_Following pobj_Following_Klein cc_Klein_and nn__Manning conj_Klein_ nsubj_use_we dep_Formulation_use amod_hypergraphs_weighted amod_hypergraphs_directed dobj_use_hypergraphs partmod_hypergraphs_ prep__as det_abstraction_an pobj_as_abstraction prep_abstraction_of det_problem_the amod_problem_probabilistic nn_problem_parsing pobj_of_problem ' +o,7358,'(Mann and Yarowsky, 2003; Chen and Martin, 2007; Baron and Freedman, 2008).',Chen,'' +p,7359,'This representation , being contiguous on both sides , successfully reduces the decoding complexity to a low polynomial and significantly improved the search quality ',Zhang,'det_representation_This nsubj_reduces_representation cop_contiguous_being partmod_representation_contiguous prep_contiguous_on det_sides_both pobj_on_sides advmod_reduces_successfully det_complexity_the amod_complexity_decoding dobj_reduces_complexity prep_reduces_to det_polynomial_a amod_polynomial_low pobj_to_polynomial cc_reduces_and advmod_improved_significantly conj_reduces_improved det_quality_the nn_quality_search dobj_improved_quality ' +n,7360,'While we have shown an increase in performance over a purely syntactic baseline model -LRB- the algorithm of -RRB- , there are a number of avenues to pursue in extending this work ',Brown,'mark_shown_While nsubj_shown_we aux_shown_have advcl_are_shown det_increase_an dobj_shown_increase prep_increase_in pobj_in_performance prep_shown_over det_model_a advmod_syntactic_purely amod_model_syntactic nn_model_baseline pobj_over_model det_algorithm_the dep_model_algorithm prep_algorithm_of pobj_of_ expl_are_there det_number_a nsubj_are_number prep_number_of pobj_of_avenues aux_pursue_to infmod_avenues_pursue prep_pursue_in pcomp_in_extending det_work_this dobj_extending_work ' +o,7361,' directly estimate DPs tbr : ~ given int -RRB- ut , whereas other models su -LRB- \' h as PCFOt -RRB- ased tel -RRB- down generation mod -LRB- ; ls P -LRB- H , , , s -RRB- do not , -LRB- Charnink , 1997 ; , 1997 ; Shir ~ fi et ~ rl ',Collins,'advmod_estimate_directly rcmod__estimate nn_tbr_DPs dobj_estimate_tbr punct_tbr_~ amod_ut_given nn_ut_int dep_tbr_ut mark_ased_whereas amod_su_other nn_su_models nsubj_ased_su dep_su_h prep_h_as pobj_as_PCFOt dep__ased dobj_ased_tel prep__down punct_mod_generation pobj_down_mod amod_P_ls dep_fi_P prep_P_H aux_do_s rcmod_P_do dobj_do_not nn_1997_Charnink dep_1997_1997 dep_1997_ dep_not_1997 cc_do_Shir conj_do_~ prep_mod_fi pobj_fi_et dep_fi_~ pobj_~_rl ' +o,7362,'4 Global Transliteration Modeling In global transliteration modeling , we directly model the agreement function between f and e We follow and consider the global feature representation : F \\* E \\* R d 613 Each global feature corresponds to a condition on the pair of strings ',Collins,'num_Modeling_4 nn_Modeling_Global nn_Modeling_Transliteration dep_In_Modeling prep_model_In amod_modeling_global nn_modeling_transliteration pobj_In_modeling nsubj_model_we advmod_model_directly det_function_the nn_function_agreement dobj_model_function prep_function_between nn__f cc_f_and dep_follow_e nsubj_follow_We conj_f_follow pobj_between_ cc_model_and conj_model_consider det_F_the amod_F_global nn_F_feature nn_F_representation dobj_consider_F dep_E_\\* dep_corresponds_E dep_R_\\* dep_d_R dep_613_d dep_corresponds_613 det_feature_Each amod_feature_global nsubj_corresponds_feature dep_model_corresponds prep_corresponds_to det_condition_a pobj_to_condition prep_condition_on det_pair_the pobj_on_pair prep_pair_of pobj_of_strings ' +p,7363,'We report on ROUGE-1 -LRB- unigrams -RRB- , ROUGE-2 -LRB- bigrams -RRB- , ROUGE W-12 -LRB- weighted LCS -RRB- , and ROUGE-S \\* -LRB- skip bigrams -RRB- as they have been shown to correlate well with human judgments for longer multidocument summaries ',Lin,'nsubj_report_We prep_report_on pobj_on_ROUGE-1 appos_ROUGE-1_unigrams conj_ROUGE-1_ROUGE-2 appos_ROUGE-2_bigrams nn_W-12_ROUGE conj_ROUGE-1_W-12 amod_LCS_weighted abbrev_W-12_LCS cc_ROUGE-1_and nn_\\*_ROUGE-S conj_ROUGE-1_\\* num_bigrams_skip appos_\\*_bigrams mark_shown_as nsubjpass_shown_they aux_shown_have auxpass_shown_been advcl_report_shown aux_correlate_to xcomp_shown_correlate advmod_correlate_well prep_correlate_with amod_judgments_human pobj_with_judgments prep_judgments_for amod_summaries_longer nn_summaries_multidocument pobj_for_summaries ' +o,7364,'The baseline hierarchical phrase-based system is trained using standard max-BLEU training -LRB- MERT -RRB- without sparse features ',Och,'det_system_The amod_system_baseline amod_system_hierarchical amod_system_phrase-based nsubjpass_trained_system auxpass_trained_is xcomp_trained_using amod_training_standard amod_training_max-BLEU dobj_using_training abbrev_training_MERT prep_using_without amod_features_sparse pobj_without_features ' +o,7365,'The lexicalized parsing experiments were run using Dan Bikels probabilistic parsing engine which in addition to replicating the models described by also provides a convenient interface to develop corresponding parsing models for other languages ',Collins,'det_experiments_The amod_experiments_lexicalized nn_experiments_parsing nsubjpass_run_experiments auxpass_run_were xcomp_run_using nn_Bikels_Dan nsubj_parsing_Bikels dep_parsing_probabilistic xcomp_using_parsing nn__engine dobj_parsing_ nsubj_provides_which prep_provides_in pobj_in_addition prep_addition_to pcomp_to_replicating det_models_the dobj_replicating_models partmod_models_described prep_described_by pobj_by_ advmod_provides_also rcmod__provides det_interface_a amod_interface_convenient dobj_provides_interface aux_develop_to infmod_interface_develop amod_models_corresponding amod_models_parsing dobj_develop_models prep_models_for amod_languages_other pobj_for_languages ' +o,7366,'Labelling was carried out by three computational linguistics graduate students with 89 \% agreement resulting in a Kappa statistic of 087 , which is a satisfactory indication that our corpus can be labeled with high reliability using our tag set ',Carletta,'auxpass_carried_was dep_Labelling_carried prt_carried_out prep_carried_by num_linguistics_three amod_linguistics_computational pobj_by_linguistics amod_students_graduate nsubj_Labelling_students prep_students_with num_agreement_89 nn_agreement_\% pobj_with_agreement partmod_agreement_resulting prep_resulting_in det_statistic_a nn_statistic_Kappa pobj_in_statistic prep_statistic_of pobj_of_087 nsubj_indication_which cop_indication_is det_indication_a amod_indication_satisfactory rcmod_students_indication complm_labeled_that poss_corpus_our nsubjpass_labeled_corpus aux_labeled_can auxpass_labeled_be ccomp_indication_labeled prep_labeled_with amod_reliability_high pobj_with_reliability purpcl_labeled_using poss_set_our nn_set_tag dobj_using_set ' +o,7367,'Speaker ranking accuracy Table 2 summarizes the accuracy of our statistical ranker on the test data with different feature sets : the performance is 8939 \% when using all feature sets , and reaches 902 \% after applying Gaussian smoothing and using incremental feature selection as described in and implemented in the yasmetFS package6 Note that restricting ourselves to only backward looking features decreases the performance significantly , as we can see in Table 2 ',Berger,'nn_accuracy_Speaker amod_accuracy_ranking nsubj_Table_accuracy num_summarizes_2 iobj_Table_summarizes det_accuracy_the dobj_Table_accuracy prep_accuracy_of poss_ranker_our amod_ranker_statistical pobj_of_ranker prep_ranker_on det_data_the nn_data_test pobj_on_data prep_data_with amod_sets_different nn_sets_feature pobj_with_sets det_performance_the nsubj_\%_performance cop_\%_is num_\%_8939 parataxis_Table_\% advmod_sets_when csubj_sets_using det_feature_all dobj_using_feature advcl_\%_sets cc_sets_and conj_sets_reaches num_\%_902 dobj_reaches_\% prep_reaches_after pcomp_after_applying nsubj_Note_Gaussian partmod_Gaussian_smoothing cc_smoothing_and conj_smoothing_using amod_selection_incremental nn_selection_feature dobj_smoothing_selection advmod_described_as amod_selection_described prep_described_in pobj_in_ cc_described_and conj_described_implemented prep_implemented_in det_package6_the amod_package6_yasmetFS pobj_in_package6 ccomp_applying_Note complm_decreases_that csubj_decreases_restricting dobj_restricting_ourselves prep_restricting_to advmod_backward_only pobj_to_backward partmod_backward_looking dobj_looking_features ccomp_Note_decreases det_performance_the dobj_decreases_performance advmod_decreases_significantly mark_see_as nsubj_see_we aux_see_can advcl_\%_see prep_see_in pobj_in_Table num_Table_2 ' +o,7368,'5 Perceptron Training We optimize feature weights using a modification of averaged perceptron learning as described by ',Collins,'num_Training_5 nn_Training_Perceptron nsubj_optimize_We dep_Training_optimize nn_weights_feature nsubj_using_weights dep_optimize_using det_modification_a dobj_using_modification prep_modification_of amod_perceptron_averaged pobj_of_perceptron partmod_perceptron_learning advmod_described_as acomp_learning_described prep_described_by ' +o,7369,'This way of creating classified data is similar to that in ',Yarowsky,'det_way_This nsubj_similar_way prep_way_of pcomp_of_creating amod_data_classified dobj_creating_data cop_similar_is prep_similar_to pobj_to_that prep_similar_in ' +o,7370,'This includes the automatic generation of sense-tagged data using monosemous relatives , automatically bootstrapped disambiguation patterns , parallel texts as a way to point out word senses bearing different translations in a second language , and the use of volunteer contributions over the Web ',Yarowsky,'nsubj_includes_This det_generation_the amod_generation_automatic dobj_includes_generation prep_generation_of amod_data_sense-tagged pobj_of_data partmod_data_using amod_relatives_monosemous dobj_using_relatives advmod_using_ advmod_generation_automatically amod_patterns_bootstrapped nn_patterns_disambiguation dep_generation_patterns partmod_patterns_ amod_texts_parallel conj_patterns_texts prep_texts_as det_way_a pobj_as_way aux_point_to infmod_way_point prt_point_out nn_senses_word dobj_point_senses xcomp_point_bearing amod_translations_different dobj_bearing_translations prep_bearing_in det__a amod__second nn__language pobj_in_ cc_patterns_and det_use_the conj_patterns_use prep_use_of nn_contributions_volunteer pobj_of_contributions prep_contributions_over det__the nn__Web pobj_over_ ' +p,7371,' , but we use a maximum entropy classifier to determine parser actions , which makes parsing considerably faster ',Berger,'cc_use_but nsubj_use_we ccomp__use det_classifier_a amod_classifier_maximum nn_classifier_entropy dobj_use_classifier dep__ aux_determine_to xcomp__determine nn_actions_parser nsubj__actions nsubj_makes_which rcmod_actions_makes xcomp_makes_parsing advmod_parsing_considerably advmod_parsing_faster ' +o,7372,'To model p -LRB- t , a s -RRB- , we use a standard loglinear approach : p -LRB- t , a s -RRB- exp bracketleftBiggsummationdisplay i ifi -LRB- s , t , a -RRB- bracketrightBigg where each fi -LRB- s , t , a -RRB- is a feature function , and weights i are set using Ochs algorithm to maximize the systems BLEU score -LRB- Papineni et aal ',Och,'aux_model_To dep_use_model dobj_model_p dep_p_t det_s_a appos_t_s nsubj_use_we det_approach_a amod_approach_standard nn_approach_loglinear dobj_use_approach nsubj_exp_p dep_p_t det_s_a appos_t_s parataxis_use_exp advmod_i_bracketleftBiggsummationdisplay amod_ifi_i dobj_exp_ifi appos_ifi_s dep_s_t dep_s_a xcomp_exp_bracketrightBigg advmod_function_where det_fi_each nsubj_function_fi appos_fi_s dep_s_t dep_s_a cop_function_is det_function_a nn_function_feature advcl_bracketrightBigg_function cc_use_and nn_i_weights nsubjpass_set_i auxpass_set_are conj_use_set xcomp_set_using nn__Ochs nn__algorithm nsubj_maximize_ aux_maximize_to xcomp_using_maximize det_score_the nn_score_systems nn_score_BLEU nsubj_aal_score nn_aal_Papineni nn_aal_et xcomp_maximize_aal ' +n,7373,'Presently , there exist methods for learning oppositional terms and paraphrase learning has been thoroughly studied , but successfully extending these techniques to learn incompatible phrases poses difficulties because of the data distribution ',Marcu,'advmod_exist_Presently expl_exist_there nsubjpass_studied_methods prep_methods_for pcomp_for_learning amod_terms_oppositional nsubj_learning_terms dep_learning_ cc__and conj__paraphrase xcomp_learning_learning aux_studied_has auxpass_studied_been advmod_studied_thoroughly dep_exist_studied cc_studied_but advmod_poses_successfully csubj_poses_extending det_techniques_these nsubj_learn_techniques aux_learn_to xcomp_extending_learn amod_phrases_incompatible dobj_learn_phrases conj_studied_poses dobj_poses_difficulties dep_of_because prep_difficulties_of det_distribution_the nn_distribution_data pobj_of_distribution ' +o,7374,' describe an approach that targets translation of French phrases of the form NOUN de NOUN -LRB- eg , conflit dinteret -RRB- ',Berger,'advmod_describe_ det_approach_an dobj_describe_approach nsubj_targets_that rcmod_approach_targets dobj_targets_translation prep_translation_of amod_phrases_French pobj_of_phrases prep_phrases_of det_form_the pobj_of_form nn_NOUN_NOUN dep_NOUN_de nsubj_describe_NOUN appos_NOUN_eg nn_dinteret_conflit dep_eg_dinteret ' +o,7375,'First as the configuration space we can use only the reference nodes -LRB- w -RRB- from the lattice which makes it similar to the method of et al 1996 described in section 21 ',Berger,'advmod_use_First dep_First_as det_space_the nn_space_configuration pobj_as_space nsubj_use_we aux_use_can advmod_nodes_only det_nodes_the nn_nodes_reference dobj_use_nodes appos_nodes_w prep_use_from det_lattice_the pobj_from_lattice nsubj_makes_which rcmod_lattice_makes nsubj_similar_it xcomp_makes_similar prep_similar_to det_method_the pobj_to_method prep_method_of pobj_of_ cc__et conj__al nsubj_described_1996 dep_use_described prep_described_in pobj_in_section num_section_21 ' +o,7376,'For the log-linear model training , we take minimum-error-rate training method as described in ',Och,'prep_take_For det_training_the amod_training_log-linear nn_training_model pobj_For_training nsubj_take_we amod_method_minimum-error-rate nn_method_training nsubj_described_method advmod_described_as xcomp_take_described prep_described_in ' +o,7377,'Minimum error-rate -LRB- MER -RRB- training was applied to obtain weights -LRB- m in Equation 2 -RRB- for these features ',Och,'nn__Minimum amod__error-rate nn__MER nn__training nsubjpass_applied_ auxpass_applied_was aux_obtain_to xcomp_applied_obtain dobj_obtain_weights dep_weights_m prep_m_in pobj_in_Equation num_Equation_2 prep_weights_for det_features_these pobj_for_features ' +o,7378,'12 Recent work A few publications , so far , deal with POS-tagging of Northern Sotho ; most prominently , de Schryver and de Pauw have presented the MaxTag method , a tagger based on Maximum Entropy 38 Learning as implemented in the machine learning package Maxent ',Berger,'num_work_12 amod_work_Recent det_publications_A amod_publications_few nsubj_deal_publications advmod_far_so advmod_deal_far dep_work_deal prep_deal_with pobj_with_POS-tagging prep_POS-tagging_of nn_Sotho_Northern pobj_of_Sotho advmod_prominently_most advmod_presented_prominently nn_Schryver_de nsubj_presented_Schryver cc_Schryver_and nn__de nn__Pauw conj_Schryver_ aux_presented_have parataxis_deal_presented det_method_the nn_method_MaxTag dobj_presented_method det_tagger_a appos_method_tagger partmod_tagger_based prep_based_on nn_Entropy_Maximum pobj_on_Entropy number_Learning_38 num__Learning dep_as_ advmod_implemented_as partmod_Entropy_implemented prep_implemented_in det_machine_the pobj_in_machine xcomp_implemented_learning nn__package nn__Maxent dobj_learning_ ' +o,7379,'In the WSJ PennTreebank corpus is analyzed and a very detailed list of syntactic patterns that correspond to different roles of commas is created ',Marcus,'prep_analyzed_In pobj_In_ det__the nn__WSJ nn__PennTreebank nn__corpus nsubjpass_analyzed_ auxpass_analyzed_is dep_list_and det_list_a advmod_detailed_very amod_list_detailed dep_analyzed_list prep_list_of amod_patterns_syntactic pobj_of_patterns complm_created_that csubjpass_created_correspond prep_correspond_to amod_roles_different pobj_to_roles prep_roles_of pobj_of_commas auxpass_created_is ccomp_analyzed_created ' +n,7380,'Formal complexity analysis has not been carried out , but my algorithm is simpler , at least conceptually , than the variable-word-order parsers of Johnson , , and Abramson and Dahl -LRB- 1989 -RRB- ',Kashket,'amod_analysis_Formal nn_analysis_complexity nsubjpass_carried_analysis aux_carried_has neg_carried_not auxpass_carried_been prt_carried_out cc_carried_but poss_algorithm_my nsubj_simpler_algorithm cop_simpler_is conj_carried_simpler advmod_conceptually_at dep_at_least dep_simpler_conceptually prep_simpler_than det_parsers_the amod_parsers_variable-word-order pobj_than_parsers prep_parsers_of pobj_of_Johnson num_Johnson_ num_Johnson_ cc_parsers_and conj_parsers_Abramson cc_Abramson_and conj_Abramson_Dahl appos_Abramson_1989 ' +o,7381,'Our methods are most influenced by IBMs Model 1 ',Brown,'poss_methods_Our nsubjpass_influenced_methods auxpass_influenced_are advmod_influenced_most prep_influenced_by nn_Model_IBMs pobj_by_Model num_Model_1 ' +n,7382,'These scores are higher than those of several other parsers , but remain behind tim scores of Charniak -LRB- 2000 -RRB- who obtains 901 \% LP and 901 \% LR for sentences _ -LRB- 40 words ',Collins,'det_scores_These nsubj_higher_scores cop_higher_are mark__than nsubj__those prep_those_of amod_parsers_several amod_parsers_other pobj_of_parsers ccomp_higher_ cc__but conj__remain prep_remain_behind nn_scores_tim pobj_behind_scores prep_scores_of pobj_of_Charniak appos_Charniak_2000 nsubj_obtains_who rcmod_Charniak_obtains num_\%_901 dobj_obtains_\% nsubj___LP cc_LP_and num_\%_901 conj_LP_\% advmod___LR dep_LR_for pobj_for_sentences dep_higher__ num_words_40 dobj___words ' +o,7383,'This can also be interpreted as a generalization of standard class-based models ',Brown,'nsubjpass_interpreted_This aux_interpreted_can advmod_interpreted_also auxpass_interpreted_be prep_interpreted_as det_generalization_a pobj_as_generalization prep_generalization_of amod_models_standard amod_models_class-based pobj_of_models ' +p,7384,'Recently , have successfully applied self-training to various parser adaptation scenarios using the reranking parser of ',McClosky,'advmod_applied_Recently nsubj_applied_ aux_applied_have advmod_applied_successfully dobj_applied_self-training prep_applied_to amod_scenarios_various nn_scenarios_parser nn_scenarios_adaptation pobj_to_scenarios xcomp_applied_using det_parser_the amod_parser_reranking dobj_using_parser prep_parser_of pobj_of_ ' +p,7385,'An efficient algorithm for performing this tuning for a larger number of model parameters can be found in ',Och,'det_algorithm_An amod_algorithm_efficient nsubjpass_found_algorithm prep_algorithm_for pcomp_for_performing det_tuning_this dobj_performing_tuning prep_performing_for det_number_a amod_number_larger pobj_for_number prep_number_of nn_parameters_model pobj_of_parameters aux_found_can auxpass_found_be prep_found_in ' +o,7386,'The data was segmented into baseNP parts and nonbaseNP parts in a similar fashion as the data used by ',Ramshaw,'det_data_The nsubjpass_segmented_data auxpass_segmented_was prep_segmented_into nn_parts_baseNP pobj_into_parts cc_parts_and nn_parts_nonbaseNP conj_parts_parts prep_segmented_in det_fashion_a amod_fashion_similar pobj_in_fashion prep_fashion_as det_data_the pobj_as_data partmod_data_used prep_used_by ' +p,7387,'Beam-search has been successful in many NLP tasks -LRB- Koehn et al , 2003 ; 562 Inputs : training examples -LRB- xi , yi -RRB- Initialization : set vectorw = 0 Algorithm : \\/ \\/ R training iterations ; N examples for t = 1R , i = 1N : zi = argmaxyGEN -LRB- xi -RRB- -LRB- y -RRB- vectorw if zi negationslash = yi : vectorw = vectorw + -LRB- yi -RRB- -LRB- zi -RRB- Outputs : vectorw Figure 1 : The perceptron learning algorithm and Roark , 2004 -RRB- , and can achieve accuracy that is close to exact inference ',Collins,'nsubj_successful_Beam-search aux_successful_has dep_successful_been prep_successful_in amod_tasks_many nn_tasks_NLP pobj_in_tasks dep_tasks_Koehn cc_Koehn_et conj_Koehn_al appos_Koehn_2003 nn_Inputs_562 dep_2003_Inputs nn_examples_training dep_2003_examples appos_examples_xi dep_xi_yi dep_examples_Initialization nn_vectorw_set dep_2003_vectorw dep_Algorithm_= nn_Algorithm_0 rcmod_vectorw_Algorithm nn_\\/_\\/ dep_2003_\\/ nn_iterations_R nn_iterations_training dep_\\/_iterations nn_examples_N dep_2003_examples prep_examples_for pobj_for_t dep_1R_= rcmod_t_1R appos_t_i dep_1N_= rcmod_i_1N nn_argmaxyGEN_zi nn_argmaxyGEN_= nn_vectorw_argmaxyGEN appos_argmaxyGEN_xi appos_argmaxyGEN_y dep_2003_vectorw mark_yi_if nn_negationslash_zi nsubj_yi_negationslash dep_yi_= dep_vectorw_yi dep_2003_vectorw dep_vectorw_= iobj_=_vectorw dobj_=_+ appos_+_yi appos_+_zi dep_vectorw_Outputs nn_Figure_vectorw dep_2003_Figure dep_Figure_1 det__The amod__perceptron nn__learning nn__algorithm dep_2003_ cc__and conj__Roark appos_Koehn_2004 cc_successful_and aux_achieve_can conj_successful_achieve dobj_achieve_accuracy nsubj_close_that cop_close_is rcmod_accuracy_close prep_close_to amod_inference_exact pobj_to_inference ' +o,7388,'To make feature ranking computationally tractable in Della Pietra et al 1995 and et al 1996 a simplified process proposed : at the feature ranking stage when adding a new feature to the model all previously computed parameters are kept fixed and , thus , we have to fit only one new constraint imposed by a candidate feature ',Berger,'aux_make_To csubj_et_make nn_computationally_feature amod_computationally_ranking dobj_make_computationally amod_computationally_tractable prep_tractable_in nn_Pietra_Della pobj_in_Pietra cc_Pietra_et conj_Pietra_al tmod_tractable_1995 dep_1995_and dep_1995_ iobj_et_al num_al_1996 det_process_a amod_process_simplified dobj_et_process partmod_process_proposed prep_et_at det_stage_the nn_stage_feature amod_stage_ranking pobj_at_stage advmod_adding_when dep_computed_adding det_feature_a amod_feature_new dobj_adding_feature prep_adding_to det_model_the pobj_to_model nsubj_computed_all advmod_computed_previously dep_et_computed nsubjpass_kept_parameters auxpass_kept_are dep_computed_kept ccomp_kept_fixed cc_kept_and dep_kept_thus nsubj_have_we conj_kept_have aux_fit_to xcomp_have_fit advmod_constraint_only num_constraint_one amod_constraint_new dobj_fit_constraint partmod_constraint_imposed prep_imposed_by det_feature_a nn_feature_candidate pobj_by_feature ' +o,7389,'We draw on and extend the work of ',Marcu,'nsubj_draw_We prt_draw_on cc_draw_and conj_draw_extend det_work_the dobj_extend_work prep_work_of ' +o,7390,'The translations were generated by the alignment template system of ',Och,'det_translations_The nsubjpass_generated_translations auxpass_generated_were prep_generated_by det_system_the amod_system_alignment nn_system_template pobj_by_system prep_system_of ' +o,7391,'4 Evaluation 41 Experimental Setup For evaluation , we use five sentiment classification datasets , including the widely-used movie review dataset -LRB- MOV -RRB- as well as four datasets that contain reviews of four different types of product from Amazon -LRB- books -LRB- BOO -RRB- , DVDs -LRB- DVD -RRB- , electronics -LRB- ELE -RRB- , and kitchen appliances -LRB- KIT -RRB- -RRB- ',Blitzer,'num_Setup_4 num_Setup_Evaluation num_Setup_41 nn_Setup_Experimental prep_use_For pobj_For_evaluation nsubj_use_we dep_Setup_use num_datasets_five nn_datasets_sentiment nn_datasets_classification dobj_use_datasets prep_datasets_including det_dataset_the amod_dataset_widely-used nn_dataset_movie nn_dataset_review pobj_including_dataset abbrev_dataset_MOV partmod_dataset_ dep_well_as cc_dataset_well dep_well_as num_datasets_four conj_dataset_datasets nsubj_contain_that rcmod_datasets_contain dobj_contain_reviews prep_reviews_of num_types_four amod_types_different pobj_of_types prep_types_of pobj_of_product prep_contain_from pobj_from_Amazon dep_datasets_books abbrev_books_BOO conj_books_DVDs abbrev_DVDs_DVD appos_DVDs_electronics abbrev_electronics_ELE cc_books_and nn_appliances_kitchen conj_books_appliances abbrev_appliances_KIT ' +p,7392,'By having the advantage of leveraging large parallel corpora , the statistical MT approach outperforms the traditional transfer based approaches in tasks for which adequate parallel corpora is available ',Och,'prep_outperforms_By pcomp_By_having det_advantage_the dobj_having_advantage prep_advantage_of pcomp_of_leveraging amod_corpora_large amod_corpora_parallel dobj_leveraging_corpora det_approach_the amod_approach_statistical nn_approach_MT nsubj_outperforms_approach det_transfer_the amod_transfer_traditional dobj_outperforms_transfer partmod_transfer_based dobj_based_approaches prep_based_in pobj_in_tasks rel_available_for pobj_for_which amod_corpora_adequate amod_corpora_parallel nsubj_available_corpora cop_available_is rcmod_tasks_available ' +o,7393,'For word alignment accuracy , F-measure is reported , ie , the harmonic mean of precision and recall against a gold-standard reference set ; for translation quality , Bleu and its variation of NIST scores are reported ',Papineni,'dep_set_For nn_accuracy_word nn_accuracy_alignment pobj_For_accuracy nsubjpass_reported_F-measure auxpass_reported_is dep_set_reported nsubj_set_ie det_mean_the amod_mean_harmonic appos_ie_mean prep_mean_of pobj_of_precision cc_precision_and conj_precision_recall prep_mean_against det_reference_a amod_reference_gold-standard pobj_against_reference prep__for nn_quality_translation pobj_for_quality nsubj__Bleu parataxis_set_ cc_set_and poss_variation_its nsubjpass_reported_variation prep_variation_of amod_scores_NIST pobj_of_scores auxpass_reported_are conj_set_reported ' +o,7394,'Minimum Error Rate Training -LRB- MERT -RRB- under BLEU criterion is used to estimate 20 feature function weights over the larger development set -LRB- dev1 -RRB- ',Och,'nn__Minimum nn__Error nn__Rate nn__Training abbrev__MERT nsubjpass_used_ prep__under amod_criterion_BLEU pobj_under_criterion auxpass_used_is aux_estimate_to xcomp_used_estimate num_weights_20 nn_weights_feature nn_weights_function dobj_estimate_weights prep_estimate_over det_set_the amod_set_larger nn_set_development pobj_over_set appos_set_dev1 ' +o,7395,'We then parse both sides of the corpus with syntactic parsers ',Collins,'nsubj_parse_We advmod_parse_then det_sides_both dobj_parse_sides prep_sides_of det_corpus_the pobj_of_corpus prep_corpus_with amod_parsers_syntactic pobj_with_parsers ' +o,7396,'S S0 , n Si , k Si , j Sj , k Si1 , i pii Figure 1 : A grammar for a large neighborhood of permutations , given one permutation pi of length n The Si , k rules are instantiated for each 0 i -LRB- j -LRB- k n , and the Si1 , i rules for each 0 , which is a familiar reordering device in machine translation ',Wu,'nn_S0_S nsubjpass_instantiated_S0 nn_Si_n appos_S0_Si nn_Si_k appos_S0_Si nn_Sj_j appos_S0_Sj nn_Si1_k appos_S0_Si1 amod_Figure_i nn_Figure_pii appos_S0_Figure dep_Figure_1 det_grammar_A dep_Figure_grammar prep_grammar_for det_neighborhood_a amod_neighborhood_large pobj_for_neighborhood prep_neighborhood_of pobj_of_permutations prep_instantiated_given num_pi_one nn_pi_permutation dep_given_pi prep_pi_of nn_Si_length nn_Si_n nn_Si_The pobj_of_Si nn_rules_k nsubjpass_instantiated_rules auxpass_instantiated_are prep_instantiated_for det_n_each amod_n_0 nn_n_i nn_n_j nn_n_k pobj_for_n cc_n_and det_Si1_the conj_n_Si1 dep_rules_i appos_n_rules prep_rules_for det__each num__0 pobj_for_ nsubj_device_which cop_device_is det_device_a amod_device_familiar nn_device_reordering rcmod__device prep_device_in nn_translation_machine pobj_in_translation ' +o,7397,'Interestingly , similar conclusions were also reached in the area of Machine Translation evaluation ; in their experiments , Zhang and Vogel show that adding an additional reference translation compensates the effects of removing 1015 \% of the testing data , and state that , therefore , it seems more cost effective to have more test sentences but fewer reference translations ',Lin,'advmod_reached_Interestingly amod_conclusions_similar nsubjpass_reached_conclusions auxpass_reached_were advmod_reached_also prep_reached_in det_area_the pobj_in_area prep_area_of nn_evaluation_Machine nn_evaluation_Translation pobj_of_evaluation prep_show_in poss_experiments_their pobj_in_experiments nn__Zhang cc_Zhang_and conj_Zhang_Vogel nsubj_show_ parataxis_reached_show complm_compensates_that csubj_compensates_adding det_translation_an amod_translation_additional nn_translation_reference dobj_adding_translation ccomp_show_compensates det_effects_the dobj_compensates_effects prep_effects_of pcomp_of_removing num_\%_1015 dobj_removing_\% prep_\%_of det_data_the amod_data_testing pobj_of_data cc_\%_and conj_\%_state complm_seems_that advmod_seems_therefore nsubj_seems_it dep_state_seems amod_cost_more nsubj_effective_cost xcomp_seems_effective aux_have_to xcomp_effective_have amod_sentences_more nn_sentences_test dobj_have_sentences prep_sentences_but amod_translations_fewer nn_translations_reference dep_but_translations ' +o,7398,'Many-to-many word alignments are induced by running a one-to-many word alignment model , such as GIZA + + , in both directions and by combining the results based on a heuristic ',Och,'amod_alignments_Many-to-many nn_alignments_word nsubjpass_induced_alignments auxpass_induced_are prep_induced_by pcomp_by_running det_model_a amod_model_one-to-many nn_model_word nn_model_alignment dobj_running_model dep_as_such prep_model_as pobj_as_GIZA amod_GIZA_+ cc_+_+ conj_+_ prep_running_in det_directions_both pobj_in_directions cc_by_and conj_by_by pcomp_by_combining det_results_the dobj_combining_results prep_combining_based dep_based_on det__a amod__heuristic pobj_on_ ' +p,7399,'More sophisticated first-order accounts may be extendable to bear this load ',Hobbs,'advmod_sophisticated_More amod_accounts_sophisticated amod_accounts_first-order nsubj_extendable_ aux_extendable_may cop_extendable_be dep_accounts_extendable aux_bear_to xcomp_extendable_bear det_load_this dobj_bear_load ' +o,7400,' -RRB- , and the phrase-based approach to Statistical Machine Translation has led to the development of heuristics for obtaining alignments between phrases of any number of words ',Koehn,'csubj_led_ dep_approach_and dep_and_the nn_approach_phrase-based dobj__approach prep_approach_to nn_Machine_Statistical pobj_to_Machine tmod__Translation num_Translation_ aux_led_has prep_led_to det_development_the pobj_to_development prep_development_of pobj_of_heuristics prep_led_for pcomp_for_obtaining dobj_obtaining_alignments prep_alignments_between pobj_between_phrases prep_phrases_of det_number_any pobj_of_number prep_number_of pobj_of_words ' +o,7401,'1 Introduction Conditional Maximum Entropy -LRB- maxent -RRB- models have been widely used for a variety of tasks , including language modeling , part-of-speech tagging , prepositional phrase attachment , and parsing , word selection for machine translation , and finding sentence boundaries ',Berger,'num_models_1 amod_models_Introduction nn_models_Conditional nn_models_Maximum nn_models_Entropy appos_models_maxent aux_used_have auxpass_used_been advmod_used_widely dep_models_used prep_used_for det_variety_a pobj_for_variety prep_variety_of pobj_of_tasks prep_variety_including nn__language nn__modeling pobj_including_ amod_tagging_part-of-speech conj__tagging amod_attachment_prepositional nn_attachment_phrase conj__attachment cc__and amod__parsing conj__ nn_selection_word appos__selection prep_selection_for nn__machine nn__translation pobj_for_ cc_used_and conj_used_finding nn_boundaries_sentence dobj_finding_boundaries advmod_finding_ ' +o,7402,'The training and test set were derived by finding all instances of the confusable words in the Brown Corpus , using the Penn Treebank parts of speech and tokenization , and then dividing this set into 80 \% for training and 20 \% for testing ',Marcus,'det_set_The nn_set_training cc_training_and conj_training_test nsubjpass_derived_set auxpass_derived_were prep_derived_by pcomp_by_finding det_instances_all dobj_finding_instances prep_instances_of det_words_the amod_words_confusable pobj_of_words prep_words_in det_Corpus_the nn_Corpus_Brown pobj_in_Corpus conj_finding_using det_parts_the nn_parts_Penn nn_parts_Treebank dobj_using_parts prep_parts_of nn__speech cc_speech_and conj_speech_tokenization pobj_of_ cc_finding_and advmod_dividing_then conj_finding_dividing det_set_this dobj_dividing_set prep_dividing_into num_\%_80 pobj_into_\% prep_\%_for pobj_for_training cc_\%_and num_\%_20 conj_\%_\% prep_\%_for pobj_for_testing ' +o,7403,'1 Introduction Word alignment is an important step of most modern approaches to statistical machine translation ',Koehn,'num_alignment_1 nn_alignment_Introduction nn_alignment_Word nsubj_step_alignment cop_step_is det_step_an amod_step_important prep_step_of amod_approaches_most amod_approaches_modern pobj_of_approaches prep_step_to amod_translation_statistical nn_translation_machine pobj_to_translation ' +o,7404,'This is similar to results in the literature ',Ramshaw,'nsubj_similar_This cop_similar_is prep_similar_to pobj_to_results prep_results_in det_literature_the pobj_in_literature ' +o,7405,'4 The Experiments For the experiments , we used PropBank -LRB- wwwcisupennedu\\/ace -RRB- along with PennTreeBank5 2 -LRB- wwwcisupennedu\\/treebank -RRB- ',Marcus,'dep_used_4 det_Experiments_The dep_For_Experiments prep_used_For det_experiments_the pobj_For_experiments nsubj_used_we measure_along_PropBank appos_PropBank_wwwcisupennedu\\/ace advmod_used_along prep_used_with pobj_with_PennTreeBank5 num_PennTreeBank5_2 dep_used_wwwcisupennedu\\/treebank ' +o,7406,'Previous workonsentimentanalysishascoveredawiderange of tasks , including polarity classification , opinion extraction , and opinion source assignment ',Pang,'amod_workonsentimentanalysishascoveredawiderange_Previous prep_workonsentimentanalysishascoveredawiderange_of pobj_of_tasks prep_tasks_including amod__polarity nn__classification pobj_including_ nn__opinion nn__extraction conj__ cc__and nn__opinion nn__source nn__assignment conj__ ' +o,7407,'They developed a simple heuristic function for Model 2 from which was non admissible ',Brown,'nsubj_developed_They det_function_a amod_function_simple amod_function_heuristic dobj_developed_function prep_developed_for pobj_for_Model num_Model_2 prep_developed_from pobj_from_ nsubjpass_non_which auxpass_non_was rcmod__non dep_non_admissible ' +p,7408,'It us widely acknowledged that word sense d~samblguatmn (WSD) us a central problem m natural language processing In order for computers to be able to understand and process natural language beyond simple keyword matching, the problem of d~samblguatmg word sense, or dlscermng the meamng of a word m context, must be effectively dealt with Advances in WSD v, ill have slgmficant Impact on apphcatlons hke information retrieval and machine translation For natural language subtasks hke part-of-speech tagging or s)ntactm parsing, there are relatlvely well defined and agreed-upon cnterm of what it means to have the \'correct\' part of speech or syntactic structure assigned to a word or sentence For instance, the Penn Treebank corpus (Marcus et al, 1993) pro~ide~,t large repo.~tory of texts annotated w~th partof-speech and s}ntactm structure mformatlon Tv.o independent human annotators can achieve a high rate of agreement on assigning part-of-speech tags to words m a g~ven sentence Unfortunately, th~s us not the case for word sense assignment F~rstly, it is rarely the case that any two dictionaries will have the same set of sense defimtmns for a g~ven word Different d~ctlonanes tend to carve up the \'semantic space\' m a different way, so to speak Secondly, the hst of senses for a word m a typical dmtmnar~ tend to be rather refined and comprehensive This is especmlly so for the commonly used words which have a large number of senses The sense dustmctmn between the different senses for a commonly used word m a d~ctmnary hke WoRDNET (Miller, 1990) tend to be rather fine Hence, two human annotators may genuinely dusagree m their sense assignment to a word m context The agreement rate between human annotators on word sense assignment us an Important concern for the evaluatmn of WSD algorithms One would prefer to define a dusamblguatlon task for which there us reasonably hlgh agreement between human annotators The agreement rate between human annotators will then form the upper ceiling against whmh to compare the performance of WSD algorithms For instance, the SENSEVAL exerclse has performed a detaded study to find out the raterannotator agreement among ~ts lexicographers taggrog the word senses (Kllgamff, 1998c, Kllgarnff, 1998a, Kflgarrlff, 1998b) 2 A Case Study In this-paper, we examine the ~ssue of raterannotator agreement by comparing the agreement rate of human annotators on a large sense-tagged corpus of more than 30,000 instances of the most frequently occurring nouns and verbs of Enghsh This corpus is the intersection of the WORDNET Semcor corpus (Miller et al, 1993) and the DSO corpus (Ng and Lee, 1996, Ng, 1997), which has been independently tagged wlth the refined senses of WORDNET by two separate groups of human annotators The Semcor corpus us a subset of the Brown corpus tagged with ~VoRDNET senses, and consists of more than 670,000 words from 352 text files Sense taggmg was done on the content words (nouns, ~erbs, adjectives and adverbs) m this subset The DSO corpus consists of sentences drawn from the Brown corpus and the Wall Street Journal For each word w from a hst of 191 frequently occurring words of Enghsh (121 nouns and 70 verbs), sentences containing w (m singular or plural form, and m its various reflectional verb form) are selected and each word occurrence w ~s tagged w~th a sense from WoRDNET There ~s a total of about 192,800 sentences in the DSO corpus m which one word occurrence has been sense-tagged m each sentence The intersection of the Semcor corpus and the DSO corpus thus consists of Brown corpus sentences m which a word occurrence w is sense-tagged m each sentence, where w Is one of.the 191 frequently oc-,currmg English nouns or verbs Since this common pomon has been sense-tagged by two independent groups of human annotators, ~t serves as our data set for investigating inter-annotator agreement in this paper 3 Sentence Matching To determine the extent of inter-annotator agreement, the first step ~s to match each sentence m Semcor to its corresponding counterpart In the DSO corpus This step ~s comphcated by the following factors 1 Although the intersected portion of both corpora came from Brown corpus, they adopted different tokemzatmn convention, and segmentartan into sentences differed sometimes 2 The latest versmn of Semcor makes use of the senses from WORDNET 1 6, whereas the senses used m the DSO corpus were from WoRDNET 15 1 To match the sentences, we first converted the senses m the DSO corpus to those of WORDNET 1 6 We ignored all sentences m the DSO corpus m which a word is tagged with sense 0 or -1 (A word is tagged with sense 0 or -1 ff none of the given senses m WoRDNFT applies ) 4, sentence from Semcor is considered to match one from the DSO corpus ff both sentences are exactl) ldent~cal or ff the~ differ only m the pre~ence or absence of the characters \' (permd) or -\' (hyphen) For each remaining Semcor sentence, taking into account word ordering, ff 75\% or more of the words m the sentence match those in a DSO corpus sentence, then a potential match ~s recorded These i -kctua[ly, the WORD~q\'ET senses used m the DSO corpus were from a shght variant of the official WORDNE\'I 1 5 release Th~s ssas brought to our attention after the pubhc release of the DSO corpus potential matches are then manually verffied to ensure that they are true matches and to ~eed out any false matches Using this method of matching, a total of 13,188 sentence-palrs contasnmg nouns and 17,127 sentence-pa~rs containing verbs are found to match from both corpora, ymldmg 30,315 sentences which form the intersected corpus used m our present study 4 The Kappa Statistic Suppose there are N sentences m our corpus where each sentence contains the word w Assume that w has M senses Let 4 be the number of sentences which are assigned identical sense b~ two human annotators Then a simple measure to quantify the agreement rate between two human annotators Is Pc, where Pc, = A/N The drawback of this simple measure is that it does not take into account chance agreement between two annotators The Kappa statistic a (Cohen, 1960) is a better measure of rater-annotator agreement which takes into account the effect of chance agreement It has been used recently w~thm computatmnal hngu~stlcs to measure raterannotator agreement (Bruce and Wmbe, 1998, Carletta, 1996, Veroms, 1998) Let Cj be the sum of the number of sentences which have been assigned sense 3 by annotator 1 and the number of sentences whmh have been assigned sense 3 by annotator 2 Then P~-P~ 1-P~ where M j=l and Pe measures the chance agreement between two annotators A Kappa ~alue of 0 indicates that the agreement is purely due to chance agreement, whereas a Kappa ~alue of 1 indicates perfect agreement A Kappa ~alue of 0 8 and above is considered as mdmatmg good agreement (Carletta, 1996) Table 1 summarizes the inter-annotator agreement on the mtersected corpus The first (becond) row denotes agreement on the nouns (xerbs), wh~le the lass row denotes agreement on all words combined The a~erage ~ reported m the table is a s~mpie average of the individual ~ value of each word The agreement rate on the 30,315 sentences as measured by P= is 57\% This tallies with the figure reported ~n our earlier paper (Ng and Lee, 1996) where we performed a quick test on a subset of 5,317 sentences,n the intersection of both the Semcor corpus and the DSO corpus 10 [] mm m m m m m mm m m m m mm m m m Type Num of v, ords A N [ P~ Avg Nouns 121 7,676 13,188 I 0 582 0 300 Verbs 70 9,520 17,127 I 0 555 0 347 All I 191 I 17,196 30,315 I 056T 0317 Table 1 Raw inter-annotator agreement 5 Algorithm Since the rater-annotator agreement on the intersected corpus is not high, we would like to find out how the agreement rate would be affected if different sense classes were in use In this section, we present a greedy search algorithm that can automatmalb derive coarser sense classes based on the sense tags assigned by two human annotators The resulting derived coarse sense classes achmve a higher agreement rate but we still maintain as many of the original sense classes as possible The algorithm is given m Figure 1 The algorithm operates on a set of sentences where each sentence contains an occurrence of the word w whmh has been sense-tagged by two human annotators At each Iteration of the algorithm, tt finds the pair of sense classes Ct and Cj such that merging these two sense classes results in the highest t~ value for the resulting merged group of sense classes It then proceeds to merge Cz and C~ Thin process Is repeated until the ~ value reaches a satisfactory value ~,~t,~, which we set as 0 8 Note that this algorithm is also applicable to deriving any coarser set of classes from a refined set for any NLP tasks in which prior human agreement rate may not be high enough Such NLP tasks could be discourse tagging, speech-act categorization, etc 6 Results For each word w from the list of 121 nouns and 70 verbs, ~e applied the greedy search algorithm to each set of sentences in the intersected corpus contaming w For a subset of 95 words (53 nouns and 42 verbs), the algorithm was able to derive a coarser set of 2 or more senses for each of these 95 words such that the resulting Kappa ~alue reaches 0 8 or higher For the other 96 words, m order for the Kappa value to reach 0 8 or higher, the algorithm collapses all senses of the ~ord to a single (trivial) class Table 2 and 3 summarizes the results for the set of 53 nouns and 42 ~erbs, respectively Table 2 md~cates that before the collapse of sense classes, these 53 nouns have an average of 7 6 senses per noun There is a total of 5,339 sentences in the intersected corpus containing these nouns, of which 3,387 sentences were assigned the same sense by the two groups of human annotators The average Kappa statistic (computed as a simple average of the Kappa statistic of ~he mdlwdual nouns) is 0 463 After the collapse of sense classes by the greedy search algorithm, the average number of senses per noun for these 53 nouns drops to 40 Howe~er, the number of sentences which have been asmgned the same coarse sense by the annotators increases to 5,033 That is, about 94 3\% of the sentences have been assigned the same coarse sense, and that the average Kappa statistic has improved to 0 862, mgmfymg high rater-annotator agreement on the derived coarse senses Table3 gl~es the analogous figures for the 42 verbs, agmn mdmatmg that high agreement is achieved on the coarse sense classes den~ed for verbs 7 Discussion Our findings on rater-annotator agreement for word sense tagging indicate that for average language users, it is quite dl~cult to achieve high agreement when they are asked to assign refned sense tags (such as those found in WORDNET) given only the scanty definition entries m the WORDNET dlctionary and a few or no example sentences for the usage of each word sense Thin observation agrees wlth that obtmned m a recent study done by (Veroms, 1998), where the agreement on sense-tagging by naive users was also not hlgh Thus It appears that an average language user is able to process language wlthout needing to perform the task of dlsamblguatmg word sense to a very fine-grained resolutmn as formulated m a tradltlonal dmtlonary In contrast, expert lexicographers tagged the ~ ord sense in the sentences used m the SENSEVAL exerclse, where high rater-annotator agreement was reported There are also fuller dlctlonary entries m the HECTOR dlctlonary used and more e ~* then ~\' +~(C~,,C~_t), z* +~, ~* +end for merge the sense class C,.',Marcus,'' +o,7409,'The de-facto answer came during the 1990s from the research community on Statistical Machine Translation , who made use of statistical tools based on a noisy channel model originally developed for speech recognition ',Och,'det_answer_The amod_answer_de-facto nsubj_came_answer prep_came_during det_1990s_the pobj_during_1990s prep_came_from det_community_the nn_community_research pobj_from_community prep_came_on nn_Translation_Statistical nn_Translation_Machine pobj_on_Translation nsubj_made_who rcmod_Translation_made dobj_made_use prep_use_of amod_tools_statistical pobj_of_tools prep_made_based dep_based_on det_model_a amod_model_noisy nn_model_channel pobj_on_model advmod_developed_originally partmod_model_developed prep_developed_for nn_recognition_speech pobj_for_recognition ' +o,7410,'The preci781 start Palestinian suicide bomberblew himself up in SLOT1 on SLOT2 killing SLOT3 other people and injuring wounding SLOT4 end detroit the \\* e \\* a s \\* e \\* building buildingin detroit flattened ground levelled to blasted leveled \\* e \\* was reduced razed leveled to down rubble into ashes \\* e \\* to \\* e \\* -LRB- 1 -RRB- -LRB- 2 -RRB- Figure 1 : Examples of paraphrase patterns extracted by and Pang et al ',Barzilay,'det_preci781_The nsubj_start_preci781 amod_suicide_Palestinian nsubj_bomberblew_suicide ccomp_start_bomberblew dobj_bomberblew_himself prt_bomberblew_up prep_bomberblew_in pobj_in_SLOT1 prep_bomberblew_on pobj_on_SLOT2 dep_SLOT2_killing amod_people_SLOT3 amod_people_other dobj_killing_people cc_killing_and conj_killing_injuring xcomp_injuring_wounding nn_end_SLOT4 nsubj_detroit_end ccomp_wounding_detroit dobj_detroit_the dep_the_\\* dep_\\*_e dep_\\*_\\* det_s_a nsubj_\\*_s dep_start_\\* dep_flattened_e dep_flattened_\\* nn_detroit_building nn_detroit_buildingin nsubj_flattened_detroit csubj_leveled_flattened dobj_flattened_ground partmod_ground_levelled prep_levelled_to pobj_to_blasted ccomp_\\*_leveled dep_reduced_\\* dep_reduced_e nsubjpass_reduced_\\* auxpass_reduced_was dep_leveled_reduced ccomp_reduced_razed dobj_razed_leveled prep_leveled_to amod_rubble_down pobj_to_rubble prep_rubble_into pobj_into_ashes dep_into_\\* dep_leveled_e dep_leveled_\\* prep_razed_to nn_\\*_\\* nn_\\*_e pobj_to_\\* appos_\\*_1 appos_\\*_2 tmod_razed_Figure num_Figure_1 nsubj_extracted_Examples prep_Examples_of amod_patterns_paraphrase pobj_of_patterns parataxis_reduced_extracted prep_extracted_by pobj_by_ cc__and conj__Pang cc_Pang_et conj_Pang_al ' +o,7411,'4 Parameter Optimization We optimize the feature weights using a modified version of averaged perceptron learning as described by ',Collins,'num_Optimization_4 nn_Optimization_Parameter nsubj_optimize_We dep_Optimization_optimize det_weights_the nn_weights_feature nsubj_using_weights dep_optimize_using det_version_a amod_version_modified dobj_using_version prep_version_of amod_perceptron_averaged pobj_of_perceptron partmod_perceptron_learning advmod_described_as acomp_learning_described prep_described_by ' +p,7412,'There are basically two kinds of systems working at these segmentation levels : the most widespread rely on statistical models , in particular the IBM ones ; others combine simpler association measures with different kinds of linguistic information ',Brown,'expl_are_There advmod_are_basically num_kinds_two nsubj_are_kinds prep_kinds_of pobj_of_systems partmod_systems_working prep_working_at det_levels_these nn_levels_segmentation pobj_at_levels det_widespread_the advmod_widespread_most nsubj_rely_widespread parataxis_are_rely prep_rely_on amod_models_statistical pobj_on_models prep_rely_in pobj_in_particular det_ones_the nn_ones_IBM nsubj__ones rcmod_particular_ nsubj_combine_others ccomp__combine amod_measures_simpler nn_measures_association dobj_combine_measures prep_combine_with amod_kinds_different pobj_with_kinds prep_kinds_of amod__linguistic nn__information pobj_of_ ' +o,7413,'There are many method proposed to extract rigid expressions from corpora such as a method of focusing on the binding strength of two words ; the distance between words ; and the number of combined words and frequency of appearance ',Church,'expl_are_There amod_method_many nsubj_are_method partmod_method_proposed aux_extract_to xcomp_proposed_extract amod_expressions_rigid dobj_extract_expressions prep_extract_from pobj_from_corpora dep_as_such prep_corpora_as det_method_a pobj_as_method prep_method_of pcomp_of_focusing prep_focusing_on det_strength_the amod_strength_binding pobj_on_strength prep_strength_of num__two nn__words pobj_of_ det_distance_the dep_method_distance prep_distance_between nn__words pobj_between_ cc_method_and det_number_the conj_method_number prep_number_of amod_words_combined pobj_of_words cc_words_and conj_words_frequency prep_words_of nn__appearance pobj_of_ ' +o,7414,'The Dublin Core Metadata Initiative3 established a de facto standard for the Semantic Web4 For -LRB- computational -RRB- linguistics proper , syntactic annotation schemes , such as the one from the Penn Treebank , or semantic annotations , such as the one underlying ACE , are increasingly being used in a quasi standard way ',Marcus,'det_Initiative3_The nn_Initiative3_Dublin nn_Initiative3_Core nn_Initiative3_Metadata nsubj_established_Initiative3 det_standard_a amod_standard_de dep_de_facto dobj_established_standard prep_standard_for det_Web4_the nn_Web4_Semantic pobj_for_Web4 prep_used_For appos_linguistics_computational pobj_For_linguistics amod_schemes_proper amod_schemes_syntactic nn_schemes_annotation nsubjpass_used_schemes dep_as_such prep_schemes_as det_one_the pobj_as_one prep_one_from det__the nn__Penn nn__Treebank pobj_from_ cc__or amod_annotations_semantic conj__annotations dep_as_such prep__as det__the num__one amod__underlying nn__ACE pobj_as_ aux_used_are advmod_used_increasingly auxpass_used_being dep_established_used prep_used_in det_way_a nn_way_quasi nn_way_standard pobj_in_way ' +n,7415,'In contrast to existing approaches , the context of the whole corpus rather than a single sentence is considered in this iterative , unsupervised procedure , yielding a more reliable alignment ',Rosti,'prep_considered_In dep_In_contrast dep_In_to amod_approaches_existing pobj_In_approaches amod_approaches_ det_context_the nsubjpass_considered_context prep_context_of det_corpus_the amod_corpus_whole pobj_of_corpus cc_corpus_rather dep_rather_than det_sentence_a amod_sentence_single conj_corpus_sentence auxpass_considered_is prep_considered_in det_iterative_this nsubj_yielding_iterative amod_procedure_unsupervised appos_iterative_procedure pcomp_in_yielding det_alignment_a advmod_reliable_more amod_alignment_reliable dobj_yielding_alignment ' +o,7416,'Our approach is to use finite-state approximations of long-distance dependencies , as they are described in for Dependency Grammar -LRB- DG -RRB- and for Lexical Functional Grammar -LRB- LFG -RRB- ',Cahill,'poss_approach_Our nsubj_is_approach aux_approximations_to dep_approximations_use amod_approximations_finite-state xcomp_is_approximations prep_approximations_of amod_dependencies_long-distance pobj_of_dependencies mark_described_as nsubjpass_described_they auxpass_described_are advcl_approximations_described prep_described_in pcomp_in_ prep__for nn_Grammar_Dependency pobj_for_Grammar abbrev_Grammar_DG cc__and conj__ prep__for nn_Grammar_Lexical nn_Grammar_Functional pobj_for_Grammar abbrev_Grammar_LFG ' +o,7417,'Alignment is often used in training both generative and discriminative models ',Brown,'nsubjpass_used_Alignment auxpass_used_is advmod_used_often prep_used_in pcomp_in_training preconj_generative_both amod_models_generative cc_generative_and conj_generative_discriminative dobj_training_models ' +o,7418,'2 Related Work One of the first works that use statistical methods to detect implicit discourse relations is that of ',Marcu,'num_Work_2 nn_Work_Related nsubj_that_Work num_Work_One prep_Work_of det_works_the amod_works_first pobj_of_works nsubj_use_that rcmod_works_use amod_methods_statistical nsubj_detect_methods aux_detect_to xcomp_use_detect amod_relations_implicit nn_relations_discourse dobj_detect_relations cop_that_is prep_that_of ' +o,7419,'Both for the training and for the testing of our algorithm , we used the syntactically analyzed sentences of the Brown Corpus , which have been manually semantically tagged into semantic concordance files -LRB- SemCor -RRB- ',Marcus,'dep_Both_for det_training_the pobj_for_training cc_training_and prep_used_for det_testing_the pobj_for_testing prep_testing_of poss_algorithm_our pobj_of_algorithm nsubj_used_we conj_training_used det_syntactically_the nsubj_analyzed_syntactically ccomp_used_analyzed dobj_analyzed_sentences prep_sentences_of det__the nn__Brown nn__Corpus pobj_of_ nsubjpass_tagged_which aux_tagged_have auxpass_tagged_been advmod_tagged_manually advmod_tagged_semantically rcmod__tagged dep_tagged_ prep_tagged_into amod_files_semantic nn_files_concordance pobj_into_files appos_files_SemCor ' +o,7420,'We used a non-projective model , trained using an application of the matrix-tree theorem for the first-order Czech models , and projective parsers for all other models ',Smith,'nsubj_used_We det_model_a amod_model_non-projective dobj_used_model partmod_model_trained xcomp_trained_using det_application_an dobj_using_application prep_application_of det__the amod__matrix-tree nn__theorem pobj_of_ prep__for det_models_the amod_models_first-order amod_models_Czech pobj_for_models cc_application_and amod_parsers_projective conj_application_parsers prep_parsers_for det_models_all amod_models_other pobj_for_models ' +o,7421,'N-best results for phrasal alignment and ordering models in the decoder were optimized by lambda training via Maximum Bleu , along the lines described in ',Och,'amod_results_N-best nsubjpass_optimized_results prep_results_for amod_alignment_phrasal pobj_for_alignment cc_alignment_and nn_models_ordering conj_alignment_models prep_alignment_in det_decoder_the pobj_in_decoder auxpass_optimized_were prep_optimized_by nn_training_lambda pobj_by_training prep_training_via nn_Bleu_Maximum pobj_via_Bleu prep_optimized_along det_lines_the pobj_along_lines partmod_lines_described prep_described_in ' +o,7422,'Various machine learning approaches have been proposed for chunking ',Ramshaw,'amod_approaches_Various nn_approaches_machine nn_approaches_learning nsubjpass_proposed_approaches aux_proposed_have auxpass_proposed_been prep_proposed_for pcomp_for_chunking ' +o,7423,'1 Introduction Parallel corpora have been shown to provide an extremely rich source of constraints for statistical analysis ',Brown,'num_corpora_1 nn_corpora_Introduction amod_corpora_Parallel nsubjpass_shown_corpora aux_shown_have auxpass_shown_been aux_provide_to xcomp_shown_provide det_source_an advmod_rich_extremely amod_source_rich dobj_provide_source prep_source_of pobj_of_constraints prep_constraints_for amod_analysis_statistical pobj_for_analysis ' +o,7424,'eg BLEU for machine translation , ROUGE for summarization ',Lin,'nn__BLEU dobj_eg_ prep__for pobj_for_machine nsubj_eg_translation nn__ROUGE appos_translation_ prep__for pobj_for_summarization ' +n,7425,'While minimum error training has by now become a standard tool for interpolating a small number of aggregate scores , it is not well suited for learning in high-dimensional feature spaces ',Och,'mark_tool_While amod__minimum nn__error nn__training nsubj_tool_ aux_tool_has prep_tool_by pobj_by_now cop_tool_become det_tool_a amod_tool_standard advcl_suited_tool prep_tool_for pcomp_for_interpolating det_number_a amod_number_small dobj_interpolating_number prep_number_of amod_scores_aggregate pobj_of_scores nsubjpass_suited_it auxpass_suited_is neg_suited_not advmod_suited_well prep_suited_for pcomp_for_learning prep_learning_in amod_spaces_high-dimensional nn_spaces_feature pobj_in_spaces ' +o,7426,'One example is the algorithm for word sense disambiguation in ',Yarowsky,'num_example_One nsubj_algorithm_example cop_algorithm_is det_algorithm_the prep_algorithm_for nn_sense_word pobj_for_sense partmod_sense_disambiguation prep_disambiguation_in ' +o,7427,'The phrase-based decoder extracts phrases from the word alignments produced by GIZA + + , and computes translation probabilities based on the frequency of one phrase being aligned with another ',Koehn,'det_phrases_The amod_phrases_phrase-based nn_phrases_decoder nn_phrases_extracts nsubj_produced_phrases prep_phrases_from det_alignments_the nn_alignments_word pobj_from_alignments prep_produced_by nn_+_GIZA nn_+_+ pobj_by_+ cc_produced_and conj_produced_computes nn_probabilities_translation dobj_computes_probabilities prep_computes_based dep_based_on det_frequency_the pobj_on_frequency prep_frequency_of num_phrase_one pobj_of_phrase auxpass_aligned_being partmod_phrase_aligned prep_aligned_with pobj_with_another ' +o,7428,'The grow-diag-final -LRB- GDF -RRB- combination heuristic adds links so that each new link connects a previously unlinked token ',Koehn,'det_heuristic_The amod_heuristic_grow-diag-final abbrev_heuristic_GDF nn_heuristic_combination nsubj_adds_ dep_heuristic_adds dobj_adds_links advmod_connects_so dep_connects_that det_link_each amod_link_new nsubj_connects_link dep_adds_connects det_token_a advmod_unlinked_previously amod_token_unlinked dobj_connects_token ' +p,7429,'1 Introduction For statistical machine translation -LRB- SMT -RRB- , phrasebased methods and syntax-based methods outperform word-based methods ',Wu,'num_Introduction_1 dep_Introduction_For amod_translation_statistical nn_translation_machine pobj_For_translation abbrev_translation_SMT amod__phrasebased nn__methods conj_translation_ cc_translation_and amod_methods_syntax-based conj_translation_methods partmod_methods_ ccomp__outperform amod_methods_word-based iobj_outperform_methods dobj_outperform_ ' +p,7430,'Using the IBM translation models IBM-1 to IBM-5 , as well as the Hidden-Markov alignment model , we can produce alignments of good quality ',Brown,'dep_produce_Using det_IBM-1_the nn_IBM-1_IBM nn_IBM-1_translation nn_IBM-1_models dobj_Using_IBM-1 prep_IBM-1_to nn__IBM-5 pobj_to_ dep_well_as cc_IBM-1_well dep_well_as det__the amod__Hidden-Markov amod__alignment nn__model conj_IBM-1_ nsubj_produce_we aux_produce_can dobj_produce_alignments prep_alignments_of amod_quality_good pobj_of_quality ' +o,7431,'Training discriminative parsers is notoriously slow , especially if it requires generating examples by repeatedly parsing the treebank ',Collins,'amod_parsers_Training nn_parsers_discriminative nsubj_slow_parsers cop_slow_is advmod_slow_notoriously advmod_slow_especially mark_requires_if nsubj_requires_it advcl_slow_requires xcomp_requires_generating dobj_generating_examples prep_generating_by advmod_parsing_repeatedly pcomp_by_parsing det_treebank_the dobj_parsing_treebank ' +o,7432,'Class based models distinguish between unobserved cooccurrences using classes of ` similar \' words ',Brown,'dep_Class_based dep_based_models nn_distinguish_ nsubj_cooccurrences_distinguish prep_distinguish_between pobj_between_unobserved rcmod_models_cooccurrences xcomp_cooccurrences_using dobj_using_classes prep_classes_of amod_words_similar pobj_of_words ' +o,7433,'Bilingual lexicographers can work with bilingual concordancing software that can point them to instances of any link type induced from a bitext and display these instances sorted by their contexts ',Dunning,'amod_lexicographers_Bilingual nsubj_work_lexicographers aux_work_can prep_work_with amod_software_bilingual amod_software_concordancing pobj_with_software nsubj_point_that aux_point_can rcmod_software_point dobj_point_them prep_point_to pobj_to_instances prep_instances_of det_type_any nn_type_link pobj_of_type partmod_type_induced prep_induced_from det_bitext_a pobj_from_bitext cc_point_and conj_point_display det_instances_these dobj_display_instances partmod_instances_sorted prep_sorted_by poss__their amod__contexts pobj_by_ ' +o,7434,'Neural networks have been used in NLP in the past , eg for machine translation and constituent parsing ',Titov,'amod_networks_Neural nsubjpass_used_networks aux_used_have auxpass_used_been prep_used_in pobj_in_NLP prep_used_in det_past_the pobj_in_past conj_past_eg prep_eg_for nn__machine nn__translation pobj_for_ cc_past_and amod_parsing_constituent conj_past_parsing ' +o,7435,'In the following , we summarize the optimization algorithm for the unsmoothed error counts presented in and the implementation detailed in ',Och,'prep_summarize_In det_following_the pobj_In_following nsubj_summarize_we det_algorithm_the nn_algorithm_optimization dobj_summarize_algorithm prep_summarize_for det_counts_the amod_counts_unsmoothed nn_counts_error pobj_for_counts partmod_counts_presented prep_presented_in pobj_in_ cc__and det_implementation_the conj__implementation partmod_implementation_detailed prep_detailed_in pobj_in_ ' +o,7436,'To derive the joint counts c -LRB- s , t -RRB- from which p -LRB- s t -RRB- and p -LRB- t s -RRB- are estimated , we use the phrase induction algorithm described in , with symmetrized word alignments generated using IBM model 2 ',Brown,'aux_derive_To dep_use_derive det_counts_the amod_counts_joint dobj_derive_counts dep_counts_c appos_counts_s dep_s_t rel_estimated_from pobj_from_which nsubjpass_estimated_p amod_t_s appos_p_t cc_p_and conj_p_p nn_s_t appos_p_s auxpass_estimated_are rcmod_counts_estimated nsubj_use_we det_algorithm_the nn_algorithm_phrase nn_algorithm_induction nsubj_described_algorithm ccomp_use_described prep_described_in pobj_in_ prep_described_with amod_alignments_symmetrized nn_alignments_word pobj_with_alignments partmod_alignments_generated xcomp_generated_using nsubj_model_IBM ccomp_using_model dobj_model_2 ' +o,7437,'In this approach we extend the denition overlap by considering the distributional similarity rather than identify of the words in the two denitions ',Lin,'prep_extend_In det_approach_this pobj_In_approach nsubj_extend_we det_denition_the nsubj_overlap_denition ccomp_extend_overlap prep_overlap_by pcomp_by_considering det__the amod__distributional nn__similarity dobj_considering_ cc_overlap_rather dep_rather_than conj_overlap_identify prep_identify_of det_words_the pobj_of_words prep_words_in det_denitions_the num_denitions_two pobj_in_denitions ' +o,7438,'Table 2: The set of tags used to mark explicit morphemes in English Tag Meaning JJR Adjective, comparative JJS Adjective, superlative NNS Noun, plural POS Possessive ending RBR Adverb, comparative RBS Adverb, superlative VB Verb, base form VBD Verb, past tense VBG Verb, gerund or present participle VBN Verb, past participle VBP Verb, non3rd person singular present VBZ Verb, 3rd person singular present Figure 2: Morpheme alignment between a Turkish and an English sentence 4 Experiments We proceeded with the following sequence of experiments: (1) Baseline: As a baseline system, we used a pure word-based approach and used Pharaoh Training tool (2004), to train on the 22,500 sentences, and decoded using Pharaoh (Koehn et al. , 2003) to obtain translations for a test set of 50 sentences.',Koehn,'' +o,7439,'Successful approaches aimed at trying to overcome the sparse data limitation include backoff , Turing-Good variants , interpolation , deleted estimation , similarity-based models , Pos-language models and decision tree models ',Brown,'amod_approaches_Successful nsubj_include_approaches partmod_approaches_aimed prep_aimed_at pcomp_at_trying aux_overcome_to xcomp_trying_overcome det_limitation_the amod_limitation_sparse nn_limitation_data dobj_overcome_limitation nn__backoff dobj_include_ amod__Turing-Good nn__variants conj__ nn__interpolation conj__ amod__deleted nn__estimation conj__ amod__similarity-based nn__models conj__ amod__Pos-language nn__models conj__ cc__and nn__decision nn__tree nn__models conj__ ' +o,7440,'53 Snow has extended the WordNet 21 by adding thousands of entries -LRB- synsets -RRB- at a relatively high precision ',Snow,'num__53 nn__ nn__Snow nsubj_extended_ aux_extended_has det_WordNet_the dobj_extended_WordNet num_WordNet_21 prep_extended_by pcomp_by_adding dobj_adding_thousands prep_thousands_of pobj_of_entries appos_entries_synsets prep_adding_at det_precision_a advmod_high_relatively amod_precision_high pobj_at_precision ' +p,7441,'A popular metric for evaluating machine translation quality is the Bleu score ',Papineni,'det_metric_A amod_metric_popular nsubj_score_metric prep_metric_for pcomp_for_evaluating nn_quality_machine nn_quality_translation dobj_evaluating_quality cop_score_is det_score_the nn_score_Bleu ' +o,7442,'First , as originally advocated by , we adopt an ONTOLOGICALLY PROMISCUOUS representation that includes a wide variety of types of entities ',Hobbs,'advmod_adopt_First mark_advocated_as advmod_advocated_originally dep_adopt_advocated prep_advocated_by pobj_by_ nsubj_adopt_we det_representation_an nn_representation_ONTOLOGICALLY nn_representation_PROMISCUOUS dobj_adopt_representation nsubj_includes_that rcmod_representation_includes det_variety_a amod_variety_wide dobj_includes_variety prep_variety_of pobj_of_types prep_types_of pobj_of_entities ' +o,7443,'We have used three different algorithms : the nearest neighbor algorithm IB1IG , which is part of the Timbl software package , the decision tree learner IGTREE , also from Timbl , and C50 , a commercial version of the decision tree learner C45 ',Marcus,'nsubj_used_We aux_used_have num_algorithms_three amod_algorithms_different dobj_used_algorithms det_IB1IG_the amod_IB1IG_nearest nn_IB1IG_neighbor nn_IB1IG_algorithm dep_algorithms_IB1IG nsubj_part_which cop_part_is rcmod_IB1IG_part prep_part_of det__the nn__Timbl nn__software nn__package pobj_of_ det_learner_the nn_learner_decision nn_learner_tree nsubj_IGTREE_learner ccomp_used_IGTREE advmod_IGTREE_also prep_IGTREE_from pobj_from_Timbl cc_used_and nsubj__C50 det_version_a amod_version_commercial appos_C50_version prep_version_of det_C45_the nn_C45_decision nn_C45_tree nn_C45_learner pobj_of_C45 conj_used_ ' +o,7444,'Even for semantically predictable phrases , the fact that the words occur in fixed patterns can be very useful for the purposes of disambiguation , as demonstrated by ',Yarowsky,'dep_for_Even advcl_useful_for advmod_predictable_semantically amod_phrases_predictable pobj_for_phrases det_fact_the nsubj_useful_fact complm_occur_that det_words_the nsubj_occur_words dep_fact_occur prep_occur_in amod_patterns_fixed pobj_in_patterns aux_useful_can cop_useful_be advmod_useful_very prep_useful_for det_purposes_the pobj_for_purposes prep_purposes_of pobj_of_disambiguation mark_demonstrated_as advcl_useful_demonstrated prep_demonstrated_by ' +p,7445,'2 Method Maximum Entropy Markov Models -LRB- MEMMs -RRB- and their extensions have been successfully applied to English POS tagging ',Ratnaparkhi,'dep_Method_2 amod__Method dep_Method_Maximum nn__Entropy nn__Markov nn__Models abbrev__MEMMs nsubjpass_applied_ cc__and poss__their nn__extensions conj__ aux_applied_have auxpass_applied_been advmod_applied_successfully prep_applied_to nn_POS_English pobj_to_POS partmod_POS_tagging ' +o,7446,'Precision and recall rates were 924 \% on the same data used in ',Ramshaw,'nn_rates_Precision cc_Precision_and conj_Precision_recall nsubj_\%_rates cop_\%_were num_\%_924 prep_\%_on det_data_the amod_data_same pobj_on_data partmod_data_used prep_used_in ' +o,7447,'We would like to apply our learning approach to the large data set mentioned in : Wall Street Journal corpus sections 2-21 as training material and section 0 as test material ',Ramshaw,'nsubj_like_We aux_like_would aux_apply_to xcomp_like_apply poss_approach_our nn_approach_learning dobj_apply_approach prep_apply_to det_set_the amod_set_large nn_set_data pobj_to_set partmod_set_mentioned prep_mentioned_in pobj_in_ nn_2-21_Wall nn_2-21_Street nn_2-21_Journal nn_2-21_corpus nn_2-21_sections dep_set_2-21 prep_2-21_as nn_material_training pobj_as_material cc_set_and conj_set_section num_section_0 prep_section_as nn_material_test pobj_as_material ' +n,7448,' and Collins and Duffy -LRB- 2002 -RRB- rerank the top N parses from an existing generative parser , but this kind of approach 1Dynamic programming methods can sometimes be used for both training and decoding , but this requires fairly strong restrictions on the features in the model ',Collins,'nsubj_rerank_ cc__and conj__Collins cc_Collins_and conj_Collins_Duffy appos_Collins_2002 det_N_the amod_N_top nsubj_parses_N dep_rerank_parses prep_parses_from det_parser_an amod_parser_existing nn_parser_generative pobj_from_parser cc_parses_but det_kind_this nsubjpass_used_kind prep_kind_of nn_methods_approach nn_methods_1Dynamic nn_methods_programming pobj_of_methods advmod_used_ aux_used_can advmod_used_sometimes auxpass_used_be conj_parses_used prep_used_for preconj_training_both pobj_for_training cc_training_and conj_training_decoding cc_used_but nsubj_requires_this conj_used_requires advmod_strong_fairly amod_restrictions_strong dobj_requires_restrictions prep_restrictions_on det_features_the pobj_on_features prep_features_in det_model_the pobj_in_model ' +o,7449,'1 Introduction This paper reports on our experience hand tagging the senses of 25 of the most frequent verbs in 12,925 sentences of the Wall Street Journal Treebank corpus ',Marcus,'num_paper_1 nn_paper_Introduction nn_paper_This nsubj_reports_paper prep_reports_on poss_hand_our nn_hand_experience pobj_on_hand partmod_hand_tagging det_senses_the dobj_tagging_senses prep_senses_of pobj_of_25 prep_senses_of det_verbs_the advmod_verbs_most amod_verbs_frequent pobj_of_verbs prep_verbs_in num_sentences_12,925 pobj_in_sentences prep_sentences_of det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nn_corpus_Treebank pobj_of_corpus ' +o,7450,'(See also Kaplan et al. , 1988, on the latter point).',Hobbs,'' +o,7451,'However , most parsers still tend to show low performance on the long sentences ',Marcus,'advmod_tend_However amod_parsers_most nsubj_tend_parsers advmod_tend_still aux_show_to xcomp_tend_show amod_performance_low dobj_show_performance prep_show_on det__the amod__long nn__sentences pobj_on_ ' +o,7452,'41 Corpora set-up The above kernels were experimented over two corpora : PropBank -LRB- wwwcisupennedu \\/ ace -RRB- along with Penn TreeBank5 2 and FrameNet ',Marcus,'num_set-up_41 nn_set-up_Corpora det_kernels_The amod_kernels_above nsubjpass_experimented_kernels auxpass_experimented_were dep_set-up_experimented prep_experimented_over num_corpora_two pobj_over_corpora dep_along_PropBank amod_ace_wwwcisupennedu amod_ace_\\/ appos_PropBank_ace advmod_corpora_along dep_along_with nn_TreeBank5_Penn pobj_with_TreeBank5 num__2 dep_corpora_ cc__and conj__FrameNet ' +o,7453,'The corpus used for training our models was on the order of 100,000 words , whereas that used by was around 1,000 times this size ',Brown,'det_corpus_The partmod_corpus_used prep_used_for pobj_for_training poss_models_our nsubj_was_models dep_corpus_was prep_was_on det_order_the pobj_on_order prep_order_of num_words_100,000 pobj_of_words dep_was_whereas nsubj_size_that partmod_that_used prep_used_by pobj_by_ cop_size_was quantmod_times_around number_times_1,000 num_size_times det_size_this dep_was_size ' +o,7454,'Not only many combinations are found in the corpus, many of them have very similar mutual information values to that of 318 Table 2: economic impact verb economic financial political social budgetary ecological economic economic economic economic economic economic economic economic economic object impact impact impact impact impact impact effect implication consequence significance fallout repercussion potential ramification risk mutual freq info 171 1.85 127 1.72 46 0.50 15 0.94 8 3.20 4 2.59 84 0.70 17 0.80 59 1.88 10 0.84 7 1.66 7 1.84 27 1.24 8 2.19 17 -0.33 nomial distribution can be accurately approximated by a normal distribution (Dunning, 1993).',Dunning,'' +o,7455,'For instance , constructed a graph which represented various types of word relations from WordNet , and compared random-walk similarity to similarity assessments from humansubject trials ',Hughes,'prep__For pobj_For_instance dep__constructed det_graph_a dobj_constructed_graph nsubj_represented_which rcmod_graph_represented amod_types_various dobj_represented_types prep_types_of nn_relations_word pobj_of_relations prep_represented_from pobj_from_WordNet cc_represented_and conj_represented_compared amod_similarity_random-walk dobj_compared_similarity prep_compared_to nn_assessments_similarity pobj_to_assessments prep_compared_from amod_trials_humansubject pobj_from_trials ' +o,7456,' , modified from , which is an average of pairwise word translation probabilities ',Koehn,'dep__modified prep_modified_from pobj_from_ nsubj_average_which cop_average_is det_average_an rcmod__average prep_average_of amod_probabilities_pairwise nn_probabilities_word nn_probabilities_translation pobj_of_probabilities ' +o,7457,'While bound compositions are not predictable , ie , their reasonableness can not be derived from the syntactic and semantic properties of the words in the ',Smadja,'mark_predictable_While amod_compositions_bound nsubj_predictable_compositions cop_predictable_are neg_predictable_not advcl_derived_predictable attr_predictable_ie poss_reasonableness_their nsubjpass_derived_reasonableness aux_derived_can neg_derived_not auxpass_derived_be prep_derived_from det_properties_the amod_properties_syntactic cc_syntactic_and conj_syntactic_semantic pobj_from_properties prep_properties_of det_words_the pobj_of_words prep_words_in pobj_in_the ' +o,7458,'22 Unsupervised Parameter Estimation We can perform maximum likelihood estimation of the parameters of this model in a similar fashion to that of Model 4 , described thoroughly in ',Brown,'num_Estimation_22 nn_Estimation_Unsupervised nn_Estimation_Parameter nsubj_described_Estimation nsubj_perform_We aux_perform_can rcmod_Estimation_perform amod_estimation_maximum nn_estimation_likelihood dobj_perform_estimation prep_estimation_of det_parameters_the pobj_of_parameters prep_parameters_of det_model_this pobj_of_model prep_model_in det_fashion_a amod_fashion_similar pobj_in_fashion prep_perform_to pobj_to_that prep_that_of nn__Model num__4 pobj_of_ advmod_described_thoroughly prep_described_in pobj_in_ ' +o,7459,'Similarly , tested his WSD algorithm on a dozen words ',Yarowsky,'advmod_tested_Similarly nsubj_tested_ poss_algorithm_his nn_algorithm_WSD dobj_tested_algorithm prep_tested_on quantmod_dozen_a num_words_dozen pobj_on_words ' +p,7460,'2 The Problem of Coverage in SMT Statistical machine translation made considerable advances in translation quality with the introduction of phrase-based translation ',Koehn,'det_Problem_The nsubj_made_Problem prep_Problem_of pobj_of_Coverage prep_Coverage_in nn_translation_SMT nn_translation_Statistical nn_translation_machine pobj_in_translation dep_2_made amod_advances_considerable dobj_made_advances prep_made_in nn_quality_translation pobj_in_quality prep_made_with det_introduction_the pobj_with_introduction prep_introduction_of amod_translation_phrase-based pobj_of_translation ' +o,7461,'In comparison , most corpus-based algorithms employ substantially larger corpora -LRB- eg , 1 million words , 25 million words , 6 million words , 13 million words -RRB- ',Hindle,'prep_employ_In pobj_In_comparison advmod_algorithms_most amod_algorithms_corpus-based nsubj_employ_algorithms advmod_larger_substantially amod_corpora_larger dobj_employ_corpora dep_corpora_eg number_million_1 num__million nn__words appos_eg_ number_million_25 num__million nn__words appos__ number_million_6 num__million nn__words appos__ number_million_13 num_words_million measure__words amod__ ' +o,7462,'We delete all links in the set -LCB- a , an , the -RCB- -LCB- DF , GI -RCB- from Ainitial as a preprocessing step7 24 Perceptron Training We set the feature weights using a modified version of averaged perceptron learning with structured outputs ',Collins,'nsubj_delete_We dobj_delete_all dep_all_links prep_links_in det_set_the pobj_in_set dep_set_a dep_links_an appos_all_the appos_all_DF dep_DF_GI prep_delete_from pobj_from_Ainitial prep_delete_as det_Training_a amod_Training_preprocessing nn_Training_step7 num_Training_24 nn_Training_Perceptron pobj_as_Training nsubj_set_We rcmod_Training_set det_weights_the nn_weights_feature dobj_set_weights xcomp_set_using det_version_a amod_version_modified dobj_using_version prep_version_of amod_perceptron_averaged pobj_of_perceptron partmod_perceptron_learning prep_learning_with amod_outputs_structured pobj_with_outputs ' +o,7463,'The mapping typically is made to try to give the most favorable mapping in terms of accuracy , typically using a greedy assignment ',Haghighi,'det_mapping_The nsubjpass_made_mapping advmod_mapping_typically auxpass_made_is aux_try_to xcomp_made_try aux_give_to xcomp_try_give det_mapping_the advmod_favorable_most amod_mapping_favorable dobj_give_mapping prep_give_in pobj_in_terms prep_terms_of pobj_of_accuracy advmod_using_typically xcomp_give_using det_assignment_a amod_assignment_greedy dobj_using_assignment ' +o,7464,'s e, the window to consider when extracting words related to word w, should span from postttuon w-5 to w+5 Maarek also defines the resolwng power of a parr m a document d as P = ~\'Pd log Pc where Pd is the observed probabshty of appearance of the pan\' m document d, Pc the observed probabdny of the pmr recorpus, and -log Pc the quantity of mformauon assocmted to the pmr It Is easdy seen that p wall be h|gher, the higher the frequency of the pmr m the document and the lower sts frequency m the corpus, which agrees wlth the sdea presented at the begmnmg of this sectton Church and Hanks (1990) propose the apphcatlon of the concept of mutual mformatton e(x,y) ~,(x.y) = hog2 ecx)e(y) 51 to the retrieval, ro a corpus, of pairs of lextcally related words They alsoconslder a word span of :e5 words and observe that \'roterestrog\' pmr, s generally present a mutual mformatxon above 3 Salton and.Allan (1995) foc~as on paragraph level Each paragraph Is represented by a weighed vector, where each element is a term (typically.',Church,'' +o,7465,'In the second experiment , the basic learning model is Model 2 parser , which uses a history-based learning algorithm that takes statistics directly over the treebank ',Collins,'prep__In det_experiment_the amod_experiment_second pobj_In_experiment det_model_the amod_model_basic nn_model_learning nsubj__model aux__is nn_parser_Model num_parser_2 dobj__parser nsubj_uses_which rcmod_parser_uses det_algorithm_a amod_algorithm_history-based nn_algorithm_learning dobj_uses_algorithm nsubj_takes_that rcmod_algorithm_takes dobj_takes_statistics advmod_takes_directly prep_takes_over det_treebank_the pobj_over_treebank ' +o,7466,' use sections 2-21 of the WSJ PennTreebank as seed data and between 50K to 2,500 K unlabeled NANC corpus sentences as self-training data ',McClosky,'amod_sections_ nn_sections_use dep_sections_2-21 prep_2-21_of det_sentences_the nn_sentences_WSJ nn_sentences_PennTreebank dep_sentences_as nn_data_seed pobj_as_data cc_as_and conj_as_between dep_between_50K dep_between_to num_corpus_2,500 nn_corpus_K amod_corpus_unlabeled nn_corpus_NANC pobj_to_corpus pobj_of_sentences prep_2-21_as amod_data_self-training pobj_as_data ' +o,7467,'Approaches have been proposed recently towards getting better word alignment and thus better TTS templates , such as encoding syntactic structure information into the HMM-based word alignment model , and build62 ing a syntax-based word alignment model May and Knight -LRB- 2007 -RRB- with TTS templates ',DeNero,'nsubjpass_proposed_Approaches aux_proposed_have auxpass_proposed_been advmod_proposed_recently prep_proposed_towards pcomp_towards_getting amod_alignment_better nn_alignment_word dobj_getting_alignment cc_proposed_and advmod_templates_thus amod_templates_better nn_templates_TTS conj_proposed_templates dep_as_such conj_proposed_as pcomp_as_encoding amod_information_syntactic nn_information_structure dobj_encoding_information prep_encoding_into det__the amod__HMM-based nn__word nn__alignment nn__model pobj_into_ cc_proposed_and nsubj_ing_build62 conj_proposed_ing det_model_a amod_model_syntax-based nn_model_word nn_model_alignment dobj_ing_model dep_model_May cc_model_and conj_model_Knight appos_Knight_2007 prep_Knight_with nn_templates_TTS pobj_with_templates ' +o,7468,'Models describing these types of dependencies are referred to as alignment mappings : alignment mapping : j ! i = aj ; which assigns a source word fj in position j to a target word ei in position i = aj ',Brown,'advmod_describing_Models det_types_these dobj_describing_types prep_types_of pobj_of_dependencies auxpass_referred_are dep_describing_referred prep_referred_to advmod_alignment_as amod__alignment nn__mappings pobj_to_ nn_mapping_alignment dep_j_mapping dep_referred_j nn_aj_i nn_aj_= nsubj_describing_aj nsubj_assigns_which rcmod_aj_assigns det_fj_a nn_fj_source nn_fj_word dobj_assigns_fj prep_fj_in nn_j_position pobj_in_j prep_assigns_to det_ei_a nn_ei_target nn_ei_word pobj_to_ei prep_ei_in pobj_in_position xcomp_assigns_i dep_i_= dobj_i_aj ' +o,7469,'All topic models utilize Gibbs sampling for inference ',Lin,'det_models_All nn_models_topic nn_Gibbs_utilize nsubj_sampling_Gibbs dep_models_sampling prep_sampling_for amod__inference pobj_for_ ' +o,7470,'sp and pt are feature weights set by performing minimum error rate training as described in ',Och,'nsubj_weights_sp cc_sp_and conj_sp_pt cop_weights_are nn_weights_feature partmod_weights_set prep_set_by pcomp_by_performing amod_training_minimum nn_training_error nn_training_rate dobj_performing_training advmod_described_as amod_training_described prep_performing_in ' +o,7471,'We follow and Sha and Pereira -LRB- 2003 -RRB- in using section 21 as a heldout set ',Collins,'nsubj_follow_We dobj_follow_ cc__and conj__Sha cc_Sha_and conj_Sha_Pereira appos_Sha_2003 prep_follow_in pcomp_in_using dobj_using_section num_section_21 prep_section_as det_set_a amod_set_heldout pobj_as_set ' +o,7472,'All words occurring less than 3 times in the training data , and words in test data that were not seen in training , are unknown words and are replaced with the UNKNOWN token Note this threshold is smaller than the one used in since the corpora used in our experiments are smaller ',Collins,'dep_words_All nsubjpass_words_words partmod_words_occurring dep_than_less quantmod_times_than number_times_3 dobj_occurring_times prep_times_in det_data_the nn_data_training pobj_in_data cc_words_and conj_words_words prep_words_in nn_data_test pobj_in_data nsubjpass_seen_that auxpass_seen_were neg_seen_not rcmod_words_seen prep_seen_in pobj_in_training cop_words_are amod_words_unknown cc_words_and auxpass_replaced_are conj_words_replaced prep_replaced_with det_Note_the nn_Note_UNKNOWN amod_Note_token pobj_with_Note det_threshold_this nsubj_smaller_threshold cop_smaller_is dep_words_smaller mark_smaller_than det_one_the nsubj_smaller_one partmod_one_used prep_used_in pobj_in_ prep_used_since det_corpora_the pobj_since_corpora partmod_corpora_used prep_used_in poss_experiments_our pobj_in_experiments cop_smaller_are advcl_smaller_smaller ' +o,7473,'Figure 1 : SCL algorithm ',Blitzer,'num_Figure_1 nn_algorithm_SCL dep_Figure_algorithm ' +o,7474,'3For decoding , loc is averaged over the training iterations as in ',Collins,'amod_decoding_3For nsubjpass_averaged_decoding appos_decoding_loc auxpass_averaged_is prep_averaged_over det_iterations_the nn_iterations_training pobj_over_iterations prep_averaged_as dep_as_in ' +p,7475,'It has been used in a variety of difficult classification tasks such as part-of-speech tagging , prepositional phrase attachment and named entity tagging , and achieves state of the art performance ',Ratnaparkhi,'nsubjpass_used_It aux_used_has auxpass_used_been prep_used_in det_variety_a pobj_in_variety prep_variety_of amod_tasks_difficult nn_tasks_classification pobj_of_tasks dep_as_such prep_tasks_as amod__part-of-speech amod__tagging pobj_as_ amod__prepositional nn__phrase nn__attachment conj__ cc__and amod__named nn__entity amod__tagging conj__ cc_used_and conj_used_achieves dobj_achieves_state prep_state_of det_performance_the nn_performance_art pobj_of_performance ' +o,7476,'However , this may still be too expensive as part of an MT model that directly optimizes some performance measure , eg , minimum error rate training ',Och,'advmod_expensive_However nsubj_expensive_this aux_expensive_may advmod_expensive_still cop_expensive_be advmod_expensive_too prep_expensive_as pobj_as_part prep_part_of det_model_an nn_model_MT pobj_of_model nsubj_optimizes_that advmod_optimizes_directly rcmod_part_optimizes det_training_some nn_training_performance nn_training_measure dep_training_eg amod_training_minimum nn_training_error nn_training_rate dobj_optimizes_training ' +o,7477,'In these experiments we used the MXPOST tagger combined withCollinsparser toassignparse trees to the corpus ',Ratnaparkhi,'prep_used_In det_experiments_these pobj_In_experiments nsubj_used_we det_tagger_the nn_tagger_MXPOST nsubj__tagger ccomp_used_ amod_trees_combined nn_trees_withCollinsparser nn_trees_ nn_trees_toassignparse dobj__trees prep__to det_corpus_the pobj_to_corpus ' +p,7478,'Recently , Cabezas and Resnik experimented with incorporating WSD translations into Pharaoh , a state-of-the-art phrase-based MT system ',Koehn,'advmod_experimented_Recently nn__Cabezas cc_Cabezas_and conj_Cabezas_Resnik nsubj_experimented_ prep_experimented_with pcomp_with_incorporating nn_translations_WSD dobj_incorporating_translations prep_incorporating_into pobj_into_Pharaoh det_system_a amod_system_state-of-the-art amod_system_phrase-based nn_system_MT appos_Pharaoh_system ' +o,7479,'The class labeling system in our experiment is IOB2 , which is a variation of IOB ',Ramshaw,'det_system_The nn_system_class nn_system_labeling nsubj__system prep_system_in poss_experiment_our pobj_in_experiment cop__is nn__IOB2 nsubj_variation_which cop_variation_is det_variation_a rcmod__variation prep_variation_of pobj_of_IOB ' +p,7480,'In particular , presents very strong results using a distributional-similarity module and achieve impressive tagging accuracy while starting with a mere 116 prototypical words ',Haghighi,'prep_presents_In pobj_In_particular nsubj_presents_ advmod_strong_very amod_results_strong dobj_presents_results xcomp_presents_using det_module_a amod_module_distributional-similarity dobj_using_module cc_presents_and conj_presents_achieve amod_accuracy_impressive amod_accuracy_tagging dobj_achieve_accuracy prep_achieve_while pcomp_while_starting prep_starting_with det_words_a amod_words_mere amod_words_116 amod_words_prototypical pobj_with_words ' +o,7481,'-LRB- owenOcogentexcom -RRB- 1 Introduction Dependency grammar has a long tradition in syntactic theory , dating back to at least Tesni ~ re \'s work from the thirties3 Recently , it has gained renewed attention as empirical methods in parsing are discovering the importance of relations between words -LRB- see , eg , -RRB- , which is what dependency grammars model explicitly do , but context-free phrasestructure grammars do not ',Collins,'dep_has_owenOcogentexcom num_grammar_1 nn_grammar_Introduction nn_grammar_Dependency nsubj_has_grammar dep_attention_has det_tradition_a amod_tradition_long dobj_has_tradition prep_tradition_in amod_theory_syntactic pobj_in_theory partmod_tradition_dating prt_dating_back prep_dating_to quantmod_Tesni_at dep_at_least num_work_Tesni amod_re_~ poss_work_re possessive_re_\'s pobj_to_work prep_work_from det_thirties3_the pobj_from_thirties3 advmod_dating_Recently nsubj_attention_it aux_attention_has dep_attention_gained amod_attention_renewed mark_discovering_as amod_methods_empirical nsubj_discovering_methods prep_methods_in pobj_in_parsing aux_discovering_are advcl_attention_discovering det_importance_the dobj_discovering_importance prep_importance_of pobj_of_relations prep_relations_between pobj_between_words dep_importance_see dep_see_eg dep_see_ nsubj_is_which rcmod_importance_is dep_do_what amod_model_dependency nn_model_grammars nsubj_do_model advmod_do_explicitly dep_is_do cc_do_but amod_grammars_context-free nn_grammars_phrasestructure nsubj_do_grammars conj_do_do neg_do_not ' +p,7482,'A pioneer work in online training is the perceptron-like algorithm used in training a hidden Markov model -LRB- HMM -RRB- ',Collins,'det_work_A nn_work_pioneer nsubj_algorithm_work prep_work_in amod_training_online pobj_in_training cop_algorithm_is det_algorithm_the amod_algorithm_perceptron-like partmod_algorithm_used prep_used_in pcomp_in_training det_model_a amod_model_hidden nn_model_Markov dobj_training_model abbrev_model_HMM ' +o,7483,'Negation was processed in a similar way as previous works ',Pang,'nsubjpass_processed_Negation auxpass_processed_was prep_processed_in det_way_a amod_way_similar pobj_in_way prep_way_as amod_works_previous pobj_as_works ' +o,7484,'6 Phrase Recognition with a Maximum Entropy Classifier For the candidates which are not filtered out in the above two phases , we perform classification with maximum entropy classifiers ',Berger,'num_Recognition_6 nn_Recognition_Phrase prep_Recognition_with det_Classifier_a nn_Classifier_Maximum nn_Classifier_Entropy pobj_with_Classifier dep_Recognition_For det_candidates_the pobj_For_candidates nsubjpass_filtered_which auxpass_filtered_are neg_filtered_not rcmod_candidates_filtered prt_filtered_out prep_filtered_in det_phases_the amod_phases_above num_phases_two pobj_in_phases nsubj_perform_we ccomp_filtered_perform dobj_perform_classification prep_perform_with amod_classifiers_maximum nn_classifiers_entropy pobj_with_classifiers ' +o,7485,'For example , when applying their approach to a different domain with somewhat less rigid syntax , need to introduce new combinators and new forms of candidate lexical entries ',Zettlemoyer,'prep_need_For pobj_For_example advmod_applying_when dep_need_applying poss_approach_their dobj_applying_approach prep_applying_to det_domain_a amod_domain_different pobj_to_domain prep_domain_with advmod_less_somewhat amod_syntax_less amod_syntax_rigid pobj_with_syntax nsubj_need_ aux_introduce_to xcomp_need_introduce amod_combinators_new dobj_introduce_combinators cc_combinators_and amod_forms_new conj_combinators_forms prep_forms_of nn_entries_candidate nn_entries_lexical pobj_of_entries ' +o,7486,'For related work on the voted perceptron algorithm applied to NLP problems , see and ',Collins,'prep_applied_For amod_work_related pobj_For_work prep_work_on det_algorithm_the amod_algorithm_voted nn_algorithm_perceptron pobj_on_algorithm aux_NLP_to xcomp_applied_NLP dobj_NLP_problems dep_NLP_see nn_and_ nsubj_applied_and ' +o,7487,'As an overall decoding performance measure , we used the BLEU metric ',Papineni,'prep_used_As det_measure_an amod_measure_overall nn_measure_decoding nn_measure_performance pobj_As_measure nsubj_used_we det_metric_the amod_metric_BLEU dobj_used_metric ' +o,7488,'For further information on these parameter settings , confer ',Koehn,'prep_confer_For amod_information_further pobj_For_information prep_information_on det_settings_these nn_settings_parameter pobj_on_settings ' +o,7489,'While it was initially believed that lexicalization of PCFG parsers is crucial for obtaining good parsing results , Gildea -LRB- 2001 -RRB- demonstrated that the lexicalized Model-1 parser of Collins does not benefit from bilexical information when tested on a new text domain , and only marginally benefits from such information when tested on the same text domain as the training corpora ',Collins,'mark_believed_While nsubjpass_believed_it auxpass_believed_was advmod_believed_initially advcl_demonstrated_believed complm_parsers_that nsubj_parsers_lexicalization prep_lexicalization_of pobj_of_PCFG ccomp_believed_parsers nsubj_crucial_ cop_crucial_is ccomp_parsers_crucial prep_crucial_for pcomp_for_obtaining amod_results_good amod_results_parsing dobj_obtaining_results nsubj_demonstrated_Gildea appos_Gildea_2001 complm_benefit_that det_parser_the amod_parser_lexicalized amod_parser_Model-1 nsubj_benefit_parser prep_parser_of nn__Collins pobj_of_ aux_benefit_does neg_benefit_not ccomp_demonstrated_benefit prep_benefit_from amod_information_bilexical pobj_from_information advmod_tested_when advcl_benefit_tested dep_benefits_on det_domain_a amod_domain_new nn_domain_text pobj_on_domain cc_on_and advmod_marginally_only conj_on_marginally dobj_tested_benefits prep_tested_from amod_information_such pobj_from_information advmod_tested_when advcl_tested_tested prep_tested_on det_domain_the amod_domain_same nn_domain_text pobj_on_domain prep_tested_as det_corpora_the nn_corpora_training pobj_as_corpora ' +o,7490,'Alternative Class-Based Estimation Methods The approaches used for comparison are that of Resnik , subsequently developed by Ribas , and that of Li and Abe , which has been adopted by McCarthy ',Dunning,'nn_Methods_Alternative nn_Methods_Class-Based nn_Methods_Estimation det_approaches_The nsubj_that_approaches partmod_approaches_used prep_used_for pobj_for_comparison cop_that_are dep_Methods_that prep_that_of nn__Resnik pobj_of_ advmod_developed_subsequently conj__developed prep_developed_by nn__Ribas pobj_by_ cc__and conj__that prep_that_of nn__Li cc_Li_and conj_Li_Abe pobj_of_ nsubjpass_adopted_which aux_adopted_has auxpass_adopted_been rcmod__adopted prep_adopted_by nn__McCarthy pobj_by_ ' +o,7491,'For detailed descriptions of SMT models see for example ',Brown,'prep_see_For amod_descriptions_detailed pobj_For_descriptions prep_descriptions_of pobj_of_SMT nsubj_see_models prep_see_for pobj_for_example ' +o,7492,'With the exception of , most unsupervised work on PP attachment is based on superficial analysis of the unlabeled corpus without the use of partial parsing ',Marcus,'prep_based_With det_exception_the pobj_With_exception prep_exception_of pobj_of_ advmod_unsupervised_most amod_work_unsupervised nsubjpass_based_work prep_work_on nn_attachment_PP pobj_on_attachment auxpass_based_is prep_based_on amod_analysis_superficial pobj_on_analysis prep_analysis_of det_corpus_the amod_corpus_unlabeled pobj_of_corpus prep_based_without det_use_the pobj_without_use prep_use_of amod__partial amod__parsing pobj_of_ ' +o,7493,'Introduction Recently , there has been an increased interest in approaches to automatically learning to recognize shallow linguistic patterns in text -LRB- and Marcus , 1995 , Vilain and Day , 1996 , Argamon et al , 1998 , Buchholz , 1998 , Cardie and Pierce , 1998 , Veenstra , 1998 , Daelemans et aI ',Ramshaw,'advmod_been_Recently expl_been_there aux_been_has dep_Introduction_been det_interest_an amod_interest_increased dobj_been_interest prep_been_in pobj_in_approaches prep_been_to advmod_learning_automatically pcomp_to_learning aux_recognize_to xcomp_learning_recognize nn_patterns_shallow amod_patterns_linguistic dobj_recognize_patterns prep_recognize_in amod__text pobj_in_ cc__and nn_Vilain_Marcus num_Vilain_1995 conj__Vilain cc__and nn_Argamon_Day num_Argamon_1996 conj__Argamon cc__et nn_Cardie_al num_Cardie_1998 dep_Cardie_Buchholz num_Cardie_1998 conj__Cardie cc__and nn_aI_Pierce num_aI_1998 dep_aI_Veenstra num_aI_1998 dep_aI_Daelemans num_aI_et conj__aI ' +n,7494,'Ever since its introduction in general and in computational linguistics , many researchers have pointed out that there are quite some problems in using -LRB- eg ',Carletta,'dep_since_Ever poss_introduction_its pobj_since_introduction prep_introduction_in amod__general pobj_in_ cc_since_and conj_since_in amod__computational nn__linguistics pobj_in_ amod_researchers_many dep_since_researchers aux_pointed_have rcmod_researchers_pointed prt_pointed_out complm_are_that expl_are_there ccomp_pointed_are predet_problems_quite det_problems_some nsubj_are_problems prep_problems_in pcomp_in_using dobj_using_eg ' +n,7495,'1 Introduction The dominance of traditional phrase-based statistical machine translation -LRB- PBSMT -RRB- models has recently been challenged by the development and improvement of a number of new models that explicity take into account the syntax of the sentences being translated ',Koehn,'partmod_1_Introduction det_dominance_The dobj_Introduction_dominance prep_dominance_of amod_translation_traditional amod_translation_phrase-based amod_translation_statistical nn_translation_machine pobj_of_translation abbrev_translation_PBSMT nsubjpass_challenged_models advmod_challenged_ aux_challenged_has advmod_challenged_recently auxpass_challenged_been dep_1_challenged prep_challenged_by det_development_the pobj_by_development cc_development_and conj_development_improvement prep_development_of det_number_a pobj_of_number prep_number_of amod_models_new pobj_of_models nsubj_take_that advmod_take_explicity rcmod_development_take prep_take_into pobj_into_account det_syntax_the dobj_take_syntax prep_syntax_of det_sentences_the pobj_of_sentences auxpass_translated_being partmod_sentences_translated ' +p,7496,'Aligning tokens in parallel sentences using the IBM Models , may require less information than full-blown translation since the task is constrained by the source and target tokens present in each sentence pair ',Brown,'dep_information_Aligning dobj_Aligning_tokens prep_Aligning_in amod_sentences_parallel pobj_in_sentences partmod_sentences_using det__the nn__IBM nn__Models dobj_using_ nsubj_information_ aux_information_may dep_information_require amod_information_less prep_information_than amod_translation_full-blown pobj_than_translation mark_constrained_since det_task_the nsubjpass_constrained_task auxpass_constrained_is advcl_information_constrained prep_constrained_by det_source_the pobj_by_source cc_source_and nn_present_target nn_present_tokens conj_source_present prep_constrained_in det_pair_each nn_pair_sentence pobj_in_pair ' +o,7497,'A period should therefore be interpreted as an abbreviation marker and not as a sentence boundary marker if the two tokens surrounding it can indeed be considered as a collocation according to original log-likelihood ratio amended with the one-sidedness constraint introduced in Section 22 ',Dunning,'det_period_A nsubjpass_interpreted_period aux_interpreted_should advmod_interpreted_therefore auxpass_interpreted_be prep_interpreted_as det_marker_an nn_marker_abbreviation pobj_as_marker cc_as_and conj_as_not dep_as_as det_marker_a nn_marker_sentence nn_marker_boundary pobj_as_marker mark_considered_if det_tokens_the num_tokens_two nsubjpass_considered_tokens partmod_tokens_surrounding dobj_surrounding_it aux_considered_can advmod_considered_indeed auxpass_considered_be advcl_interpreted_considered prep_considered_as det_collocation_a pobj_as_collocation prep_considered_according dep_according_to amod_ratio_ amod_ratio_original amod_ratio_log-likelihood pobj_to_ratio amod_ratio_amended prep_ratio_with det_constraint_the amod_constraint_one-sidedness pobj_with_constraint partmod_constraint_introduced prep_introduced_in pobj_in_Section num_Section_22 ' +o,7498,'415-458 , Stochastic inversion transduction grammars and bilingual parsing of parallel corpora ',Wu,'num_grammars_ amod_grammars_Stochastic nn_grammars_inversion nn_grammars_transduction conj_415-458_grammars cc_415-458_and amod_parsing_bilingual conj_415-458_parsing prep_415-458_of amod_corpora_parallel pobj_of_corpora ' +p,7499,'To improve the unknown word model , featurebased approach such as the maximum entropy method might be useful , because we don \' t have to divide the training data into several disjoint sets -LRB- like we did by part of speech and word type -RRB- and we can incorporate more linguistic and morphological knowledge into the same probabilistic framework ',Ratnaparkhi,'aux_improve_To dep_useful_improve det_model_the amod_model_unknown nn_model_word dobj_improve_model amod_approach_featurebased nsubj_useful_approach dep_as_such prep_approach_as det__the amod__maximum amod__entropy nn__method pobj_as_ aux_useful_might cop_useful_be prep_have_because pobj_because_we poss_t_don nsubj_have_t conj_useful_have aux_divide_to xcomp_have_divide det_data_the nn_data_training dobj_divide_data prep_divide_into amod_sets_several nn_sets_disjoint pobj_into_sets mark_did_like nsubj_did_we parataxis_divide_did prep_did_by pobj_by_part prep_part_of nn_type_speech cc_speech_and conj_speech_word pobj_of_type cc_useful_and nsubj_incorporate_we aux_incorporate_can conj_useful_incorporate advmod_linguistic_more amod_knowledge_linguistic cc_linguistic_and conj_linguistic_morphological dobj_incorporate_knowledge prep_incorporate_into det_framework_the amod_framework_same amod_framework_probabilistic pobj_into_framework ' +o,7500,'1999 -RRB- , OpenCCG and XLE , or created semi-automatically , or fully automatically extracted from annotated corpora , like the HPSG , LFG and CCG resources derived from the Penn-II Treebank -LRB- PTB -RRB- ',Cahill,'dep__1999 nn__OpenCCG nsubj__ cc__and conj__XLE dep__ cc__or conj__created amod__semi-automatically dobj__ cc__or advmod__fully advmod_extracted_automatically conj__extracted prep_extracted_from amod_corpora_annotated pobj_from_corpora prep_extracted_like det__the nn__HPSG nn__ appos__LFG pobj_like_ cc__and nn_resources_CCG nn_resources_ conj__resources partmod__derived prep_derived_from det_Treebank_the nn_Treebank_Penn-II pobj_from_Treebank abbrev__PTB ' +o,7501,'0 500 1000 1500 2000 5000 10000 15000 20000 25000 30000 Number of interlanguage links Vector length aren ares arro enes enro esro Figure 5: Number of interlanguage links vs. vector length for the Miller-Charles data set 0 500 1000 1500 2000 2500 3000 3500 4000 5000 10000 15000 20000 25000 30000 Number of interlanguage links Vector length aren ares arro enes enro esro Figure 6: Number of interlanguage links vs. vector length for the WordSimilarity-353 data set edge bases (Lesk, 1986; Wu and Palmer, 1994; Resnik, 1995; Jiang and Conrath, 1997; Hughes and Ramage, 2007) or on large corpora (Salton et al., 1997; Landauer et al., 1998; Turney, 2001; Gabrilovich and Markovitch, 2007).',Hughes,'' +o,7502,'1 Introduction Early works , , and to a certain extent , presented methods to ex ~ : \' ~ ct bi \' _ ` i ~ gua ! ',Brown,'num_works_1 nn_works_Introduction nn_works_Early nsubj_presented_works dep_works_ cc__and conj__to det__a amod__certain nn__extent pobj_to_ dobj_presented_methods prep_presented_to nn_~_ex pobj_to_~ amod___~ nn___ct nn___bi dobj_presented__ dep_gua_i amod_gua_~ dep___gua ' +o,7503,'In the Penn Treebank , null elements , or empty categories , are used to indicate non-local dependencies , discontinuous constituents , and certain missing elements ',Marcus,'prep_used_In det__the nn__Penn nn__Treebank pobj_In_ amod_elements_null conj__elements cc__or amod_categories_empty conj__categories auxpass_used_are aux_indicate_to xcomp_used_indicate amod_dependencies_non-local nsubjpass_used_dependencies amod_constituents_discontinuous appos_dependencies_constituents cc_constituents_and amod_elements_certain amod_elements_missing conj_constituents_elements ' +o,7504,'Different approaches have been proposed to measure matches using words or more meaningful semantic units , for example , ROUGE , factoid analysis , pyramid method , and Basic Element -LRB- BE -RRB- ',Lin,'amod_approaches_Different nsubjpass_proposed_approaches aux_proposed_have auxpass_proposed_been aux_measure_to xcomp_proposed_measure ccomp_measure_matches xcomp_matches_using dobj_using_words cc_words_or advmod_meaningful_more amod_units_meaningful amod_units_semantic conj_words_units conj_words_for pobj_for_example nn__ROUGE conj_example_ amod__factoid nn__analysis conj_example_ nn__pyramid nn__method conj_example_ cc_example_and nn__Basic nn__Element abbrev__BE conj_example_ ' +o,7505,'To make things worse , languages are non-isomorphic , ie , there is no 1to-1 mapping between tree nodes , thus in practice one has to use more expressive formalisms such as synchronous tree-substitution grammars ',Galley,'aux_make_To dep_is_make nsubj_ie_things dep_ie_worse nsubj_non-isomorphic_languages cop_non-isomorphic_are dep_ie_non-isomorphic xcomp_make_ie expl_is_there det_mapping_no amod_mapping_1to-1 nsubj_is_mapping prep_mapping_between nn_nodes_tree pobj_between_nodes dep_in_thus prep_is_in pobj_in_practice nsubj_has_one rcmod_practice_has aux_use_to xcomp_has_use advmod_expressive_more amod_formalisms_expressive dobj_use_formalisms dep_as_such prep_formalisms_as amod_grammars_synchronous amod_grammars_tree-substitution pobj_as_grammars ' +o,7506,'Following , we first generated CTB 30 from CTB 40 using sentence IDs 110364 ',Zhang,'prep_generated_Following pobj_Following_ nsubj_generated_we advmod_generated_first dobj_generated_CTB num_CTB_30 prep_CTB_from pobj_from_CTB num_CTB_40 xcomp_generated_using nn_IDs_sentence dobj_using_IDs num_IDs_110364 ' +p,7507,'Recently , many syntax-based models have been proposed to address the above deficiencies ',Wu,'advmod_proposed_Recently amod_models_many amod_models_syntax-based nsubjpass_proposed_models aux_proposed_have auxpass_proposed_been aux_address_to xcomp_proposed_address det_deficiencies_the amod_deficiencies_above dobj_address_deficiencies ' +o,7508,'The examples represent seven-word windows of words and their respective -LRB- predicted -RRB- part-of-speech tags , and each example is labeled with a class using the IOB type of segmentation coding as introduced by , marking whether the middle word is inside -LRB- I -RRB- , outside -LRB- O -RRB- , or at the beginning -LRB- B -RRB- of a chunk ',Ramshaw,'det_examples_The nsubj_represent_examples amod_windows_seven-word dobj_represent_windows prep_windows_of pobj_of_words cc_words_and poss_tags_their amod_tags_respective appos_tags_predicted nn_tags_part-of-speech conj_words_tags cc_represent_and det_example_each nsubjpass_labeled_example auxpass_labeled_is conj_represent_labeled prep_labeled_with det_class_a pobj_with_class partmod_class_using det_type_the nn_type_IOB dobj_using_type prep_type_of pobj_of_segmentation partmod_segmentation_coding mark_introduced_as advcl_coding_introduced prep_introduced_by pobj_by_ dep_introduced_marking complm_inside_whether det_word_the amod_word_middle nsubj_inside_word cop_inside_is ccomp_marking_inside dep_inside_I advmod_inside_outside appos_type_O cc_with_or conj_with_at det_beginning_the pobj_at_beginning appos_beginning_B prep_beginning_of det_chunk_a pobj_of_chunk ' +p,7509,'The field of statistical machine translation has been blessed with a long tradition of freely available software tools such as GIZA + + and parallel corpora such as the Canadian Hansards2 ',Och,'det_field_The nsubjpass_blessed_field prep_field_of amod_translation_statistical nn_translation_machine pobj_of_translation aux_blessed_has auxpass_blessed_been prep_blessed_with det_tradition_a amod_tradition_long pobj_with_tradition prep_tradition_of advmod_available_freely amod_tools_available nn_tools_software pobj_of_tools dep_as_such prep_tools_as nn_corpora_GIZA nn_corpora_+ cc__+ amod_corpora_ cc__and conj__parallel pobj_as_corpora dep_as_such prep_tools_as det_Hansards2_the amod_Hansards2_Canadian pobj_as_Hansards2 ' +n,7510,'1 Introduction Hierarchical approaches to machine translation have proven increasingly successful in recent years , and often outperform phrase-based systems on target-language fluency and adequacy ',Och,'num_approaches_1 nn_approaches_Introduction nn_approaches_Hierarchical nsubj_proven_approaches prep_approaches_to nn_translation_machine pobj_to_translation aux_proven_have advmod_successful_increasingly acomp_proven_successful prep_proven_in amod__recent nn__years pobj_in_ cc_proven_and advmod_outperform_often conj_proven_outperform amod__phrase-based nn__systems dobj_outperform_ prep__on amod_fluency_target-language pobj_on_fluency cc_fluency_and conj_fluency_adequacy ' +o,7511,' proposed a new algorithm for parameter estimation as an alternate to CRF ',Collins,'nsubj_proposed_ det_algorithm_a amod_algorithm_new dobj_proposed_algorithm prep_proposed_for nn_estimation_parameter pobj_for_estimation prep_estimation_as det_alternate_an pobj_as_alternate prep_proposed_to pobj_to_CRF ' +o,7512,'However , in the coarse-grained task , the sense inventory was first clustered semi-automatically with each cluster representing an equivalence class over senses ',Navigli,'advmod_clustered_However prep_clustered_in det_task_the amod_task_coarse-grained pobj_in_task det_inventory_the nn_inventory_sense nsubjpass_clustered_inventory auxpass_clustered_was advmod_clustered_first advmod_clustered_semi-automatically prep_clustered_with det_cluster_each pobj_with_cluster xcomp_clustered_representing det_class_an amod_class_equivalence dobj_representing_class prep_representing_over pobj_over_senses ' +o,7513,'A detailed discussion on the use of kappa in natural language processing is presented in ',Carletta,'det_discussion_A amod_discussion_detailed nsubjpass_presented_discussion prep_discussion_on det_use_the pobj_on_use prep_use_of pobj_of_kappa prep_kappa_in amod_processing_natural nn_processing_language pobj_in_processing auxpass_presented_is prep_presented_in ' +o,7514,'SSee for explanation of this notation for events ',Hobbs,'nn__SSee prep__for pobj_for_explanation prep__of det_notation_this pobj_of_notation prep_notation_for pobj_for_events ' +o,7515,'To determine the target distribution we classified 171 -LRB- approximately 5 \% -RRB- randomly selected utterances from the TownInfo data , that were used as a development set2 In Table 1 we can see that 152 \% of the trees in the artificial corpus will be NP NSUs3 4 Data generation We constructed our artificial corpus from sections 2 to 21 of the Wall Street Journal -LRB- WSJ -RRB- section of the Penn Treebank corpus 2We discarded very short utterances -LRB- yes , no , and greetings -RRB- since they dont need parsing ',Marcus,'aux_determine_To dep_classified_determine det_distribution_the nn_distribution_target dobj_determine_distribution nsubj_classified_we amod_utterances_171 quantmod_5_approximately num_\%_5 appos_utterances_\% advmod_selected_randomly amod_utterances_selected dobj_classified_utterances prep_utterances_from det_data_the nn_data_TownInfo pobj_from_data nsubjpass_used_that auxpass_used_were rcmod_utterances_used prep_used_as det_set2_a nn_set2_development pobj_as_set2 prep_see_In pobj_In_Table num_Table_1 nsubj_see_we aux_see_can rcmod_set2_see complm_generation_that num_\%_152 nsubj_generation_\% prep_\%_of det_trees_the pobj_of_trees prep_trees_in det_corpus_the amod_corpus_artificial pobj_in_corpus aux_generation_will cop_generation_be nn_generation_NP nn_generation_NSUs3 num_generation_4 nn_generation_Data ccomp_see_generation nsubj_constructed_We dep_classified_constructed poss_corpus_our amod_corpus_artificial dobj_constructed_corpus prep_constructed_from pcomp_from_sections number_21_2 dep_21_to dobj_sections_21 prep_21_of det_Journal_the nn_Journal_Wall nn_Journal_Street pobj_of_Journal abbrev_21_WSJ punct_21_section prep_21_of det_2We_the nn_2We_Penn nn_2We_Treebank nn_2We_corpus nn_2We_ pobj_of_2We partmod_2We_discarded advmod_short_very amod_utterances_short dobj_discarded_utterances dep_utterances_yes conj_yes_no cc_yes_and conj_yes_greetings mark_dont_since nsubj_dont_they advcl_discarded_dont nn_parsing_need dobj_dont_parsing ' +o,7516,'This problem can be cast as an instance of synchronous ITG parsing ',Wu,'det_problem_This nsubjpass_cast_problem aux_cast_can auxpass_cast_be prep_cast_as det_instance_an pobj_as_instance prep_instance_of amod_parsing_synchronous nn_parsing_ITG pobj_of_parsing ' +o,7517,'Following , we describe the original parsing architecture and our modifications to it as a Dynamic Bayesian network ',Titov,'prep_describe_Following pobj_Following_ nsubj_describe_we det_architecture_the amod_architecture_original amod_architecture_parsing dobj_describe_architecture cc_architecture_and poss_modifications_our conj_architecture_modifications prep_modifications_to pobj_to_it prep_modifications_as det_network_a nn_network_Dynamic nn_network_Bayesian pobj_as_network ' +o,7518,' -RRB- , the tagger for grammatical functions works with lexical and contextual probability measures Pq -LRB- -RRB- ',Brown,'dep_works_ det_tagger_the nsubj_works_tagger prep_tagger_for amod_functions_grammatical pobj_for_functions prep_works_with amod_measures_lexical cc_lexical_and conj_lexical_contextual nn_measures_probability pobj_with_measures tmod_works_Pq ' +o,7519,'As multiple derivations are used for finding optimal translations , we extend the minimum error rate training -LRB- MERT -RRB- algorithm to tune feature weights with respect to BLEU score for max-translation decoding -LRB- Section 4 -RRB- ',Och,'mark_used_As amod_derivations_multiple nsubjpass_used_derivations auxpass_used_are advcl_extend_used prep_used_for pcomp_for_finding amod_translations_optimal dobj_finding_translations nsubj_extend_we det_training_the amod_training_minimum nn_training_error nn_training_rate nn__training abbrev__MERT nn__algorithm dobj_extend_ prep__to nn_weights_tune nn_weights_feature pobj_to_weights prep_weights_with pobj_with_respect prep_extend_to amod_score_BLEU pobj_to_score prep_score_for amod_decoding_max-translation pobj_for_decoding appos_decoding_Section num_Section_4 ' +o,7520,'This is the way the Maximum Entropy tagger runs if one uses the binary version from the website -LRB- see the comparison in Section 5 -RRB- ',Ratnaparkhi,'nsubj_way_This cop_way_is det_way_the det__the nn__Maximum nn__Entropy nn__tagger nsubj_runs_ rcmod_way_runs mark_uses_if nsubj_uses_one advcl_runs_uses det_version_the amod_version_binary dobj_uses_version prep_uses_from det_website_the pobj_from_website parataxis_runs_see det_comparison_the dobj_see_comparison prep_comparison_in pobj_in_Section num_Section_5 ' +p,7521,'3 Space-Efficient Approximate Frequency Estimation Prior work on approximate frequency estimation for language models provide a no-false-negative guarantee , ensuring that counts for n-grams in the model are returned exactly , while working to make sure the false-positive rate remains small ',Talbot,'dep_provide_3 nn_Estimation_Space-Efficient nn_Estimation_Approximate nn_Estimation_Frequency nsubj_provide_Estimation prep_Estimation_Prior dep_Prior_work prep_work_on amod_estimation_approximate nn_estimation_frequency pobj_on_estimation prep_estimation_for nn_models_language pobj_for_models det_guarantee_a amod_guarantee_no-false-negative dobj_provide_guarantee xcomp_provide_ensuring complm_returned_that nsubjpass_returned_counts prep_counts_for pobj_for_n-grams prep_n-grams_in det_model_the pobj_in_model auxpass_returned_are ccomp_ensuring_returned advmod_returned_exactly dep_working_while parataxis_returned_working aux_make_to xcomp_working_make acomp_make_sure det_rate_the amod_rate_false-positive nsubj_small_rate cop_small_remains dep_provide_small ' +o,7522,'Proceedings of the Conference on Empirical Methods in Natural 2 Automatic Thesaurus Extraction The development of large thesauri and semantic resources , such as WordNet , has allowed lexical semantic information to be leveraged to solve NLP tasks , including collocation discovery , model estimation and text classi cation ',Brown,'dep_development_Proceedings prep_Proceedings_of det_Conference_the pobj_of_Conference prep_Conference_on nn_Methods_Empirical pobj_on_Methods prep_Methods_in pobj_in_Natural num_Natural_2 nn_Extraction_Automatic nn_Extraction_Thesaurus dep_development_Extraction det_development_The dep_as_development prep_development_of amod_thesauri_large pobj_of_thesauri cc_thesauri_and amod_resources_semantic conj_thesauri_resources dep_as_such nn__WordNet pobj_as_ aux_allowed_has dep__allowed amod_information_lexical amod_information_semantic dobj_allowed_information aux_leveraged_to cop_leveraged_be xcomp_allowed_leveraged aux_solve_to xcomp_leveraged_solve nn_tasks_NLP dobj_solve_tasks prep__including nn__collocation nn__discovery nn__ dep__model nn__estimation pobj_including_ cc__and nn__text nn__classi nn__cation conj__ ' +o,7523,'In order to filter the noise caused by the error alignment links , we only retain those translation pairs whose log-likelihood ratio scores are above a threshold ',Dunning,'mark_filter_In dep_filter_order aux_filter_to dep_retain_filter det_noise_the dobj_filter_noise partmod_noise_caused prep_caused_by det_links_the nn_links_error nn_links_alignment pobj_by_links nsubj_retain_we advmod_retain_only det_pairs_those nn_pairs_translation dobj_retain_pairs poss_ratio_whose amod_ratio_log-likelihood rel_are_ratio nn__scores nsubj_are_ rcmod_pairs_are prep_are_above det_threshold_a pobj_above_threshold ' +o,7524,'POS tagging and phrase chunking in English were done using the trained systems provided with the fnTBL Toolkit ; both were trained from the annotated Penn Treebank corpus ',Brown,'dep_tagging_POS nsubjpass_done_tagging cc_tagging_and nn_chunking_phrase conj_tagging_chunking prep_tagging_in pobj_in_English auxpass_done_were xcomp_done_using det_systems_the amod_systems_trained dobj_using_systems prep_using_provided dep_provided_with det__the amod__fnTBL nn__Toolkit pobj_with_ nsubjpass_trained_both auxpass_trained_were parataxis_done_trained prep_trained_from det__the amod__annotated nn__Penn nn__Treebank nn__corpus pobj_from_ ' +o,7525,'We use the distributed training and application infrastructure described in with modifications to allow the training of predictive class-based models and their application in the decoder of the machine translation system ',Brants,'nsubj_use_We det_training_the amod_training_distributed nsubj_described_training cc_training_and nn_infrastructure_application conj_training_infrastructure ccomp_use_described prep_described_in pobj_in_ prep_described_with pobj_with_modifications aux_allow_to xcomp_described_allow det_training_the dobj_allow_training prep_training_of amod_models_predictive amod_models_class-based pobj_of_models cc_models_and poss_application_their conj_models_application prep_allow_in det_decoder_the pobj_in_decoder prep_decoder_of det_system_the nn_system_machine nn_system_translation pobj_of_system ' +o,7526,'We borrow this useful term from the Core Language Engine project ',Hobbs,'nsubj_borrow_We det_term_this amod_term_useful dobj_borrow_term prep_borrow_from det__the nn__Core nn__Language nn__Engine nn__project pobj_from_ ' +o,7527,'4 Phrase-Based Translation In phrase-based translation , the translation process is modeled by splitting the source sentence into phrases -LRB- a contiguous string of words -RRB- and translating the phrases as a unit ',Koehn,'num_Translation_4 amod_Translation_Phrase-Based prep_modeled_In amod_translation_phrase-based pobj_In_translation det_process_the nn_process_translation nsubjpass_modeled_process auxpass_modeled_is dep_Translation_modeled prep_modeled_by pcomp_by_splitting det_sentence_the nn_sentence_source dobj_splitting_sentence prep_splitting_into pobj_into_phrases det_string_a amod_string_contiguous dep_phrases_string prep_string_of pobj_of_words cc_splitting_and conj_splitting_translating det_phrases_the dobj_translating_phrases prep_translating_as det_unit_a pobj_as_unit ' +n,7528,'While we do not have a direct comparison , we note that performs worse on movie reviews than on his other datasets , the same type of data as the polarity dataset ',Turney,'mark_have_While nsubj_have_we aux_have_do neg_have_not advcl_note_have det_comparison_a amod_comparison_direct dobj_have_comparison nsubj_note_we complm_performs_that nsubj_performs_ ccomp_note_performs acomp_performs_worse prep_worse_on nn_reviews_movie pobj_on_reviews dep_on_than prep_performs_on poss_datasets_his amod_datasets_other pobj_on_datasets det_type_the amod_type_same appos_datasets_type prep_type_of pobj_of_data prep_data_as det_dataset_the amod_dataset_polarity pobj_as_dataset ' +o,7529,'In contrast , semi-supervised domain adaptation is the scenario in which , in addition to the labeled source data , we only have unlabeled and no labeled target domain data ',Dredze,'prep_scenario_In pobj_In_contrast amod__semi-supervised nn__domain nn__adaptation nsubj_scenario_ cop_scenario_is det_scenario_the rel_labeled_in dep_in_which pobj_in_in pobj_in_addition prep_addition_to det_data_the amod_data_labeled nn_data_source pobj_to_data nsubj_labeled_we advmod_labeled_only aux_labeled_have advmod_labeled_unlabeled cc_unlabeled_and conj_unlabeled_no rcmod_scenario_labeled nn_data_target nn_data_domain dobj_labeled_data ' +p,7530,' gave a good description of ME model ',Berger,'nsubj_gave_ det_description_a amod_description_good dobj_gave_description prep_description_of nn_model_ME pobj_of_model ' +o,7531,' provide four sets of annotation principles , one for non-coordinate configurations , one for coordinate configurations , one for traces -LRB- long distance dependencies -RRB- and a final catch all and clean up phase ',Cahill,'advmod_provide_ num_sets_four dobj_provide_sets prep_sets_of nn_principles_annotation pobj_of_principles conj_sets_one prep_one_for amod_configurations_non-coordinate pobj_for_configurations conj_sets_one prep_one_for amod_configurations_coordinate pobj_for_configurations conj_sets_one prep_one_for pobj_for_traces amod_dependencies_long nn_dependencies_distance appos_traces_dependencies cc_sets_and det_catch_a amod_catch_final conj_sets_catch dep_catch_all cc_provide_and conj_provide_clean prt_clean_up dobj_clean_phase ' +o,7532,'In our approach , we take into account both the relative positions of the nearby context words as well as the mutual information associated with the occurrence of a particular context word ',Church,'prep_take_In poss_approach_our pobj_In_approach nsubj_take_we prep_take_into pobj_into_account predet_positions_both det_positions_the amod_positions_relative dobj_take_positions prep_positions_of det_words_the amod_words_nearby nn_words_context pobj_of_words dep_well_as cc_positions_well dep_well_as det__the amod__mutual nn__information conj_positions_ partmod__associated prep_associated_with det_occurrence_the pobj_with_occurrence prep_occurrence_of det_word_a amod_word_particular nn_word_context pobj_of_word ' +o,7533,'1 Introduction Formal grammar used in statistical machine translation -LRB- SMT -RRB- , such as Bracketing Transduction Grammar -LRB- BTG -RRB- proposed by and the synchronous CFG presented by , provides a natural platform for integrating linguistic knowledge into SMT because hierarchical structures produced by the formal grammar resemble linguistic structures ',Wu,'num_grammar_1 nn_grammar_Introduction nn_grammar_Formal nsubj_provides_grammar partmod_grammar_used prep_used_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT dep_as_such prep_translation_as nn_Grammar_Bracketing nn_Grammar_Transduction pobj_as_Grammar abbrev_Grammar_BTG partmod_Grammar_proposed prep_proposed_by pobj_by_ cc__and det_CFG_the amod_CFG_synchronous conj__CFG partmod_CFG_presented prep_presented_by pobj_by_ det_platform_a amod_platform_natural dobj_provides_platform prep_provides_for dep_knowledge_integrating amod_knowledge_linguistic pcomp_for_knowledge prep_knowledge_into pobj_into_SMT mark_structures_because amod_structures_hierarchical nsubj_structures_structures partmod_structures_produced prep_produced_by det_grammar_the amod_grammar_formal pobj_by_grammar cop_structures_resemble amod_structures_linguistic advcl_knowledge_structures ' +o,7534,'Compared to the Penn Treebank , the POS tagset of the French Treebank is smaller -LRB- 13 tags vs 36 tags -RRB- : all punctuation marks are represented as the single PONCT tag , there are no separate tags for modal verbs , whwords , and possessives ',Marcus,'prep_smaller_Compared dep_Compared_to det__the nn__Penn nn__Treebank pobj_to_ det_tagset_the dep_tagset_POS nsubj_smaller_tagset prep_tagset_of det_Treebank_the amod_Treebank_French pobj_of_Treebank cop_smaller_is num_tags_13 dep_smaller_tags cc_tags_vs num_tags_36 conj_tags_tags det_marks_all nn_marks_punctuation nsubjpass_represented_marks auxpass_represented_are ccomp_tags_represented prep_represented_as det_tag_the amod_tag_single nn_tag_PONCT pobj_as_tag expl_tags_there cop_tags_are det_tags_no amod_tags_separate parataxis_smaller_tags prep_tags_for amod_verbs_modal pobj_for_verbs conj_tags_whwords cc_tags_and conj_tags_possessives ' +o,7535,'We perform minimum-error-rate training to tune the feature weights of the translation model to maximize the BLEU score on development set ',Och,'nsubj_perform_We amod__minimum-error-rate nn__training nsubj_tune_ aux_tune_to xcomp_perform_tune det_weights_the nn_weights_feature dobj_tune_weights prep_weights_of det_model_the nn_model_translation pobj_of_model aux_maximize_to xcomp_tune_maximize det_score_the amod_score_BLEU dobj_maximize_score prep_maximize_on nn_set_development pobj_on_set ' +n,7536,'At any rate , regularized conditional loglinear models have not previously been applied to the problem of producing a high quality part-of-speech tagger : Ratnaparkhi , Toutanova and Manning , and all present unregularized models ',Collins,'prep_applied_At det_rate_any pobj_At_rate amod_models_regularized amod_models_conditional nn_models_loglinear nsubjpass_applied_models aux_applied_have neg_applied_not advmod_applied_previously auxpass_applied_been prep_applied_to det_problem_the pobj_to_problem prep_problem_of pcomp_of_producing det_quality_a amod_quality_high dobj_producing_quality amod_tagger_part-of-speech dep_quality_tagger nn__Ratnaparkhi dep_quality_ conj__Toutanova cc__and nn__Manning conj__ cc_quality_and advmod_all_ amod_models_all amod_models_present amod_models_unregularized conj_quality_models ' +o,7537,'The domain axioms will bind the body variables to their most likely referents during unification with facts , and previously assumed and proven propositions similarly to ',Hobbs,'det_axioms_The nn_axioms_domain nsubj_bind_axioms aux_bind_will det_variables_the nn_variables_body dobj_bind_variables prep_bind_to poss_referents_their advmod_likely_most amod_referents_likely pobj_to_referents prep_bind_during pobj_during_unification prep_unification_with pobj_with_facts cc_facts_and advmod_facts_previously amod_propositions_assumed cc_assumed_and conj_assumed_proven dep_facts_propositions advmod_facts_similarly dep_bind_to ' +p,7538,'One of the largest and earliest such efforts is the Penn Treebank , which contains a one-million word Institute for Research in Cognitive Science , University of Pennsylvania , 3401 Walnut Street , Suite 400A , Philadelphia , PA 19104-6228 , USA ',Marcus,'nsubj__One prep_One_of dep_largest_the pobj_of_largest cc_largest_and conj_largest_earliest amod_efforts_such dep_largest_efforts cop__is det__the nn__Penn nn__Treebank nsubj_contains_which rcmod__contains det_Institute_a amod_Institute_one-million nn_Institute_word dobj_contains_Institute prep_Institute_for pobj_for_Research prep_contains_in nn_Science_Cognitive pobj_in_Science appos_Science_University prep_University_of pobj_of_Pennsylvania num_Street_3401 nn_Street_Walnut appos_Science_Street appos_Science_Suite num_Suite_400A appos_Science_Philadelphia appos_Science_PA num_PA_19104-6228 appos_Science_USA ' +o,7539,'When the training text is adequate to estimate the tagger parameters , more efficient stochastic taggers and training methods can be implemented ',Marcus,'advmod_adequate_When det_text_the nn_text_training nsubj_adequate_text cop_adequate_is dep_implemented_adequate aux_estimate_to xcomp_adequate_estimate det_parameters_the nn_parameters_tagger dobj_estimate_parameters advmod_efficient_more amod__efficient amod__stochastic amod__taggers nsubjpass_implemented_ cc__and nn_methods_training conj__methods aux_implemented_can auxpass_implemented_be dep_implemented_ ' +o,7540,'In this study , we use the Google Web 1T 5gram Corpus ',Brants,'prep_use_In det_study_this pobj_In_study nsubj_use_we det_Corpus_the nn_Corpus_Google nn_Corpus_Web num_Corpus_1T amod_Corpus_5gram dobj_use_Corpus ' +o,7541,'Techniques that analyze n-gram precision such as BLEU score have been developed with the goal of comparing candidate translations against references provided by human experts in order to determine accuracy ; although in our application the candidate translator is a student and not a machine , the principle is the same , and we wish to adapt their technique to our context ',Papineni,'nsubjpass_developed_Techniques nsubj_analyze_that rcmod_Techniques_analyze amod_precision_n-gram dobj_analyze_precision dep_as_such prep_precision_as amod__BLEU nn__score pobj_as_ aux_developed_have auxpass_developed_been prep_developed_with det_goal_the pobj_with_goal prep_goal_of pcomp_of_comparing nn_translations_candidate dobj_comparing_translations prep_comparing_against pobj_against_references partmod_references_provided prep_provided_by amod_experts_human pobj_by_experts mark_determine_in dep_determine_order aux_determine_to purpcl_provided_determine dobj_determine_accuracy mark_student_although prep_student_in poss_application_our pobj_in_application det_translator_the nn_translator_candidate nsubj_student_translator cop_student_is det_student_a advcl_same_student cc_student_and conj_student_not det_machine_a dep_student_machine det_principle_the nsubj_same_principle cop_same_is det_same_the parataxis_developed_same cc_developed_and nsubj_wish_we conj_developed_wish aux_adapt_to xcomp_wish_adapt poss_technique_their dobj_adapt_technique prep_adapt_to poss_context_our pobj_to_context ' +o,7542,'Above the phrase level , these models typically have a simple distortion model that reorders phrases independently of their content , or not at all ',Koehn,'prep_have_Above det_level_the nn_level_phrase pobj_Above_level det_models_these nsubj_have_models advmod_have_typically det_model_a amod_model_simple nn_model_distortion dobj_have_model nsubj_reorders_that rcmod_model_reorders dobj_reorders_phrases dep_of_independently prep_reorders_of poss__their nn__content pobj_of_ cc_of_or conj_of_not dep_of_at det__all pobj_at_ ' +o,7543,'7However , the algorithms shares many common points with iterative algorithm that are known to converge and that have been proposed to find maximum entropy probability distributions under a set of constraints ',Berger,'advmod_points_7However det_algorithms_the nsubj_points_algorithms dep_points_shares amod_points_many amod_points_common prep_points_with amod_algorithm_iterative pobj_with_algorithm nsubjpass_known_that auxpass_known_are dep_points_known aux_converge_to xcomp_known_converge cc_known_and nsubjpass_proposed_that aux_proposed_have auxpass_proposed_been conj_known_proposed aux_find_to xcomp_proposed_find amod_distributions_maximum amod_distributions_entropy nn_distributions_probability dobj_find_distributions prep_find_under det_set_a pobj_under_set prep_set_of pobj_of_constraints ' +o,7544,'255 Meteor , Precision and Recall , and other such automatic metrics may also be affected to a greater or lesser degree because they are all quite rough measures of translation similarity , and have inexact models of allowable variation in translation ',Banerjee,'nn__255 nn__Meteor nsubjpass_affected_ nn__Precision cc_Precision_and conj_Precision_Recall conj__ cc__and amod_metrics_other amod_metrics_such amod_metrics_automatic conj__metrics aux_affected_may advmod_affected_also auxpass_affected_be prep_affected_to det_degree_a amod_degree_greater cc_greater_or conj_greater_lesser pobj_to_degree mark_measures_because nsubj_measures_they cop_measures_are det_measures_all advmod_rough_quite amod_measures_rough advcl_affected_measures prep_measures_of nn_similarity_translation pobj_of_similarity cc_affected_and conj_affected_have amod_models_inexact dobj_have_models prep_models_of amod_variation_allowable pobj_of_variation prep_variation_in pobj_in_translation ' +o,7545,'In addition to tfidf scores , Hulth uses part-of-speech tags and NP chunks and complements this with machine learning ; the latter has been used to good results in similar cases ',Lin,'prep_uses_In pobj_In_addition prep_addition_to nn_scores_tfidf pobj_to_scores nn__Hulth nsubj_uses_ amod_tags_part-of-speech dobj_uses_tags cc_tags_and nn_chunks_NP conj_tags_chunks cc_uses_and conj_uses_complements dobj_complements_this prep_complements_with nn_learning_machine pobj_with_learning det_latter_the nsubjpass_used_latter aux_used_has auxpass_used_been parataxis_uses_used prep_used_to amod_results_good pobj_to_results prep_results_in amod__similar nn__cases pobj_in_ ' +o,7546,'Our method revises and considerably extends the approach of originally designed for English , and , to the best of our knowledge , is the first NLD recovery algorithm for Chinese ',Cahill,'poss_method_Our nsubj_revises_method cc_revises_and advmod_revises_considerably conj_revises_extends det_approach_the nsubj_algorithm_approach prep_approach_of pobj_of_ advmod_designed_originally partmod__designed prep_designed_for pobj_for_English cc_for_and conj_for_to det_best_the pobj_to_best prep_best_of poss_knowledge_our pobj_of_knowledge cop_algorithm_is det_algorithm_the amod_algorithm_first nn_algorithm_NLD nn_algorithm_recovery dep_revises_algorithm prep_algorithm_for pobj_for_Chinese ' +o,7547,' gives indirect experimental evidence that this difference affects performance ',Lopez,'advmod_gives_ amod_evidence_indirect amod_evidence_experimental dobj_gives_evidence dep_affects_that det_difference_this nsubj_affects_difference dep_gives_affects dobj_affects_performance ' +o,7548,'We have used the optimal experiment configurations that we had obtained from the fourth experiment series for processing the complete data set ',Ramshaw,'nsubj_used_We aux_used_have det_configurations_the amod_configurations_optimal nn_configurations_experiment dobj_used_configurations complm_obtained_that nsubj_obtained_we aux_obtained_had ccomp_used_obtained prep_obtained_from det_series_the amod_series_fourth nn_series_experiment pobj_from_series prep_obtained_for pcomp_for_processing det_data_the amod_data_complete num_data_ dobj_processing_data partmod_data_set ' +o,7549,'5 Comparison with other approaches In some sense , this approach is similar to the notion of ` ambiguity classes \' explained in and where words that belong to the same part-of-speech figure together ',Cutting,'num_Comparison_5 nsubj_similar_Comparison prep_Comparison_with amod_approaches_other pobj_with_approaches prep_similar_In det_sense_some pobj_In_sense det_approach_this nsubj_similar_approach cop_similar_is prep_similar_to det_notion_the pobj_to_notion prep_notion_of amod_classes_ambiguity pobj_of_classes partmod_classes_explained prep_explained_in pobj_in_ cc__and conj__ advmod_figure_where nsubj_figure_words nsubj_belong_that rcmod_words_belong prep_belong_to det_part-of-speech_the amod_part-of-speech_same pobj_to_part-of-speech advcl_explained_figure advmod_figure_together ' +o,7550,'Essentially , we follow in using a rich ontology and a representation scheme that makes explicit all the individuals and abstract objects -LRB- ie , propositions , facts\\/beliefs , and eventualities -RRB- involved in the LF interpretation of an utterance ',Hobbs,'advmod_follow_Essentially nsubj_follow_we acomp_follow_ prep__in pcomp_in_using det_ontology_a amod_ontology_rich dobj_using_ontology cc_ontology_and det_scheme_a nn_scheme_representation conj_ontology_scheme nsubj_makes_that rcmod_scheme_makes acomp_makes_explicit predet_individuals_all det_individuals_the nsubj_involved_individuals cc_individuals_and amod_objects_abstract conj_individuals_objects dep_objects_ie conj_ie_propositions conj_ie_facts\\/beliefs cc_ie_and conj_ie_eventualities advmod_involved_ ccomp_explicit_involved prep_involved_in det_interpretation_the nn_interpretation_LF pobj_in_interpretation prep_interpretation_of det_utterance_an pobj_of_utterance ' +o,7551,'Third , we hope that the improved parses of bitext will serve as higher quality training data for improving monolingual parsing using a process similar to self-training ',McClosky,'advmod_hope_Third nsubj_hope_we complm_serve_that det_parses_the amod_parses_improved nsubj_serve_parses prep_parses_of pobj_of_bitext aux_serve_will ccomp_hope_serve prep_serve_as amod_data_higher nn_data_quality nn_data_training pobj_as_data prep_serve_for pcomp_for_improving amod_parsing_monolingual dobj_improving_parsing xcomp_improving_using det_process_a dobj_using_process amod_process_similar dep_similar_to dep_similar_self-training ' +o,7552,'As an alternative to the often used sourcechannel approach , we directly model the posterior probability Pr -LRB- e I 1 f J 1 -RRB- ',Brown,'prep_model_As det_alternative_an pobj_As_alternative prep_alternative_to det__the advmod_used_often amod__used amod__sourcechannel nn__approach pobj_to_ nsubj_model_we advmod_model_directly det_Pr_the nn_Pr_posterior nn_Pr_probability nsubj__Pr dep_I_e dep_f_I num_I_1 dep_-LRB-_f num_1_J dep_f_1 xcomp_model_ ' +o,7553,' O :981 \% C :982 \% 924 \% 931 \% IOB1 :9737 \% 9180 \% 9227 \% Argamon et al ',Ramshaw,'nn_O_ dep_C_O num_\%_:981 dep_O_\% num_\%_:982 dobj_C_\% num_\%_924 tmod_C_\% advmod_\%_931 dep_931_\% dep_C_ number_:9737_IOB1 num_\%_:9737 dobj__\% num_\%_9180 amod_\%_\% num_\%_9227 dep_\%_\% nn_al_Argamon nn_al_et nsubj_C_al ' +o,7554,'2 The Inversion Transduction Grammar The Inversion Transduction Grammar of can be thought as a a generative process which simultaneously produces strings in both languages through a series of synchronous context-free grammar productions ',Wu,'nsubjpass_thought_2 det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction dep_2_Grammar det_Grammar_The nn_Grammar_Inversion nn_Grammar_Transduction dep_2_Grammar prep_Grammar_of pobj_of_ aux_thought_can auxpass_thought_be prep_thought_as det_process_a dep_generative_a amod_process_generative pobj_as_process nsubj_produces_which advmod_produces_simultaneously rcmod_process_produces dobj_produces_strings prep_strings_in det_languages_both pobj_in_languages prep_produces_through det_series_a pobj_through_series prep_series_of amod_productions_synchronous amod_productions_context-free nn_productions_grammar pobj_of_productions ' +o,7555,'6 Parameter Estimation From the duality of ME and maximum likelihood , optimal parameters for model -LRB- 3 -RRB- can be found by maximizing the log-likelihood function over a training sample -LCB- -LRB- xt , yt -RRB- : t = 1 , , N -RCB- , ie : = argmax Nsummationdisplay t = 1 logp -LRB- yt xt -RRB- ',Berger,'num_Estimation_6 nn_Estimation_Parameter prep_found_From det_duality_the pobj_From_duality prep_duality_of nn__ME cc_ME_and conj_ME_maximum nn__likelihood pobj_of_ amod_parameters_optimal nsubjpass_found_parameters prep_parameters_for pobj_for_model appos_model_3 aux_found_can auxpass_found_be rcmod_Estimation_found prep_found_by pcomp_by_maximizing det_function_the amod_function_log-likelihood dobj_maximizing_function prep_function_over det_sample_a nn_sample_training pobj_over_sample nn_yt_xt dep_sample_yt dep_sample_t dep_1_= rcmod_t_1 appos_function_N appos_function_ie dep_found_= ccomp_=_argmax nn_t_Nsummationdisplay nsubj_1_t dep_1_= ccomp_argmax_1 dep_Estimation_logp nn_xt_yt appos_logp_xt ' +o,7556,'In the first set of experiments , we compare two settings of our UALIGN system with other aligners , GIZA + + -LRB- Union -RRB- and LEAF -LRB- with 2 iterations -RRB- ',Fraser,'prep_compare_In det_set_the amod_set_first pobj_In_set prep_set_of pobj_of_experiments nsubj_compare_we num_settings_two dobj_compare_settings prep_settings_of poss_system_our nn_system_UALIGN pobj_of_system prep_compare_with amod_aligners_other pobj_with_aligners nn__GIZA nn__+ nn__+ appos__Union appos_aligners_ cc__and conj__LEAF dep_LEAF_with num_iterations_2 pobj_with_iterations ' +o,7557,'23 Experiment The training set for these experiments was sections 01-21 of the Penn Treebank ',Marcus,'num_training_23 nn_training_Experiment nn_training_The dep_set_training mark_sections_for det_experiments_these nsubj_sections_experiments cop_sections_was advcl_set_sections num_sections_01-21 prep_sections_of det_Treebank_the nn_Treebank_Penn pobj_of_Treebank ' +p,7558,' has recently proposed a simpler SVM-based algorithm for analogical classification called PairClass ',Turney,'nsubj_proposed_ aux_proposed_has advmod_proposed_recently det_algorithm_a amod_algorithm_simpler amod_algorithm_SVM-based dobj_proposed_algorithm prep_proposed_for amod_classification_analogical pobj_for_classification partmod_classification_called dobj_called_PairClass ' +o,7559,'Many existing systems tbr SMT -LRB- Wang and Waibel , 1997 ; Niefien et al , 1 make use of a special way of structuring the string translation model : ` l ? he correspondence between the words in the source and the target string is described by aligmuents that assign one target word position to each source word position ',Brown,'amod_systems_Many amod_systems_existing nsubj_tbr_systems nn_Wang_SMT nsubj_make_Wang cc_Wang_and conj_Wang_Waibel appos_Waibel_1997 dep_Waibel_Niefien cc_Niefien_et conj_Niefien_al num__1 appos_Niefien_ ccomp_tbr_make nsubj__use prep_use_of det_way_a amod_way_special pobj_of_way prep_way_of pcomp_of_structuring det_model_the nn_model_string nn_model_translation dobj_structuring_model ccomp_make_ dep_correspondence_l nsubj_correspondence_he dep__correspondence prep_correspondence_between det_words_the pobj_between_words prep_correspondence_in det_source_the pobj_in_source cc_correspondence_and det_string_the nn_string_target nsubjpass_described_string auxpass_described_is conj_correspondence_described prep_described_by pobj_by_aligmuents nsubj_assign_that rcmod_aligmuents_assign num_position_one nn_position_target nn_position_word dobj_assign_position prep_assign_to det_position_each nn_position_source nn_position_word pobj_to_position ' +o,7560,'Here , following , we use a weighted , quasi-synchronous dependency grammar Apart from the obvious difference in application task , there are a few important differences with their model ',Smith,'advmod_use_Here prep_use_following pobj_following_ nsubj_use_we det_grammar_a amod_grammar_weighted amod_grammar_quasi-synchronous nn_grammar_dependency dobj_use_grammar dep_from_Apart prep_use_from det_difference_the amod_difference_obvious pobj_from_difference prep_are_in nn_task_application pobj_in_task expl_are_there dep_use_are det_differences_a amod_differences_few amod_differences_important nsubj_are_differences prep_differences_with poss_model_their pobj_with_model ' +o,7561,'Depending on the type of input , these efforts can be divided into two broad categories : the string-based systems whose input is a string to be simultaneously parsed and translated by a synchronous grammar , and the tree-based systems whose input is already a parse tree to be directly converted into a target tree or string ',Wu,'prep_divided_Depending dep_Depending_on det_type_the pobj_on_type prep_type_of pobj_of_input det_efforts_these nsubjpass_divided_efforts aux_divided_can auxpass_divided_be prep_divided_into num_categories_two amod_categories_broad pobj_into_categories det_systems_the amod_systems_string-based dep_categories_systems poss_input_whose nsubj_string_input cop_string_is det_string_a rcmod_systems_string aux_parsed_to auxpass_parsed_be advmod_parsed_simultaneously infmod_string_parsed cc_parsed_and conj_parsed_translated prep_parsed_by det__a amod__synchronous nn__grammar pobj_by_ cc_systems_and det_systems_the amod_systems_tree-based conj_systems_systems poss_input_whose nsubj_tree_input cop_tree_is advmod_tree_already det_tree_a amod_tree_parse rcmod_systems_tree aux_converted_to auxpass_converted_be advmod_converted_directly infmod_tree_converted prep_converted_into det_tree_a nn_tree_target pobj_into_tree cc_tree_or nn__string conj_tree_ ' +o,7562,'Moses used the development data for minimum error-rate training of its small number of parameters ',Och,'nsubj_used_Moses det_data_the nn_data_development dobj_used_data prep_used_for amod__minimum amod__error-rate nn__training pobj_for_ prep__of poss_number_its amod_number_small pobj_of_number prep_number_of pobj_of_parameters ' +o,7563,'He used the representation as well -LRB- IOB1 -RRB- ',Ramshaw,'nsubj_used_He det_representation_the amod_representation_ dobj_used_representation dep_well_as cc_representation_well conj_representation_IOB1 ' +o,7564,'5 Related work Cutting introduced grouping of words into equivalence classes based on the set of possible tags to reduce the number of the parameters Schmid used tile equivaleuce classes for smoothing ',Cutting,'num_Cutting_5 amod_Cutting_Related nn_Cutting_work nsubj_introduced_Cutting dobj_introduced_grouping prep_grouping_of pobj_of_words prep_introduced_into amod_classes_equivalence pobj_into_classes partmod_classes_based prep_based_on det_set_the pobj_on_set prep_set_of amod_tags_possible pobj_of_tags aux_reduce_to xcomp_based_reduce det_number_the dobj_reduce_number prep_number_of det_parameters_the pobj_of_parameters nn_Schmid_ nsubj_used_Schmid dep_introduced_used nn_classes_tile nn_classes_equivaleuce dobj_used_classes prep_used_for pobj_for_smoothing ' +o,7565,'We propose quasi-synchronous grammar -LRB- Section 3 -RRB- as a general solution and the Jeopardy model -LRB- Section 4 -RRB- as a specific instance ',Smith,'nsubj_propose_We amod_grammar_ amod_grammar_quasi-synchronous dobj_propose_grammar appos_grammar_Section num_Section_3 prep_grammar_as det_solution_a amod_solution_general pobj_as_solution cc_grammar_and det_model_the nn_model_Jeopardy conj_grammar_model appos_model_Section num_Section_4 prep_model_as det_instance_a amod_instance_specific pobj_as_instance ' +o,7566,'Our approach permits an alternative to minimum error-rate training ; it is discriminativebuthandleslatentstructureandregularization in more principled ways ',Och,'poss_approach_Our nsubj_permits_approach det_alternative_an dobj_permits_alternative prep_permits_to amod__minimum amod__error-rate nn__training pobj_to_ nsubjpass_discriminativebuthandleslatentstructureandregularization_it auxpass_discriminativebuthandleslatentstructureandregularization_is parataxis_permits_discriminativebuthandleslatentstructureandregularization prep_discriminativebuthandleslatentstructureandregularization_in advmod_principled_more amod_ways_principled pobj_in_ways ' +o,7567,'Recently , introduced an approach for incorporating a dependency-based language model into SMT ',Shen,'advmod_introduced_Recently nsubj_introduced_ det_approach_an dobj_introduced_approach prep_introduced_for pcomp_for_incorporating det_model_a amod_model_dependency-based nn_model_language dobj_incorporating_model prep_incorporating_into pobj_into_SMT ' +p,7568,'We chose a dataset that would be enjoyable to reannotate : the movie review dataset of 3 The dataset consists of 1000 positive and 1000 negative movie reviews obtained from the Internet Movie Database -LRB- IMDb -RRB- review archive , all written before 2002 by a total of 312 authors , with a cap of 20 reviews per author per 2Taking Ccontrast to be constant means that all rationales are equally valuable ',Pang,'nsubj_chose_We det_dataset_a dobj_chose_dataset nsubj_enjoyable_that aux_enjoyable_would cop_enjoyable_be rcmod_dataset_enjoyable aux_reannotate_to xcomp_enjoyable_reannotate det_dataset_the nn_dataset_movie nn_dataset_review nsubj_means_dataset prep_dataset_of pobj_of_ num__3 det_dataset_The nsubj_consists_dataset rcmod_dataset_consists prep_dataset_of num_reviews_1000 amod_reviews_positive cc_positive_and conj_positive_1000 amod_reviews_negative nn_reviews_movie pobj_of_reviews partmod_dataset_obtained prep_obtained_from det_archive_the nn_Database_Internet nn_Database_Movie nn_archive_Database abbrev_Database_IMDb nn_archive_review pobj_from_archive nsubj_written_all rcmod_archive_written prep_written_before pobj_before_2002 prep_written_by det_total_a pobj_by_total prep_total_of num_authors_312 pobj_of_authors prep_written_with det_cap_a pobj_with_cap prep_cap_of num_reviews_20 pobj_of_reviews prep_reviews_per pobj_per_author prep_author_per amod_Ccontrast_2Taking pobj_per_Ccontrast aux_constant_to cop_constant_be infmod_cap_constant parataxis_chose_means complm_valuable_that det_rationales_all nsubj_valuable_rationales cop_valuable_are advmod_valuable_equally ccomp_means_valuable ' +o,7569,'Let a183a49a48a50 a69 a188 a50 a51a181a51a181a51a212a188 a50a7a51a24a52 a48a54a53 a185a56a55 be a substring of a183 from the word a188 a50 with length a57 Note this notation is different from ',Brown,'aux_substring_Let num_a185a56a55_a183a49a48a50 amod_a185a56a55_a69 amod_a185a56a55_a188 amod_a185a56a55_a50 amod_a185a56a55_a51a181a51a181a51a212a188 amod_a185a56a55_a50a7a51a24a52 amod_a185a56a55_a48a54a53 nsubj_substring_a185a56a55 cop_substring_be det_substring_a prep_substring_of pobj_of_a183 prep_a183_from det_word_the pobj_from_word amod_a50_a188 nsubj_Note_a50 prep_a50_with nn_a57_length pobj_with_a57 rcmod_word_Note det_notation_this nsubj_different_notation cop_different_is ccomp_Note_different prep_different_from ' +o,7570,'1 Introduction Previous work on sentiment categorization makes an implicit assumption that a single score can express the polarity of an opinion text ',Turney,'num_work_1 amod_work_Introduction amod_work_Previous nsubj_makes_work prep_work_on nn_categorization_sentiment pobj_on_categorization det_assumption_an amod_assumption_implicit dobj_makes_assumption complm_express_that det_score_a amod_score_single nsubj_express_score aux_express_can ccomp_makes_express det_polarity_the dobj_express_polarity prep_polarity_of det_text_an nn_text_opinion pobj_of_text ' +o,7571,' , and extend it to structured shape descriptions of visual data ',Brown,'advmod__ cc__and conj__extend dobj_extend_it prep_extend_to amod_descriptions_structured nn_descriptions_shape pobj_to_descriptions prep_descriptions_of amod_data_visual pobj_of_data ' +o,7572,'We demonstrate that allowing different values for these hyperparameters significantly improves performance over both a strong baseline and within both a conditional random field sequence model for named entity recognition and a discriminatively trained dependency parser ','Daume III','nsubj_demonstrate_We complm_improves_that csubj_improves_allowing amod_values_different dobj_allowing_values prep_allowing_for det_hyperparameters_these pobj_for_hyperparameters advmod_improves_significantly ccomp_demonstrate_improves dobj_improves_performance prep_performance_over preconj_baseline_both det_baseline_a amod_baseline_strong pobj_over_baseline cc_baseline_and conj_baseline_ prep_baseline_within predet_model_both det_model_a amod_model_conditional amod_model_random nn_model_field nn_model_sequence pobj_within_model prep_improves_for amod_recognition_named nn_recognition_entity pobj_for_recognition cc_recognition_and det_parser_a advmod_trained_discriminatively amod_parser_trained nn_parser_dependency conj_recognition_parser ' +o,7573,'Idiom 0 0 1 1 0 2 V Doubt 3 0 4 0 0 7 Total A 294 160 546 39 1 1,040 In order to measure the agreement in a more precise way , we used the Kappa statistic , recently proposed by Carletta as a measure of agreement for discourse analysis ',Carletta,'nn_Doubt_Idiom num_Doubt_0 num_Doubt_0 num_Doubt_1 num_Doubt_1 num_Doubt_0 num_Doubt_2 nn_Doubt_V nsubj_294_Doubt num_A_3 num_A_0 num_A_4 num_A_0 num_A_0 num_A_7 nn_A_Total dep_Doubt_A number_546_160 num_39_546 iobj_294_39 number_1,040_1 dobj_294_1,040 mark_measure_In dep_measure_order aux_measure_to dep_used_measure det_agreement_the dobj_measure_agreement prep_agreement_in det_way_a advmod_precise_more amod_way_precise pobj_in_way nsubj_used_we dep_294_used det__the nn__Kappa nn__statistic dobj_used_ advmod_proposed_recently dep_used_proposed prep_proposed_by pobj_by_Carletta prep_proposed_as det_measure_a pobj_as_measure prep_measure_of pobj_of_agreement prep_proposed_for nn_analysis_discourse pobj_for_analysis ' +p,7574,'To reduce the time complexity , we adapted the lazy update proposed in , which was also used in ',Collins,'aux_reduce_To dep_adapted_reduce det_complexity_the nn_complexity_time dobj_reduce_complexity nsubj_adapted_we det_update_the amod_update_lazy dobj_adapted_update partmod_update_proposed prep_proposed_in pobj_in_ nsubjpass_used_which auxpass_used_was advmod_used_also rcmod__used prep_used_in pobj_in_ ' +o,7575,'With the availability of large natural language corpora annotated for syntactic structure , the treebanks , eg , , automatic grammar extraction became possible ',Marcus,'prep__With det_availability_the pobj_With_availability prep_availability_of amod_corpora_large amod_corpora_natural nn_corpora_language pobj_of_corpora partmod_corpora_annotated prep_annotated_for amod_structure_syntactic pobj_for_structure det_extraction_the nn_extraction_treebanks dep_extraction_eg dep_extraction_ amod_extraction_automatic nn_extraction_grammar nsubj__extraction cop__became amod__possible ' +o,7576,'Likelihood ratios are particularly useful when comparing common and rare events , making them natural here given the rareness of most question categories and the frequency of contributions ',Dunning,'amod_ratios_Likelihood nsubj_useful_ratios cop_useful_are advmod_useful_particularly advmod_comparing_when dep_useful_comparing amod__common cc_common_and conj_common_rare nn__events dobj_comparing_ dep_comparing_making nsubj_given_them dep_given_natural advmod_given_here xcomp_making_given det_rareness_the dobj_given_rareness prep_rareness_of amod_categories_most nn_categories_question pobj_of_categories cc_categories_and det_frequency_the conj_categories_frequency prep_given_of pobj_of_contributions ' +o,7577,'Using these patterns , we introduced verb form errors into AQUAINT , then re-parsed the corpus , and compiled the changes in the disturbed trees into a catalog ',Collins,'dep_introduced_Using det_patterns_these dobj_Using_patterns nsubj_introduced_we amod_errors_verb nn_errors_form dobj_introduced_errors prep_introduced_into pobj_into_AQUAINT advmod_re-parsed_then dep_AQUAINT_re-parsed det__the nn__corpus dobj_re-parsed_ cc_re-parsed_and conj_re-parsed_compiled det_changes_the dobj_compiled_changes prep_changes_in det_trees_the amod_trees_disturbed pobj_in_trees prep_compiled_into det_catalog_a pobj_into_catalog ' +o,7578,'ALM does this by using alignment models from the statistical machine translation literature ',Brown,'nsubj_does_ALM dobj_does_this prep_does_by pcomp_by_using amod_models_alignment dobj_using_models prep_using_from det_literature_the amod_literature_statistical nn_literature_machine nn_literature_translation pobj_from_literature ' +o,7579,'in their treatment of chunk-initial and chunk-final -LRB- + -RRB- words : IOB1 IOB2 IOE1 IOE2 The first word inside a baseNP immediately following another baseNP receives a B tag ',Ramshaw,'poss_treatment_their pobj_in_treatment prep_treatment_of amod_words_chunk-initial cc_chunk-initial_and conj_chunk-initial_chunk-final appos_words_+ pobj_of_words nn_IOE2_IOB1 nn_IOE2_IOB2 nn_IOE2_IOE1 dep_treatment_IOE2 det_word_The amod_word_first nsubj_receives_word prep_word_inside det_baseNP_a pobj_inside_baseNP advmod_following_immediately partmod_baseNP_following det_baseNP_another dobj_following_baseNP rcmod_IOE2_receives det_tag_a nn_tag_B dobj_receives_tag ' +o,7580,'1http : \\/ \\/ chasenorg \\/ taku\\/software\\/yamcha \\/ 2http : \\/ \\/ chasenorg \\/ taku\\/software\\/TinySVM \\/ 197 a0 Bracketed representation of roles was converted into IOB2 representation ',Ramshaw,'nn_\\/_\\/ nsubj_chasenorg_\\/ dep_1http_chasenorg amod_2http_\\/ nn_2http_taku\\/software\\/yamcha nn_2http_\\/ dobj_chasenorg_2http nn_\\/_\\/ nsubj_chasenorg_\\/ dep_2http_chasenorg amod_\\/_\\/ nn_\\/_taku\\/software\\/TinySVM dobj_chasenorg_\\/ num_representation_197 amod_representation_a0 nn_representation_Bracketed nsubjpass_converted_representation prep_representation_of pobj_of_roles auxpass_converted_was rcmod_\\/_converted prep_converted_into nn__IOB2 nn__representation nn__ pobj_into_ ' +o,7581,'Since no such corpus exists , researchers have used coarser features learned from smaller sets through supervised learning , manually-de ned coreference patterns to mine speci c kinds of data , or accepted the noise inherent in unsupervised schemes ',Bean,'mark_exists_Since det_corpus_no amod_corpus_such nsubj_exists_corpus advcl_used_exists nsubj_used_researchers aux_used_have nn_features_coarser dobj_used_features partmod_features_learned prep_learned_from amod_sets_smaller pobj_from_sets prep_learned_through pobj_through_supervised xcomp_learned_learning amod_patterns_ amod_patterns_manually-de nn_patterns_ned nn_patterns_coreference dobj_learning_patterns prep_patterns_to nn_speci_mine pobj_to_speci dep__c nsubj__kinds prep_kinds_of pobj_of_data dep_used_ cc__or conj__accepted det_noise_the nsubj_inherent_noise xcomp_accepted_inherent prep_inherent_in amod__unsupervised nn__schemes pobj_in_ ' +o,7582,'A richer set of features besides n-grams should be checked , and we should not ignore the potential effectiveness of unigrams in this task ',Pang,'det_set_A amod_set_richer nsubjpass_checked_set prep_set_of pobj_of_features prep_features_besides pobj_besides_n-grams aux_checked_should auxpass_checked_be cc_checked_and nsubj_ignore_we aux_ignore_should neg_ignore_not conj_checked_ignore det_effectiveness_the amod_effectiveness_potential dobj_ignore_effectiveness prep_effectiveness_of pobj_of_unigrams prep_unigrams_in det_task_this pobj_in_task ' +o,7583,'Following we also use the early-update strategy , where an update happens whenever the goldstandard action-sequence falls off the beam , with the rest of the sequence neglected ',Collins,'xcomp_Following_ dobj__we advmod_use_also dep_Following_use det_strategy_the amod_strategy_early-update nsubj_Following_strategy advmod_happens_where det_update_an nsubj_happens_update rcmod_strategy_happens advmod_falls_whenever det_action-sequence_the amod_action-sequence_goldstandard nsubj_falls_action-sequence advcl_happens_falls prt_falls_off det_beam_the dobj_falls_beam prep_beam_with det_rest_the pobj_with_rest prep_rest_of det_sequence_the pobj_of_sequence partmod_sequence_neglected ' +o,7584,'For example , Smith and Smith and Burkett and Klein show that joint parsing -LRB- or reranking -RRB- on a bitext improves accuracies on either or both sides by leveraging bilingual constraints , which is very promising for syntax-based machine translation which requires -LRB- good-quality -RRB- parse trees for rule extraction ',Galley,'prep_show_For pobj_For_example nn__Smith cc_Smith_and conj_Smith_Smith nsubj_show_ cc__and nn__Burkett cc_Burkett_and conj_Burkett_Klein conj__ complm_improves_that amod_parsing_joint nsubj_improves_parsing dep_reranking_or dep_parsing_reranking prep_parsing_on det_bitext_a pobj_on_bitext ccomp_show_improves dobj_improves_accuracies prep_accuracies_on amod_sides_either cc_either_or conj_either_both pobj_on_sides prep_improves_by pcomp_by_leveraging amod_constraints_bilingual dobj_leveraging_constraints nsubj_promising_which cop_promising_is advmod_promising_very rcmod_constraints_promising prep_promising_for amod_translation_syntax-based nn_translation_machine pobj_for_translation nsubj_requires_which rcmod_translation_requires appos_trees_good-quality amod_trees_parse dobj_requires_trees prep_requires_for nn_extraction_rule pobj_for_extraction ' +o,7585,'6 Related Work Other work combining supervised and unsupervised learning for parsing includes , , and ',Collins,'num_work_6 nn_work_Related nn_work_Work amod_work_Other nsubj_includes_work partmod_work_combining amod_learning_supervised cc_supervised_and conj_supervised_unsupervised dobj_combining_learning prep_combining_for pcomp_for_parsing acomp_includes_ conj__ cc__and conj__ ' +o,7586,'Although the training algorithm can handle realvalued features as used in the current paper intentionally excludes them ',Och,'mark_handle_Although det_algorithm_the nn_algorithm_training nsubj_handle_algorithm aux_handle_can amod_features_realvalued dobj_handle_features mark_excludes_as csubj_excludes_used prep_used_in pcomp_in_ det_paper_the amod_paper_current dobj__paper advmod_excludes_intentionally advcl_handle_excludes dobj_excludes_them ' +p,7587,' showed that adding a small set of prototypes to the unlabeled data can improve tagging accuracy significantly ',Haghighi,'nsubj_showed_ complm_improve_that csubj_improve_adding det_set_a amod_set_small dobj_adding_set prep_set_of pobj_of_prototypes prep_adding_to det_data_the amod_data_unlabeled pobj_to_data aux_improve_can ccomp_showed_improve amod_accuracy_tagging dobj_improve_accuracy advmod_improve_significantly ' +o,7588,' -RRB- , can be summarized as : -LRB- -RRB- C -LRB- such as including -RRB- I -LRB- and , -RRB- , where I is a potential instance -LRB- eg , Venezuelan equine encephalitis -RRB- and C is a potential class label for the instance -LRB- eg , zoonotic diseases -RRB- , for example in the sentence : The expansion of the farms increased the spread of zoonotic diseases such as Venezuelan equine encephalitis -LRB- -RRB- ',Snow,'nsubj_increased_ aux_be_can csubj_as_be dep_as_summarized dep__as prep_-RRB-_C dep_C_such prep_such_as pcomp_as_including pobj_C_I dep_I_and advmod_instance_where nsubj_instance_I cop_instance_is det_instance_a amod_instance_potential advcl_as_instance appos_instance_eg amod_encephalitis_Venezuelan nn_encephalitis_equine dep_eg_encephalitis cc_as_and nsubj_label_C cop_label_is det_label_a amod_label_potential nn_label_class conj_as_label prep_label_for det_instance_the pobj_for_instance dep_instance_eg amod_diseases_zoonotic appos_eg_diseases prep_label_for pobj_for_example prep_example_in det_sentence_the pobj_in_sentence det_expansion_The dep__expansion prep_expansion_of det_farms_the pobj_of_farms det_spread_the dobj_increased_spread prep_spread_of amod_diseases_zoonotic pobj_of_diseases dep_as_such prep_diseases_as amod_encephalitis_Venezuelan nn_encephalitis_equine pobj_as_encephalitis ' +o,7589,'2 Related Work Syntax-based translation models engaged with SCFG have been actively investigated in the literature ',Wu,'num_models_2 amod_models_Related nn_models_Work nn_models_Syntax-based nn_models_translation nsubjpass_investigated_models partmod_models_engaged prep_engaged_with pobj_with_SCFG aux_investigated_have auxpass_investigated_been advmod_investigated_actively prep_investigated_in det_literature_the pobj_in_literature ' +o,7590,'In cut-and-paste summarization , sentence combination operations were implemented manually following the study of a set of professionally written abstracts ; however the particular pasting operation presented here was not implemented ',Jing,'prep_implemented_In amod__cut-and-paste nn__summarization pobj_In_ nn_operations_sentence nn_operations_combination nsubjpass_implemented_operations auxpass_implemented_were advmod_implemented_manually prep_implemented_following det_study_the pobj_following_study prep_study_of det_set_a pobj_of_set prep_set_of amod_abstracts_professionally amod_abstracts_written pobj_of_abstracts advmod_implemented_however det_operation_the amod_operation_particular nn_operation_pasting nsubjpass_implemented_operation partmod_operation_presented advmod_implemented_here auxpass_implemented_was neg_implemented_not parataxis_implemented_implemented ' +o,7591,'In addition , the calculation cost for estimating parameters of embedded joint PMs -LRB- HMMs -RRB- is independent of the number of HMMs , J , that we used ',Suzuki,'prep_independent_In pobj_In_addition det_cost_the nn_cost_calculation nsubj_independent_cost prep_cost_for pcomp_for_estimating dobj_estimating_parameters prep_parameters_of amod_PMs_embedded amod_PMs_joint pobj_of_PMs abbrev_PMs_HMMs cop_independent_is prep_independent_of det_number_the pobj_of_number prep_number_of nn_J_HMMs pobj_of_J complm_used_that nsubj_used_we ccomp_independent_used ' +o,7592,'These estimates are usually heuristic and inconsistent ',Koehn,'det_estimates_These nsubj_heuristic_estimates cop_heuristic_are advmod_heuristic_usually cc_heuristic_and conj_heuristic_inconsistent ' +o,7593,'Apart from BLEU , a standard automatic measure METEOR was used for evaluation ',Banerjee,'dep_from_Apart pobj_from_BLEU det_measure_a amod_measure_standard amod_measure_automatic dep_from_measure nn__METEOR nsubjpass_used_ auxpass_used_was rcmod_measure_used prep_used_for pobj_for_evaluation ' +o,7594,'Parsers that attempt to disambiguate the input completely full parsing typically first employ some kind of dynamic programming algorithm to derive a packed parse forest and then applies a probabilistic top-down model in order to select the most probable analysis ',Collins,'nsubj_employ_Parsers nsubj_attempt_that rcmod_Parsers_attempt aux_disambiguate_to xcomp_attempt_disambiguate det_input_the nsubj_parsing_input advmod_full_completely dep_parsing_full xcomp_disambiguate_parsing advmod_parsing_typically advmod_parsing_first det_kind_some dobj_employ_kind prep_kind_of amod_algorithm_dynamic nn_algorithm_programming pobj_of_algorithm aux_derive_to xcomp_employ_derive det_forest_a amod_forest_packed nn_forest_parse dobj_derive_forest cc_employ_and advmod_applies_then conj_employ_applies det_model_a amod_model_probabilistic amod_model_top-down dobj_applies_model prep_applies_in pobj_in_order aux_select_to xcomp_applies_select det_analysis_the advmod_probable_most amod_analysis_probable dobj_select_analysis ' +p,7595,'The statistical machine translation community relies on the Bleu metric for the purposes of evaluating incremental system changes and optimizing systems through minimum error rate training ',Och,'det_community_The amod_community_statistical nn_community_machine nn_community_translation nsubj_relies_community prep_relies_on det_metric_the nn_metric_Bleu pobj_on_metric prep_metric_for det_purposes_the pobj_for_purposes prep_purposes_of pcomp_of_evaluating amod_changes_incremental nn_changes_system dobj_evaluating_changes cc_evaluating_and conj_evaluating_optimizing dobj_optimizing_systems prep_optimizing_through amod_training_minimum nn_training_error nn_training_rate pobj_through_training ' +o,7596,'Second , movie reviews are apparently harder to classify than reviews of other products ',Turney,'advmod_harder_Second nn_reviews_movie nsubj_harder_reviews cop_harder_are advmod_harder_apparently aux_classify_to xcomp_harder_classify prep_classify_than pobj_than_reviews prep_reviews_of amod_products_other pobj_of_products ' +o,7597,'We split the returned documents into classes encompassing n-grams -LRB- terms of word length n -RRB- , adjectives -LRB- using a part-of-speech tagger -RRB- and noun phrases -LRB- using a lexical chunker -RRB- ',Ramshaw,'nsubj_split_We det_documents_the amod_documents_returned dobj_split_documents prep_split_into nn_n-grams_classes amod_n-grams_encompassing pobj_into_n-grams dep_n-grams_terms prep_terms_of nn_n_word nn_n_length pobj_of_n conj_n-grams_adjectives dep_adjectives_using det__a amod__part-of-speech nn__tagger dobj_using_ cc_n-grams_and nn_phrases_noun conj_n-grams_phrases dep_phrases_using det__a amod__lexical nn__chunker dobj_using_ ' +p,7598,'A detailed description of the popular translation models IBM-1 to IBM-5 , aswellastheHidden-Markovalignmentmodel -LRB- HMM -RRB- can be found in ',Brown,'det_description_A amod_description_detailed nsubjpass_found_description prep_description_of det_IBM-1_the amod_IBM-1_popular nn_IBM-1_translation nn_IBM-1_models pobj_of_IBM-1 prep_description_to nn__IBM-5 pobj_to_ amod__aswellastheHidden-Markovalignmentmodel abbrev__HMM appos__ aux_found_can auxpass_found_be prep_found_in pobj_in_ ' +o,7599,'Thus we rank each sense wsi WSw using Prevalence Score wsi = -LRB- 11 -RRB- njNw dssnj wnss -LRB- wsi , nj -RRB- wsiWSw wnss -LRB- wsi , nj -RRB- where the WordNet similarity score -LRB- wnss -RRB- is defined as : wnss -LRB- wsi , nj -RRB- = max nsxNSnj -LRB- wnss -LRB- wsi , nsx -RRB- -RRB- 22 Building the Thesaurus The thesaurus was acquired using the method described by ',Lin,'advmod_rank_Thus nsubj_rank_we det_WSw_each nn_WSw_sense nn_WSw_wsi dobj_rank_WSw partmod_WSw_using nn_wsi_Prevalence nn_wsi_Score nsubj_=_wsi ccomp_using_= dep_=_11 amod_wnss_njNw dep_njNw_dssnj amod_wnss_wnss appos_wnss_wsi dep_wsi_nj nn_wnss_wsiWSw dobj_=_wnss appos_wnss_wsi dep_wsi_nj advmod_score_where det_similarity_the nn_similarity_WordNet nsubj_score_similarity advcl_using_score nsubjpass_defined_wnss auxpass_defined_is ccomp_score_defined mark_=_as nsubj_=_wnss appos_wnss_wsi dep_wsi_nj advcl_defined_= nn_nsxNSnj_max dobj_=_nsxNSnj dep_nsxNSnj_wnss appos_nsxNSnj_wsi dep_wsi_nsx dep_nsxNSnj_22 dep_nsxNSnj_Building det_Thesaurus_the dobj_Building_Thesaurus det_thesaurus_The nsubjpass_acquired_thesaurus auxpass_acquired_was dep_rank_acquired xcomp_acquired_using det_method_the dobj_using_method partmod_method_described prep_described_by ' +o,7600,'Because views the adaptation as merely augmenting the feature space , each of his features has the same prior mean and variance , regardless of whether it is domain specific or independent ','Daume III','mark_views_Because nsubj_views_ advcl_has_views det_adaptation_the dobj_views_adaptation prep_views_as advmod_augmenting_merely pcomp_as_augmenting det_space_the nn_space_feature dobj_augmenting_space nsubj_has_each prep_each_of poss_features_his pobj_of_features det_mean_the amod_mean_same amod_mean_prior dobj_has_mean cc_mean_and conj_mean_variance advmod_has_regardless dep_regardless_of complm_domain_whether nsubj_domain_it cop_domain_is pcomp_of_domain amod_domain_specific cc_specific_or conj_specific_independent ' +o,7601,'On the other hand , purely statistical systems extract discriminating MWUs from text corpora by means of association measure regularities ',Smadja,'prep__On det_hand_the amod_hand_other pobj_On_hand advmod_systems_purely amod_systems_statistical nsubj__systems dep__extract xcomp_extract_discriminating dobj_discriminating_MWUs prep_MWUs_from nn_corpora_text pobj_from_corpora prep_discriminating_by pobj_by_means prep_means_of nn_regularities_association nn_regularities_measure pobj_of_regularities ' +o,7602,'In fact , researchers in sentiment analysis have realized benefits by decomposing the problem into S\\/O and polarity classification ',Pang,'prep_realized_In pobj_In_fact nsubj_realized_researchers prep_researchers_in nn_analysis_sentiment pobj_in_analysis aux_realized_have dobj_realized_benefits prep_realized_by pcomp_by_decomposing det_problem_the dobj_decomposing_problem prep_decomposing_into nn_classification_S\\/O cc_S\\/O_and conj_S\\/O_polarity pobj_into_classification ' +o,7603,'The summaries from the above algorithm for the QF-MDS were evaluated based on ROUGE metrics ',Lin,'det_summaries_The nsubjpass_evaluated_summaries prep_summaries_from det_algorithm_the amod_algorithm_above pobj_from_algorithm prep_algorithm_for det_QF-MDS_the pobj_for_QF-MDS auxpass_evaluated_were prep_evaluated_based dep_based_on amod_metrics_ROUGE pobj_on_metrics ' +o,7604,'In the latter case , we use an unsupervised attachment disambiguation method , based on the log-likelihood ratio -LRB- LLR \' , -RRB- ',Dunning,'prep_use_In det_case_the amod_case_latter pobj_In_case nsubj_use_we det_method_an amod_method_unsupervised nn_method_attachment nn_method_disambiguation dobj_use_method prep_use_based dep_based_on det_ratio_the amod_ratio_log-likelihood pobj_on_ratio dep_ratio_LLR appos_LLR_ ' +o,7605,'To perform code generalization , Li adopted to Smadjas work and defined the code strength using a code frequency and a standard deviation in each level of the concept hierarchy ',Smadja,'aux_perform_To dep_adopted_perform nn_generalization_code dobj_perform_generalization nsubj_adopted_Li prep_adopted_to nn__Smadjas nn__work pobj_to_ cc_adopted_and conj_adopted_defined det_strength_the nn_strength_code dobj_defined_strength xcomp_defined_using det_frequency_a nn_frequency_code dobj_using_frequency cc_frequency_and det_deviation_a amod_deviation_standard conj_frequency_deviation prep_using_in det_level_each pobj_in_level prep_level_of det_hierarchy_the nn_hierarchy_concept pobj_of_hierarchy ' +o,7606,' used iterative scaling algorithms for CRF training , following earlier work on maximumentropy models for natural language ',Berger,'advmod_used_ amod_algorithms_used nn_algorithms_iterative amod_algorithms_scaling prep_algorithms_for nn_training_CRF pobj_for_training prep_training_following amod_work_earlier pobj_following_work prep_work_on amod_models_maximumentropy pobj_on_models prep_models_for amod_language_natural pobj_for_language ' +p,7607,'One of the best efforts to quantify the performance of a term-recognition system does so only for one processing stage , leaving unassessed the text-to-output performance of the system ',Smadja,'nsubj_does_One prep_One_of det_efforts_the amod_efforts_best pobj_of_efforts aux_quantify_to infmod_One_quantify det_performance_the dobj_quantify_performance prep_performance_of det__a amod__term-recognition nn__system pobj_of_ advmod_only_so advmod_does_only prep_does_for num_stage_one nn_stage_processing pobj_for_stage xcomp_does_leaving amod_performance_unassessed det_performance_the amod_performance_text-to-output dobj_leaving_performance prep_performance_of det_system_the pobj_of_system ' +o,7608,'Results This algorithm was applied to a fragment of the Canadian Hansards that has been used in a number of other studies : Church and Simard et al ',Brown,'det_algorithm_This nsubjpass_applied_algorithm auxpass_applied_was dep_Results_applied prep_applied_to det_fragment_a pobj_to_fragment prep_fragment_of det_Hansards_the amod_Hansards_Canadian pobj_of_Hansards nsubjpass_used_that aux_used_has auxpass_used_been rcmod_Hansards_used prep_used_in det_number_a pobj_in_number prep_number_of amod_studies_other pobj_of_studies nn__Church dep_Hansards_ cc__and nn__Simard cc_Simard_et conj_Simard_al conj__ ' +o,7609,'W -LRB- S , T -RRB- = summationdisplay uS , vT w -LRB- u , v -RRB- Globally optimal minimum cuts can be found in polynomial time and near-linear running time in practice , using the maximum flow algorithm ',Pang,'nsubj_summationdisplay_S dep_S_T dep_summationdisplay_= ccomp_W_summationdisplay nsubj_W_uS nn_w_vT appos_uS_w appos_w_u dep_u_v nn_cuts_Globally amod_cuts_optimal nn_cuts_minimum nsubjpass_found_cuts aux_found_can auxpass_found_be rcmod_w_found prep_found_in amod_time_polynomial pobj_in_time cc_time_and conj_time_near-linear ccomp_found_running dobj_running_time prep_time_in pobj_in_practice dep_running_using det_algorithm_the amod_algorithm_maximum nn_algorithm_flow dobj_using_algorithm ' +o,7610,'A third of the corpus is syntactically parsed as part of the Penn Treebank 2This type corresponds to Princes -LRB- 1981 ; 1992 -RRB- inferrables ',Marcus,'det_third_A nsubj_parsed_third prep_third_of det_corpus_the pobj_of_corpus cop_parsed_is advmod_parsed_syntactically mark_corresponds_as nsubj_corresponds_part prep_part_of det_type_the nn_type_Penn nn_type_Treebank nn_type_ amod_type_2This pobj_of_type advcl_parsed_corresponds aux_Princes_to xcomp_corresponds_Princes dep_inferrables_1981 dep_1981_1992 dobj_Princes_inferrables ' +o,7611,' applied two different classifiers to perform sentiment annotation in two sequential steps : the first classifier separated subjective -LRB- sentiment-laden -RRB- texts from objective -LRB- neutral -RRB- ones and then they used the second classifier to classify the subjective texts into positive and negative ',Pang,'nsubj_applied_ num_classifiers_two amod_classifiers_different dobj_applied_classifiers aux_perform_to xcomp_applied_perform nn_annotation_sentiment dobj_perform_annotation prep_perform_in num_steps_two amod_steps_sequential pobj_in_steps det_classifier_the amod_classifier_first nsubj_separated_classifier parataxis_applied_separated amod_texts_subjective appos_texts_sentiment-laden dobj_separated_texts prep_separated_from amod_ones_objective dep_ones_neutral pobj_from_ones cc_applied_and advmod_used_then nsubj_used_they conj_applied_used det_classifier_the amod_classifier_second dobj_used_classifier aux_classify_to xcomp_used_classify det_texts_the amod_texts_subjective dobj_classify_texts prep_classify_into pobj_into_positive cc_positive_and conj_positive_negative ' +o,7612,'Bayesian approaches can also improve performance ',Johnson,'nn_approaches_Bayesian nsubj_improve_approaches aux_improve_can advmod_improve_also dobj_improve_performance ' +o,7613,'For the Penn Treebank , our research and the work of others have shown that such a correspondence exists in most cases ',Cahill,'prep_shown_For det_Treebank_the nn_Treebank_Penn pobj_For_Treebank poss_research_our nsubj_shown_research cc_research_and det_work_the conj_research_work prep_work_of nn__others pobj_of_ aux_shown_have complm_exists_that predet_correspondence_such det_correspondence_a nsubj_exists_correspondence ccomp_shown_exists prep_exists_in amod_cases_most pobj_in_cases ' +o,7614,'It is worth noting , however , that even in the choice of seed words is explicitly motivated by domain properties of movie reviews ',Turney,'nsubj_worth_It cop_worth_is xcomp_worth_noting advmod_noting_however complm_motivated_that dep_in_even advcl_motivated_in pcomp_in_ det_choice_the dobj__choice prep_choice_of pobj_of_seed nsubjpass_motivated_words auxpass_motivated_is advmod_motivated_explicitly ccomp_noting_motivated prep_motivated_by nn_properties_domain pobj_by_properties prep_properties_of nn_reviews_movie pobj_of_reviews ' +o,7615,'This might prove beneficial for various discriminative training methods ',Tillmann,'nsubj_prove_This aux_prove_might acomp_prove_beneficial prep_prove_for amod_methods_various amod_methods_discriminative nn_methods_training pobj_for_methods ' +o,7616,'The feature weights are learned by maximizing the BLEU score on held-out data,usingminimum-error-ratetraining as implemented by Koehn ',Papineni,'det_weights_The nn_weights_feature nsubjpass_learned_weights auxpass_learned_are prep_learned_by pcomp_by_maximizing det__the amod__BLEU nn__score dobj_maximizing_ prep_maximizing_on amod__held-out amod__data,usingminimum-error-ratetraining pobj_on_ mark_implemented_as advcl_maximizing_implemented prep_implemented_by pobj_by_Koehn ' +o,7617,'Most importantly , whereas the one-sense-per-discourse assumption also applies to discriminating images , there is no guarantee of a local collocational or co-occurrence context around the target image ',Yarowsky,'advmod_importantly_Most advmod_is_importantly mark_applies_whereas det__the amod__one-sense-per-discourse nn__assumption nsubj_applies_ advmod_applies_also dep_is_applies prep_applies_to pcomp_to_discriminating dobj_discriminating_images advmod_is_there det_guarantee_no nsubj_is_guarantee prep_guarantee_of det_context_a amod_context_local amod_context_collocational cc_collocational_or conj_collocational_co-occurrence pobj_of_context prep_context_around det_image_the nn_image_target pobj_around_image ' +o,7618,'Probabilistic models where probabilities are assigned to the CFG backbone of the unification-based grammar have been developed , and the most probable parse is found by PCFG parsing ',Marcus,'amod_models_Probabilistic nsubjpass_developed_models advmod_assigned_where nsubjpass_assigned_probabilities auxpass_assigned_are rcmod_models_assigned prep_assigned_to det_backbone_the nn_backbone_CFG pobj_to_backbone prep_backbone_of det_grammar_the amod_grammar_unification-based pobj_of_grammar aux_developed_have auxpass_developed_been dep_developed_ cc_developed_and det_parse_the advmod_probable_most amod_parse_probable nsubjpass_found_parse auxpass_found_is conj_developed_found prep_found_by pcomp_by_PCFG dobj_PCFG_parsing ' +o,7619,'To compare the performance of different taggers learned by different mechanisms, one can measure the precision, recall and F-measure, given by precision = # correct predictions# predicted gene mentions recall = # correct predictions# true gene mentions F-measure = a96a15a14 precision a14 recallprecision a44 recall In our evaluation, we compared the proposed semi-supervised learning approach to the state of the art supervised CRF of McDonald and Pereira (2005), and also to self-training (Celeux and Govaert 1992; Yarowsky 1995), using the same feature set as (McDonald and Pereira 2005).',Yarowsky,'' +o,7620,' present an incremental perceptron algorithm for parsing that uses early update to update the parameters when an error is encountered ',Collins,'advmod_algorithm_ dep_algorithm_present det_algorithm_an amod_algorithm_incremental nn_algorithm_perceptron prep_algorithm_for pcomp_for_parsing complm_uses_that ccomp_parsing_uses amod_update_early dobj_uses_update aux_update_to infmod_update_update det_parameters_the dobj_update_parameters advmod_encountered_when det_error_an nsubjpass_encountered_error auxpass_encountered_is advcl_algorithm_encountered ' +o,7621,'33 Perceptron learning of feature weights As we saw above , our model is a linear model with the global weight vector w acting as the coefficient vector , and hence various existing techniques can be exploited to optimize w In this paper , we use the averaged perceptron learning to optimize w on a training corpus , so that the system assigns the highest score to the correct coordination tree among all possible trees for each training sentence ',Collins,'nsubj_learning_Perceptron dep_33_learning prep_learning_of nn_weights_feature pobj_of_weights mark_saw_As nsubj_saw_we advcl_model_saw advmod_saw_above poss_model_our nsubj_model_model cop_model_is det_model_a amod_model_linear dep_33_model prep_model_with det_w_the amod_w_global nn_w_weight nn_w_vector pobj_with_w partmod_w_acting prep_acting_as det_vector_the amod_vector_coefficient pobj_as_vector cc_model_and advmod_techniques_hence amod_techniques_various amod_techniques_existing nsubjpass_exploited_techniques aux_exploited_can auxpass_exploited_be conj_model_exploited aux_optimize_to purpcl_exploited_optimize dobj_optimize_w prep_use_In det_paper_this pobj_In_paper nsubj_use_we dep_optimize_use det_perceptron_the amod_perceptron_averaged dobj_use_perceptron partmod_perceptron_learning advmod_learning_ aux_optimize_to xcomp_learning_optimize dobj_optimize_w prep_optimize_on det_corpus_a nn_corpus_training pobj_on_corpus cc_use_so dep_assigns_that det_system_the nsubj_assigns_system conj_use_assigns det_score_the amod_score_highest dobj_assigns_score prep_assigns_to det_tree_the amod_tree_correct nn_tree_coordination pobj_to_tree prep_tree_among det_trees_all amod_trees_possible pobj_among_trees prep_trees_for det_sentence_each nn_sentence_training pobj_for_sentence ' +o,7622,'The input is POS-tagged using the tagger of ',Ratnaparkhi,'det_input_The nsubj_POS-tagged_input cop_POS-tagged_is xcomp_POS-tagged_using det_tagger_the dobj_using_tagger prep_tagger_of ' +o,7623,'Our strategy for choosing heads is similar to the one in ',Collins,'poss_strategy_Our nsubj_similar_strategy prep_strategy_for nn_heads_choosing pobj_for_heads cop_similar_is prep_similar_to det_one_the pobj_to_one prep_similar_in ' +o,7624,'The relationship between Kneser-Ney smoothing to the Bayesian approach have been explored in using Pitman-Yor processes ',Teh,'det_relationship_The nsubjpass_explored_relationship prep_relationship_between pobj_between_Kneser-Ney partmod_Kneser-Ney_smoothing prep_smoothing_to det_approach_the nn_approach_Bayesian pobj_to_approach aux_explored_have auxpass_explored_been prep_explored_in pobj_in_ xcomp_explored_using amod_processes_Pitman-Yor dobj_using_processes ' +o,7625,'In particular we work with dependency paths that can reach beyond direct dependencies as opposed to but in the line of Pado and Lapata -LRB- 2007 -RRB- ',Lin,'prep_work_In pobj_In_particular nsubj_work_we prep_work_with amod_paths_dependency pobj_with_paths nsubj_reach_that aux_reach_can rcmod_paths_reach prep_reach_beyond amod_dependencies_direct pobj_beyond_dependencies advmod_opposed_as amod_dependencies_opposed aux__to xcomp_reach_ cc_in_but prep__in det_line_the pobj_in_line prep_line_of pobj_of_Pado cc_Pado_and conj_Pado_Lapata appos_paths_2007 ' +o,7626,'64 Feature Selection Methods A number of previous papers describe feature selection approaches for log-linear models applied to NLP problems ',Berger,'num_Methods_64 nn_Methods_Feature nn_Methods_Selection det_number_A nsubj_describe_number prep_number_of amod_papers_previous pobj_of_papers advmod_describe_ dep_Methods_describe nn_approaches_feature nn_approaches_selection dobj_describe_approaches prep_describe_for amod_models_log-linear pobj_for_models partmod_models_applied prep_applied_to nn_problems_NLP pobj_to_problems ' +o,7627,'Under certain precise conditions , as described in , we can analyze Algorithm 1 as minimizing the entropy of the distribution over translations of U However , this is true only when the functions Estimate , Score and Select have very prescribed definitions ',Abney,'prep_true_Under amod_conditions_certain amod_conditions_precise pobj_Under_conditions mark_analyze_as ccomp_analyze_described prep_described_in pobj_in_ nsubj_analyze_we aux_analyze_can dep_true_analyze dobj_analyze_Algorithm num_Algorithm_1 prep_analyze_as pcomp_as_minimizing det_entropy_the dobj_minimizing_entropy prep_entropy_of det_distribution_the pobj_of_distribution prep_distribution_over pobj_over_translations prep_translations_of pobj_of_U advmod_minimizing_However nsubj_true_this cop_true_is advmod_true_only advmod_have_when det_Estimate_the nn_Estimate_functions nsubj_have_Estimate conj_Estimate_Score cc_Estimate_and conj_Estimate_Select advcl_true_have advmod_definitions_very amod_definitions_prescribed dobj_have_definitions ' +o,7628,'Applications of word clustering include language modeling , text classification , thesaurus construction and so on ',Lin,'nsubj_include_Applications prep_Applications_of nn_clustering_word pobj_of_clustering nn__language nn__modeling dobj_include_ nn__text nn__classification conj__ nn__thesaurus nn__construction conj__ cc__and advmod_on_so conj__on ' +o,7629,'It is a variant of the batch-based Bloomier filter LM of which we refer to as the TB-LM henceforth ',Talbot,'nsubj_variant_It cop_variant_is det_variant_a prep_variant_of det_LM_the amod_LM_batch-based nn_LM_Bloomier nn_LM_filter pobj_of_LM prep_LM_of pobj_of_ rel_refer_which nsubj_refer_we rcmod_variant_refer aux_henceforth_to advmod_henceforth_as det_TB-LM_the dep_as_TB-LM xcomp_refer_henceforth ' +p,7630,'More suitable ways could be bilingual chunk parsing , and refining the bracketing grammar as described in ',Wu,'advmod_suitable_More amod_ways_suitable nsubj_chunk_ways aux_chunk_could cop_chunk_be amod_chunk_bilingual dep_chunk_parsing cc_parsing_and conj_parsing_refining det_grammar_the amod_grammar_bracketing dobj_refining_grammar advmod_described_as amod_grammar_described prep_refining_in ' +o,7631,'Statistical dependency parsers of English must therefore rely on dependency structures automatically converted from a constituent corpus such as the Penn Treebank ',Marcus,'nn_parsers_Statistical nn_parsers_dependency nsubj_rely_parsers prep_parsers_of pobj_of_English aux_rely_must advmod_rely_therefore prep_rely_on amod_structures_dependency pobj_on_structures advmod_converted_automatically partmod_structures_converted prep_converted_from det_corpus_a amod_corpus_constituent pobj_from_corpus dep_as_such prep_corpus_as det_Treebank_the nn_Treebank_Penn pobj_as_Treebank ' +o,7632,'Also , attribute classi cation is a hard problem and there is no existing classi cation scheme that can be used for open domains like newswire ; for example , WordNet organises adjectives as concepts that are related by the non-hierarchical relations of synonymy and antonymy -LRB- unlike nouns that are related through hierarchical links such as hyponymy , hypernymy and metonymy -RRB- ',Marcus,'advmod_problem_Also nn_cation_attribute nn_cation_classi nsubj_problem_cation cop_problem_is det_problem_a amod_problem_hard cc_problem_and expl_is_there conj_problem_is det_scheme_no amod_scheme_existing amod_scheme_classi nn_scheme_cation nsubj_is_scheme nsubjpass_used_that aux_used_can auxpass_used_be rcmod_scheme_used prep_used_for amod_domains_open pobj_for_domains prep_domains_like pobj_like_newswire prep_adjectives_for pobj_for_example nn_organises_WordNet nn_organises_ nsubj_adjectives_organises parataxis_problem_adjectives prep_adjectives_as pobj_as_concepts nsubjpass_related_that auxpass_related_are rcmod_concepts_related prep_related_by det_relations_the amod_relations_non-hierarchical pobj_by_relations prep_relations_of pobj_of_synonymy cc_synonymy_and conj_synonymy_antonymy dep_adjectives_unlike pobj_unlike_nouns nsubjpass_related_that auxpass_related_are rcmod_nouns_related prep_related_through amod_links_hierarchical pobj_through_links dep_as_such prep_links_as pobj_as_hyponymy conj_hyponymy_hypernymy cc_hyponymy_and conj_hyponymy_metonymy ' +o,7633,'43 Baseline We use a standard log-linear phrase-based statistical machine translation system as a baseline : GIZA + + implementation of IBM word alignment model 4 ,8 the refinement and phrase-extraction heuristics described in , minimum-error-rate training 7More specifically , we choose the first English reference from the 7 references and the Chinese sentence to construct new sentence pairs ',Och,'num_Baseline_43 nsubj_use_We dep_Baseline_use det_system_a amod_system_standard amod_system_log-linear amod_system_phrase-based amod_system_statistical nn_system_machine nn_system_translation dobj_use_system prep_system_as det_baseline_a pobj_as_baseline amod_implementation_GIZA cc_GIZA_+ conj_GIZA_+ dep_system_implementation prep_implementation_of nn_model_IBM nn_model_word nn_model_alignment pobj_of_model number__4 num_,8_ prep_model_,8 det_refinement_the dep_,8_refinement cc_refinement_and amod_heuristics_phrase-extraction conj_refinement_heuristics partmod_heuristics_described prep_described_in amod_7More_ amod_7More_minimum-error-rate nn_7More_training pobj_in_7More advmod_choose_specifically nsubj_choose_we dep_Baseline_choose det_reference_the amod_reference_first amod_reference_English dobj_choose_reference prep_choose_from det_references_the num_references_7 pobj_from_references cc_references_and det_sentence_the amod_sentence_Chinese conj_references_sentence aux_construct_to xcomp_choose_construct amod_pairs_new nn_pairs_sentence dobj_construct_pairs ' +o,7634,'Features that consider only target-side syntax and words without considering s can be seen as syntactic language model features ',Shen,'nsubjpass_seen_Features nsubj_consider_that rcmod_Features_consider advmod_syntax_only amod_syntax_target-side dobj_consider_syntax cc_syntax_and conj_syntax_words prep_consider_without pcomp_without_considering dobj_considering_s aux_seen_can auxpass_seen_be prep_seen_as amod_features_syntactic nn_features_language nn_features_model pobj_as_features ' +o,7635,'As a follow-up to the work described in this paper we developed a method that utilizes the unlabeled NPs in the corpus using a structured rule learner ',Stoyanov,'det_learner_a amod_learner_follow-up prep_follow-up_to det_work_the pobj_to_work partmod_work_described prep_described_in det_paper_this pobj_in_paper nsubj_developed_we rcmod_paper_developed det_method_a dobj_developed_method nsubj_utilizes_that rcmod_method_utilizes det_NPs_the amod_NPs_unlabeled dobj_utilizes_NPs prep_NPs_in det_corpus_the pobj_in_corpus partmod_corpus_using det_rule_a amod_rule_structured dobj_using_rule pobj_As_learner ' +o,7636,'Logics for the IBM Models would be similar to our logics for phrase-based models ',Brown,'nsubj_similar_Logics prep_Logics_for det__the nn__IBM nn__Models pobj_for_ aux_similar_would cop_similar_be prep_similar_to poss_logics_our pobj_to_logics prep_logics_for amod_models_phrase-based pobj_for_models ' +p,7637,'-LRB- 2 -RRB- We note that these posterior probabilities can be computed efficiently for some alignment models such as the HMM , Models 1 and 2 ',Och,'dep_note_2 nsubj_note_We complm_computed_that det_probabilities_these nn_probabilities_posterior nsubjpass_computed_probabilities aux_computed_can auxpass_computed_be ccomp_note_computed advmod_computed_efficiently prep_computed_for det_alignment_some pobj_for_alignment nsubj__models dep_as_such prep_models_as det__the nn__HMM pobj_as_ appos__Models num_Models_1 cc_Models_and conj_Models_2 dep_note_ ' +o,7638,'Table 1 shows theresultsalongwithB andthethreemetricsthat achieved higher correlations than B : semantic role overlap , ParaEval recall , and METEOR ',Banerjee,'number_1_Table num_shows_1 nsubj_achieved_shows amod_shows_theresultsalongwithB dep_theresultsalongwithB_andthethreemetricsthat amod_correlations_higher dobj_achieved_correlations prep_achieved_than pobj_than_B amod_role_semantic nsubj_overlap_role parataxis_achieved_overlap dobj_overlap_ nn__ParaEval nn__recall conj__ cc__and conj__METEOR ' +p,7639,'Widely used alignment models , such as IBM Model serial and HMM , all assume one-to-many alignments ',Brown,'advmod_used_Widely amod_models_alignment nsubj_used_models dep_as_such dep_models_as nn__IBM nn__Model nn__serial pobj_as_ cc__and conj__HMM nsubj_assume_all rcmod_models_assume amod_alignments_one-to-many dobj_assume_alignments ' +n,7640,'The most commonly used metric , BLEU , correlates well over large test sets with human judgments , but does not perform as well on sentence-level evaluation ',Papineni,'det_used_The advmod_used_most advmod_used_commonly nsubj_correlates_used amod_used_metric appos_used_BLEU advmod_correlates_well prep_correlates_over amod_sets_large nn_sets_test pobj_over_sets prep_sets_with amod__human nn__judgments pobj_with_ cc_correlates_but aux_perform_does neg_perform_not conj_correlates_perform advmod_well_as advmod_perform_well prep_perform_on amod__sentence-level nn__evaluation pobj_on_ ' +o,7641,'Whereas used feature support cutoffs and early stopping to stop overfitting of the model , and Collins contends that including low support features harms a maximum entropy model , our results show that low support features are useful in a regularized maximum entropy model ',Ratnaparkhi,'mark_used_Whereas nsubj_used_ advcl_show_used nn_cutoffs_feature nn_cutoffs_support nsubj_stop_cutoffs cc_cutoffs_and amod_stopping_early conj_cutoffs_stopping aux_stop_to xcomp_used_stop dobj_stop_overfitting prep_overfitting_of det_model_the pobj_of_model cc_used_and nn__Collins nsubj_contends_ conj_used_contends complm_harms_that prep_harms_including amod_support_low pobj_including_support nsubj_harms_features ccomp_contends_harms det_model_a amod_model_maximum nn_model_entropy dobj_harms_model poss_results_our nsubj_show_results complm_useful_that amod_features_low nn_features_support nsubj_useful_features cop_useful_are ccomp_show_useful prep_useful_in det_model_a amod_model_regularized amod_model_maximum nn_model_entropy pobj_in_model ' +p,7642,'This approach has been shown to be accurate , relatively efficient , and robust using both generative and discriminative models ',Collins,'det_approach_This nsubjpass_shown_approach aux_shown_has auxpass_shown_been aux_accurate_to cop_accurate_be xcomp_shown_accurate advmod_efficient_relatively conj_accurate_efficient cc_accurate_and conj_accurate_robust xcomp_accurate_using preconj_generative_both amod_models_generative cc_generative_and conj_generative_discriminative dobj_using_models ' +o,7643,'They are part of an effort to better integrate a linguistic , rule-based system and the statistical correcting layer also illustrated in ',Ueffing,'nsubj_part_They cop_part_are prep_part_of det_effort_an pobj_of_effort aux_integrate_to advmod_integrate_better infmod_effort_integrate det_system_a amod_system_linguistic amod_system_rule-based dobj_integrate_system cc_system_and det_layer_the amod_layer_statistical amod_layer_correcting conj_system_layer advmod_illustrated_also partmod_layer_illustrated prep_illustrated_in ' +o,7644,'1PMI is subject to overestimation for low frequency items , thus we require a minimum frequency of occurrence for the expressions under study ',Dunning,'nsubj_subject_1PMI cop_subject_is aux_overestimation_to xcomp_subject_overestimation prep_overestimation_for amod_items_low nn_items_frequency pobj_for_items nsubj_require_ advmod_require_thus nsubj_require_we dep_subject_require det_frequency_a amod_frequency_minimum dobj_require_frequency prep_frequency_of pobj_of_occurrence prep_occurrence_for det_expressions_the pobj_for_expressions prep_require_under pobj_under_study ' +o,7645,'Translation results are given in terms of the automaticBLEUevaluation metric as well as the TER metric ',Papineni,'nn_results_Translation nsubjpass_given_results auxpass_given_are prep_given_in pobj_in_terms prep_terms_of det__the amod__automaticBLEUevaluation amod__metric pobj_of_ dep_well_as cc__well dep_well_as det__the nn__TER amod__metric conj__ ' +o,7646,' -RRB- , concordancing for bilingual lexicography , computerassisted language learning , corpus linguistics -LRB- Melby ',Dunning,'dep__concordancing prep_concordancing_for pobj_for_bilingual nn__lexicography dobj__ amod_learning_computerassisted nn_learning_language appos__learning advmod_linguistics_corpus dep__linguistics dobj_linguistics_Melby ' +o,7647,'Factored models are introduced in for better integration of morphosyntactic information ',Koehn,'amod_models_Factored nsubjpass_introduced_models auxpass_introduced_are prep_introduced_in pobj_in_ prep_introduced_for amod_integration_better pobj_for_integration prep_integration_of amod_information_morphosyntactic pobj_of_information ' +o,7648,'The cohesion between two words is measured as in by an estimation of the mutual information based on their collocation frequency ',Church,'det_cohesion_The nsubjpass_measured_cohesion prep_cohesion_between num_words_two pobj_between_words auxpass_measured_is dep_in_as prep_measured_in pcomp_in_ prep__by det_estimation_an pobj_by_estimation prep_estimation_of det_information_the amod_information_mutual pobj_of_information prep__based dep_based_on poss_frequency_their nn_frequency_collocation pobj_on_frequency ' +o,7649,'More recently , however , showed that a 1 Conditional maximum entropy models provide somewhat of a counter-example , but there , too , many kinds of global and non-local features are difficult to use ',Okanohara,'advmod_recently_More advmod_showed_recently advmod_showed_however nsubj_showed_ complm_provide_that det__a num__1 nn__Conditional amod__maximum nn__entropy nn__models nsubj_provide_ ccomp_showed_provide advmod_provide_somewhat prep_provide_of det_counter-example_a pobj_of_counter-example cc_provide_but expl_are_there advmod_are_too amod_kinds_many nsubj_are_kinds prep_kinds_of amod_features_global cc_global_and conj_global_non-local pobj_of_features conj_provide_are acomp_are_difficult aux_use_to xcomp_difficult_use dobj_use_ ' +o,7650,' , et al ',Rosti,'appos__ cc__et conj__al ' +o,7651,'Our baseline model follows Chiangs hierarchical model in conjunction with additional features : conditional probabilities in both directions : P -LRB- -RRB- and P -LRB- -RRB- ; lexical weights in both directions : Pw -LRB- -RRB- and Pw -LRB- -RRB- ; 21 word counts e ; rule counts D ; target n-gram language model PLM -LRB- e -RRB- ; glue rule penalty to learn preference of nonterminal rewriting over serial combination through Eq ',Koehn,'poss_model_Our nn_model_baseline nsubj_follows_model nn__Chiangs amod__hierarchical nn__model dobj_follows_ prep__in pobj_in_conjunction prep_conjunction_with amod_features_additional pobj_with_features amod_probabilities_conditional dep__probabilities prep_probabilities_in det_directions_both pobj_in_directions dep__P cc_-RRB-_and conj_-RRB-_P amod__lexical nn__weights dep__ prep__in det_directions_both pobj_in_directions dep__Pw cc_-RRB-_and conj_-RRB-_Pw num_counts_21 nn_counts_word dep__counts dep_counts_e nn_counts_rule dep__counts dep_counts_D dep__target amod_PLM_n-gram nn_PLM_language nn_PLM_model dep_target_PLM dep_-LRB-_e nn_penalty_glue nn_penalty_rule dep__penalty aux_learn_to infmod_penalty_learn dobj_learn_preference prep_preference_of amod_rewriting_nonterminal pobj_of_rewriting prep_learn_over amod_combination_serial pobj_over_combination prep_learn_through pobj_through_Eq ' +o,7652,'3 Data Sets We used three opinion-related data sets for our analyses and experiments : the OP data set created by , the Polarity data set5 created by , and the MPQA data set created by 6 The OP and Polarity data sets involve document-level opinion classi cation , while the MPQA data set involves 5Version v20 , which is available at : http://wwwcscornelledu/people/pabo/movie-review-data/ 6Available at http://wwwcspittedu/mpqa/databaserelease/ sentence-level classi cation ',Pang,'num_Sets_3 nn_Sets_Data nsubj_used_We dep_Sets_used num_sets_three amod_sets_opinion-related nn_sets_data dobj_used_sets prep_sets_for poss_analyses_our pobj_for_analyses cc_analyses_and conj_analyses_experiments det_set_the nn_set_OP nn_set_data dep_sets_set partmod_set_created prep_created_by pobj_by_ det_set5_the nn_set5_Polarity nn_set5_data conj_set_set5 partmod_set5_created prep_created_by pobj_by_ cc_set_and det_set_the nn_set_MPQA nn_set_data conj_set_set partmod_set_created prep_created_by pobj_by_ num__6 det_data_The nn_data_OP cc_OP_and conj_OP_Polarity nsubj_sets_data dep_Sets_sets ccomp_sets_involve amod_classi_document-level nn_classi_opinion nsubj_cation_classi xcomp_involve_cation mark_involves_while det_set_the nn_set_MPQA nn_set_data nsubj_involves_set advcl_sets_involves amod_v20_5Version dobj_involves_v20 nsubj_available_which cop_available_is rcmod_v20_available prep_available_at amod_6Available_http://wwwcscornelledu/people/pabo/movie-review-data/ dep_v20_6Available prep_6Available_at amod_cation_http://wwwcspittedu/mpqa/databaserelease/ amod_cation_sentence-level nn_cation_classi pobj_at_cation ' +o,7653,'We use the Europarl corpus , and the statistical word alignment was performed with the GIZA + + toolkit 1 For the current experiments we assume no preexisting parser for any of the languages , contrary to the information projection scenario ',Koehn,'nsubj_use_We det__the nn__Europarl nn__corpus dobj_use_ cc_use_and det_alignment_the amod_alignment_statistical nn_alignment_word nsubjpass_performed_alignment auxpass_performed_was conj_use_performed prep_performed_with det_toolkit_the amod_toolkit_GIZA cc_GIZA_+ conj_GIZA_+ pobj_with_toolkit amod_toolkit_ tmod__1 prep_assume_For det_experiments_the amod_experiments_current pobj_For_experiments nsubj_assume_we dep_use_assume det_parser_no amod_parser_preexisting dobj_assume_parser prep_assume_for pobj_for_any prep_any_of det_languages_the pobj_of_languages dep_assume_contrary prep_contrary_to det_scenario_the nn_scenario_information nn_scenario_projection pobj_to_scenario ' +n,7654,'We will show that some achieve significantly better results than the standard minimum error rate training of ',Och,'nsubj_show_We aux_show_will complm_achieve_that nsubj_achieve_some ccomp_show_achieve advmod_better_significantly amod_results_better dobj_achieve_results prep_achieve_than det_training_the amod_training_standard amod_training_minimum nn_training_error nn_training_rate pobj_than_training prep_training_of ' +o,7655,'Computing the phrase translation probability is trivial in the training corpora , but lexical weighting needs lexical-level alignment ',Koehn,'det_probability_the nn_probability_phrase nn_probability_translation nsubj_trivial_probability cop_trivial_is dep_Computing_trivial prep_trivial_in det_corpora_the nn_corpora_training pobj_in_corpora cc_trivial_but amod__lexical nn__weighting nsubj_needs_ conj_trivial_needs amod_alignment_lexical-level dobj_needs_alignment ' +o,7656,'The first work on SMT done at IBM , used a noisy-channel model , resulting in what Brown et al ',Brown,'det_work_The amod_work_first dep_used_work prep_used_on pobj_on_SMT partmod_SMT_done prep_done_at nn__IBM pobj_at_ det_model_a amod_model_noisy-channel dobj_used_model partmod_model_resulting prep_resulting_in pobj_in_what nn_al_Brown nn_al_et nsubj_used_al ' +o,7657,'For each candidate triple , the log-likelihood and salience scores were calculated ',Dunning,'prep_calculated_For det_candidate_each pobj_For_candidate advmod_candidate_triple det__the amod__log-likelihood nsubjpass_calculated_ cc__and amod_scores_salience amod_scores_ conj__scores auxpass_calculated_were ' +o,7658,'Other works based on this scheme like have shown promising results ',Marcus,'amod_works_Other nsubj_shown_works prep_works_based dep_based_on det_scheme_this pobj_on_scheme prep_scheme_like pobj_like_ aux_shown_have xcomp_shown_promising dobj_promising_results ' +o,7659,'Brown et al proposed a class-based n-gram model , which generalizes the n-gram model , to predict a word from previous words in a text ',Brown,'nsubj_proposed_Brown cc_Brown_et conj_Brown_al det_model_a amod_model_class-based amod_model_n-gram dobj_proposed_model nsubj_generalizes_which rcmod_model_generalizes det_model_the amod_model_n-gram dobj_generalizes_model aux_predict_to xcomp_proposed_predict det_word_a dobj_predict_word prep_predict_from amod_words_previous pobj_from_words prep_words_in det_text_a pobj_in_text ' +o,7660,'TheauthorsapplySO-PMI-IR to extract and determine the polarity of adjectives ',Turney,'nn__TheauthorsapplySO-PMI-IR nsubj_extract_ aux_extract_to cc_extract_and conj_extract_determine det_polarity_the dobj_extract_polarity prep_polarity_of pobj_of_adjectives ' +o,7661,'et al , 2007 -RRB- -RRB- and unigrams -LRB- used by many researchers , eg , -RRB- ',Pang,'nn_al_et appos_al_2007 cc_al_and conj_al_unigrams dep_al_used prep_used_by amod_researchers_many pobj_by_researchers appos_researchers_eg dep_researchers_ ' +o,7662,'Automatic subjectivity analysis would also be useful to perform flame recognition , e-mail classification , intellectual attribution in text , recognition of speaker role in radio broadcasts , review mining , review classification , style in generation , and clustering documents by ideological point of view ',Pang,'nn_analysis_Automatic nn_analysis_subjectivity nsubj_useful_analysis aux_useful_would advmod_useful_also cop_useful_be aux_perform_to xcomp_useful_perform nn__flame nn__recognition nsubj_review_ amod__e-mail nn__classification appos__ amod_attribution_intellectual appos__attribution prep_attribution_in nn__text pobj_in_ appos__recognition prep_recognition_of nn_role_speaker pobj_of_role prep_role_in nn__radio nn__broadcasts pobj_in_ nn__review nn__mining appos__ ccomp_perform_review nn__classification dobj_review_ conj__style prep_style_in nn__generation pobj_in_ cc__and amod_documents_clustering conj__documents prep_documents_by amod_point_ideological pobj_by_point prep_documents_of nn__view pobj_of_ ' +o,7663,'For a sequential learning algorithm , we make use of the Collins Perceptron Learner ',Collins,'prep_make_For det_algorithm_a amod_algorithm_sequential nn_algorithm_learning pobj_For_algorithm nsubj_make_we dobj_make_use prep_use_of det_Learner_the nn_Learner_Collins nn_Learner_Perceptron pobj_of_Learner ' +o,7664,'determining document orientation -LRB- or polarity -RRB- , as in deciding if a given Subjective text expresses a Positive or a Negative opinion on its subject matter ; 3 ',Turney,'nn_orientation_document dobj_determining_orientation dep_polarity_or dep_orientation_polarity dep_in_as prep_determining_in pcomp_in_deciding mark_expresses_if det_text_a amod_text_given nn_text_Subjective nsubj_expresses_text advcl_deciding_expresses dep_Positive_a dobj_expresses_Positive cc_Positive_or det_opinion_a amod_opinion_Negative conj_Positive_opinion prep_opinion_on poss__its nn__subject nn__matter pobj_on_ dep_Positive_3 ' +o,7665,'We used pointwise mutual information to obtain these distances ',Church,'nsubj_used_We amod__pointwise amod__mutual nn__information dobj_used_ aux_obtain_to xcomp_used_obtain det_distances_these dobj_obtain_distances ' +o,7666,'Monte Carlo sampling methods and Variational Bayes are two kinds of approximate inference methods that have been applied to Bayesian inference of unsupervised HMM POS taggers ',Johnson,'nn_Carlo_Monte dep_Carlo_sampling dobj_sampling_methods cc_sampling_and nn_Bayes_Variational nsubj_kinds_Bayes cop_kinds_are num_kinds_two conj_sampling_kinds prep_kinds_of amod_methods_approximate nn_methods_inference pobj_of_methods nsubjpass_applied_that aux_applied_have auxpass_applied_been rcmod_kinds_applied prep_applied_to amod_inference_Bayesian pobj_to_inference prep_inference_of amod_taggers_unsupervised nn_taggers_HMM dep_taggers_POS pobj_of_taggers ' +o,7667,'In addition to reducing the original sentences , use a number of manually compiled rules to aggregate reduced sentences ; for example , reduced clauses might be conjoined with and ',Jing,'prep_use_In pobj_In_addition prep_addition_to pcomp_to_reducing det_sentences_the amod_sentences_original dobj_reducing_sentences nsubj_use_ det_number_a dobj_use_number prep_number_of dep_of_manually partmod_number_compiled dobj_compiled_rules prep_compiled_to amod_sentences_aggregate amod_sentences_reduced pobj_to_sentences prep_conjoined_for pobj_for_example amod_clauses_reduced nsubjpass_conjoined_clauses aux_conjoined_might auxpass_conjoined_be parataxis_use_conjoined prep_conjoined_with pobj_with_and ' +o,7668,'In fact , we still have a question as to whether SS-CRF-MER is really scalable in practical time for such a large amount of unlabeled data as used in our experiments , which is about 680 times larger than that of ',Jiao,'prep_have_In pobj_In_fact nsubj_have_we advmod_have_still det_question_a dobj_have_question prep_question_as dep_as_to complm_scalable_whether nsubj_scalable_SS-CRF-MER cop_scalable_is advmod_scalable_really pcomp_as_scalable prep_scalable_in amod_time_practical pobj_in_time prep_time_for predet_amount_such det_amount_a amod_amount_large pobj_for_amount prep_amount_of amod_data_unlabeled pobj_of_data mark_used_as advcl_scalable_used prep_used_in poss_experiments_our pobj_in_experiments nsubj_larger_which cop_larger_is quantmod_times_about number_times_680 num_larger_times rcmod_experiments_larger dep_of_than dep_of_that dep_larger_of ' +o,7669,'The second one needs no labeled data for the new domain ',Blitzer,'det_needs_The amod_needs_second num_needs_one det_data_no amod_data_labeled dep_needs_data prep_data_for det_domain_the amod_domain_new pobj_for_domain ' +p,7670,'3 The Syntactic and Semantic Parser Architecture To achieve the complex task of joint syntactic and semantic parsing , we extend a current state-of-theart statistical parser to learn semantic role annotation as well as syntactic structure ',Titov,'det_Parser_The nn_Parser_Syntactic cc_Syntactic_and conj_Syntactic_Semantic nsubj_Architecture_Parser ccomp_extend_Architecture aux_achieve_To xcomp_Architecture_achieve det_task_the amod_task_complex dobj_achieve_task prep_task_of amod_parsing_joint amod_parsing_syntactic cc_syntactic_and conj_syntactic_semantic pobj_of_parsing nsubj_extend_we dep_3_extend det__a amod__current amod__state-of-theart amod__statistical nn__parser nsubj_learn_ aux_learn_to xcomp_extend_learn amod_annotation_semantic nn_annotation_role dobj_learn_annotation dep_well_as cc_annotation_well dep_well_as amod_structure_syntactic conj_annotation_structure ' +o,7671,' proposed a distributional method in detecting non-anaphoric pronouns by first extracting the surrounding textual context of the pronoun , then gathering the distribution of words that occurred within that context from a large corpus and finally learning to classify these distributions as representing either anaphoric and non-anaphoric pronoun instances ',Bergsma,'nsubj_proposed_ det_method_a amod_method_distributional dobj_proposed_method prep_proposed_in pcomp_in_detecting amod_pronouns_non-anaphoric dobj_detecting_pronouns prep_detecting_by advmod_extracting_first pcomp_by_extracting det_context_the amod_context_surrounding amod_context_textual dobj_extracting_context prep_context_of det_pronoun_the pobj_of_pronoun advmod_gathering_then conj_detecting_gathering det_distribution_the dobj_gathering_distribution prep_distribution_of pobj_of_words nsubj_occurred_that rcmod_words_occurred prep_occurred_within det_context_that pobj_within_context prep_occurred_from det_corpus_a amod_corpus_large pobj_from_corpus cc_detecting_and advmod_learning_finally conj_detecting_learning aux_classify_to xcomp_learning_classify det_distributions_these dobj_classify_distributions prep_classify_as pcomp_as_representing preconj_anaphoric_either amod_instances_anaphoric cc_anaphoric_and conj_anaphoric_non-anaphoric nn_instances_pronoun dobj_representing_instances ' +o,7672,'In this work we use the following contextual information : a3 Target context : As in we consider a window of 3 words to the left and to the right of the target word considered ',Berger,'prep_use_In det_work_this pobj_In_work nsubj_use_we det_context_the amod_context_following amod_context_contextual nn_context_information nn_context_a3 nn_context_Target dobj_use_context prep_consider_As dep_As_in pobj_in_ nsubj_consider_we parataxis_use_consider det_window_a dobj_consider_window prep_window_of num_words_3 pobj_of_words prep_consider_to det_left_the pobj_to_left cc_to_and conj_to_to det_right_the pobj_to_right prep_right_of det_word_the nn_word_target pobj_of_word partmod_word_considered ' +o,7673,'The tagger is a Hidden Markov Model trained with the perceptron algorithm introduced in , which applies Viterbi decoding and is regularized using averaging ',Collins,'det_tagger_The nsubj_Model_tagger cop_Model_is det_Model_a nn_Model_Hidden nn_Model_Markov partmod_Model_trained prep_trained_with det_algorithm_the nn_algorithm_perceptron pobj_with_algorithm partmod_algorithm_introduced prep_introduced_in pobj_in_ nsubj_applies_which rcmod__applies nsubj_decoding_Viterbi xcomp_applies_decoding cc_applies_and auxpass_regularized_is conj_applies_regularized xcomp_regularized_using dobj_using_averaging ' +o,7674,'Decoding is carried-out using the Moses decoder ',Koehn,'nsubj_carried-out_Decoding cop_carried-out_is xcomp_carried-out_using det_decoder_the nn_decoder_Moses dobj_using_decoder ' +o,7675,'The lexicalized PCFG that sits behind Model 2 of has rules of the form P ~ LnLn-I \' \' \' LIHRI ` Rn-IRn -LRB- 1 -RRB- S -LRB- will-MD -RRB- NP -LRB- AppI , ~ NNP -RRB- VP -LRB- wilI-MD -RRB- NNP I Apple MD VP -LRB- buy-VB -RRB- VB PRT -LRB- out-RP -RRB- NP -LRB- Microsoft -- NNP -RRB- I -LRB- I buy RP NNP I I out Microsoft Figure 1 : A sample sentence with parse tree ',Collins,'det_PCFG_The amod_PCFG_lexicalized nsubj_has_PCFG nsubj_sits_that rcmod_PCFG_sits prep_sits_behind pobj_behind_Model num_Model_2 prep_Model_of pobj_of_ dobj_has_rules prep_rules_of det_LnLn-I_the nn_LnLn-I_form nn_LnLn-I_P nn_LnLn-I_~ poss_VP_LnLn-I amod_Rn-IRn_LIHRI measure_S_Rn-IRn appos_Rn-IRn_1 amod_VP_S dep_S_will-MD nn_VP_NP appos_VP_AppI nn_NNP_~ dep_AppI_NNP pobj_of_VP dep_rules_wilI-MD nn_VP_NNP nn_VP_I nn_VP_Apple nn_VP_MD dep_rules_VP appos_VP_buy-VB nn_NP_VB nn_NP_PRT appos_NP_out-RP prep_VP_NP abbrev_NP_Microsoft dep_Microsoft_NNP dep_NP_I nsubj_buy_I rcmod_I_buy nn_NNP_RP nsubj_I_NNP dep_NNP_I xcomp_buy_I prep_buy_out nn_Figure_Microsoft pobj_out_Figure num_Figure_1 det_sentence_A nn_sentence_sample dep_VP_sentence prep_sentence_with amod_tree_parse pobj_with_tree ' +o,7676,'To extract semantic information of words such as synonyms and antonyms from corpora , previous research used syntactic structures , response time to associate synonyms and antonyms in psychological experiments , or extracting related words automatically from corpora ',Hindle,'aux_extract_To dep_used_extract amod_information_semantic dobj_extract_information prep_information_of pobj_of_words dep_as_such prep_words_as pobj_as_synonyms cc_synonyms_and conj_synonyms_antonyms prep_extract_from pobj_from_corpora amod_research_previous nsubj_used_research amod__syntactic nn__structures dobj_used_ nn_time_response appos__time prep_time_to amod_synonyms_associate pobj_to_synonyms cc_synonyms_and conj_synonyms_antonyms prep_synonyms_in amod_experiments_psychological pobj_in_experiments amod_time_ cc__or conj__extracting amod_words_related dobj_extracting_words advmod_extracting_automatically prep_extracting_from nn__corpora pobj_from_ ' +o,7677,'The feature weights are learned by maximizing the BLEU score on held-out data,usingminimum-error-ratetraining as implemented by Koehn ',Och,'det_weights_The nn_weights_feature nsubjpass_learned_weights auxpass_learned_are prep_learned_by pcomp_by_maximizing det__the amod__BLEU nn__score dobj_maximizing_ prep_maximizing_on amod__held-out amod__data,usingminimum-error-ratetraining pobj_on_ mark_implemented_as advcl_maximizing_implemented prep_implemented_by pobj_by_Koehn ' +p,7678,'This results also agree with Dunning \'s argument about overestimation on the infrequent occurrences in which many infrequent pairs tend to get higher estimation ',Dunning,'nsubj_results_This advmod_results_also dep_results_agree prep_agree_with poss_argument_Dunning possessive_Dunning_\'s pobj_with_argument prep_argument_about pobj_about_overestimation prep_agree_on det_occurrences_the amod_occurrences_infrequent pobj_on_occurrences rel_tend_in pobj_in_which amod_pairs_many dep_many_infrequent nsubj_tend_pairs rcmod_occurrences_tend aux_get_to xcomp_tend_get amod_estimation_higher dobj_get_estimation ' +o,7679,'The need for some way to model aspects of syntactic behavior , such as the tendency of constituents to move together as a unit , is widely recognizedthe role of syntactic units is well attested in recent systematic studies of translation , and their absence in phrase-based models is quite evident when looking at MT system output ',Koehn,'det_need_The nsubj_recognizedthe_need prep_need_for det_way_some pobj_for_way prep_way_to nn_aspects_model pobj_to_aspects prep_aspects_of amod_behavior_syntactic pobj_of_behavior dep_as_such prep_need_as det_tendency_the pobj_as_tendency prep_tendency_of pobj_of_constituents aux_move_to infmod_constituents_move advmod_move_together prep_move_as det_unit_a pobj_as_unit cop_recognizedthe_is advmod_recognizedthe_widely nsubjpass_attested_role prep_role_of amod_units_syntactic pobj_of_units auxpass_attested_is advmod_attested_well dep_recognizedthe_attested prep_attested_in amod_studies_recent amod_studies_systematic pobj_in_studies prep_studies_of nn__translation pobj_of_ cc_attested_and poss_absence_their nsubj_evident_absence prep_absence_in amod_models_phrase-based pobj_in_models cop_evident_is advmod_evident_quite conj_attested_evident advmod_looking_when advcl_evident_looking prep_looking_at nn_output_MT nn_output_system pobj_at_output ' +o,7680,'It is possible to prove that , provided the training set -LRB- xi , zi -RRB- is separable with margin -RRB- 0 , the algorithm is assured to converge after a finite number of iterations to a model with zero training errors ',Collins,'nsubj_possible_It cop_possible_is aux_prove_to xcomp_possible_prove dobj_prove_that prep_prove_provided det_set_the nn_set_training nsubj_separable_set appos_set_xi dep_xi_zi cop_separable_is dep_provided_separable prep_prove_with pobj_with_margin advmod_assured_0 det_algorithm_the nsubjpass_assured_algorithm auxpass_assured_is dep_possible_assured aux_converge_to xcomp_assured_converge prep_converge_after det_number_a amod_number_finite pobj_after_number prep_number_of pobj_of_iterations prep_converge_to det_model_a pobj_to_model prep_model_with num_errors_zero nn_errors_training pobj_with_errors ' +o,7681,' presented a linear parsing model trained with an averaged perceptron algorithm ',Collins,'nsubj_presented_ det_model_a amod_model_linear amod_model_parsing dobj_presented_model partmod_model_trained prep_trained_with det_algorithm_an amod_algorithm_averaged nn_algorithm_perceptron pobj_with_algorithm ' +o,7682,'Since in these LVCs the complement is a predicative noun in stem form identical to a verb , we form development and test expressions by combining give or take with verbs from selected semantic classes of Levin , taken from Stevenson et al ',Dunning,'mark_noun_Since prep_noun_in det_LVCs_these pobj_in_LVCs det_complement_the nsubj_noun_complement cop_noun_is det_noun_a amod_noun_predicative advcl_form_noun prep_noun_in nn_form_stem pobj_in_form amod_form_identical prep_identical_to det_verb_a pobj_to_verb nsubj_form_we nn_expressions_development cc_development_and conj_development_test dobj_form_expressions prep_form_by pcomp_by_combining ccomp_combining_give cc_give_or conj_give_take prep_give_with pobj_with_verbs prep_give_from amod_classes_selected amod_classes_semantic pobj_from_classes prep_classes_of nn__Levin pobj_of_ partmod__taken prep_taken_from pobj_from_Stevenson cc_Stevenson_et conj_Stevenson_al ' +o,7683,'Currently , the scheme supports PhraseChunks with subtypes such as NP , VP , PP , or ADJP ',Marcus,'advmod_supports_Currently det_scheme_the nsubj_supports_scheme dobj_supports_PhraseChunks prep_supports_with pobj_with_subtypes dep_as_such prep_subtypes_as nn_PP_NP appos_PP_VP pobj_as_PP cc_supports_or conj_supports_ADJP ' +o,7684,'Both models have been used to achieve state-of-the-art accuracy for a wide range of languages , as shown in the CoNLL shared tasks on dependency parsing , but showed that a detailed error analysis reveals important differences in the distribution of errors associated with the two models ',McDonald,'det_models_Both nsubjpass_used_models aux_used_have auxpass_used_been aux_achieve_to xcomp_used_achieve amod_accuracy_state-of-the-art dobj_achieve_accuracy prep_achieve_for det_range_a amod_range_wide pobj_for_range prep_range_of pobj_of_languages mark_shared_as csubj_shared_shown prep_shown_in det_CoNLL_the pobj_in_CoNLL advcl_achieve_shared dobj_shared_tasks prep_tasks_on amod__dependency amod__parsing pobj_on_ cc_shared_but nsubj_showed_ conj_shared_showed complm_reveals_that det_analysis_a amod_analysis_detailed nn_analysis_error nsubj_reveals_analysis ccomp_showed_reveals amod_differences_important dobj_reveals_differences prep_reveals_in det_distribution_the pobj_in_distribution prep_distribution_of pobj_of_errors partmod_errors_associated prep_associated_with det_models_the num_models_two pobj_with_models ' +p,7685,'An online learning algorithm considers a single training instance for each update to the weight vector w We use the common method of setting the final weight vector as the average of the weight vectors after each iteration , which has been shown to alleviate overfitting ',Collins,'det_algorithm_An amod_algorithm_online nn_algorithm_learning nsubj_considers_algorithm det_instance_a amod_instance_single nn_instance_training dobj_considers_instance prep_instance_for det_update_each pobj_for_update prep_considers_to det_w_the nn_w_weight nn_w_vector pobj_to_w nsubj_use_We dep_considers_use det_method_the amod_method_common dobj_use_method prep_method_of pcomp_of_setting det_vector_the amod_vector_final nn_vector_weight dobj_setting_vector prep_setting_as det_average_the pobj_as_average prep_average_of det_vectors_the nn_vectors_weight pobj_of_vectors prep_setting_after det__each nn__iteration pobj_after_ nsubjpass_shown_which aux_shown_has auxpass_shown_been rcmod__shown aux_alleviate_to xcomp_shown_alleviate dobj_alleviate_overfitting ' +n,7686,'Allomorphs -LRB- eg , deni and deny -RRB- are also automatically identified in , but the general problem of recognizing highly irregular forms is examined more extensively in ',Dasgupta,'nsubjpass_identified_Allomorphs dep_Allomorphs_eg appos_eg_deni cc_deni_and conj_deni_deny auxpass_identified_are advmod_identified_also advmod_identified_automatically prep_identified_in pobj_in_ cc_identified_but det_problem_the amod_problem_general nsubjpass_examined_problem prep_problem_of pcomp_of_recognizing advmod_irregular_highly amod_forms_irregular dobj_recognizing_forms auxpass_examined_is conj_identified_examined advmod_extensively_more dep_examined_extensively prep_extensively_in pobj_in_ ' +o,7687,'The annotation guidelines for the Penn Treebank flattened noun phrases to simplify annotation , so there is no complex structure to NPs ',Marcus,'det_annotation_The nsubj_flattened_guidelines prep_guidelines_for det_Treebank_the nn_Treebank_Penn pobj_for_Treebank dep_annotation_flattened nn_phrases_noun dobj_flattened_phrases aux_simplify_to xcomp_flattened_simplify nn__annotation dobj_simplify_ dep_flattened_so expl_is_there dep_flattened_is det_structure_no amod_structure_complex nsubj_is_structure prep_structure_to pobj_to_NPs ' +o,7688,'The machine translation literature is littered with various attempts to learn a phrase-based string transducer directly from aligned sentence pairs , doing away with the separate word alignment step ',Zhang,'det_literature_The nn_literature_machine nn_literature_translation nsubjpass_littered_literature auxpass_littered_is prep_littered_with amod_attempts_various pobj_with_attempts aux_learn_to xcomp_littered_learn det_transducer_a amod_transducer_phrase-based nn_transducer_string dobj_learn_transducer advmod_learn_directly prep_learn_from amod_pairs_aligned nn_pairs_sentence pobj_from_pairs xcomp_learn_doing prt_doing_away prep_doing_with det_step_the amod_step_separate nn_step_word nn_step_alignment pobj_with_step ' +o,7689,'Either pruning or lossy randomizing approaches may result in a compact representation for the application run-time ',Brants,'dep_result_Either nn__pruning nsubj_result_ cc__or nn_approaches_lossy nn_approaches_randomizing conj__approaches advmod_result_ aux_result_may prep_result_in det_representation_a amod_representation_compact pobj_in_representation prep_representation_for det_run-time_the nn_run-time_application pobj_for_run-time ' +n,7690,'Most recently , published their Semi-supervised sequential labeling method , whose results on POS tagging seem to be optically better than , but no significance tests were given and the tool is not available for download , ie for repeating the results and significance testing ',Shen,'advmod_recently_Most advmod_published_recently nsubj_published_ poss_method_their amod_method_Semi-supervised amod_method_sequential nn_method_labeling dobj_published_method poss_results_whose nsubj_seem_results prep_results_on dep_tagging_POS pobj_on_tagging rcmod_method_seem aux_better_to cop_better_be advmod_better_optically xcomp_seem_better prep_better_than pobj_than_ cc_published_but det_tests_no nn_tests_significance nsubjpass_given_tests auxpass_given_were conj_published_given cc_published_and det_tool_the nsubj_available_tool cop_available_is neg_available_not conj_published_available prep_available_for pobj_for_download appos_download_ie prep_ie_for pcomp_for_repeating det_results_the dobj_repeating_results cc_results_and nn_testing_significance conj_results_testing ' +o,7691,'The implementation is similar to the idea of lexical weight in : all points in the alignment matrices of the entire training corpus are collected to calculate the probabilistic distribution , P -LRB- t s -RRB- , of some TL word 3Some readers may prefer the expression the subtree rooted at node N to node N The latter term is used in this paper for simplicity ',Koehn,'det_implementation_The nsubj_similar_implementation cop_similar_is prep_similar_to det_idea_the pobj_to_idea prep_idea_of amod_weight_lexical pobj_of_weight prep_weight_in pobj_in_ advmod_similar_all nsubjpass_collected_points prep_points_in det_matrices_the amod_matrices_alignment pobj_in_matrices prep_matrices_of det_corpus_the amod_corpus_entire nn_corpus_training pobj_of_corpus auxpass_collected_are dep_similar_collected aux_calculate_to xcomp_collected_calculate det_distribution_the amod_distribution_probabilistic dobj_calculate_distribution appos_distribution_P nn_s_t appos_P_s prep_distribution_of det_word_some nn_word_TL pobj_of_word amod_readers_3Some nsubj_prefer_readers aux_prefer_may dep_similar_prefer det_expression_the nsubj_subtree_expression det_subtree_the xcomp_prefer_subtree partmod_subtree_rooted prep_rooted_at amod_N_node pobj_at_N aux_node_to xcomp_rooted_node dobj_node_N det_term_The amod_term_latter nsubjpass_used_term auxpass_used_is rcmod_N_used prep_used_in det_paper_this pobj_in_paper prep_paper_for pobj_for_simplicity ' +o,7692,'In this work , we focus on learning bilingual word phrases by using Stochastic Inversion Transduction Grammars -LRB- SITGs -RRB- ',Wu,'prep_focus_In det_work_this pobj_In_work nsubj_focus_we prep_focus_on pcomp_on_learning amod_phrases_bilingual nn_phrases_word dobj_learning_phrases prep_learning_by pcomp_by_using nn_Grammars_Stochastic nn_Grammars_Inversion nn_Grammars_Transduction dobj_using_Grammars abbrev_Grammars_SITGs ' +o,7693,'21 Keywords As our starting point , we calculated the keywords of the Belgian corpus with respect to the Netherlandic corpus , both on the basis of a chi-square test -LRB- with Yates continuity correction -RRB- and the log-likelihood ratio ',Dunning,'num_Keywords_21 prep_calculated_As poss_point_our amod_point_starting pobj_As_point nsubj_calculated_we dep_Keywords_calculated det_keywords_the dobj_calculated_keywords prep_keywords_of det_corpus_the amod_corpus_Belgian pobj_of_corpus prep_calculated_with pobj_with_respect prep_calculated_to det_corpus_the nn_corpus_Netherlandic pobj_to_corpus dep_on_both prep_calculated_on det_basis_the pobj_on_basis prep_basis_of det__a amod__chi-square nn__test dep__with nn_correction_Yates nn_correction_continuity pobj_with_correction pobj_of_ cc__and det_ratio_the amod_ratio_log-likelihood conj__ratio ' +p,7694,'ROUGE-L , ROUGE-W , and ROUGE-S have also been applied in automatic evaluation of summarization and achieved very promising results ',Lin,'nn_ROUGE-S_ROUGE-L conj_ROUGE-S_ROUGE-W cc_ROUGE-S_and nsubjpass_applied_ROUGE-S aux_applied_have advmod_applied_also auxpass_applied_been prep_applied_in amod_evaluation_automatic pobj_in_evaluation prep_evaluation_of pobj_of_summarization cc_applied_and conj_applied_achieved advmod_results_very amod_results_promising dobj_achieved_results ' +o,7695,'Maximum entropy estimation for translation of individual words dates back to Berger et al , and the idea of using multi-class classifiers to sharpen predictions normally made through relative frequency estimates has been recently reintroducedundertherubricofwordsensedisambiguation and generalized to substrings ',Carpuat,'amod_estimation_Maximum nn_estimation_entropy nsubj_dates_estimation prep_estimation_for pobj_for_translation prep_translation_of amod_words_individual pobj_of_words prt_dates_back prep_dates_to nn__Berger cc_Berger_et conj_Berger_al pobj_to_ cc_dates_and det_idea_the nsubjpass_reintroducedundertherubricofwordsensedisambiguation_idea prep_idea_of pcomp_of_using amod_classifiers_multi-class dobj_using_classifiers aux_sharpen_to xcomp_using_sharpen dobj_sharpen_predictions advmod_made_normally partmod_predictions_made prep_made_through amod_estimates_relative nn_estimates_frequency pobj_through_estimates aux_reintroducedundertherubricofwordsensedisambiguation_has auxpass_reintroducedundertherubricofwordsensedisambiguation_been advmod_reintroducedundertherubricofwordsensedisambiguation_recently conj_dates_reintroducedundertherubricofwordsensedisambiguation cc_reintroducedundertherubricofwordsensedisambiguation_and conj_reintroducedundertherubricofwordsensedisambiguation_generalized prep_reintroducedundertherubricofwordsensedisambiguation_to pobj_to_substrings ' +o,7696,'Starting from an initial point M1 , computing the most probable sentence hypothesis out of a set of K candidate translations Cs AG D8e1,,eKD9 along the line M1 A0 A4 dM1 results in the following optimization problem (Och, 2003): e D4fs;D5 AG argmax eC8Cs AX D4 M 1 A0 A4 d M 1 D5 C2 A4 hM1 D4e,fsD5 B5 AG argmax eC8Cs AY F4 m mhmD4e,fsD5 D0D3D3D3D3D3D3D3D3D1D3D3D3D3D3D3D3D3D2 AGaD4e,fsD5 A0 A4 F4 m dmhmD4e,fsD5 D0D3D3D3D3D3D3D3D3D1D3D3D3D3D3D3D3D3D2 AGbD4e,fsD5 B6 AG argmax eC8Cs AWa D4e,fsD5 A0 A4 bD4e,fsD5 D0D3D3D3D3D3D3D3D3D3D3D3D1D3D3D3D3D3D3D3D3D3D3D3D2 D4A6D5 B4 (5) Hence, the total score D4A6D5 for any candidate translation corresponds to a line in the plane with as the independent variable.',Och,'' +o,7697,'BLEU : BLEU score , which computes the ratio of n-gram for the translation results found in reference translations ',Papineni,'amod_score_BLEU dep_BLEU_score nsubj_computes_which rcmod_score_computes det_ratio_the dobj_computes_ratio prep_ratio_of pobj_of_n-gram prep_score_for det_results_the nn_results_translation pobj_for_results partmod_results_found prep_found_in nn_translations_reference pobj_in_translations ' +o,7698,'These models can be tuned using minimum error rate training ',Och,'det_models_These nsubjpass_tuned_models aux_tuned_can auxpass_tuned_be dep_tuned_using amod_training_minimum nn_training_error nn_training_rate dobj_using_training ' +o,7699,'We performed feature selection by incrementally growing a log-linear model with order0 features f -LRB- x , yt -RRB- using a forward feature selection procedure similar to ',Berger,'nsubj_performed_We nn_selection_feature dobj_performed_selection prep_performed_by advmod_growing_incrementally pcomp_by_growing det_model_a amod_model_log-linear dobj_growing_model prep_model_with amod_features_order0 pobj_with_features dep_model_f dep_growing_x appos_x_yt xcomp_growing_using det_procedure_a amod_procedure_forward nn_procedure_feature nn_procedure_selection nsubj_similar_procedure xcomp_using_similar prep_similar_to ' +o,7700,'However , they use the data set in a different training-test division -LRB- 10-fold cross validation -RRB- which makes it -LRB- tifficult to compare their results with others ',Ramshaw,'advmod_use_However nsubj_use_they det_data_the amod_data_ dobj_use_data partmod_data_set prep_set_in det_division_a amod_division_different amod_division_training-test pobj_in_division amod_validation_10-fold amod_validation_cross appos_division_validation nsubj_makes_which ccomp_division_makes nsubj_compare_it advmod_compare_tifficult aux_compare_to ccomp_makes_compare poss_results_their dobj_compare_results prep_compare_with pobj_with_others ' +o,7701,'31 Translation Model Form We first assume the general hypergraph setting of , namely , that derivations under our translation model form a hypergraph ',Huang,'num_Form_31 amod_Form_Translation nn_Form_Model nsubj_assume_We advmod_assume_first dep_Form_assume det_setting_the amod_setting_general nn_setting_hypergraph dobj_assume_setting prep_setting_of pcomp_of_ advmod__namely complm_form_that nsubj_form_derivations prep_derivations_under poss_model_our nn_model_translation pobj_under_model ccomp__form det_hypergraph_a dobj_form_hypergraph ' +o,7702,'For the WMT 2009 Workshop , we selected a linear combination of BLEU and TER as optimization criterion , : = argmax -LCB- -LRB- 2BLEU -RRB- TER -RCB- , based on previous experience ',Papineni,'prep_selected_For det_Workshop_the nn_Workshop_WMT num_Workshop_2009 pobj_For_Workshop nsubj_selected_we det_combination_a amod_combination_linear dobj_selected_combination prep_combination_of nn__BLEU pobj_of_ cc__and nn__TER conj__ prep__as nn_criterion_optimization pobj_as_criterion dep_argmax_= dep_combination_argmax nn_TER_2BLEU dep_combination_TER partmod_combination_based prep_based_on amod__previous nn__experience pobj_on_ ' +o,7703,'In fact , many attempts have recently been made to develop semi-supervised SOL methods ',Jiao,'prep_made_In pobj_In_fact amod_attempts_many nsubjpass_made_attempts aux_made_have advmod_made_recently auxpass_made_been aux_develop_to xcomp_made_develop amod_methods_semi-supervised nn_methods_SOL dobj_develop_methods ' +o,7704,'Previous research has focused on classifying subjective-versus-objective expressions , and also on accurate sentiment polarity assignment ',Turney,'amod_research_Previous nsubj_focused_research aux_focused_has prep_focused_on pcomp_on_classifying amod_expressions_subjective-versus-objective dobj_classifying_expressions advmod_classifying_ cc_focused_and conj_focused_also prep_also_on amod_assignment_accurate nn_assignment_sentiment nn_assignment_polarity pobj_on_assignment ' +o,7705,'Although various approaches to SMT system combination have been explored , including enhanced combination model structure , better word alignment between translations and improved confusion network construction , most previous work simply used the ensemble of SMT systems based on different models and paradigms at hand and did not tackle the issue of how to obtain the ensemble in a principled way ',Ayan,'mark_explored_Although amod_approaches_various nsubj_explored_approaches prep_approaches_to nn_combination_SMT nn_combination_system pobj_to_combination aux_explored_have dep_explored_been advcl_used_explored prep_explored_including amod__enhanced nn__combination nn__model nn__structure pobj_including_ amod_alignment_better nn_alignment_word appos__alignment prep_alignment_between nn__translations pobj_between_ cc__and amod__improved nn__confusion nn__network nn__construction conj__ advmod_work_most amod_work_previous nsubj_used_work advmod_used_simply det_ensemble_the dobj_used_ensemble prep_ensemble_of nn_systems_SMT pobj_of_systems partmod_systems_based prep_based_on amod_models_different pobj_on_models cc_models_and conj_models_paradigms prep_based_at pobj_at_hand cc_used_and aux_tackle_did neg_tackle_not conj_used_tackle det_issue_the dobj_tackle_issue prep_issue_of advmod_obtain_how aux_obtain_to pcomp_of_obtain det_ensemble_the dobj_obtain_ensemble prep_ensemble_in det_way_a amod_way_principled pobj_in_way ' +o,7706,'Moreover , this evaluation concern dovetails with a frequent engineering concern , that sentence-level scores are useful at various points in the MT pipeline : for example , minimum Bayes risk decoding , selecting oracle translations for discriminative reranking , and sentenceby-sentence comparisons of outputs during error analysis ',Liang,'advmod_dovetails_Moreover det_concern_this nn_concern_evaluation nsubj_dovetails_concern prep_dovetails_with det_concern_a amod_concern_frequent nn_concern_engineering pobj_with_concern complm_useful_that amod_scores_sentence-level nsubj_useful_scores cop_useful_are ccomp_dovetails_useful prep_useful_at amod_points_various pobj_at_points prep_points_in det_pipeline_the nn_pipeline_MT pobj_in_pipeline prep_useful_for nn__example amod__minimum nn__Bayes nn__risk amod__decoding pobj_for_ partmod__selecting amod_translations_oracle dobj_selecting_translations prep_translations_for amod__discriminative nn__reranking pobj_for_ cc_translations_and amod_comparisons_sentenceby-sentence conj_translations_comparisons prep_comparisons_of pobj_of_outputs prep_selecting_during nn_analysis_error pobj_during_analysis ' +o,7707,'The 45 stochastic word mapping is trained on a FrenchEnglish parallel corpus containing 700,000 sentence pairs , and , following , we only keep the top 100 most similar words for each English word ',Liu,'det_mapping_The num_mapping_45 amod_mapping_stochastic nn_mapping_word nsubj_trained_mapping aux_trained_is prep_trained_on det_corpus_a amod_corpus_FrenchEnglish amod_corpus_parallel pobj_on_corpus partmod_corpus_containing num_pairs_700,000 nn_pairs_sentence dobj_containing_pairs cc_trained_and xcomp_,_following nsubj_keep_ nsubj_keep_we advmod_keep_only ccomp_following_keep det_words_the amod_words_top num_words_100 advmod_similar_most amod_words_similar dobj_keep_words prep_words_for det_word_each amod_word_English pobj_for_word ' +o,7708,' uses selftraining to perform this step -RRB- -LRB- 2 -RRB- smoothing , usually this is performed using a markovization procedure and -LRB- 3 -RRB- make the data more coarse -LRB- ie clustering -RRB- ',McClosky,'nsubj_uses_ acomp_uses_selftraining prep_uses_to pobj_to_perform det_step_this dep_perform_step dep_performed_2 nsubjpass_performed_smoothing advmod_smoothing_usually dep_smoothing_this auxpass_performed_is dep_uses_performed xcomp_performed_using det_procedure_a nn_procedure_markovization nsubj__procedure ccomp_using_ cc__and dep__3 conj__make det_data_the dobj_make_data advmod_coarse_more amod_data_coarse nn_clustering_ie appos_data_clustering ' +o,7709,'It has been shown that phrasal machine translation systems are not affected by the quality of the input word alignments ',Koehn,'nsubjpass_shown_It aux_shown_has auxpass_shown_been complm_affected_that amod_systems_phrasal nn_systems_machine nn_systems_translation nsubjpass_affected_systems auxpass_affected_are neg_affected_not ccomp_shown_affected prep_affected_by det_quality_the pobj_by_quality prep_quality_of det_alignments_the nn_alignments_input nn_alignments_word pobj_of_alignments ' +o,7710,'When evaluated against the state-of-the-art, phrase-based decoder Pharaoh (Koehn, 2004), using the same experimental conditions translation table trained on the FBIS corpus (7.2M Chinese words and 9.2M English words of parallel text), trigram language model trained on 155M words of English newswire, interpolation weights a65 (Equation 2) trained using discriminative training (Och, 2003) (on the 2002 NIST MT evaluation set), probabilistic beam a90 set to 0.01, histogram beam a58 set to 10 and BLEU (Papineni et al. , 2002) as our metric, the WIDL-NGLM-Aa86 a129 algorithm produces translations that have a BLEU score of 0.2570, while Pharaoh translations have a BLEU score of 0.2635.',Papineni,'' +o,7711,'Clustering can be done statistically by analyzing text corpora and usually results in a set of words or word senses ',Brown,'nsubjpass_done_Clustering aux_done_can auxpass_done_be advmod_done_statistically prep_done_by pcomp_by_analyzing nn__text nn__corpora dobj_analyzing_ cc__and advmod__usually dep__results prep_results_in det_set_a pobj_in_set prep_set_of pobj_of_words cc_words_or nn_senses_word conj_words_senses ' +o,7712,'Therefore , other machine learning techniques such as perceptron could also be applied for this problem ',Collins,'advmod_applied_Therefore amod_machine_other nsubjpass_applied_machine partmod_machine_learning dobj_learning_techniques dep_as_such prep_techniques_as nn__perceptron pobj_as_ aux_applied_could advmod_applied_also auxpass_applied_be prep_applied_for det_problem_this pobj_for_problem ' +o,7713,'Most work in machine learning literature on utilizing labeled features has focused on using them to generate weakly labeled examples that are then used for standard supervised learning : propose one such framework for boosting logistic regression ; build a modi ed SVM and use a combination of clustering and EM based methods to instantiate similar frameworks ',Pang,'amod_work_Most nsubj_labeled_work prep_work_in pobj_in_machine partmod_machine_learning dobj_learning_literature prep_learning_on pcomp_on_utilizing nsubj_focused_features aux_focused_has ccomp_labeled_focused prep_focused_on pcomp_on_using nsubj_generate_them aux_generate_to xcomp_using_generate amod_examples_weakly amod_examples_labeled dobj_generate_examples nsubjpass_used_that auxpass_used_are advmod_used_then rcmod_examples_used prep_used_for amod_learning_standard amod_learning_supervised pobj_for_learning nsubj_propose_ parataxis_labeled_propose num_framework_one amod_framework_such dobj_propose_framework prep_propose_for pcomp_for_boosting amod_regression_logistic dobj_boosting_regression nsubj_build_ parataxis_labeled_build det_SVM_a nn_SVM_modi nn_SVM_ed dobj_build_SVM cc_labeled_and nsubj_use_ conj_labeled_use det_combination_a dobj_use_combination prep_combination_of pobj_of_clustering cc_clustering_and conj_clustering_EM partmod_clustering_based dobj_based_methods aux_instantiate_to xcomp_based_instantiate amod_frameworks_similar dobj_instantiate_frameworks ' +p,7714,'Given the parameters -LCB- pi0 , pi , , K -RCB- of the HMM , the joint distribution over hidden states s and observationsy can be written -LRB- with s0 = 0 -RRB- : p -LRB- s , y pi0 , pi , , K -RRB- = Tproductdisplay t = 1 p -LRB- st st1 -RRB- p -LRB- yt st -RRB- As clearly explained , training the HMM with EM leads to poor results in PoS tagging ',Johnson,'prep_s_Given det_parameters_the dep_Given_parameters dep_parameters_pi0 appos_pi0_pi appos_pi0_K prep_parameters_of det_HMM_the pobj_of_HMM det_distribution_the amod_distribution_joint nsubjpass_s_distribution prep_distribution_over amod_states_hidden pobj_over_states cc_s_and advmod_written_observationsy aux_written_can auxpass_written_be conj_s_written dep_s0_with nsubj_0_s0 dep_0_= parataxis_written_0 dobj_written_p dep_p_s nn_pi0_y appos_s_pi0 appos_pi0_pi appos_s_K dep_p_= nn_t_Tproductdisplay iobj_=_t number_1_= num_p_1 nn_p_p nn_st1_st appos_p_st1 dobj_=_p nn_st_yt appos_p_st mark_explained_As nsubj_explained_ advmod_explained_clearly csubj_leads_explained xcomp_explained_training det_HMM_the dobj_training_HMM prep_training_with pobj_with_EM dep_s_leads prep_leads_to amod_results_poor pobj_to_results prep_results_in nn_tagging_PoS pobj_in_tagging ' +o,7715,'The supervised methods are based on Maximum Entropy -LRB- ME -RRB- , neural network using the Learning Vector Quantization algorithm and Specialized Hidden Markov Models ',Berger,'det_methods_The amod_methods_supervised nsubjpass_based_methods auxpass_based_are prep_based_on nn__Maximum nn__Entropy abbrev__ME pobj_on_ amod_network_neural appos__network partmod_network_using det__the nn__Learning nn__Vector nn__Quantization nn__algorithm dobj_using_ cc__and nn__Specialized nn__Hidden nn__Markov nn__Models conj__ ' +o,7716,'The search is based on the property that when computing sim -LRB- wl , w2 -RRB- , words that have high mutual information values 5The nominator in our metric resembles the similarity metric in ',Hindle,'det_search_The nsubjpass_based_search auxpass_based_is prep_based_on det_property_the pobj_on_property complm_metric_that advmod_values_when amod_sim_computing nsubj_values_sim appos_sim_wl dep_wl_w2 dep_sim_words nsubj_have_that rcmod_words_have amod_information_high amod_information_mutual dobj_have_information csubj_metric_values amod_nominator_5The dobj_values_nominator prep_nominator_in poss_metric_our pobj_in_metric cop_metric_resembles det_metric_the nn_metric_similarity ccomp_based_metric prep_metric_in ' +o,7717,'In previous alignment methods , some researchers modeled the alignments with different statistical models ',Wu,'prep_modeled_In amod_methods_previous nn_methods_alignment pobj_In_methods det_researchers_some nsubj_modeled_researchers det_alignments_the dobj_modeled_alignments prep_modeled_with amod_models_different amod_models_statistical pobj_with_models ' +p,7718,'Unlike , one interesting idea proposed by is to cluster similar pairs of paraphrases to apply multiplesequence alignment ',Barzilay,'prep_is_Unlike pobj_Unlike_ num_idea_one amod_idea_interesting nsubj_is_idea partmod_idea_proposed prep_proposed_by pobj_by_ aux_cluster_to xcomp_is_cluster amod_pairs_similar dobj_cluster_pairs prep_pairs_of pobj_of_paraphrases aux_apply_to xcomp_cluster_apply amod_alignment_multiplesequence dobj_apply_alignment ' +o,7719,'For natural language engineers , the problem bears on information management systems like abstractive summarizers that must measure semantic overlap between sentences , question answering modules and machine translation ',Barzilay,'prep_bears_For amod_engineers_natural nn_engineers_language pobj_For_engineers det_problem_the nsubj_bears_problem prep_bears_on nn_systems_information nn_systems_management pobj_on_systems prep_systems_like amod_summarizers_abstractive pobj_like_summarizers nsubj_measure_that aux_measure_must rcmod_systems_measure amod_overlap_semantic dobj_measure_overlap prep_overlap_between nn__sentences pobj_between_ nn__question nn__answering nn__modules conj__ cc__and nn__machine nn__translation conj__ ' +o,7720,'This new model leads to significant improvements in MT quality as measured by BLEU ',Papineni,'det_model_This amod_model_new nsubj_leads_model prep_leads_to amod_improvements_significant pobj_to_improvements prep_improvements_in nn_quality_MT pobj_in_quality mark_measured_as advcl_leads_measured prep_measured_by pobj_by_BLEU ' +o,7721,'We design special inference algorithms , instead of general-purpose inference algorithms used in previous works , by taking advantage of special properties of our task ',Ding,'nsubj_design_We amod_algorithms_special nn_algorithms_inference dobj_design_algorithms dep_of_instead prep_design_of amod_algorithms_general-purpose nn_algorithms_inference pobj_of_algorithms partmod_algorithms_used prep_used_in amod__previous nn__works pobj_in_ prep_design_by pcomp_by_taking dobj_taking_advantage prep_advantage_of amod_properties_special pobj_of_properties prep_properties_of poss_task_our pobj_of_task ' +o,7722,'Knight and Marcu treat reduction as a translation process using a noisychannel model ',Brown,'nn__Knight cc_Knight_and conj_Knight_Marcu nsubj_treat_ dobj_treat_reduction prep_treat_as det_process_a nn_process_translation pobj_as_process partmod_process_using det_model_a amod_model_noisychannel dobj_using_model ' +o,7723,' looked at Golomb Coding and et al ',Brants,'nsubj_looked_ prep_looked_at nn_Coding_Golomb pobj_at_Coding cc_Coding_and conj_Coding_ cc__et conj__al ' +o,7724,' and , among others , look at using a large amount of unlabeled data to classify relations between words ',Nakov,'nsubj_look_ cc__and conj__ prep__among pobj_among_others prep_look_at pcomp_at_using det_amount_a amod_amount_large dobj_using_amount prep_amount_of amod_data_unlabeled pobj_of_data aux_classify_to xcomp_using_classify dobj_classify_relations prep_relations_between pobj_between_words ' +p,7725,'Automated metrics such as BLEU , RED , Weighted N-gram model -LRB- WNM -RRB- , syntactic relation \\/ semantic vector model have been shown to correlate closely with scoring or ranking by different human evaluation parameters ',Papineni,'amod_metrics_Automated nsubjpass_shown_metrics dep_as_such prep_metrics_as nn__BLEU pobj_as_ nn_model_RED nn_model_ appos_model_Weighted nn_model_N-gram appos__model abbrev__WNM amod__ amod__syntactic nn__relation amod__\\/ amod__semantic nn__vector nn__model dep__ aux_shown_have auxpass_shown_been aux_correlate_to xcomp_shown_correlate advmod_correlate_closely prep_correlate_with pcomp_with_scoring cc_scoring_or conj_scoring_ranking prep_scoring_by amod_parameters_different amod_parameters_human nn_parameters_evaluation pobj_by_parameters ' +o,7726,'First , we noted how frequently WordNet gets used compared to other resources , such as FrameNet or the Penn Treebank ',Marcus,'advmod_noted_First nsubj_noted_we advmod_used_how advmod_how_frequently nn__WordNet nsubjpass_used_ auxpass_used_gets ccomp_noted_used prep_used_compared dep_compared_to amod_resources_other pobj_to_resources dep_as_such prep_resources_as nn__FrameNet pobj_as_ cc__or det_Treebank_the nn_Treebank_Penn conj__Treebank ' +o,7727,'We are currently investigating caching and optimizations that will enable the use of our metric for MT parameter tuning in a Minimum Error Rate Training setup ',Och,'nsubj_investigating_We aux_investigating_are advmod_investigating_currently dobj_investigating_caching cc_caching_and conj_caching_optimizations nsubj_enable_that aux_enable_will rcmod_caching_enable det_use_the dobj_enable_use prep_use_of poss_metric_our pobj_of_metric prep_enable_for nn_parameter_MT pobj_for_parameter partmod_parameter_tuning prep_tuning_in det_setup_a nn_setup_Minimum nn_setup_Error nn_setup_Rate nn_setup_Training pobj_in_setup ' +o,7728,'Discriminatively trained parsers that score entire trees for a given sentence have only recently been investigated ',Collins,'advmod_trained_Discriminatively amod_parsers_trained nsubjpass_investigated_parsers nsubj_score_that rcmod_parsers_score amod_trees_entire dobj_score_trees prep_trees_for det_sentence_a amod_sentence_given pobj_for_sentence aux_investigated_have advmod_recently_only advmod_investigated_recently auxpass_investigated_been ' +o,7729,'Previous studies defined a probabilistic model of unification-based grammars including HPSG as a log-linear model or maximum entropy model ',Berger,'amod__Previous nn__studies nsubj_defined_ det_model_a amod_model_probabilistic dobj_defined_model prep_model_of amod_grammars_unification-based pobj_of_grammars prep_grammars_including pcomp_including_HPSG prep_HPSG_as det_model_a amod_model_log-linear pobj_as_model cc_model_or amod_model_maximum nn_model_entropy conj_model_model ' +o,7730,'Regressive FLM -LRB- rFLM -RRB- h -LRB- FLM -LRB- e , j -RRB- -RRB- = w1 FLM -LRB- e , j -RRB- + b Regressive ALM -LRB- rALM -RRB- h -LRB- ALM -LRB- e , j -RRB- -RRB- = w1 ALM -LRB- e , j -RRB- + b Notice that h -LRB- -RRB- here is supposed to relate FLM or ALM to some independent evaluation metric such as BLEU , not the log likelihood of a translation ',Papineni,'nn_FLM_Regressive appos_FLM_rFLM nsubjpass_supposed_h dep_h_FLM appos_FLM_e dep_e_j nn_FLM_= nn_FLM_w1 nsubj_+_FLM appos_FLM_e dep_e_j rcmod_FLM_+ dep_Regressive_b dep_+_Regressive nn_h_ALM nn_h_rALM dobj_+_h dep_h_ALM appos_h_e dep_e_j dep_h_= iobj_=_w1 nn_+_ALM appos_+_e dep_e_j dobj_=_+ dep_+_b prep_FLM_Notice det_-LRB-_that nn_-LRB-_h advmod_supposed_here auxpass_supposed_is dep_FLM_supposed aux_relate_to xcomp_supposed_relate dobj_relate_FLM cc_FLM_or conj_FLM_ALM prep_relate_to det_metric_some amod_metric_independent nn_metric_evaluation pobj_to_metric dep_as_such prep_metric_as nn__BLEU pobj_as_ neg_likelihood_not det_likelihood_the nn_likelihood_log appos__likelihood prep_likelihood_of det_translation_a pobj_of_translation ' +o,7731,'To tune feature weights minimum error rate training is used , optimized against the Neva metric ',Och,'prep_used_To nn_weights_tune nn_weights_feature pobj_To_weights amod_training_minimum nn_training_error nn_training_rate nsubjpass_used_training auxpass_used_is dep_used_ ccomp_used_optimized prep_optimized_against det__the nn__Neva amod__metric pobj_against_ ' +o,7732,' later addressed the same problem using 8000 automatically generated patterns ',Turney,'nsubj_addressed_ advmod_addressed_later det_problem_the amod_problem_same dobj_addressed_problem partmod_problem_using dobj_using_8000 advmod_generated_automatically partmod_8000_generated dobj_generated_patterns ' +o,7733,'Unlike previous annotations of sentiment or subjectivity , which typically relied on binary 0\\/1 annotations , we decided to use a finer-grained scale , hence allowing the annotators to select different degrees of emotional load ',Pang,'prep_decided_Unlike amod_annotations_previous pobj_Unlike_annotations prep_annotations_of nn__sentiment cc_sentiment_or conj_sentiment_subjectivity pobj_of_ nsubj_relied_which advmod_relied_typically rcmod_annotations_relied prep_relied_on amod_annotations_binary num_annotations_0\\/1 pobj_on_annotations nsubj_decided_we aux_use_to xcomp_decided_use det_scale_a amod_scale_finer-grained dobj_use_scale advmod_allowing_hence xcomp_decided_allowing det_annotators_the nsubj_select_annotators aux_select_to xcomp_allowing_select amod_degrees_different dobj_select_degrees prep_degrees_of amod_load_emotional pobj_of_load ' +o,7734,'It is clear that Appendix B contains far fewer true non-compositional phrases than Appendix A 7 Related Work There have been numerous previous research on extracting collocations from corpus , eg , and ',Smadja,'nsubj_clear_It cop_clear_is complm_contains_that nn_B_Appendix nsubj_contains_B ccomp_clear_contains advmod_fewer_far amod_phrases_fewer amod_phrases_true amod_phrases_non-compositional dobj_contains_phrases prep_phrases_than nn_Work_Appendix nn_Work_A num_Work_7 nn_Work_Related pobj_than_Work expl_been_There aux_been_have dep_clear_been amod_research_numerous amod_research_previous dobj_been_research prep_research_on pcomp_on_extracting dobj_extracting_collocations prep_extracting_from pobj_from_corpus appos_corpus_eg nn_and_ appos_eg_and ' +o,7735,'Once an acceptable rate of interjudge agreement was verified on the first nine clusters -LRB- Kappa of 068 -RRB- , the remaining 11 clusters were annotated by one judge each ',Carletta,'mark_verified_Once det_rate_an amod_rate_acceptable nsubjpass_verified_rate prep_rate_of amod_agreement_interjudge pobj_of_agreement auxpass_verified_was advcl_annotated_verified prep_verified_on det_clusters_the amod_clusters_first num_clusters_nine pobj_on_clusters nn__Kappa dep_clusters_ prep__of pobj_of_068 det_clusters_the amod_clusters_remaining num_clusters_11 nsubjpass_annotated_clusters auxpass_annotated_were prep_annotated_by num_judge_one pobj_by_judge dep_judge_each ' +n,7736,'By increasing the size of the basic unit of translation , phrase-based machine translation does away with many of the problems associated with the original word-based formulation of statistical machine translation , in particular : The Brown et al ',Brown,'prep_does_By pcomp_By_increasing det_size_the dobj_increasing_size prep_size_of det_unit_the amod_unit_basic pobj_of_unit prep_unit_of pobj_of_translation amod_translation_phrase-based nn_translation_machine nsubj_does_translation prt_does_away prep_does_with pobj_with_many prep_many_of det_problems_the pobj_of_problems partmod_problems_associated prep_associated_with det_formulation_the amod_formulation_original amod_formulation_word-based pobj_with_formulation prep_formulation_of amod__statistical nn__machine nn__translation pobj_of_ prep_associated_in pobj_in_particular det_al_The nn_al_Brown nn_al_et dep_particular_al ' +o,7737,'1998 ; Goldman and Zhou , 2000 -RRB- that has been used previously to train classifiers in applications like word-sense disambiguation , document classification and named-entity recognition and apply this method to the more complex domain of statistical parsing ',Yarowsky,'dep_1998_Goldman cc_Goldman_and conj_Goldman_Zhou appos_Goldman_2000 nsubjpass_used_that aux_used_has auxpass_used_been rcmod_Goldman_used advmod_used_previously aux_train_to xcomp_used_train dobj_train_classifiers prep_train_in pobj_in_applications prep_applications_like amod__word-sense nn__disambiguation pobj_like_ conj_train_document dep__classification dobj_document_ cc__and amod__named-entity nn__recognition conj__ cc_train_and conj_train_apply det_method_this dobj_apply_method prep_apply_to det_domain_the advmod_complex_more amod_domain_complex pobj_to_domain prep_domain_of amod_parsing_statistical pobj_of_parsing ' +o,7738,'ROUGE , a recall-oriented evaluation package for automatic summarization ',Lin,'nsubj_ROUGE_ det_package_a amod_package_recall-oriented nn_package_evaluation appos__package prep_package_for amod_summarization_automatic pobj_for_summarization ' +o,7739,'Automatically determining the degree of antonymy between words has many uses including detecting and generating paraphrases -LRB- The dementors caught Sirius Black \\/ Black could not escape the dementors -RRB- and detecting contradictions -LRB- Kyoto has a predominantly wet climate \\/ It is mostly dry in Kyoto -RRB- ',Marneffe,'advmod_determining_Automatically det_degree_the dobj_determining_degree prep_degree_of pobj_of_antonymy prep_antonymy_between pobj_between_words dep_determining_has amod_uses_many dobj_has_uses prep_uses_including pcomp_including_detecting cc_detecting_and conj_detecting_generating dobj_detecting_paraphrases det_dementors_The nsubj_caught_dementors dep_paraphrases_caught nn_Black_Sirius nn_Black_Black nn_Black_\\/ nsubj_escape_Black aux_escape_could neg_escape_not ccomp_caught_escape det_dementors_the dobj_escape_dementors cc_detecting_and conj_detecting_detecting nsubj_determining_contradictions amod_contradictions_ nsubj_has_Kyoto dep__has det_\\/_a advmod_wet_predominantly amod_\\/_wet nn_\\/_climate dobj_has_\\/ nsubj_dry_It cop_dry_is advmod_dry_mostly rcmod_\\/_dry prep_dry_in pobj_in_Kyoto ' +p,7740,'Arguably the most widely used is the mutual information ',Hindle,'advmod_most_Arguably det_most_the nsubj_information_most advmod_used_widely amod_most_used cop_information_is det_information_the amod_information_mutual ' +p,7741,'An especially well-founded framework is maximum entropy ',Berger,'det_framework_An advmod_well-founded_especially amod_framework_well-founded nsubj_entropy_framework cop_entropy_is amod_entropy_maximum ' +o,7742,'It is used , as tagging mode -LRB- in English and morphological analysis nlodel -LRB- word segmentation and tagging -RRB- in Japanese ',Cutting,'nsubjpass_used_It auxpass_used_is prep_used_as pcomp_as_tagging dobj_tagging_mode dep_mode_in nn__English pobj_in_ cc__and amod_segmentation_morphological nn_segmentation_analysis nn_segmentation_nlodel nn_segmentation_word conj__segmentation cc_segmentation_and conj_segmentation_tagging prep_mode_in amod__Japanese pobj_in_ ' +o,7743,'This dependency graph is partitioned into treelets ; like , we assume a uniform probability distribution over all partitions ',Koehn,'det_graph_This amod_graph_dependency nsubjpass_partitioned_graph auxpass_partitioned_is prep_partitioned_into pobj_into_treelets prep_assume_like pobj_like_ nsubj_assume_we parataxis_partitioned_assume det_distribution_a amod_distribution_uniform nn_distribution_probability dobj_assume_distribution prep_distribution_over det_partitions_all pobj_over_partitions ' +o,7744,'Much of the recent work in word alignment has focussed on improving the word alignment quality through better modeling or alternative approaches to training ',Och,'nsubj_focussed_Much prep_Much_of det_work_the amod_work_recent pobj_of_work prep_work_in nn_alignment_word pobj_in_alignment aux_focussed_has prep_focussed_on pcomp_on_improving det_quality_the nn_quality_word nn_quality_alignment dobj_improving_quality prep_improving_through amod__better nn__modeling pobj_through_ cc__or amod_approaches_alternative conj__approaches prep_improving_to nn__training pobj_to_ ' +o,7745,'Setting the gradient to zero yields the usual maximum entropy constraints , except that in this case the empirical values are themselves expectations -LRB- over all derivations leading to each gold standard dependency structure -RRB- ',Berger,'det_gradient_the dobj_Setting_gradient prep_Setting_to pobj_to_zero dep_Setting_yields det__the amod__usual amod__maximum amod__entropy nn__constraints nsubj_Setting_ mark_expectations_except dep_expectations_that prep_expectations_in det_case_this pobj_in_case det_values_the amod_values_empirical nsubj_expectations_values cop_expectations_are advmod_expectations_themselves dep__expectations dep_expectations_over det_derivations_all nsubj_leading_derivations pcomp_over_leading prep_leading_to det_structure_each amod_structure_gold amod_structure_standard nn_structure_dependency pobj_to_structure ' +n,7746,' examine the FS of the weighted log-likelihood ratio -LRB- WLLR -RRB- on the movie review dataset and achieves an accuracy of 871 \% , which is higher than the result reported by with the same dataset ',Pang,'nsubj_examine_ det_FS_the dobj_examine_FS prep_FS_of det_ratio_the amod_ratio_weighted amod_ratio_log-likelihood pobj_of_ratio abbrev_ratio_WLLR prep_ratio_on det_dataset_the nn_dataset_movie nn_dataset_review pobj_on_dataset cc_examine_and conj_examine_achieves det_accuracy_an dobj_achieves_accuracy prep_accuracy_of num_\%_871 pobj_of_\% nsubj_higher_which cop_higher_is rcmod_\%_higher mark_reported_than det_result_the nsubj_reported_result advcl_higher_reported prep_reported_by pobj_by_ prep_reported_with det_dataset_the amod_dataset_same pobj_with_dataset ' +o,7747,'Entropy , used in some part-of-speech tagging systems , is a measure of how much information is necessary to separate data ',Ratnaparkhi,'nsubj_measure_Entropy partmod_Entropy_used prep_used_in det__some amod__part-of-speech nn__tagging nn__systems pobj_in_ cop_measure_is det_measure_a prep_measure_of advmod_much_how dep_necessary_much nsubj_necessary_information cop_necessary_is pcomp_of_necessary prep_necessary_to amod_data_separate pobj_to_data ' +o,7748,'1 Introduction Most state-of-the-art wide-coverage parsers are based on the Penn Treebank , making such parsers highly tuned to newspaper text ',Marcus,'num_Introduction_1 amod_parsers_Most amod_parsers_state-of-the-art amod_parsers_wide-coverage nsubjpass_based_parsers auxpass_based_are dep_Introduction_based prep_based_on det__the nn__Penn nn__Treebank pobj_on_ xcomp_based_making amod_parsers_such nsubj_tuned_parsers advmod_tuned_highly xcomp_making_tuned prep_tuned_to nn_text_newspaper pobj_to_text ' +p,7749,'2 Related Work Supervised machine learning methods including Support Vector Machines -LRB- SVM -RRB- are often used in sentiment analysis and shown to be very promising ',Pang,'num_machine_2 nn_machine_Related nn_machine_Work nn_machine_Supervised nsubjpass_used_machine partmod_machine_learning dobj_learning_methods prep_methods_including nn_Machines_Support nn_Machines_Vector pobj_including_Machines abbrev_Machines_SVM auxpass_used_are advmod_used_often prep_used_in nn_analysis_sentiment pobj_in_analysis cc_used_and conj_used_shown aux_promising_to cop_promising_be advmod_promising_very xcomp_shown_promising ' +o,7750,'The component features are weighted to minimize a translation error criterion on a development set ',Och,'det_features_The nn_features_component nsubjpass_weighted_features auxpass_weighted_are aux_minimize_to xcomp_weighted_minimize det_criterion_a nn_criterion_translation nn_criterion_error dobj_minimize_criterion prep_minimize_on det_set_a nn_set_development pobj_on_set ' +p,7751,'Turney also reported good result without domain customization ',Turney,'advmod_reported_Turney advmod_reported_also amod_result_good dobj_reported_result prep_reported_without nn_customization_domain pobj_without_customization ' +o,7752,'The node mapping function f for the entire tree thus has a different role from the alignment function in the IBM statistical translation model ; the role of the latter includes the linear ordering of words in the target string ',Brown,'det_mapping_The amod_mapping_node nsubj_function_mapping nsubj_has_f prep_f_for det_tree_the amod_tree_entire pobj_for_tree advmod_has_thus ccomp_function_has det_role_a amod_role_different dobj_has_role prep_role_from det_function_the amod_function_alignment pobj_from_function prep_function_in det__the nn__IBM amod__statistical nn__translation nn__model pobj_in_ det_role_the nsubj_includes_role prep_role_of det_latter_the pobj_of_latter parataxis_function_includes det_ordering_the amod_ordering_linear dobj_includes_ordering prep_ordering_of pobj_of_words prep_words_in det_string_the nn_string_target pobj_in_string ' +o,7753,'8 Conclusions In this paper , we developed probability models for the multi-level transfer rules presented in , showed how to acquire larger rules that crucially condition on more syntactic context , and how to pack multiple derivations , including interpretations of unaligned words , into derivation forests ',Galley,'num_Conclusions_8 prep_developed_In det_paper_this pobj_In_paper nsubj_developed_we dep_Conclusions_developed nn_models_probability dobj_developed_models prep_developed_for det_rules_the amod_rules_multi-level nn_rules_transfer pobj_for_rules partmod_rules_presented prep_presented_in pobj_in_ dep_presented_showed advmod_acquire_how aux_acquire_to ccomp_showed_acquire amod_rules_larger dobj_acquire_rules complm_condition_that advmod_condition_crucially dep_acquire_condition prep_condition_on advmod_syntactic_more amod_context_syntactic pobj_on_context cc_showed_and advmod_pack_how aux_pack_to conj_showed_pack amod_derivations_multiple dobj_pack_derivations conj_showed_including pobj_including_interpretations prep_interpretations_of amod_words_unaligned pobj_of_words prep_presented_into nn_forests_derivation pobj_into_forests ' +n,7754,'While both and propose models which use the parameters of the generative model but train to optimize a discriminative criteria , neither proposes training algorithms which are computationally tractable enough to be used for broad coverage parsing ',Klein,'mark_propose_While preconj__both nsubj_propose_ cc__and conj__ advcl_proposes_propose dobj_propose_models nsubj_use_which rcmod_models_use det_parameters_the dobj_use_parameters prep_parameters_of det_model_the amod_model_generative pobj_of_model cc_use_but conj_use_train aux_optimize_to xcomp_train_optimize det_criteria_a amod_criteria_discriminative dobj_optimize_criteria nsubj_proposes_neither nn_algorithms_training dobj_proposes_algorithms nsubj_tractable_which cop_tractable_are advmod_tractable_computationally rcmod_algorithms_tractable advmod_tractable_enough aux_used_to auxpass_used_be xcomp_tractable_used prep_used_for amod_parsing_broad nn_parsing_coverage pobj_for_parsing ' +o,7755,'In an attempt to provide a quantitative evaluation of our results , for each of the 12 ambiguous words shown in table 1 we manually assigned the top 30 first-order associations to one of the two senses provided by ',Yarowsky,'det_attempt_an pobj_In_attempt aux_provide_to infmod_attempt_provide det_evaluation_a amod_evaluation_quantitative dobj_provide_evaluation prep_evaluation_of poss_results_our pobj_of_results prep_provide_for pobj_for_each prep_each_of det_words_the num_words_12 amod_words_ambiguous pobj_of_words partmod_words_shown prep_shown_in pobj_in_table num_table_1 nsubj_assigned_we advmod_assigned_manually rcmod_table_assigned det_associations_the amod_associations_top num_associations_30 amod_associations_first-order dobj_assigned_associations prep_associations_to pobj_to_one prep_associations_of det_senses_the num_senses_two pobj_of_senses prep_assigned_provided dep_provided_by ' +p,7756,'In our experiments , we follow Lowe and McDonald in using the well-known log-likelihood ratio G 2 ',Dunning,'prep_follow_In poss_experiments_our pobj_In_experiments nsubj_follow_we nn__Lowe cc_Lowe_and conj_Lowe_McDonald dobj_follow_ prep_follow_in pcomp_in_using det_ratio_the amod_ratio_well-known amod_ratio_log-likelihood dobj_using_ratio tmod_using_G num_G_2 ' +o,7757,'In future work , we will expand all of the above types of features and employ techniques to reduce dimensionality along the lines suggested in and ',Yarowsky,'prep_expand_In amod_work_future pobj_In_work nsubj_expand_we aux_expand_will dobj_expand_all prep_all_of det_types_the amod_types_above pobj_of_types prep_types_of pobj_of_features cc_expand_and conj_expand_employ nsubj_reduce_techniques aux_reduce_to xcomp_employ_reduce advmod_reduce_dimensionality prep_reduce_along det_lines_the pobj_along_lines partmod_lines_suggested prep_suggested_in nn_and_ pobj_in_and ' +o,7758,'Similar to , each word in the hypothesis is assigned with a rank-based score of 1 \\/ -LRB- 1 -RRB- r + , where r is the rank of the hypothesis ',Rosti,'aux__to xcomp_Similar_ det_word_each nsubjpass_assigned_word prep_word_in det_hypothesis_the pobj_in_hypothesis auxpass_assigned_is ccomp__assigned prep_assigned_with det_score_a amod_score_rank-based pobj_with_score prep_score_of number_\\/_1 num_+_\\/ appos_+_1 nn_+_r pobj_of_+ advmod_rank_where nsubj_rank_r cop_rank_is det_rank_the rcmod_+_rank prep_rank_of det_hypothesis_the pobj_of_hypothesis ' +o,7759,'22 Closed Challenge Setting The organization provided training , development and test sets derived from the standard sections of the Penn TreeBank and PropBank corpora ',Marcus,'num_organization_22 amod_organization_Closed nn_organization_Challenge nn_organization_Setting nn_organization_The nsubj_provided_organization nn_sets_training conj_training_development cc_training_and conj_training_test dobj_provided_sets partmod_sets_derived prep_derived_from det_sections_the amod_sections_standard pobj_from_sections prep_sections_of det__the nn__Penn nn__TreeBank pobj_of_ cc__and nn_corpora_PropBank nn_corpora_ conj__corpora ' +o,7760,' and used POS-trigrams as well ',Collins,'cc__and conj__used dobj_used_POS-trigrams advmod_well_as advmod_used_well ' +p,7761,'It is often straightforward to obtain large amounts of unlabeled data , making semi-supervised approaches appealing ; previous work on semisupervised methods for dependency parsing includes ',Smith,'nsubj_straightforward_It cop_straightforward_is advmod_straightforward_often aux_obtain_to xcomp_straightforward_obtain amod_amounts_large dobj_obtain_amounts prep_amounts_of amod_data_unlabeled pobj_of_data xcomp_obtain_making amod_approaches_semi-supervised nsubj_appealing_approaches xcomp_making_appealing amod_work_previous nsubj_includes_work prep_work_on amod_methods_semisupervised pobj_on_methods prep_methods_for amod_parsing_dependency pobj_for_parsing parataxis_straightforward_includes ' +o,7762,'We build phrase translations by first acquiring bidirectional GIZA + + alignments , and using Moses grow-diag alignment symmetrization heuristic1 We set the maximum phrase length to a large value -LRB- 10 -RRB- , because some segmenters described later in this paper will result in shorter 1In our experiments , this heuristic consistently performed better than the default , grow-diag-final ',Och,'nsubj_build_We ccomp_performed_build nn_translations_phrase dobj_build_translations prep_build_by advmod_acquiring_first pcomp_by_acquiring amod_alignments_bidirectional amod_alignments_GIZA cc_GIZA_+ conj_GIZA_+ nn_alignments_ dobj_acquiring_alignments cc_acquiring_and conj_acquiring_using nn_heuristic1_Moses amod_heuristic1_grow-diag amod_heuristic1_alignment nn_heuristic1_symmetrization dobj_using_heuristic1 nsubj_set_We rcmod_heuristic1_set det_length_the amod_length_maximum nn_length_phrase dobj_set_length prep_set_to det_value_a amod_value_large pobj_to_value appos_value_10 mark_result_because det_segmenters_some nsubj_result_segmenters partmod_segmenters_described advmod_described_later prep_described_in det_paper_this pobj_in_paper aux_result_will advcl_set_result prep_result_in amod_1In_shorter pobj_in_1In poss_experiments_our dobj_acquiring_experiments det_heuristic_this nsubj_performed_heuristic advmod_performed_consistently advmod_performed_better prep_performed_than det_default_the pobj_than_default advmod_performed_grow-diag-final ' +o,7763,'The model is defined mathematically as following : p -LRB- f e -RRB- = 1Zexp nsummationdisplay i = 1 ihi -LRB- f , e -RRB- -LRB- 1 -RRB- where i is a vector of weights determined during a tuning process , and hi is the feature function ',Koehn,'det_model_The nsubjpass_defined_model auxpass_defined_is advmod_defined_mathematically xcomp_defined_ prep__as pobj_as_following nsubj_=_p dep_-LRB-_f dep_f_e parataxis_defined_= amod_i_1Zexp nn_i_nsummationdisplay nsubj_=_i ccomp_=_= num_ihi_1 dobj_=_ihi dep_ihi_f appos_f_e appos_ihi_1 advmod_vector_where nsubj_vector_i cop_vector_is det_vector_a rcmod_ihi_vector prep_vector_of pobj_of_weights partmod_weights_determined prep_determined_during det_process_a amod_process_tuning pobj_during_process cc_defined_and nsubj_function_hi cop_function_is det_function_the nn_function_feature conj_defined_function ' +o,7764,'The next two methods are heuristic -LRB- H -RRB- in and grow-diagonal -LRB- GD -RRB- proposed in ',Koehn,'det_methods_The amod_methods_next num_methods_two nsubj_proposed_methods cop_heuristic_are rcmod_methods_heuristic dep_heuristic_H prep_heuristic_in pobj_in_ cc__and conj__grow-diagonal abbrev_methods_GD prt_proposed_in ' +o,7765,'We measure translation performance by the BLEU and METEOR scores with multiple translation references ',Papineni,'nsubj_measure_We nn_performance_translation dobj_measure_performance prep_measure_by det__the nn__BLEU pobj_by_ cc__and nn_scores_METEOR nn_scores_ conj__scores prep_measure_with amod_references_multiple nn_references_translation pobj_with_references ' +o,7766,'3 Implementation 31 Pronoun resolution model We built a machine learning based pronoun resolution engine using a Maximum Entropy ranker model , similar with Denis and Baldridges model ',Berger,'num_model_3 nn_model_Implementation num_model_31 nn_model_Pronoun nn_model_resolution nsubj_We_model partmod_We_built det_machine_a dobj_built_machine partmod_machine_learning prep_learning_based nn_engine_pronoun nn_engine_resolution dep_based_engine partmod_engine_using det__a amod__Maximum nn__Entropy nn__ranker nn__model dobj_using_ amod__similar prep_similar_with pobj_with_Denis cc_Denis_and conj_Denis_Baldridges nn__model dobj_We_ ' +o,7767,'1 Introduction In the part-of-speech hterature , whether taggers are based on a rule-based approach , , , or on a statistical one , , , , , , there is a debate as to whether more attention should be paid to lexical probabilities rather than contextual ones ',Cutting,'num_Introduction_1 dep_Introduction_In det_hterature_the amod_hterature_part-of-speech pobj_In_hterature dep_based_whether nsubjpass_based_taggers auxpass_based_are dep_hterature_based prep_based_on det__a amod__rule-based nn__approach nn__ appos__ pobj_on_ dep_Introduction_or dep_Introduction_on det__a amod__statistical num__one pobj_on_ nn__ appos__ appos__ appos__ dep_on_ expl_is_there rcmod__is det_debate_a nsubj_is_debate prep_debate_as dep_as_to complm_paid_whether amod_attention_more nsubjpass_paid_attention aux_paid_should auxpass_paid_be pcomp_as_paid prep_paid_to amod_probabilities_lexical pobj_to_probabilities cc_probabilities_rather dep_rather_than amod_ones_contextual conj_probabilities_ones ' +o,7768,'Accordingly, in this section we describe a set of experiments which extends the work of (Way and Gough, 2005) by evaluating the Marker-based EBMT system of (Gough & Way, 2004b) against a phrase-based SMT system built using the following components: Giza++, to extract the word-level correspondences; The Giza++ word alignments are then refined and used to extract phrasal alignments ((Och & Ney, 2003); or (Koehn et al. , 2003) for a more recent implementation); Probabilities of the extracted phrases are calculated from relative frequencies; The resulting phrase translation table is passed to the Pharaoh phrase-based SMT decoder which along with SRI language modelling toolkit5 performs translation.',Koehn,'' +p,7769,'In the II , OO , and OI scenarios , succeeded in improving the parser performance only when a reranker was used to reorder the 50-best list of the generative parser , with a seed size of 40K sentences ',McClosky,'prep_succeeded_In det_OO_the num_OO_II pobj_In_OO cc_OO_and nn_scenarios_OI conj_OO_scenarios nsubj_succeeded_ prep_succeeded_in dep_performance_improving det_performance_the nn_performance_parser pcomp_in_performance advmod_performance_only advmod_used_when det_reranker_a nsubjpass_used_reranker auxpass_used_was advcl_performance_used aux_reorder_to xcomp_used_reorder det_list_the amod_list_50-best dobj_reorder_list prep_list_of det_parser_the amod_parser_generative pobj_of_parser prep_succeeded_with det_size_a nn_size_seed pobj_with_size prep_size_of num_sentences_40K pobj_of_sentences ' +o,7770,'Instead of using Inversion Transduction Grammar -LRB- ITG -RRB- directly , we will discuss an ITG extension to accommodate gapping ',Wu,'dep_of_Instead advcl_discuss_of pcomp_of_using nn_Grammar_Inversion nn_Grammar_Transduction dobj_using_Grammar abbrev_Grammar_ITG partmod_Grammar_ advmod__directly nsubj_discuss_we aux_discuss_will det_extension_an nn_extension_ITG nsubj_accommodate_extension aux_accommodate_to xcomp_discuss_accommodate dobj_accommodate_gapping ' +o,7771,'We use a simple , single parameter distribution , with = 80 throughout P -LRB- K m , e -RRB- = P -LRB- K m , l -RRB- K Word-to-Phrase Alignment Alignment is a Markov process that specifies the lengths of phrases and their alignment with source words P -LRB- aK1 , hK1 , K1 K , m , e -RRB- = Kproductdisplay k = 1 P -LRB- ak , hk , k ak1 , k1 , e -RRB- = Kproductdisplay k = 1 p -LRB- ak ak1 , hk ; l -RRB- d -LRB- hk -RRB- n -LRB- k ; eak -RRB- The actual word-to-phrase alignment -LRB- ak -RRB- is a firstorder Markov process , as in HMM-based word-toword alignment ',Berger,'nsubj_use_We det_distribution_a amod_distribution_simple amod_distribution_single nn_distribution_parameter dobj_use_distribution prep_use_with pcomp_with_= nsubj_=_80 prep_80_throughout pobj_throughout_P nn_m_K appos_P_m dep_m_e ccomp_=_= dobj_=_P nn_m_K dep_P_m dep_m_l nn_Alignment_K nn_Alignment_Word-to-Phrase nn_Alignment_Alignment nsubj_process_Alignment cop_process_is det_process_a nn_process_Markov rcmod_P_process nsubj_specifies_that rcmod_process_specifies det_lengths_the nsubj_=_lengths prep_lengths_of pobj_of_phrases cc_phrases_and poss_P_their amod_P_alignment prep_alignment_with nn_words_source pobj_with_words conj_phrases_P nn_hK1_aK1 appos_P_hK1 nn_K_K1 dep_hK1_K nn_e_m dep_hK1_e ccomp_specifies_= nn_k_Kproductdisplay nsubj_=_k ccomp_=_= num_P_1 dobj_=_P nn_hk_ak appos_P_hk nn_k1_k nn_k1_ak1 dep_hk_k1 dep_hk_e partmod_P_= nn_k_Kproductdisplay nsubj_p_k dep_p_= num_p_1 ccomp_=_p nn_ak1_ak appos_P_ak1 nn_l_hk dep_ak1_l dep_process_d appos_d_hk dep_process_n nn_eak_k appos_n_eak det_alignment_The amod_alignment_actual amod_alignment_word-to-phrase nsubj_process_alignment appos_alignment_ak cop_process_is det_process_a amod_process_firstorder nn_process_Markov rcmod_n_process amod_process_as prep_as_in amod__HMM-based amod__word-toword nn__alignment pobj_in_ ' +o,7772,'21 Synchronous derivations The derivations for syntactic dependency trees are the same as specified in , which are based on the shift-reduce style parser of ',Titov,'num_Synchronous_21 nsubj_derivations_Synchronous det_derivations_The nsubjpass_specified_derivations prep_derivations_for amod_trees_syntactic amod_trees_dependency pobj_for_trees auxpass_specified_are det_same_the dep_as_same advmod_specified_as ccomp_derivations_specified prep_specified_in pobj_in_ nsubjpass_based_which auxpass_based_are rcmod__based prep_based_on det_parser_the amod_parser_shift-reduce nn_parser_style pobj_on_parser prep_parser_of pobj_of_ ' +o,7773,'Furthermore , Bikel provides evidence that lexical information -LRB- in the form of bi-lexical dependencies -RRB- only makes a small contribution to the performance of parsing models such as ',Collins,'advmod_provides_Furthermore nn__Bikel nsubj_provides_ dobj_provides_evidence complm_makes_that amod_information_lexical nsubj_makes_information dep_information_in det_form_the pobj_in_form prep_form_of amod_dependencies_bi-lexical pobj_of_dependencies advmod_makes_only ccomp_provides_makes det_contribution_a amod_contribution_small dobj_makes_contribution prep_makes_to det_performance_the pobj_to_performance prep_performance_of pcomp_of_parsing dobj_parsing_models dep_as_such prep_parsing_as ' +o,7774,'The feature weights are tuned using minimum error rate training to optimize BLEU score on a held-out development set ',Och,'det_weights_The nn_weights_feature nsubjpass_tuned_weights auxpass_tuned_are dep_tuned_using amod__minimum nn__error nn__rate nn__training dobj_using_ aux_optimize_to xcomp_using_optimize amod_score_BLEU dobj_optimize_score prep_optimize_on det_set_a amod_set_held-out nn_set_development pobj_on_set ' +p,7775,'52 Maximum Entropy Maximum entropy classiflcation -LRB- MaxEnt , or ME , for short -RRB- is an alternative technique which has proven efiective in a number of natural language processing applications ',Berger,'num_classiflcation_52 nn_classiflcation_Maximum nn_classiflcation_Entropy nn_classiflcation_Maximum nn_classiflcation_entropy nsubj_technique_classiflcation dep_classiflcation_MaxEnt cc_MaxEnt_or conj_MaxEnt_ME prep_MaxEnt_for pobj_for_short cop_technique_is det_technique_an amod_technique_alternative nsubj_proven_which aux_proven_has rcmod_technique_proven dep_proven_efiective prep_proven_in det_number_a pobj_in_number prep_number_of amod_applications_natural nn_applications_language nn_applications_processing pobj_of_applications ' +o,7776,'The feature weights were tuned on a heldout development set so as to maximize an equally weighted linear combination of BLEU and 1-TER using the minimum error training algorithm on a packed forest representation of the decoders hypothesis space ',Papineni,'det_weights_The nn_weights_feature nsubjpass_tuned_weights auxpass_tuned_were prep_tuned_on det_development_a amod_development_heldout pobj_on_development partmod_development_set advmod_as_so advmod_set_as aux_maximize_to xcomp_set_maximize det_combination_an advmod_combination_equally amod_combination_weighted amod_combination_linear dobj_maximize_combination prep_combination_of pobj_of_BLEU cc_combination_and num__1-TER conj_combination_ partmod__using det_algorithm_the amod_algorithm_minimum nn_algorithm_error nn_algorithm_training dobj_using_algorithm prep_algorithm_on det_representation_a amod_representation_packed nn_representation_forest pobj_on_representation prep_representation_of det__the amod__decoders nn__hypothesis nn__space pobj_of_ ' +o,7777,'24 Formalization of As mentioned earlier , our model is equivalent to that presented in , and can be viewed as a formal version of his model2 In his presentation , the adapation is done through feature augmentation ','Daume III','num__24 nn__Formalization dep__of mark_mentioned_As advcl_equivalent_mentioned advmod_mentioned_earlier poss_model_our nsubjpass_equivalent_model cop_equivalent_is dep__equivalent prep_equivalent_to pobj_to_that partmod_that_presented prep_presented_in pobj_in_ cc_equivalent_and aux_viewed_can auxpass_viewed_be conj_equivalent_viewed prep_viewed_as det_version_a amod_version_formal pobj_as_version prep_version_of poss_model2_his pobj_of_model2 prep_done_In poss_presentation_his pobj_In_presentation det_adapation_the nsubjpass_done_adapation auxpass_done_is dep__done prep_done_through nn_augmentation_feature pobj_through_augmentation ' +o,7778,'In general , previous work in opinion mining includes document level sentiment classification using supervised and unsupervised methods , machine learning techniques and sentiment classification considering rating scales , and scoring of features ',Turney,'prep_includes_In amod_work_general amod_work_previous pobj_In_work prep_work_in nn_mining_opinion pobj_in_mining nn_classification_document nn_classification_level nn_classification_sentiment dobj_includes_classification partmod_classification_using dobj_using_supervised dep_supervised_ cc_supervised_and amod_methods_unsupervised conj_supervised_methods nsubj_includes_ appos__machine partmod_machine_learning dobj_learning_techniques cc_techniques_and nn_classification_sentiment conj_techniques_classification dep_learning_considering nn_scales_rating dobj_considering_scales advmod_considering_ cc_considering_and conj_considering_scoring prep_scoring_of nn__features pobj_of_ ' +o,7779,'The k-best list is also frequently used in discriminative learning to approximate the whole set of candidates which is usually exponentially large ',Och,'det_list_The amod_list_k-best nsubjpass_used_list auxpass_used_is advmod_used_also advmod_used_frequently prep_used_in pobj_in_discriminative partmod_discriminative_learning aux_approximate_to xcomp_learning_approximate det_set_the amod_set_whole dobj_approximate_set prep_set_of pobj_of_candidates nsubj_large_which cop_large_is advmod_large_usually advmod_large_exponentially rcmod_candidates_large ' +o,7780,'These dependencies differ from those used by Liu and Gildea (2005), in that they are extracted according to the rules of the LFG grammar and they are labelled with a type of grammatical relation that connects the head and the modifier, such as subject, determiner, etc. The presence of grammatical relation labels adds another layer of important linguistic information into the comparison and allows us to account for partial matches, for example when a lexical item finds itself in a correct relation but with an incorrect partner.',Liu,'' +o,7781,'We use discourse-level feature predicates in a maximum entropy classifier with binary and n-class classification to select referring expressions from a list ',Berger,'nsubj_use_We amod_predicates_discourse-level nn_predicates_feature dobj_use_predicates prep_use_in det__a amod__maximum amod__entropy nn__classifier pobj_in_ prep__with amod_classification_binary cc_binary_and conj_binary_n-class pobj_with_classification aux_select_to xcomp_use_select xcomp_select_referring dobj_referring_expressions prep_referring_from det_list_a pobj_from_list ' +p,7782,'Yarowsky has proposed an algorithm that requires as little user input as one seed word per sense to start the training process ',Yarowsky,'advmod_proposed_Yarowsky aux_proposed_has det_algorithm_an dobj_proposed_algorithm nsubj_requires_that rcmod_algorithm_requires prep_requires_as amod_input_little nn_input_user pobj_as_input mark_start_as num_word_one nn_word_seed nsubj_start_word prep_word_per pobj_per_sense aux_start_to dep_proposed_start det_process_the nn_process_training dobj_start_process ' +o,7783,'1 Introduction In this paper , we present an approach for extracting the named entities -LRB- NE -RRB- of natural language inputs which uses the maximum entropy -LRB- ME -RRB- framework ',Berger,'num_Introduction_1 prep_present_In det_paper_this pobj_In_paper nsubj_present_we dep_Introduction_present det_approach_an dobj_present_approach prep_present_for pcomp_for_extracting det_entities_the amod_entities_named dobj_extracting_entities abbrev_entities_NE prep_entities_of amod_inputs_natural nn_inputs_language pobj_of_inputs nsubj_uses_which rcmod_inputs_uses det_framework_the amod_framework_maximum nn_framework_entropy abbrev_framework_ME dobj_uses_framework ' +o,7784,'Three recent papers in this area are Church and Hanks , , and Smadja and McKeown ',Hindle,'num_papers_Three amod_papers_recent nsubj__papers prep_papers_in det_area_this pobj_in_area cop__are nn__Church cc_Church_and conj_Church_Hanks advmod__ amod__ cc__and nn__Smadja cc_Smadja_and conj_Smadja_McKeown conj__ ' +n,7785,'Turneys method did not work well although they reported 80 \% accuracy in ',Turney,'amod_method_Turneys nsubj_work_method aux_work_did neg_work_not advmod_work_well mark_reported_although nsubj_reported_they advcl_work_reported dep_\%_80 amod_accuracy_\% dobj_reported_accuracy prep_reported_in ' +o,7786,'22 Evaluation of Acquisition Algorithms Many methods for automatic acquisition of rules have been suggested in recent years , ranging from distributional similarity to finding shared contexts ',Barzilay,'num_Evaluation_22 nsubjpass_suggested_Evaluation prep_Evaluation_of nn_methods_Acquisition nn_methods_Algorithms amod_methods_Many pobj_of_methods prep_methods_for amod_acquisition_automatic pobj_for_acquisition prep_acquisition_of pobj_of_rules aux_suggested_have auxpass_suggested_been prep_suggested_in amod_years_recent pobj_in_years xcomp_suggested_ranging prep_ranging_from amod_similarity_distributional pobj_from_similarity prep_ranging_to pcomp_to_finding amod_contexts_shared dobj_finding_contexts ' +p,7787,'Thus , over the past few years , along with advances in the use of learning and statistical methods for acquisition of full parsers , significant progress has been made on the use of statistical learning methods to recognize shallow parsing patterns syntactic phrases or words that participate in a syntactic relationship ',Ramshaw,'advmod_made_Thus prep_made_over det_years_the amod_years_past amod_years_few pobj_over_years advmod_made_along dep_along_with pobj_with_advances prep_advances_in det_use_the pobj_in_use prep_use_of pobj_of_learning cc_use_and amod_methods_statistical conj_use_methods prep_methods_for pobj_for_acquisition prep_acquisition_of amod__full nn__parsers pobj_of_ amod_progress_significant nsubjpass_made_progress aux_made_has auxpass_made_been prep_made_on det_use_the pobj_on_use prep_use_of amod_methods_statistical nn_methods_learning pobj_of_methods aux_recognize_to xcomp_made_recognize nn_patterns_shallow nn_patterns_parsing dobj_recognize_patterns amod_phrases_syntactic tmod_recognize_phrases cc_phrases_or conj_phrases_words nsubj_participate_that dep_phrases_participate prep_participate_in det_relationship_a amod_relationship_syntactic pobj_in_relationship ' +o,7788,'TER-based : TER-based word alignment method is an extension of multiple string matching algorithm based on Levenshtein edit distance ',Rosti,'dep_extension_TER-based amod__TER-based nn__word nn__alignment nn__method nsubj_extension_ cop_extension_is det_extension_an prep_extension_of amod_algorithm_multiple nn_algorithm_string amod_algorithm_matching pobj_of_algorithm prep_algorithm_based dep_based_on nn__Levenshtein nn__edit nn__distance pobj_on_ ' +o,7789,'This is due to the reason that Telugu is comparitively a high entropy language than English ',Brown,'nsubj_due_This cop_due_is prep_due_to det_reason_the pobj_to_reason complm_language_that nn__Telugu nn__ nsubj_language_ cop_language_is advmod_language_comparitively det_language_a amod_language_high nn_language_entropy ccomp_due_language prep_language_than pobj_than_English ' +p,7790,'Such a method alleviates the problem of creating templates from examples which would be used in an ulterior phase of generation ',Barzilay,'predet_method_Such det_method_a nsubj_alleviates_method det_problem_the dobj_alleviates_problem prep_problem_of pcomp_of_creating dobj_creating_templates prep_creating_from pobj_from_examples nsubjpass_used_which aux_used_would auxpass_used_be rcmod_examples_used prep_used_in det_phase_an nn_phase_ulterior pobj_in_phase prep_phase_of nn__generation pobj_of_ ' +o,7791,'-LRB- 3 -RRB- s in Equation 1 are the weights of different feature functions , learned to maximize development set BLEU scores using a method similar to ',Och,'dep_weights_3 nsubj_weights_s prep_s_in pobj_in_Equation num_Equation_1 cop_weights_are det_weights_the prep_weights_of amod_functions_different nn_functions_feature pobj_of_functions partmod_functions_learned aux_maximize_to xcomp_learned_maximize nn_set_development dobj_maximize_set amod_scores_BLEU nsubj_using_scores dep_weights_using det_method_a nsubj_similar_method xcomp_using_similar prep_similar_to ' +o,7792,'Since adjectives have been a focus of previous work in sentiment detection 13 , we looked at the performance of using adjectives alone ',Turney,'mark_been_Since nsubj_been_adjectives aux_been_have advcl_looked_been det_focus_a dobj_been_focus prep_focus_of amod_work_previous pobj_of_work prep_been_in nn__sentiment nn__detection pobj_in_ num__13 nsubj_looked_we prep_looked_at det_performance_the pobj_at_performance prep_performance_of pcomp_of_using dobj_using_adjectives advmod_using_alone ' +o,7793,'The other recipe that is currently used on a large scale is to measure the performance of a parser on existing treebanks , such as WSJ , and assume that the accuracy measure will carry over to the domains of interest ',Marcus,'det_recipe_The amod_recipe_other nsubj_is_recipe nsubjpass_used_that auxpass_used_is advmod_used_currently rcmod_recipe_used prep_used_on det_scale_a amod_scale_large pobj_on_scale aux_measure_to xcomp_is_measure det_performance_the dobj_measure_performance prep_performance_of det_parser_a pobj_of_parser prep_parser_on amod_treebanks_existing pobj_on_treebanks dep_as_such prep_performance_as nn__WSJ pobj_as_ cc_measure_and conj_measure_assume complm_carry_that det_measure_the nn_measure_accuracy nsubj_carry_measure aux_carry_will ccomp_assume_carry prt_carry_over prep_carry_to det_domains_the pobj_to_domains prep_domains_of pobj_of_interest ' +o,7794,'Previous SMT systems used a word-based translation model which assumes that a sentence can be translated into other languages by translating each word into one or more words in the target language ',Brown,'amod__Previous nn__SMT nn__systems nsubj_used_ det_model_a amod_model_word-based nn_model_translation dobj_used_model nsubj_assumes_which rcmod_model_assumes complm_translated_that det_sentence_a nsubjpass_translated_sentence aux_translated_can auxpass_translated_be ccomp_assumes_translated prep_translated_into amod_languages_other pobj_into_languages prep_translated_by pcomp_by_translating det_word_each dobj_translating_word prep_translating_into num_words_one cc_one_or conj_one_more pobj_into_words prep_words_in det_language_the nn_language_target pobj_in_language ' +o,7795,'on test BLEU BP BLEU BP pair-CI 95\% BLEU BP 3 01 03 32.98 0.92 33.03 0.93 [ -0.23, +0.34] 33.60 0.93 4 01 04 33.44 0.93 33.46 0.93 [ -0.26, +0.29] 34.97 0.94 5 01 05 33.07 0.92 33.14 0.93 [ -0.29, +0.43] 34.33 0.93 6 01 06 32.86 0.92 33.53 0.93 [+0.26, +1.08] 34.43 0.93 7 01 07 33.08 0.93 33.51 0.93 [+0.04, +0.82] 34.49 0.93 8 01 08 33.12 0.93 33.47 0.93 [ -0.06, +0.75] 34.50 0.94 9 01 09 33.15 0.93 33.22 0.93 [ -0.35, +0.51] 34.68 0.93 10 01 10 33.01 0.93 33.59 0.94 [+0.18, +0.96] 34.79 0.94 11 01 11 32.84 0.94 33.40 0.94 [+0.13, +0.98] 34.76 0.94 12 01 12 32.73 0.93 33.49 0.94 [+0.34, +1.18] 34.83 0.94 13 01 13 32.71 0.93 33.54 0.94 [+0.39, +1.26] 34.91 0.94 14 01 14 32.66 0.93 33.69 0.94 [+0.58, +1.47] 34.97 0.94 15 01 15 32.47 0.93 33.57 0.94 [+0.63, +1.57] 34.99 0.94 16 01 16 32.51 0.93 33.62 0.94 [+0.62, +1.59] 35.00 0.94 3.2 Non-Uniform System Prior Weights As pointed out in Section 2.1, a useful property of the MBR-like system selection method is that system prior weights can easily be trained using the Minimum Error Rate Training (Och, 2003).',Och,'' +o,7796,'109 machine translation evaluation , paraphraserecognition , and automatic grading ',Banerjee,'num_machine_109 nn__translation nn__evaluation dep_machine_ nn__paraphraserecognition conj__ cc__and amod__automatic nn__grading conj__ ' +o,7797,' proposed the perceptron as an alternative to the CRF method for HMM-style taggers ',Collins,'nsubj_proposed_ det_perceptron_the dobj_proposed_perceptron prep_proposed_as det_alternative_an pobj_as_alternative prep_proposed_to det_method_the nn_method_CRF pobj_to_method prep_method_for amod_taggers_HMM-style pobj_for_taggers ' +o,7798,'4 Evaluation The evaluation is conducted with all four corpora from Bakeoff-3 , as summarized in Table 1 with corpus size in number of characters ',Levow,'num_Evaluation_4 det_evaluation_The nsubjpass_conducted_evaluation auxpass_conducted_is dep_Evaluation_conducted prep_conducted_with det_corpora_all num_corpora_four pobj_with_corpora prep_conducted_from amod__Bakeoff-3 pobj_from_ mark_summarized_as advcl_conducted_summarized prep_summarized_in pobj_in_Table num_Table_1 prep_summarized_with nn_size_corpus pobj_with_size prep_size_in pobj_in_number prep_number_of pobj_of_characters ' +o,7799,'Our proposal is a first order linear model that relies on an online averaged Perceptron for learning and an extended Eisner algorithm for the joint parsing inference ',Collins,'poss_proposal_Our nsubj_model_proposal cop_model_is det_model_a amod_model_first nn_model_order amod_model_linear nsubj_relies_that rcmod_model_relies prep_relies_on det_Perceptron_an amod_Perceptron_online nn_Perceptron_averaged pobj_on_Perceptron prep_relies_for pcomp_for_learning dobj_learning_ cc_model_and det_algorithm_an amod_algorithm_extended nn_algorithm_Eisner conj_model_algorithm prep_algorithm_for det_inference_the amod_inference_joint nn_inference_parsing pobj_for_inference ' +o,7800,'As pointed out by , in effect this expression adds to c a small value that asymptotically approaches 05 as c approaches , and 0 as c approaches 0 ',Johnson,'mark_pointed_As advcl_adds_pointed prt_pointed_out prep_pointed_by pobj_by_ prep_pointed_in pobj_in_effect det_expression_this nsubj_adds_expression aux_c_to xcomp_adds_c det_value_a amod_value_small dobj_c_value nsubj_approaches_that advmod_approaches_asymptotically rcmod_value_approaches dobj_approaches_05 mark_approaches_as nsubj_approaches_c advcl_approaches_approaches cc_adds_and nsubj_approaches_0 prep_0_as dep_as_c conj_adds_approaches dobj_approaches_0 ' +o,7801,'We trained and tested the parser on the Wall Street Journal corpus of the Penn Treebank using the standard split : sections 2-21 were used for training , section 22 was used for development and tuning of parameters and features , and section 23 was used for testing ',Marcus,'nsubj_trained_We cc_trained_and conj_trained_tested det_parser_the dobj_tested_parser prep_tested_on det_corpus_the nn_corpus_Wall nn_corpus_Street nn_corpus_Journal nsubj_using_corpus prep_corpus_of det__the nn__Penn nn__Treebank pobj_of_ pcomp_on_using det_split_the amod_split_standard dobj_using_split nn_2-21_sections nsubjpass_used_2-21 auxpass_used_were parataxis_trained_used prep_used_for pobj_for_training nsubjpass_used_section num_section_22 auxpass_used_was ccomp_trained_used prep_used_for pobj_for_development cc_development_and conj_development_tuning prep_development_of pobj_of_parameters cc_parameters_and conj_parameters_features cc_trained_and nsubjpass_used_section num_section_23 auxpass_used_was conj_trained_used prep_used_for pobj_for_testing ' +o,7802,' s similar word list for eat misses these but includes sleep -LRB- ranked 6 -RRB- and sit -LRB- ranked 14 -RRB- , because these have similar subjects to eat ',Lin,'nsubj_similar_ dep_similar_s nn_list_word nsubj_misses_list prep_list_for pobj_for_eat ccomp_similar_misses nsubj_includes_these cc_includes_but ccomp_misses_includes dobj_includes_sleep number_6_ranked dep_sleep_6 cc_includes_and conj_includes_sit parataxis_sit_ranked dobj_ranked_14 mark_have_because nsubj_have_these advcl_similar_have amod_subjects_similar dobj_have_subjects aux_eat_to infmod_subjects_eat ' +o,7803,' , In addition to the usual issues involved with the complex annotation of data , we have come to terms with a number of issues that are specific to a highly inflected language with a rich history of traditional grammar ',Marcus,'advmod_come_ dep_In_ prep_come_In pobj_In_addition prep_addition_to det_issues_the amod_issues_usual pobj_to_issues partmod_issues_involved prep_involved_with det_annotation_the amod_annotation_complex pobj_with_annotation prep_annotation_of pobj_of_data nsubj_come_we aux_come_have prep_come_to pobj_to_terms prep_come_with det_number_a pobj_with_number prep_number_of pobj_of_issues nsubj_specific_that cop_specific_are rcmod_issues_specific prep_specific_to det_language_a advmod_inflected_highly amod_language_inflected pobj_to_language prep_language_with det_history_a amod_history_rich pobj_with_history prep_history_of amod_grammar_traditional pobj_of_grammar ' +o,7804,'Volume 17, Number 1 March 1991 References Lakoff, George and Johnson, Mark Metaphors We Live 8y University of Chicago Press 1980 MADCOW Committee (Hirschman, Lynette et al) Multi-Site Data Collection for a Spoken Language Corpus in Proceedings Speech and Natural Language Workshop February 1992 Grice, H. P. Logic and Conversation in P. Cole and J. L. Morgan, Speech Acts, New York: Academic Press, 1975 Pustejovsky, James The Generative Lexicon Computational Linguistics Volume 17, Number 4 December 1991 Hobbs, Jerry R. and Stickel, Mark Interpretation as Abduction in Proceedings of the 26th ACL June 1988 Bobrow, R. , Ingria, R. and Stallard, D. The Mapping Unit Approach to Subcategorization in Proceedings Speech and Natural Language Workshop February 1991 Hobbs, Jerry R. , and Martin, Paul Local Pragmatics in Proceedings, 10th International Joint Conference on Artificial Intelligence (IJCAI-87).',Hobbs,'' +o,7805,'Our statistical tagging model is adjusted from standard bi-grams using the Viterbi-search plus on-the-fly extra computing of lexical probabilities for unknown morphemes ',Cutting,'poss_model_Our amod_model_statistical amod_model_tagging nsubjpass_adjusted_model auxpass_adjusted_is prep_adjusted_from amod_bi-grams_standard pobj_from_bi-grams partmod_bi-grams_using det__the amod__Viterbi-search dobj_using_ prep__plus amod_computing_on-the-fly amod_computing_extra dep_plus_computing prep_computing_of amod_probabilities_lexical pobj_of_probabilities prep_probabilities_for amod_morphemes_unknown pobj_for_morphemes ' +o,7806,'However, by examining the Algorithm 4.2: Perceptron with local and non-local features (parameters: n, Ca, Cl) 0 until no more updates do for i 1 to L do8 >> >> >> >> >> < >> >> >> >> >> : {yn} = n-bestyl(xi,y) y = argmaxy{yn}a(xi,y) y = 2nd-besty{yn}a(xi,y) if y = yi & a(xi,yi)a(xi,y) Ca then = + a(xi,yi)a(xi,y) (A) else if a(xi,yi)a(xi,y) Ca then = + a(xi,yi)a(xi,y) (A) else (B) 8> < >: if y1 = yi then (y1 represents the best in {yn}) = + l(xi,yi)l(xi,y1) else if l(xi,yi)l(xi,y2) Cl then = + l(xi,yi)l(xi,y2) proofs in Collins (2002a), we can see that the essential condition for convergence is that the weights are always updated using some y (= y) that satises: (xi,yi)(xi,y) 0 ( C in the case of a perceptron with a margin).',Collins,'' +o,7807,'3 Inversion Transduction Grammars While our approach applies in principle to a variety of machine translation systems -LRB- phrase-based or syntactic -RRB- , we will use the inversion transduction grammar -LRB- ITG -RRB- approach of to facilitate comparison with previous work aswellastofocuson language model complexity ',Wu,'num_Grammars_3 nn_Grammars_Inversion nn_Grammars_Transduction mark_applies_While poss_approach_our nsubj_applies_approach advcl_use_applies prep_applies_in pobj_in_principle prep_applies_to det_variety_a pobj_to_variety prep_variety_of nn_systems_machine nn_systems_translation pobj_of_systems dep_systems_phrase-based cc_phrase-based_or conj_phrase-based_syntactic nsubj_use_we aux_use_will dep_Grammars_use det_approach_the nn_approach_inversion nn_approach_transduction nn_approach_grammar abbrev_approach_ITG dobj_use_approach prep_approach_of pobj_of_ aux_facilitate_to xcomp_use_facilitate dobj_facilitate_comparison prep_facilitate_with amod_complexity_previous nn_complexity_work nn_complexity_ nn_complexity_aswellastofocuson nn_complexity_language nn_complexity_model pobj_with_complexity ' +o,7808,'Each word i in the context vector of w is then weighted with a measure of its association with w. We chose the loglikelihood ratio test, (Dunning, 1993), to measure this association the context vectors of the target words are then translated with our general bilingual dictionary, leaving the weights unchanged (when several translations are proposed by the dictionary, we consider all of them with the same weight) the similarity of each source word s, for each target word t, is computed on the basis of the cosine measure the similarities are then normalized to yield a probabilistic translation lexicon, P(t|s).',Dunning,'' +o,7809,'We utilize maximum entropy -LRB- MaxEnt -RRB- model to design the basic classifier used in active learning for WSD and TC tasks ',Berger,'nsubj_utilize_We amod__maximum amod__entropy appos__MaxEnt nn__model dobj_utilize_ aux_design_to infmod__design det_classifier_the amod_classifier_basic dobj_design_classifier partmod_classifier_used prep_used_in amod_learning_active pobj_in_learning prep_learning_for nn_tasks_WSD cc_WSD_and conj_WSD_TC pobj_for_tasks ' +p,7810,'The variance semiring is essential for many interesting training paradigms such as deterministic 40 annealing , minimum risk , active and semi-supervised learning ',Jiao,'det_semiring_The amod_semiring_variance nsubj_essential_semiring cop_essential_is prep_essential_for amod_paradigms_many amod_paradigms_interesting nn_paradigms_training pobj_for_paradigms dep_as_such prep_paradigms_as amod_40_deterministic pobj_as_40 xcomp_essential_annealing nn_risk_ amod_risk_minimum nsubj_learning_risk dep_learning_ conj__active cc__and conj__semi-supervised xcomp_annealing_learning ' +o,7811,'Recent advances in these approaches include the use of a fully Bayesian HMM ',Johnson,'amod_advances_Recent nsubj_include_advances prep_advances_in det_approaches_these pobj_in_approaches det_use_the dobj_include_use prep_use_of det_HMM_a advmod_HMM_fully amod_HMM_Bayesian pobj_of_HMM ' +o,7812,'In the second pass , 5-gram and 6-gram zero-cutoff stupid-backoff language models estimated using 47 billion words of English newswire text are used to generate lattices for phrasal segmentation model rescoring ',Brants,'prep_used_In det_pass_the amod_pass_second pobj_In_pass amod_models_5-gram cc_5-gram_and conj_5-gram_6-gram amod_models_zero-cutoff amod_models_stupid-backoff amod_models_ nn_models_language nsubjpass_used_models partmod_models_estimated xcomp_estimated_using number_billion_47 num_words_billion dobj_using_words prep_words_of amod_text_English nn_text_newswire pobj_of_text auxpass_used_are aux_generate_to xcomp_used_generate dobj_generate_lattices prep_lattices_for amod_rescoring_phrasal nn_rescoring_segmentation nn_rescoring_model pobj_for_rescoring ' +p,7813,'Among these methods , CRFs is the most common technique used in NLP and has been successfully applied to Part-of-Speech Tagging , Named-Entity Recognition and shallow parsing ',Collins,'prep_technique_Among det_methods_these pobj_Among_methods nsubj_technique_CRFs cop_technique_is det_technique_the advmod_common_most amod_technique_common dep_technique_used prep_used_in pobj_in_NLP cc_used_and aux_applied_has auxpass_applied_been advmod_applied_successfully conj_used_applied prep_applied_to nn__Part-of-Speech nn__Tagging pobj_to_ nn__Named-Entity nn__Recognition appos__ cc__and nn__shallow amod__parsing conj__ ' +o,7814,'using Spearmans rank correlation coefficient and Pearsons rank correlation coefficient ',Lin,'nn_coefficient_Spearmans nn_coefficient_rank nn_coefficient_correlation dobj_using_coefficient cc_coefficient_and nn_coefficient_Pearsons nn_coefficient_rank nn_coefficient_correlation conj_coefficient_coefficient ' +o,7815,'translation lexicon entries were scored according to the log likelihood ratio -LRB- cf ',Dunning,'nn_entries_translation nn_entries_lexicon nsubjpass_scored_entries auxpass_scored_were prep_scored_according dep_according_to det_ratio_the nn_ratio_log nn_ratio_likelihood pobj_to_ratio nn_cf_ dep_ratio_cf ' +p,7816,'Wikipedia first sentence -LRB- WikiFS -RRB- : used Wikipedia as an external knowledge to improve Named Entity Recognition ',Kazama,'nn_sentence_Wikipedia amod_sentence_first appos_sentence_WikiFS nsubj_used_ dep_sentence_used dobj_used_Wikipedia prep_used_as det_knowledge_an amod_knowledge_external pobj_as_knowledge aux_improve_to xcomp_used_improve nn_Recognition_Named nn_Recognition_Entity dobj_improve_Recognition ' +o,7817,'5 Parsing experiments 51 Data and setup We used the standard partitions of the Wall Street Journal Penn Treebank ; ie , sections 2-21 for training , section 22 for development and section 23 for evaluation ',Marcus,'num_experiments_5 nn_experiments_Parsing nsubj_51_experiments dobj_51_Data cc_Data_and conj_Data_setup nsubj_used_We dep_51_used det_partitions_the amod_partitions_standard dobj_used_partitions prep_partitions_of det__the nn__Wall nn__Street nn__Journal nn__Penn nn__Treebank pobj_of_ dep_partitions_ie appos_ie_sections num_sections_2-21 prep_sections_for pobj_for_training conj_partitions_section num_section_22 prep_section_for pobj_for_development cc_partitions_and conj_partitions_section num_section_23 prep_section_for pobj_for_evaluation ' +o,7818,'The adaptive approach is somehow similar to their idea of incremental learning and to the bootstrap approach proposed by ',Yarowsky,'det_approach_The amod_approach_adaptive nsubj_similar_approach cop_similar_is advmod_similar_somehow prep_similar_to poss_idea_their pobj_to_idea prep_idea_of amod_learning_incremental pobj_of_learning cc_to_and conj_to_to det_approach_the nn_approach_bootstrap pobj_to_approach partmod_approach_proposed prep_proposed_by ' +o,7819,'Another attractive property of the voted perceptron is that it can be used with kernels , for example the kernels over parse trees described in ',Collins,'det_property_Another amod_property_attractive nsubj_is_property prep_property_of det_perceptron_the amod_perceptron_voted pobj_of_perceptron complm_used_that nsubjpass_used_it aux_used_can auxpass_used_be ccomp_is_used prep_used_with pobj_with_kernels prep_used_for pobj_for_example det_kernels_the nsubj_described_kernels prep_kernels_over amod_trees_parse pobj_over_trees dep_is_described prt_described_in ' +o,7820,'alpha 0 0.1 0.2 0.3 0.4 0.5 Freq=2 13555 13093 12235 11061 10803 10458 Freq=3 4203 3953 3616 3118 2753 2384 Freq=4 1952 1839 1649 1350 1166 960 Freq=5 1091 1019 917 743 608 511 Freq>2 2869 2699 2488 2070 1666 1307 TOTAL 23670 22603 20905 18342 16996 15620 alpha 0.6 0.7 0.8 0.9 1.0 Freq=2 10011 9631 9596 9554 9031 Freq=3 2088 1858 1730 1685 1678 Freq=4 766 617 524 485 468 Freq=5 392 276 232 202 189 Freq>2 1000 796 627 517 439 TOTAL 14257 13178 12709 12443 11805 Table 7: Number of extracted MWUs by frequency 6.2 Qualitative Analysis As many authors assess (Frank Smadja, 1993; John Justeson and Slava Katz, 1995), deciding whether a sequence of words is a multiword unit or not is a tricky problem.',Smadja,'' +n,7821,' tried a different generative phrase translation model analogous to IBM word-translation Model 3 , and again found that the standard model outperformed their generative model ',Brown,'nsubj_tried_ det_model_a amod_model_different amod_model_generative nn_model_phrase nn_model_translation nsubj_analogous_model xcomp_tried_analogous prep_analogous_to nn__IBM nn__word-translation nn__Model nn__3 pobj_to_ cc_tried_and advmod_found_again conj_tried_found complm_outperformed_that det_model_the amod_model_standard nsubj_outperformed_model ccomp_found_outperformed poss_model_their amod_model_generative dobj_outperformed_model ' +n,7822,'The automatically generated patterns in PairClass are slightly more general than the patterns of ',Turney,'det_automatically_The nsubj_generated_automatically nsubj_general_patterns prep_patterns_in pobj_in_PairClass cop_general_are advmod_general_slightly advmod_general_more ccomp_generated_general prep_general_than det_patterns_the pobj_than_patterns prep_patterns_of ' +o,7823,'We can then state the following theorem (see (Collins, 2002) for a proof): Theorem 1 For any training sequence (xi;yi) that is separable with margin, for any value of T, then for the perceptron algorithm in figure 1 Ne R 2 2 where R is a constant such that 8i;8z 2 GEN(xi) jj (xi;yi) (xi;z)jj R. This theorem implies that if there is a parameter vector U which makes zero errors on the training set, then after a finite number of iterations the training algorithm will converge to parameter values with zero training error.',Collins,'' +o,7824,'In this paper a discriminative parser is proposed to implement maximum entropy -LRB- ME -RRB- models to address the learning task ',Berger,'prep_proposed_In det_paper_this pobj_In_paper det_parser_a amod_parser_discriminative nsubjpass_proposed_parser auxpass_proposed_is aux_implement_to xcomp_proposed_implement amod_models_maximum nn_models_entropy abbrev_models_ME dobj_implement_models partmod_models_ aux_address_to xcomp__address det_task_the nn_task_learning dobj_address_task ' +o,7825,'1 is based on several realvalued feature functions fi Their computation is based on the so-called IBM Model-1 ',Brown,'nsubjpass_based_1 auxpass_based_is prep_based_on amod_fi_several amod_fi_realvalued nn_fi_feature nn_fi_functions pobj_on_fi poss_computation_Their nsubjpass_based_computation auxpass_based_is dep_based_based prep_based_on det_Model-1_the amod_Model-1_so-called nn_Model-1_IBM pobj_on_Model-1 ' +o,7826,'35 Domain adaptation in Machine Translation Within MT there has been a variety of approaches dealing with domain adaption -LRB- for example ',Wu,'num_adaptation_35 amod_adaptation_Domain prep_adaptation_in nn_Translation_Machine pobj_in_Translation dep_adaptation_Within pobj_Within_MT expl_been_there aux_been_has rcmod_MT_been det_variety_a dobj_been_variety prep_variety_of pobj_of_approaches partmod_approaches_dealing prep_dealing_with punct_adaption_domain pobj_with_adaption prep_adaption_for pobj_for_example ' +o,7827,'The use of structured prediction to SMT is also investigated by ',Liang,'det_use_The nsubjpass_investigated_use prep_use_of amod_prediction_structured pobj_of_prediction prep_prediction_to pobj_to_SMT auxpass_investigated_is advmod_investigated_also prep_investigated_by ' +n,7828,'Although previous work has tackled the bootstrapping approach from both the theoretical and practical point of view , many key problems still remain unresolved , such as the selection of initial seed set ',Yarowsky,'mark_tackled_Although amod__previous nn__work nsubj_tackled_ aux_tackled_has advcl_unresolved_tackled det_approach_the amod_approach_bootstrapping dobj_tackled_approach prep_approach_from preconj_theoretical_both dep_theoretical_the pobj_from_theoretical cc_theoretical_and amod_point_practical conj_theoretical_point prep_approach_of pobj_of_view amod_problems_many amod_problems_key nsubj_unresolved_problems advmod_unresolved_still cop_unresolved_remain dep_as_such prep_unresolved_as det_selection_the pobj_as_selection prep_selection_of amod_seed_initial pobj_of_seed partmod_seed_set ' +o,7829,'The latter problem of developing methods that can work with incomplete supervisory information is addressed in a subsequent effort ',Stoyanov,'det_problem_The amod_problem_latter nsubjpass_addressed_problem prep_problem_of amod_methods_developing pobj_of_methods nsubj_work_that aux_work_can rcmod_methods_work prep_work_with amod_information_incomplete amod_information_supervisory pobj_with_information auxpass_addressed_is prep_addressed_in det_effort_a amod_effort_subsequent pobj_in_effort ' +o,7830,'The translations are evaluated in terms of BLEU score ',Papineni,'det_translations_The nsubjpass_evaluated_translations auxpass_evaluated_are prep_evaluated_in pobj_in_terms prep_terms_of amod_score_BLEU pobj_of_score ' +o,7831,'To have a fair comparison , for PR , we estimate the conditional probability of a relation given the evidence P -LRB- Rij Eij -RRB- , as in , by using the same set of features as in ME Table 3 shows precision , recall , and F1measure of each system for WordNet hypernyms -LRB- is-a -RRB- , WordNet meronyms -LRB- part-of -RRB- and ODP hypernyms -LRB- is-a -RRB- ',Snow,'aux_have_To dep_estimate_have det_comparison_a amod_comparison_fair dobj_have_comparison prep_have_for pobj_for_PR nsubj_estimate_we det_probability_the amod_probability_conditional dobj_estimate_probability prep_probability_of det_relation_a pobj_of_relation partmod_relation_given det_P_the nn_P_evidence dobj_given_P nn_Eij_Rij appos_P_Eij advmod_P_as dep_as_in pobj_in_ prep_given_by pcomp_by_using det_set_the amod_set_same dobj_using_set prep_set_of pobj_of_features mark_shows_as prep_shows_in nn_Table_ME pobj_in_Table nsubj_shows_3 advcl_using_shows nn_F1measure_precision conj_F1measure_recall cc_F1measure_and dobj_shows_F1measure prep_F1measure_of det_system_each pobj_of_system prep_system_for amod_hypernyms_WordNet pobj_for_hypernyms appos_hypernyms_is-a amod_meronyms_WordNet conj_hypernyms_meronyms appos_meronyms_part-of cc_hypernyms_and nn_hypernyms_ODP conj_hypernyms_hypernyms appos_hypernyms_is-a ' +p,7832,'However , this is not unprecedented : discriminatively weighted generative models have been shown to outperform purely discriminative competitors in various NLP classification tasks , and remain the standard approach in statistical translation modeling ',Och,'advmod_unprecedented_However nsubj_unprecedented_this cop_unprecedented_is neg_unprecedented_not amod_models_discriminatively amod_models_weighted nn_models_generative nsubjpass_shown_models aux_shown_have auxpass_shown_been parataxis_unprecedented_shown aux_outperform_to xcomp_shown_outperform advmod_discriminative_purely amod_competitors_discriminative dobj_outperform_competitors prep_outperform_in amod__various nn__NLP nn__classification nn__tasks pobj_in_ cc_outperform_and cop_approach_remain det_approach_the amod_approach_standard conj_outperform_approach prep_approach_in amod_modeling_statistical nn_modeling_translation pobj_in_modeling ' +p,7833,'A more refined algorithm , the incremental feature selection algorithm by , allows one feature being added at each selection and at the same time keeps estimated parameter values for the features selected in the previous stages ',Berger,'det_algorithm_A amod_algorithm_more amod_algorithm_refined nsubj_allows_algorithm det_algorithm_the amod_algorithm_incremental nn_algorithm_feature nn_algorithm_selection appos_algorithm_algorithm prep_algorithm_by pobj_by_ num_feature_one nsubj_keeps_feature auxpass_added_being partmod_feature_added prep_added_at det_selection_each pobj_at_selection cc_at_and conj_at_at det_time_the amod_time_same pobj_at_time ccomp_allows_keeps amod_values_estimated nn_values_parameter dobj_keeps_values prep_values_for det_features_the pobj_for_features partmod_features_selected prep_selected_in det_stages_the amod_stages_previous pobj_in_stages ' +o,7834,'1 Introduction and Motivation Parse selection constitutes an important part of many parsing systems ',McClosky,'nsubj_constitutes_1 nn_selection_Introduction cc_Introduction_and conj_Introduction_Motivation nn_selection_Parse dep_1_selection det_part_an amod_part_important dobj_constitutes_part prep_part_of amod_systems_many amod_systems_parsing pobj_of_systems ' +p,7835,'We also plan to apply self-training of n-best tagger which successfully boosted the performance of one of the best existing English syntactic parser ',McClosky,'nsubj_plan_We advmod_plan_also aux_apply_to xcomp_plan_apply dobj_apply_self-training prep_self-training_of amod_tagger_n-best pobj_of_tagger nsubj_boosted_which advmod_boosted_successfully rcmod_self-training_boosted det_performance_the dobj_boosted_performance prep_performance_of pobj_of_one prep_performance_of det_parser_the dep_existing_best amod_parser_existing dep_syntactic_English amod_parser_syntactic pobj_of_parser ' +o,7836,'This process is repeated for a number of iterations in a self-training fashion ',Yarowsky,'det_process_This nsubjpass_repeated_process auxpass_repeated_is prep_repeated_for det_number_a pobj_for_number prep_number_of pobj_of_iterations prep_iterations_in det_fashion_a amod_fashion_self-training pobj_in_fashion ' +p,7837,'In machine translation , the rankings from the automatic BLEU method have been shown to correlate well with human evaluation , and it has been widely used since and has even been adapted for summarization ',Papineni,'prep_shown_In nn_translation_machine pobj_In_translation det_rankings_the nsubjpass_shown_rankings prep_rankings_from det__the amod__automatic amod__BLEU nn__method pobj_from_ aux_shown_have auxpass_shown_been aux_correlate_to xcomp_shown_correlate advmod_correlate_well prep_correlate_with amod_evaluation_human pobj_with_evaluation cc_shown_and nsubjpass_used_it aux_used_has auxpass_used_been advmod_used_widely conj_shown_used prep_used_since cc_used_and aux_adapted_has advmod_adapted_even auxpass_adapted_been conj_used_adapted prep_adapted_for nn__summarization pobj_for_ ' +p,7838,'Several studies have demonstrated that for instance Statistical Machine Translation -LRB- SMT -RRB- benefits from incorporating a dedicated WSD module ',Carpuat,'amod_studies_Several nsubj_demonstrated_studies aux_demonstrated_have complm_benefits_that prep_benefits_for pobj_for_instance nn_Translation_Statistical nn_Translation_Machine nsubj_benefits_Translation abbrev_Translation_SMT ccomp_demonstrated_benefits prep_benefits_from pcomp_from_incorporating det_module_a amod_module_dedicated dep_dedicated_WSD dobj_incorporating_module ' +p,7839,'In our experiments , we have used Averaged Perceptron and Perceptron with margin to improve performance ',Collins,'prep_used_In poss_experiments_our pobj_In_experiments nsubj_used_we aux_used_have nn__Averaged nn__Perceptron dobj_used_ cc__and conj__Perceptron prep__with nn__margin pobj_with_ aux_improve_to xcomp_used_improve dobj_improve_performance ' +o,7840,'However , few papers in the field of computational linguistics have focused on this approach ',Yarowsky,'advmod_focused_However amod_papers_few nsubj_focused_papers prep_papers_in det_field_the pobj_in_field prep_field_of amod_linguistics_computational pobj_of_linguistics aux_focused_have prep_focused_on det_approach_this pobj_on_approach ' +o,7841,'The structure of the graphical model resembles IBM Model 1 in which each target -LRB- record -RRB- word is assigned one or more source -LRB- text -RRB- words ',Brown,'det_structure_The nsubj__structure prep_structure_of det_model_the amod_model_graphical pobj_of_model cop__resembles nn__IBM nn__Model num__1 rel_assigned_in pobj_in_which det_word_each nn_word_target appos_target_record nsubjpass_assigned_word auxpass_assigned_is rcmod__assigned num_words_one cc_one_or conj_one_more nn_words_source appos_words_text dobj_assigned_words ' +o,7842,'Recently , there have been several discriminative approaches at training large parameter sets including and ',Liang,'advmod_been_Recently expl_been_there aux_been_have amod_approaches_several nn_approaches_discriminative dobj_been_approaches prep_been_at pcomp_at_training amod_sets_large nn_sets_parameter dobj_training_sets prep_sets_including nn_and_ pobj_including_and ' +o,7843,'However , compositional approaches to lexical choice have been successful whenever detailed representations of lexical constraints can be collected and entered into the lexicon -LRB- eg , -RRB- ',Brown,'advmod_successful_However amod_approaches_compositional nsubj_successful_approaches prep_approaches_to amod_choice_lexical pobj_to_choice aux_successful_have dep_successful_been advmod_collected_whenever amod_representations_detailed nsubjpass_collected_representations prep_representations_of amod_constraints_lexical pobj_of_constraints aux_collected_can auxpass_collected_be advcl_successful_collected cc_collected_and conj_collected_entered prep_collected_into det_lexicon_the pobj_into_lexicon appos_lexicon_eg dep_eg_ ' +o,7844,'These methods have reported performance in the range of 95-99 \% ` correct \' by word ',Cutting,'det_methods_These nsubj_reported_methods aux_reported_have nsubj_correct_performance prep_performance_in det_range_the pobj_in_range prep_range_of num_\%_95-99 pobj_of_\% xcomp_reported_correct prep_correct_by pobj_by_word ' +o,7845,'We use the log likelihood ratio -LRB- LLR -RRB- given by -2 log 2 -LRB- H o -LRB- p ; k 1 , n 1 , k 2 , n 2 -RRB- \\/ H a -LRB- p 1 , p 2 ; n 1 , k 1 , n 2 , k 2 -RRB- -RRB- LLR measures the extent to which a hypothesized model of the distribution of cell counts , H a , differs from the null hypothesis , H o -LRB- namely , that the percentage of documents containing this term is the same in both corpora -RRB- ',Dunning,'nsubj_use_We det_ratio_the nn_ratio_log nn_ratio_likelihood dobj_use_ratio abbrev_ratio_LLR partmod_ratio_ prep__given dep_given_by num_measures_-2 number_2_log amod_measures_2 nn_o_H dep_H_o dep_H_p dep_p_k num_k_1 appos_k_n num_n_1 appos_k_k num_k_2 appos_k_n num_n_2 nn_H_\\/ dep_2_H amod_H_a dep_a_p num_p_1 appos_p_p num_p_2 appos_p_n num_n_1 appos_p_k num_k_1 appos_p_n num_n_2 appos_p_k num_k_2 nn_measures_LLR pobj_by_measures dep_extent_the rel_differs_extent prep_extent_to pobj_to_which det_model_a amod_model_hypothesized nsubj_differs_model prep_model_of det_distribution_the pobj_of_distribution prep_distribution_of nn_counts_cell pobj_of_counts appos_counts_H dep_H_a rcmod_ratio_differs prep_differs_from det_hypothesis_the amod_hypothesis_null pobj_from_hypothesis nn_o_H appos_hypothesis_o dep_o_namely dep_same_that det_percentage_the nsubj_same_percentage prep_percentage_of pobj_of_documents partmod_documents_containing det_term_this dobj_containing_term cop_same_is det_same_the dep_namely_same prep_same_in det_corpora_both pobj_in_corpora ' +o,7846,'A related method is multi-category perceptron , which explicitly finds a weight vector that separates correct labels from the incorrect ones in a mistake driven fashion ',Collins,'det_method_A amod_method_related nsubj_perceptron_method cop_perceptron_is amod_perceptron_multi-category nsubj_finds_which advmod_finds_explicitly rcmod_perceptron_finds det_vector_a nn_vector_weight dobj_finds_vector nsubj_separates_that rcmod_vector_separates amod_labels_correct dobj_separates_labels prep_separates_from det_ones_the amod_ones_incorrect pobj_from_ones prep_ones_in det_mistake_a pobj_in_mistake partmod_mistake_driven dobj_driven_fashion ' +o,7847,'Our work is most similar to work using discriminative log-linear models for alignment , which is similar to discriminative log-linear models used for the SMT decoding -LRB- translation -RRB- problem ',Och,'poss_work_Our nsubj_similar_work cop_similar_is advmod_similar_most aux_work_to xcomp_similar_work xcomp_work_using amod_models_discriminative amod_models_log-linear dobj_using_models prep_using_for pobj_for_alignment nsubj_similar_which cop_similar_is rcmod_alignment_similar aux_discriminative_to xcomp_similar_discriminative amod_models_log-linear dobj_discriminative_models partmod_models_used prep_used_for det_problem_the nn_problem_SMT nn_problem_decoding nn_problem_translation pobj_for_problem ' +o,7848,'We measured the accuracy of the POS tagger trained in three settings : Original : The tagger is trained with the union of Wall Street Journal -LRB- WSJ -RRB- section of Penn Treebank , GENIA , and Penn BioIE ',Marcus,'nsubj_measured_We det_accuracy_the dobj_measured_accuracy prep_accuracy_of det_tagger_the dep_tagger_POS pobj_of_tagger partmod_tagger_trained prep_trained_in num_settings_three pobj_in_settings dep_trained_Original det_tagger_The nsubjpass_trained_tagger auxpass_trained_is dep_trained_trained prep_trained_with det_union_the pobj_with_union prep_union_of nn_Journal_Wall nn_Journal_Street pobj_of_Journal abbrev_Journal_WSJ punct_Journal_section prep_Journal_of nn__Penn nn__Treebank pobj_of_ conj__GENIA cc__and nn_BioIE_Penn conj__BioIE ' +o,7849,'Unlike our technique , in most cases researchers have focused on the scenario where labeled training data is available in both the source and the target domain -LRB- eg , -RRB- ','Daume III','prep_focused_Unlike poss_technique_our pobj_Unlike_technique prep_technique_in amod_cases_most pobj_in_cases nsubj_focused_researchers aux_focused_have prep_focused_on det_scenario_the pobj_on_scenario advmod_labeled_where advcl_focused_labeled nn_data_training nsubj_available_data cop_available_is ccomp_labeled_available prep_available_in preconj_source_both det_source_the pobj_in_source cc_source_and det_domain_the nn_domain_target conj_source_domain appos_domain_eg dep_eg_ ' +o,7850,'A re nement of this model is the class-based n-gram where the words are partitioned into equivalence classes ',Brown,'det_nement_A nn_nement_re nsubj_n-gram_nement prep_nement_of det_model_this pobj_of_model cop_n-gram_is det_n-gram_the amod_n-gram_class-based advmod_partitioned_where det_words_the nsubjpass_partitioned_words auxpass_partitioned_are rcmod_n-gram_partitioned prep_partitioned_into amod_classes_equivalence pobj_into_classes ' +o,7851,'Curran and use syntactic features in the vector definition ',Lin,'nn__Curran nsubj_use_ cc__and conj__ amod_features_syntactic dobj_use_features prep_features_in det_definition_the nn_definition_vector pobj_in_definition ' +o,7852,'Both techniques implement variations on the approaches of and for the purpose of differentiating between complement and adjunct ',Collins,'det_techniques_Both nsubj_implement_techniques dobj_implement_variations prep_implement_on det_approaches_the pobj_on_approaches prep_approaches_of pobj_of_ cc__and conj__ prep_implement_for det_purpose_the pobj_for_purpose prep_purpose_of pcomp_of_differentiating prep_differentiating_between pobj_between_complement cc_complement_and conj_complement_adjunct ' +o,7853,'There are many research directions, e.g., sentiment classification (classifying an opinion document as positive or negative) (e.g., Pang, Lee and Vaithyanathan, 2002; Turney, 2002), subjectivity classification (determining whether a sentence is subjective or objective, and its associated opinion) (Wiebe and Wilson, 2002; Yu and Hatzivassiloglou, 2003; Wilson et al, 2004; Kim and Hovy, 2004; Riloff and Wiebe, 2005), feature/topic-based sentiment analysis (assigning positive or negative sentiments to topics or product features) (Hu and Liu 2004; Popescu and Etzioni, 2005; Carenini et al., 2005; Ku et al., 2006; Kobayashi, Inui and Matsumoto, 2007; Titov and McDonald.',Turney,'' +o,7854,' use an information extraction engine to extract linguistic features from documents relevant to the target term ',Lin,'nsubj_use_ det_engine_an nn_engine_information nn_engine_extraction nsubj_extract_engine aux_extract_to xcomp_use_extract amod_features_linguistic dobj_extract_features prep_extract_from pobj_from_documents amod_documents_relevant prep_relevant_to det_term_the nn_term_target pobj_to_term ' +o,7855,'Proceedings of the 22nd International Conference on Computational Linguistics , pages 585592 Manchester , August 2008 Random Restarts in Minimum Error Rate Training for Statistical Machine Translation Robert C Moore and Chris Quirk Microsoft Research Redmond , WA 98052 , USA bobmoore @ microsoftcom , chrisq @ microsoftcom Abstract minimum error rate training -LRB- MERT -RRB- procedure is the most commonly used method for training feature weights in statistical machine translation -LRB- SMT -RRB- models ',Och,'nsubj_method_Proceedings prep_Proceedings_of det_Conference_the amod_Conference_22nd nn_Conference_International pobj_of_Conference prep_Conference_on nn__Computational nn__Linguistics pobj_on_ conj_Conference_pages num_Manchester_585592 dep_pages_Manchester nn_Restarts_August num_Restarts_2008 nn_Restarts_Random conj_Conference_Restarts prep_Restarts_in nn_Training_Minimum nn_Training_Error nn_Training_Rate pobj_in_Training prep_Training_for nn_Moore_Statistical nn_Moore_Machine nn_Moore_Translation nn_Moore_Robert nn_Moore_C pobj_for_Moore cc_Conference_and nn_Redmond_Chris nn_Redmond_Quirk nn_Redmond_Microsoft nn_Redmond_Research conj_Conference_Redmond appos_Redmond_WA num_WA_98052 nn_microsoftcom_USA nn_microsoftcom_bobmoore nn_microsoftcom_@ appos_Redmond_microsoftcom nn__chrisq nn__@ nn__microsoftcom nn__Abstract appos_Redmond_ amod_procedure_minimum nn_procedure_error nn_procedure_rate nn_procedure_training abbrev_procedure_MERT dep__procedure cop_method_is det_method_the advmod_used_most advmod_used_commonly amod_method_used prep_method_for nn_weights_training nn_weights_feature nn_models_weights prep_weights_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_SMT pobj_for_models ' +o,7856,'The statistical approach involves the following : alignment of bilingual texts at the sentence level nsing statistical techniques -LRB- eg , Gale and Church -LRB- 1993 -RRB- , Chen -LRB- 1993 -RRB- , and Kay and RSscheisen -LRB- 1993 -RRB- -RRB- , statistical machine translation models -LRB- eg Brown , Cooke , Pietra , Pietra et al ',Brown,'det_approach_The amod_approach_statistical nsubj_involves_approach det_following_the dobj_involves_following dep_following_alignment prep_alignment_of amod_texts_bilingual pobj_of_texts prep_involves_at det_level_the nn_level_sentence pobj_at_level partmod_level_nsing amod_techniques_statistical dobj_nsing_techniques nn__eg dep_techniques_ conj__Gale cc__and conj__Church appos_Church_1993 conj_Church_Chen appos_Chen_1993 cc_Church_and conj_Church_Kay cc_Kay_and conj_Kay_RSscheisen appos_Kay_1993 amod_models_statistical nn_models_machine nn_models_translation appos_techniques_models nn_Brown_eg dep_models_Brown conj_Brown_Cooke conj_Brown_Pietra conj_Brown_Pietra cc_Brown_et conj_Brown_al ' +o,7857,'Let us suppose that we have two bilingual lexicons L f L p and L p L e We obtain word alignments of these lexicons by applying GIZA + + , and grow-diag-final heuristics ',Koehn,'nsubj_suppose_us ccomp_Let_suppose complm_have_that nsubj_have_we ccomp_suppose_have num_L_two amod_L_bilingual nn_L_lexicons dobj_have_L dep_p_f nn_p_L dep_obtain_p cc_p_and nn_L_L nn_L_p conj_p_L dep_obtain_e nsubj_obtain_We ccomp_Let_obtain nn_alignments_word dobj_obtain_alignments prep_alignments_of det_lexicons_these pobj_of_lexicons prep_obtain_by pcomp_by_applying nsubj_+_GIZA xcomp_applying_+ cc_+_+ conj_+_ cc_Let_and advmod_heuristics_grow-diag-final conj_Let_heuristics ' +o,7858,'The following treebanks were used for training the parser : ',Marcus,'det_treebanks_The amod_treebanks_following nsubjpass_used_treebanks auxpass_used_were prep_used_for pcomp_for_training det_parser_the dobj_training_parser ' +o,7859,'For mention detection we use approaches based on Maximum Entropy -LRB- MaxEnt henceforth -RRB- and Robust Risk Minimization -LRB- RRM henceforth -RRB- 1For a description of the ACE program see http://wwwnistgov/speech/tests/ace/ ',Berger,'prep_use_For nn_detection_mention pobj_For_detection nsubj_use_we dobj_use_approaches prep_use_based dep_based_on nn__Maximum nn__Entropy nn_henceforth_MaxEnt appos__henceforth pobj_on_ cc__and nn_1For_Robust nn_1For_Risk nn_1For_Minimization nn_henceforth_RRM abbrev_1For_henceforth conj__1For det_description_a nsubj_see_description prep_description_of det_program_the nn_program_ACE pobj_of_program dep_use_see acomp_see_http://wwwnistgov/speech/tests/ace/ ' +o,7860,'Good partof-speech results can be obtained using only the preceding category , which is what we will be using ',Marcus,'amod_results_Good amod_results_partof-speech nsubjpass_obtained_results aux_obtained_can auxpass_obtained_be purpcl_obtained_using advmod__only det__the amod__preceding nn__category dobj_using_ nsubj_is_which rcmod__is dobj_using_what nsubj_using_we aux_using_will aux_using_be ccomp_is_using ' +o,7861,'1 Introduction With the introduction of the BLEU metric for machine translation evaluation , the advantages of doing automatic evaluation for various NLP applications have become increasingly appreciated : they allow for faster implement-evaluate cycles -LRB- by by-passing the human evaluation bottleneck -RRB- , less variation in evaluation performance due to errors in human assessor judgment , and , not least , the possibility of hill-climbing on such metrics in order to improve system performance ',Och,'num_Introduction_1 prep_become_With det_introduction_the pobj_With_introduction prep_introduction_of det_metric_the amod_metric_BLEU pobj_of_metric prep_metric_for nn__machine nn__translation nn__evaluation pobj_for_ det_advantages_the nsubj_become_advantages prep_advantages_of pcomp_of_doing amod_evaluation_automatic dobj_doing_evaluation prep_doing_for amod_applications_various nn_applications_NLP pobj_for_applications aux_become_have dep_Introduction_become advmod_appreciated_increasingly dep_become_appreciated nsubj_allow_they parataxis_become_allow prep_allow_for amod_cycles_faster amod_cycles_implement-evaluate pobj_for_cycles dep_cycles_by pcomp_by_by-passing det_bottleneck_the amod_bottleneck_human nn_bottleneck_evaluation dobj_by-passing_bottleneck amod_variation_less appos_cycles_variation prep_variation_in nn_performance_evaluation pobj_in_performance amod_performance_due prep_due_to pobj_to_errors prep_errors_in amod_judgment_human nn_judgment_assessor pobj_in_judgment cc_errors_and neg_least_not conj_errors_least det_possibility_the appos_cycles_possibility prep_possibility_of pcomp_of_hill-climbing prep_hill-climbing_on amod_metrics_such pobj_on_metrics mark_improve_in dep_improve_order aux_improve_to purpcl_hill-climbing_improve nn_performance_system dobj_improve_performance ' +o,7862,'More specifically , two recent works have suggested using statistical data on lexical relations for resolving ambiguity of prepositional phrase attachment and pronoun references ',Hindle,'advmod_specifically_More advmod_suggested_specifically num_works_two amod_works_recent nsubj_suggested_works aux_suggested_have xcomp_suggested_using amod_data_statistical dobj_using_data prep_using_on amod_relations_lexical pobj_on_relations prep_using_for pcomp_for_resolving dobj_resolving_ambiguity prep_ambiguity_of amod__prepositional nn__phrase nn__attachment pobj_of_ cc__and nn__pronoun nn__references conj__ ' +o,7863,'7Our decoder lacks certain features shown to be beneficial to synchronous grammar decoding , in particular rule binarisation ',Zhang,'amod_decoder_7Our nsubj_lacks_decoder amod_features_certain dobj_lacks_features partmod_features_shown aux_beneficial_to cop_beneficial_be xcomp_shown_beneficial prep_beneficial_to amod_grammar_synchronous pobj_to_grammar amod_grammar_decoding prep_features_in amod_binarisation_particular nn_binarisation_rule pobj_in_binarisation ' +o,7864,'We selected 580 short sentences of length at most 50 characters from the 2002 NIST MT Evaluation test set as our development corpus and used it to tune s by maximizing the BLEU score , and used the 2005 NIST MT Evaluation test set as our test corpus ',Och,'nsubj_selected_We amod_sentences_580 amod_sentences_short dobj_selected_sentences prep_sentences_of pobj_of_length prep_length_at advmod_characters_most num_characters_50 pobj_at_characters prep_selected_from det_test_the num_test_2002 nn_test_NIST nn_test_MT nn_test_Evaluation pobj_from_test dep_test_set prep_set_as poss_corpus_our nn_corpus_development pobj_as_corpus cc_set_and conj_set_used dobj_used_it prep_used_to nn_s_tune pobj_to_s prep_used_by pcomp_by_maximizing det__the amod__BLEU nn__score dobj_maximizing_ cc_set_and conj_set_used det_test_the num_test_2005 nn_test_NIST nn_test_MT nn_test_Evaluation dobj_used_test partmod_test_set prep_set_as poss_corpus_our nn_corpus_test pobj_as_corpus ' +o,7865,'They are most commonly used for parsing and linguistic analysis , but are now commonly seen in applications like machine translation and question answering ',Wu,'nsubjpass_used_They auxpass_used_are advmod_commonly_most advmod_used_commonly prep_used_for amod__parsing cc_parsing_and conj_parsing_linguistic nn__analysis pobj_for_ cc_used_but auxpass_seen_are advmod_seen_now advmod_seen_commonly conj_used_seen prep_seen_in pobj_in_applications prep_applications_like nn__machine nn__translation pobj_like_ cc__and nn__question nn__answering conj__ ' +o,7866,'The mixture coefficients are trained in the usual way , so that the additional context is exploited when it is useful and ignored when it isnt The paper proceeds as follows ',Och,'det_coefficients_The nn_coefficients_mixture nsubjpass_trained_coefficients auxpass_trained_are prep_trained_in det__the amod__usual nn__way pobj_in_ advmod_exploited_so dep_exploited_that det_context_the amod_context_additional nsubjpass_exploited_context auxpass_exploited_is ccomp_trained_exploited advmod_useful_when nsubj_useful_it cop_useful_is advcl_exploited_useful cc_exploited_and conj_exploited_ignored advmod_isnt_when nsubj_isnt_it advcl_ignored_isnt det_paper_The nsubj_proceeds_paper ccomp_isnt_proceeds mark_follows_as advcl_proceeds_follows ' +o,7867,'task , originally introduced in Ramshaw and Marcus and also described in , brackets just base NP constituents5 ',Collins,'nsubj_brackets_task advmod_introduced_originally dep_task_introduced prep_introduced_in nn__Ramshaw cc_Ramshaw_and conj_Ramshaw_Marcus pobj_in_ cc_introduced_and advmod_described_also conj_introduced_described prep_described_in pobj_in_ advmod_constituents5_just nn_constituents5_base nn_constituents5_NP dobj_brackets_constituents5 ' +o,7868,'We also compared the cluster gazetteers with the Wikipedia gazetteer constructed by following the method of ',Kazama,'nsubj_compared_We advmod_compared_also det_gazetteers_the nn_gazetteers_cluster dobj_compared_gazetteers prep_compared_with det_gazetteer_the nn_gazetteer_Wikipedia pobj_with_gazetteer partmod_gazetteer_constructed prep_constructed_by pcomp_by_following det_method_the dobj_following_method prep_method_of ' +o,7869,'This paper , however , aims at the k-best tree algorithms whose packed representations are hypergraphs -LRB- equivalently , and\\/or graphs or packed forests -RRB- , which includes most parsers and parsing-based MT decoders ',Marcus,'det_paper_This nsubj_aims_paper advmod_paper_however prep_aims_at det_algorithms_the amod_algorithms_k-best nn_algorithms_tree pobj_at_algorithms poss_packed_whose dobj__packed nsubj__representations cop__are amod__hypergraphs rcmod_algorithms_ dep__equivalently amod_graphs_and\\/or conj_equivalently_graphs cc_equivalently_or amod_forests_packed conj_equivalently_forests nsubj_includes_which rcmod__includes advmod_parsers_most amod_decoders_parsers cc_parsers_and conj_parsers_parsing-based nn_decoders_MT dobj_includes_decoders ' +o,7870,'In experiments with the system of we have found that in practice a large number of complete translations are completely monotonic -LRB- ie , have a0 skips -RRB- , suggesting that the system has difficulty learning exactly what points in the translation should allow reordering ',Koehn,'prep_found_In pobj_In_experiments prep_experiments_with det_system_the pobj_with_system prep_system_of pobj_of_ nsubj_found_we aux_found_have complm_monotonic_that prep_monotonic_in pobj_in_practice det_number_a amod_number_large nsubj_monotonic_number prep_number_of amod_translations_complete pobj_of_translations cop_monotonic_are advmod_monotonic_completely ccomp_found_monotonic dep_monotonic_ie dep_ie_have amod_skips_a0 dobj_have_skips ccomp_monotonic_suggesting complm_has_that det_system_the nsubj_has_system ccomp_suggesting_has dobj_has_difficulty partmod_difficulty_learning dep_allow_exactly dep_exactly_what nsubj_allow_points prep_points_in det_translation_the pobj_in_translation aux_allow_should ccomp_learning_allow dobj_allow_reordering ' +o,7871,'Examples are the Penn Treebank for American English annotated at the University of Pennsylvania , the French treebank developed in Paris , the TIGER Corpus for German annotated at the Universities of Saarbrcurrency1ucken and This research was funded by a German Science Foundation grant -LRB- DFG SFB441-6 -RRB- ',Marcus,'nsubj__Examples cop__are det__the nn__Penn nn__Treebank prep__for nn_English_American pobj_for_English partmod_English_annotated prep_annotated_at det_University_the pobj_at_University prep_University_of pobj_of_Pennsylvania det__the amod__French nn__treebank nsubj_developed_ ccomp__developed prep_developed_in pobj_in_Paris det__the nn__TIGER nn__Corpus nsubj_annotated_ prep__for pobj_for_German conj__annotated prep_annotated_at det_Universities_the pobj_at_Universities prep_Universities_of pobj_of_Saarbrcurrency1ucken cc__and det_research_This nsubjpass_funded_research auxpass_funded_was conj__funded prep_funded_by det_grant_a amod_grant_German nn_grant_Science nn_grant_Foundation pobj_by_grant nn_SFB441-6_DFG abbrev_grant_SFB441-6 ' +o,7872,'First , two maximum entropy classifiers are applied , where the first predicts clause start labels and the second predicts clause end labels ',Berger,'advmod_applied_First num__two amod__maximum amod__entropy nn__classifiers nsubjpass_applied_ auxpass_applied_are advmod_predicts_where det_first_the nsubj_predicts_first advcl_applied_predicts nn_labels_clause nn_labels_start nsubj_predicts_labels cc_labels_and det_second_the conj_labels_second ccomp_predicts_predicts nn_labels_clause nn_labels_end dobj_predicts_labels ' +p,7873,'In 2004 , Conroy tested Maximal Marginal Relevance as well as QR decomposition ',Goldstein,'prep_tested_In pobj_In_2004 nn__Conroy nsubj_tested_ amod__Maximal amod__Marginal nn__Relevance dobj_tested_ dep_well_as cc__well dep_well_as nn_decomposition_QR conj__decomposition ' +o,7874,'-LRB- 1 -RRB- a -RRB- ~ x e \' ~ y ? read -LRB- e \' x y -RRB- & book -LRB- y -RRB- b -RRB- ~ x 3 e e \' y past -LRB- e -RRB- & enjoy -LRB- e x e \' -RRB- & ? read -LRB- e \' x y -RRB- & book -LRB- y -RRB- c -RRB- 3 e e \' y past -LRB- e -RRB- & enjoy -LRB- e j e \' -RRB- & ? read -LRB- e \' j y -RRB- & book -LRB- y -RRB- We follow , Alshawi et al ',Hobbs,'dep_read_1 dep_x_a amod_x_~ nsubj_read_x partmod_x_e advmod_e_~ dep_e_y poss_y_e amod_y_x dep_read_y cc_book_& nsubj_~_book appos_book_y dep_book_b dep_read_~ dobj_~_x dep_x_3 dep_x_e poss_past_e nn_past_y tmod_~_past dep_past_e cc_past_& conj_past_enjoy nn_e_e nn_e_x dep_enjoy_e dep_enjoy_& dep_past_read poss_y_e amod_y_x appos_read_y cc_read_& conj_read_book appos_past_y dep_e_c num_e_3 dep_past_e poss_past_e nn_past_y prep_e_past dep_-LRB-_e dep_past_& rcmod_&_enjoy amod_e_e nn_e_j dep_&_e dep_enjoy_& dep_&_read poss_y_e nn_y_j appos_e_y cc_e_& conj_e_book appos_book_y nsubj_follow_We rcmod_book_follow dep_past_ conj__Alshawi cc__et conj__al ' +o,7875,'Similarly to classical NLP tasks such as base noun phrase chunking , text chunking or named entity recognition , we formulate the mention detection problem as a classification problem , by assigning to each token in the text a label , indicating whether it starts a specific mention , is inside a specific mention , or is outside any mentions ',Ramshaw,'advmod_chunking_Similarly dep_Similarly_to amod_tasks_classical nn_tasks_NLP pobj_to_tasks dep_as_such prep_tasks_as nn_phrase_base nn_phrase_noun pobj_as_phrase dobj_chunking_ nn__text amod__chunking conj__ cc__or amod__named nn__entity nn__recognition conj__ nsubj_formulate_we conj__formulate det_problem_the nn_problem_mention nn_problem_detection dobj_formulate_problem prep_problem_as det_problem_a nn_problem_classification pobj_as_problem prep_formulate_by pcomp_by_assigning prep_assigning_to det_token_each pobj_to_token prep_token_in det_text_the pobj_in_text det_label_a nsubj_is_label partmod_label_indicating complm_starts_whether nsubj_starts_it ccomp_indicating_starts det_mention_a amod_mention_specific dobj_starts_mention ccomp_chunking_is prep_is_inside det_mention_a amod_mention_specific pobj_inside_mention cc_chunking_or cop_mentions_is amod_mentions_outside det_mentions_any conj_chunking_mentions ' +o,7876,' showed how to use the Voted Perceptron algorithm for learning W , and we use it for learning the global transliteration model ',Collins,'nsubj_showed_ advmod_use_how aux_use_to ccomp_showed_use det_algorithm_the nn_algorithm_Voted nn_algorithm_Perceptron dobj_use_algorithm prep_use_for pcomp_for_learning dobj_learning_W cc_showed_and nsubj_use_we conj_showed_use dobj_use_it prep_use_for pcomp_for_learning det_model_the amod_model_global nn_model_transliteration dobj_learning_model ' +o,7877,'Indeed , the proposed speech translation paradigm of log-linear models have been shown e ective in many applications ',Och,'advmod_shown_Indeed det_paradigm_the amod_paradigm_proposed nn_paradigm_speech nn_paradigm_translation nsubjpass_shown_paradigm prep_paradigm_of amod_models_log-linear pobj_of_models aux_shown_have auxpass_shown_been dep__e dep__ective prep_ective_in amod_applications_many pobj_in_applications dep_shown_ dobj__ ' +o,7878,'Our approach to inducing syntactic clusters is closely related to that described in which is one of the earliest papers on the subject ',Brown,'poss_approach_Our nsubjpass_related_approach prep_approach_to pcomp_to_inducing amod_clusters_syntactic dobj_inducing_clusters auxpass_related_is advmod_related_closely prep_related_to pobj_to_that partmod_that_described prep_described_in pobj_in_ nsubj_one_which cop_one_is rcmod__one prep_one_of det_papers_the amod_papers_earliest pobj_of_papers prep_papers_on det_subject_the pobj_on_subject ' +p,7879,'Automatically creating or extending taxonomies for specific domains is then a very interesting area of research ',Snow,'nsubj_area_Automatically partmod_Automatically_creating cc_creating_or conj_creating_extending dobj_creating_taxonomies prep_taxonomies_for amod_domains_specific pobj_for_domains cop_area_is advmod_area_then det_area_a advmod_interesting_very amod_area_interesting prep_area_of pobj_of_research ' +o,7880,' demonstrated that for pairs of sentences that are less than 16 words , the ITG alignment space has a good coverage over all possibilities ',Wu,'nsubj_demonstrated_ complm_has_that prep_has_for pobj_for_pairs prep_pairs_of pobj_of_sentences nsubj_words_that cop_words_are dep_than_less quantmod_16_than num_words_16 rcmod_sentences_words det_space_the nn_space_ITG nn_space_alignment nsubj_has_space ccomp_demonstrated_has det_coverage_a amod_coverage_good dobj_has_coverage prep_coverage_over det_possibilities_all pobj_over_possibilities ' +o,7881,'The use of such relations -LRB- mainly relations between verbs or nouns and their arguments and modifiers -RRB- for various purposes has received growing attention in recent research ',Hindle,'det_use_The nsubj_received_use prep_use_of amod_relations_such pobj_of_relations advmod_relations_mainly dep_relations_relations prep_relations_between pobj_between_verbs cc_verbs_or conj_verbs_nouns cc_verbs_and poss_arguments_their conj_verbs_arguments cc_arguments_and conj_arguments_modifiers prep_relations_for amod_purposes_various pobj_for_purposes aux_received_has amod_attention_growing dobj_received_attention prep_attention_in amod_research_recent pobj_in_research ' +o,7882,'For each differently tokenized corpus , we computed word alignments by a HMM translation model and by a word alignment refinement heuristic of grow-diagfinal ',Och,'prep_computed_For det_corpus_each advmod_tokenized_differently amod_corpus_tokenized pobj_For_corpus nsubj_computed_we nn_alignments_word dobj_computed_alignments prep_computed_by det__a nn__HMM nn__translation nn__model pobj_by_ cc_by_and conj_by_by det_heuristic_a nn_heuristic_word nn_heuristic_alignment nn_heuristic_refinement pobj_by_heuristic prep_heuristic_of amod__grow-diagfinal pobj_of_ ' +o,7883,'Such linguistic-preprocessing techniques could 1Various models have been constructed by the IBM team ',Brown,'amod_techniques_Such amod_techniques_linguistic-preprocessing nsubj_could_techniques amod_models_1Various nsubjpass_constructed_models aux_constructed_have auxpass_constructed_been dep_could_constructed prep_constructed_by det_team_the nn_team_IBM pobj_by_team ' +o,7884,'Secondly , we explore the possibility of designing complementary similarity metrics that exploit linguistic information at levels further than lexical Inspired in the work by , who introduced a series of metrics based on constituent\\/dependency syntactic matching , we have designed three subgroups of syntactic similarity metrics ',Liu,'advmod_explore_Secondly nsubj_explore_we det_possibility_the dobj_explore_possibility prep_possibility_of pcomp_of_designing amod_metrics_complementary nn_metrics_similarity dobj_designing_metrics nsubj_exploit_that rcmod_metrics_exploit amod_information_linguistic dobj_exploit_information prep_exploit_at pobj_at_levels advmod_exploit_further dep_further_than pobj_than_lexical partmod_lexical_Inspired prep_Inspired_in det_work_the pobj_in_work prep_Inspired_by pobj_by_ nsubj_introduced_who rcmod__introduced det_series_a dobj_introduced_series prep_series_of pobj_of_metrics prep_introduced_based dep_based_on amod_matching_constituent\\/dependency amod_matching_syntactic pobj_on_matching nsubj_designed_we aux_designed_have dep_explore_designed num_subgroups_three dobj_designed_subgroups prep_subgroups_of amod_metrics_syntactic nn_metrics_similarity pobj_of_metrics ' +o,7885,'We computed precision , recall and error rate on the entire set of sentence pairs for each data set5 To evaluate NeurAlign , we used GIZA + + in both directions -LRB- E-to-F and F-to-E , where F is either Chinese -LRB- C -RRB- or Spanish -LRB- S -RRB- -RRB- as input and a refined alignment approach that uses a heuristic combination method called grow-diagfinal for comparison ',Koehn,'nsubj_computed_We nn_rate_precision conj_precision_recall cc_precision_and conj_precision_error dobj_computed_rate prep_computed_on det_set_the amod_set_entire pobj_on_set prep_set_of nn_pairs_sentence pobj_of_pairs prep_pairs_for det_set5_each nn_set5_data pobj_for_set5 aux_evaluate_To xcomp_computed_evaluate dobj_evaluate_NeurAlign nsubj_used_we conj_computed_used nn_+_GIZA nn_+_+ dobj_used_+ prep_+_in det_directions_both pobj_in_directions dep_used_E-to-F cc_E-to-F_and conj_E-to-F_F-to-E advmod_C_where nsubj_C_F cop_C_is advmod_C_either amod_C_Chinese dep_E-to-F_C cc_C_or conj_C_Spanish appos_Spanish_S prep_used_as pobj_as_input cc_computed_and det__a amod__refined amod__alignment nn__approach nsubj_called_ nsubj_uses_that rcmod__uses det_method_a amod_method_heuristic nn_method_combination dobj_uses_method conj_computed_called amod__grow-diagfinal dobj_called_ prep_called_for pobj_for_comparison ' +o,7886,'2 The Tagger We used Ratnaparkhi \'s maximum entropybased POS tagger ',Ratnaparkhi,'num_Tagger_2 nn_Tagger_The nsubj_used_We rcmod_Tagger_used poss_tagger_Ratnaparkhi possessive_Ratnaparkhi_\'s amod_tagger_maximum amod_tagger_entropybased dep_tagger_POS dep_Tagger_tagger ' +p,7887,'SVM has been shown to be useful for text classification tasks , and has previously given good performance in sentiment classification experiments ',Pang,'nsubjpass_shown_SVM aux_shown_has auxpass_shown_been aux_useful_to cop_useful_be xcomp_shown_useful mark__for nn_tasks_text nn_tasks_classification nsubj__tasks advcl_useful_ cc__and aux_given_has advmod_given_previously conj__given amod_performance_good dobj_given_performance prep_given_in nn_experiments_sentiment nn_experiments_classification pobj_in_experiments ' +o,7888,'1 Introduction Statistical Machine Translation is a data driven machine translation technique which uses probabilistic models of natural language for automatic translation , ',Brown,'num_Translation_1 nn_Translation_Introduction nn_Translation_Statistical nn_Translation_Machine nsubj_data_Translation cop_data_is det_data_a partmod_data_driven nn_technique_machine nn_technique_translation dobj_driven_technique nsubj_uses_which rcmod_technique_uses amod_models_probabilistic dobj_uses_models prep_models_of amod_language_natural pobj_of_language prep_language_for amod__automatic nn__translation pobj_for_ advmod_uses_ ' +o,7889,'1 Introduction Co-training , and several variants of co-training , have been applied to a number of NLP problems , including word sense disambiguation , named entity recognition , noun phrase bracketing and statistical parsing ',Yarowsky,'num__1 nn__Introduction nn__Co-training nsubjpass_applied_ cc__and amod_variants_several conj__variants prep_variants_of pobj_of_co-training aux_applied_have auxpass_applied_been prep_applied_to det_number_a pobj_to_number prep_number_of nn_problems_NLP pobj_of_problems prep_applied_including nn__word nn__sense nn__disambiguation pobj_including_ amod__named nn__entity nn__recognition nn__ amod__noun nn__phrase amod__bracketing appos__ cc__and amod__statistical amod__parsing conj__ ' +o,7890,'However , one of the major limitations of these advances is the structured syntactic knowledge , which is important to global reordering , has not been well exploited ',Elming,'advmod_is_However nsubj_is_one prep_one_of det_limitations_the amod_limitations_major pobj_of_limitations prep_limitations_of det_advances_these pobj_of_advances det_knowledge_the amod_knowledge_structured amod_knowledge_syntactic nsubjpass_exploited_knowledge nsubj_important_which cop_important_is rcmod_knowledge_important prep_important_to amod__global nn__reordering pobj_to_ aux_exploited_has neg_exploited_not auxpass_exploited_been advmod_exploited_well ccomp_is_exploited ' +o,7891,'I -RRB- agan eL al proposed a similarity-based model in which each word is generalized , not to its own specific class , but to a set of words which are most similar to it ',Dagan,'nsubj_-RRB-_I nn_al_agan nn_al_eL nsubj_proposed_al ccomp_-RRB-_proposed det_model_a amod_model_similarity-based dobj_proposed_model rel_generalized_in pobj_in_which det_word_each nsubj_generalized_word cop_generalized_is rcmod_model_generalized dep_to_not prep_generalized_to poss_class_its amod_class_own amod_class_specific pobj_to_class cc_to_but conj_to_to det_set_a pobj_to_set prep_set_of pobj_of_words nsubj_similar_which cop_similar_are advmod_similar_most rcmod_words_similar prep_similar_to pobj_to_it ' +o,7892,'is a WordNet based relatedness measure ',Pedersen,'auxpass_based_is det_WordNet_a nsubjpass_based_WordNet amod_measure_relatedness dobj_based_measure ' +n,7893,'Unlike , who found optimal performance when was approximately 104 , we observed monotonic increases in performance as dropped ',Johnson,'prep_observed_Unlike pobj_Unlike_ nsubj_performance_who dep_performance_found amod_performance_optimal rcmod__performance advmod_104_when cop_104_was quantmod_104_approximately advcl_performance_104 nsubj_observed_we amod_increases_monotonic dobj_observed_increases prep_observed_in pobj_in_performance mark_dropped_as advcl_observed_dropped ' +o,7894,'First , we extend the mechanism of adding gap variables for nodes dominating a site of discontinuity ',Collins,'advmod_extend_First nsubj_extend_we det_mechanism_the dobj_extend_mechanism prep_mechanism_of pcomp_of_adding nn_variables_gap dobj_adding_variables prep_adding_for pobj_for_nodes partmod_nodes_dominating det_site_a dobj_dominating_site prep_site_of pobj_of_discontinuity ' +o,7895,'22 Statistical Approaches with a grmnnmr There have been nlally l -RRB- rOl -RRB- osals tbr statistical t ` rameworks particularly designed tbr 1 -RRB- arsers with hand-crafted grmnmars -LRB- Schal -RRB- es , 1992 ; Briscoe and Carroll , 1993 ; Abney , 1996 ; Inui et al , 1 ! -RRB- 97 -RRB- ',Berger,'num_Approaches_22 nn_Approaches_Statistical nsubj_osals_Approaches prep_Approaches_with det_grmnnmr_a pobj_with_grmnnmr expl_been_There aux_been_have rcmod_grmnnmr_been advmod_been_nlally prep_been_l pobj_l_rOl amod_t_tbr amod_t_statistical dobj_osals_t nsubj_tbr_rameworks advmod_designed_particularly amod_rameworks_designed dep_osals_tbr dep_arsers_1 nsubj_es_arsers prep_arsers_with amod_grmnmars_hand-crafted pobj_with_grmnmars appos_arsers_Schal xcomp_tbr_es appos_es_1992 dep_1992_Briscoe cc_Briscoe_and conj_Briscoe_Carroll appos_Briscoe_1993 dep_1996_Abney dep_1993_1996 dep_1996_Inui cc_Inui_et conj_Inui_al appos_Inui_1 dep_1993_97 ' +o,7896,'7 Automated Sense Labelling of Discourse Connectives The focus here is on automated sense labelling of discourse connectives (Elwell and Baldridge, 2008; Marcu and Echihabi, 2002; Pitler et al., 2009; Wellner and Pustejovsky, 2007; Wellner, 679 Total Density of Intra-Sentential Intra-Sentential Total Intra-Sentential Intra-Sentential Subordinating Coordinating Discourse Genre Sentences Connectives Connectives/Sentence Conjunctions Conjunctions Adverbials ESSAYS 4774 1397 0.293 808 (57.8\%) 438 (31.4\%) 151 (10.8\%) SUMMARIES 2118 275 0.130 166 (60.4\%) 99 (36.0\%) 10 (3.6\%) LETTERS 739 200 0.271 126 (63.0\%) 56 (28.0\%) 18 (9.0\%) NEWS 40095 9336 0.233 5514 (59.1\%) 3015 (32.3\%) 807 (8.6\%) Figure 4: Distribution of Explicit Intra-Sentential Connectives.',Marcu,'' +o,7897,'For the MER training , we modified Koehns MER trainer for our tree sequence-based system ',Och,'prep_modified_For det__the nn__MER nn__training pobj_For_ nsubj_modified_we nn__Koehns nn__MER nn__trainer dobj_modified_ prep__for poss_system_our nn_system_tree amod_system_sequence-based pobj_for_system ' +p,7898,'2 Prior Work Statistical machine translation , as pioneered by IBM , is grounded in the noisy channel model ',Brown,'num_translation_2 amod_translation_Prior nn_translation_Work nn_translation_Statistical nn_translation_machine nsubjpass_grounded_translation mark_pioneered_as dep_grounded_pioneered prep_pioneered_by nn__IBM pobj_by_ auxpass_grounded_is prep_grounded_in det_model_the amod_model_noisy nn_model_channel pobj_in_model ' +o,7899,'The discriminative training regimen is otherwise similar to ',Och,'det_regimen_The amod_regimen_discriminative nn_regimen_training nsubj_similar_regimen cop_similar_is advmod_similar_otherwise prep_similar_to ' +o,7900,'The minimum error training was used on the development data for parameter estimation ',Och,'det__The amod__minimum nn__error nn__training nsubjpass_used_ auxpass_used_was prep_used_on det_data_the nn_data_development pobj_on_data prep_data_for nn_estimation_parameter pobj_for_estimation ' +o,7901,'This statistical technique of labeling predicate argument operates on the output of the probabilistic parser reported in ',Collins,'det_technique_This amod_technique_statistical nsubj_operates_technique prep_technique_of pcomp_of_labeling amod_argument_predicate dobj_labeling_argument prep_operates_on det_output_the pobj_on_output prep_output_of det_parser_the amod_parser_probabilistic pobj_of_parser partmod_parser_reported prep_reported_in ' +o,7902,'However , developing the PDTB may help facilitate the production of more such corpora , through an initial pass of automatic annotation , followed by manual correction , much as was done in developing the PTB -RRB- ',Marcus,'advmod_followed_However dep_followed_developing det_PDTB_the nsubj_help_PDTB aux_help_may ccomp_developing_help dep_help_facilitate det_production_the dobj_help_production prep_production_of amod_corpora_more amod_corpora_such pobj_of_corpora prep_help_through det_pass_an amod_pass_initial pobj_through_pass prep_pass_of amod_annotation_automatic pobj_of_annotation prep_followed_by amod_correction_manual pobj_by_correction advmod_as_much advmod_followed_as auxpass_done_was dep_followed_done prep_done_in pcomp_in_developing det__the nn__PTB dobj_developing_ ' +o,7903,'-RRB- -LRB- maxarg \\* STPT T = -LRB- 1 -RRB- Then we assume that the tagging of one character is independent of each other , and modify formula 1 as = = = = = n i ii tttT nn tttT ctP ccctttPT n n 1 2121 \\* -RRB- -LRB- maxarg -RRB- -LRB- maxarg 21 21 -LRB- 2 -RRB- Beam search -LRB- n = 3 -RRB- is applied for tag sequence searching , but we only search the valid sequences to ensure the validity of searching result ',Ratnaparkhi,'nn_T_maxarg nn_T_\\* nn_T_STPT dep_T_= appos_T_1 advmod_assume_Then nsubj_assume_we dep_T_assume complm_independent_that det_tagging_the nsubj_independent_tagging prep_tagging_of num_character_one pobj_of_character cop_independent_is ccomp_assume_independent prep_independent_of det_other_each pobj_of_other cc_other_and amod_formula_modify conj_other_formula dep_as_1 advmod_=_as dep_formula_= nn_n_= nn_n_= nn_n_= nn_n_= nn_n_n nn_n_i nn_n_ii nn_n_tttT nn_n_nn nn_n_tttT nn_n_ctP nn_n_ccctttPT nn_n_n dobj_=_n number_2121_1 dep_n_2121 dep_n_\\* appos_other_maxarg amod_search_maxarg number_21_21 dep_maxarg_21 appos_search_2 nn_search_Beam nsubjpass_applied_search nsubj_3_n dep_3_= dep_search_3 advmod_applied_ auxpass_applied_is ccomp_independent_applied prep_applied_for nn_sequence_tag pobj_for_sequence partmod_sequence_searching cc_assume_but nsubj_search_we advmod_search_only conj_assume_search det_sequences_the amod_sequences_valid nsubj_ensure_sequences aux_ensure_to xcomp_search_ensure det_validity_the dobj_ensure_validity prep_validity_of pcomp_of_searching dobj_searching_result ' +o,7904,'Previously published approaches to reducing the rule set include : enforcing a minimum span of two words per non-terminal , which would reduce our set to 115M rules ; or a minimum count -LRB- mincount -RRB- threshold , which would reduce our set to 78M -LRB- mincount = 2 -RRB- or 57M -LRB- mincount = 3 -RRB- rules ',Zollmann,'advmod_published_Previously dobj_published_approaches prep_published_to pcomp_to_reducing det_set_the nn_set_rule dobj_reducing_set rcmod_set_include parataxis_reducing_enforcing det_span_a amod_span_minimum dobj_enforcing_span prep_span_of num_words_two pobj_of_words prep_words_per amod__non-terminal pobj_per_ nsubj_reduce_which aux_reduce_would rcmod__reduce poss_set_our dobj_reduce_set prep_reduce_to num_rules_115M pobj_to_rules cc_rules_or det_count_a amod_count_minimum conj_rules_count appos_rules_mincount nn__threshold nsubj_published_ nsubj_reduce_which aux_reduce_would rcmod__reduce poss_set_our dobj_reduce_set prep_reduce_to pobj_to_78M nsubj_2_mincount dep_2_= dep_78M_2 cc_78M_or amod_rules_57M nsubj_3_mincount dep_3_= dep_57M_3 conj_78M_rules ' +o,7905,'ROUGE-N This measure compares n-grams of two summaries , and counts the number of matches ',Lin,'nn__ROUGE-N det_measure_This nsubj_compares_measure dep__compares dobj_compares_n-grams prep_n-grams_of num_summaries_two pobj_of_summaries cc_compares_and conj_compares_counts det_number_the dobj_counts_number prep_number_of pobj_of_matches ' +o,7906,'1 Motivation Most of the noisy-channel-based models used in statistical machine translation -LRB- MT -RRB- are conditional probability models ',Brown,'dep_Most_1 nn_Most_Motivation prep_Most_of det_models_the amod_models_noisy-channel-based pobj_of_models partmod_models_used prep_used_in amod_translation_statistical nn_translation_machine pobj_in_translation abbrev_translation_MT nsubj_models_ cop_models_are amod_models_conditional nn_models_probability dep_Most_models ' +o,7907,'Taking SIGHAN Bakeoff 2006 as an example , the recall is lower about 5 \% than the precision for each submitted system on MSRA and CityU closed track ',Levow,'dep_lower_Taking nn__SIGHAN nn__Bakeoff num__2006 dobj_Taking_ prep_Taking_as det_example_an pobj_as_example det_recall_the nsubj_lower_recall cop_lower_is prep_lower_about num_\%_5 pobj_about_\% mark_closed_than det_precision_the nsubj_closed_precision prep_precision_for det_system_each amod_system_submitted pobj_for_system prep_system_on pobj_on_MSRA cc_MSRA_and conj_MSRA_CityU advcl_lower_closed dobj_closed_track ' +o,7908,'Language modeling , noun-clustering , constructing syntactic rules for SMT , and finding analogies are examples of some of the problems where we need to compute relative frequencies ',Galley,'amod__Language nn__modeling amod__noun-clustering appos__ dep__constructing amod_rules_syntactic dobj_constructing_rules prep_constructing_for nn__SMT pobj_for_ cc_constructing_and conj_constructing_finding dobj_finding_analogies nsubj_examples_ cop_examples_are rcmod_analogies_examples prep_examples_of pobj_of_some prep_some_of det_problems_the pobj_of_problems advmod_need_where nsubj_need_we rcmod_examples_need aux_compute_to xcomp_need_compute amod_frequencies_relative dobj_compute_frequencies ' +o,7909,'More recently , other approaches have investigated the use of machine learning to nd patterns in documents and the utility of parameterized modules so as to deal with dierent genres or corpor ',Goldstein,'advmod_recently_More advmod_investigated_recently amod_approaches_other nsubj_investigated_approaches aux_investigated_have det_use_the dobj_investigated_use prep_use_of pobj_of_machine partmod_machine_learning aux_nd_to xcomp_learning_nd dobj_nd_patterns prep_patterns_in nn__documents pobj_in_ cc_patterns_and det_utility_the conj_patterns_utility prep_utility_of amod_modules_parameterized pobj_of_modules advmod_as_so advmod_nd_as aux_deal_to dep_as_deal prep_deal_with amod_genres_dierent pobj_with_genres cc_genres_or conj_genres_corpor ' +o,7910,'The forest representation was obtained by adopting chart generation where ambiguous candidates are packed into an equivalence class and mapping a chart into a forest in the same way as parsing ',Berger,'det_representation_The nn_representation_forest nsubjpass_obtained_representation auxpass_obtained_was prep_obtained_by dep__adopting nn__chart nn__generation pcomp_by_ advmod_packed_where amod_candidates_ambiguous nsubjpass_packed_candidates auxpass_packed_are advcl__packed prep_packed_into det_class_an amod_class_equivalence pobj_into_class cc__and conj__mapping det_chart_a dobj_mapping_chart prep_mapping_into det_forest_a pobj_into_forest prep_forest_in det_way_the amod_way_same pobj_in_way prep_way_as pobj_as_parsing ' +o,7911,'Another body of related work is the literature on word clustering in computational linguistics and document clustering in information retrieval ',Brown,'det_body_Another nsubj_literature_body prep_body_of amod_work_related pobj_of_work cop_literature_is det_literature_the prep_literature_on pobj_on_word partmod_word_clustering prep_clustering_in amod__computational nn__linguistics pobj_in_ cc__and nn_clustering_document conj__clustering prep__in nn__information nn__retrieval pobj_in_ ' +o,7912,'With regard to the local update , -LRB- B -RRB- , in Algorithm 42 , early updates and y-good requirement in resemble our local update in that they tried to avoid the situation where the correct answer can not be output ',Collins,'prep_update_With pcomp_With_regard prep_regard_to det_update_the amod_update_local pobj_to_update appos_update_B prep_update_in pobj_in_Algorithm num_Algorithm_42 amod__early nn__updates nsubj_update_ cc__and amod_requirement_y-good conj__requirement prep_requirement_in pobj_in_ cop_update_resemble poss_update_our amod_update_local rel_tried_in pobj_in_that nsubj_tried_they rcmod_update_tried aux_avoid_to xcomp_tried_avoid det_situation_the dobj_avoid_situation advmod_output_where det_answer_the amod_answer_correct nsubj_output_answer aux_output_can neg_output_not cop_output_be advcl_avoid_output ' +o,7913,' approached the issue as a Word Sense Disambiguation problem ',Carpuat,'nsubj_approached_ det_issue_the dobj_approached_issue prep_approached_as det_problem_a nn_problem_Word nn_problem_Sense nn_problem_Disambiguation pobj_as_problem ' +o,7914,'So the sequence with a fork, which corresponds to only one nucleus is treated as a three word sequence in model C. Apart from this difference, model C directly relies on a combination of equations (10) and (12), namely conditioning by a80a7a81a49a82a9a12, a74a61a8a65a75a57a12 and a74a61a8a65a75a57a14a61a86, both the probability of generating a74a61a8a65a75 a47 and the one of generating a80a7a81a49a82 a47 . Thus, model C uses a reduced version of equation (12) and an extended version of 2Other models, as (Collins and Brooks, 1995; Merlo et al. , 1998) for PP-attachment resolution, or (Collins, 1997; Samuelsson, 2000) for probabilistic parsing, are somewhat related, but their supervised nature makes any direct comparison impossible.',Collins,'' +o,7915,'On the other hand , models that deal with structures or phrases instead of single words have also been proposed : the syntax translation models are described in , alignment templates are used in , and the alignment template approach is re-framed into the so-called phrase based translation -LRB- PBT -RRB- in ',Koehn,'prep_proposed_On det_hand_the amod_hand_other pobj_On_hand nsubjpass_proposed_models nsubj_deal_that rcmod_models_deal prep_deal_with pobj_with_structures cc_structures_or conj_structures_phrases dep_of_instead prep_structures_of amod_words_single pobj_of_words aux_proposed_have advmod_proposed_also auxpass_proposed_been det_models_the nn_models_syntax nn_models_translation nsubjpass_described_models auxpass_described_are parataxis_proposed_described prep_described_in pobj_in_ nn_templates_alignment nsubjpass_used_templates auxpass_used_are ccomp_proposed_used prep_used_in pobj_in_ cc_proposed_and det_approach_the amod_approach_alignment nn_approach_template nsubj_re-framed_approach cop_re-framed_is conj_proposed_re-framed prep_re-framed_into det_phrase_the amod_phrase_so-called pobj_into_phrase partmod_phrase_based dobj_based_translation abbrev_translation_PBT prep_based_in ' +o,7916,'These belong to two main categories based on machine learning and language or domain specific rules ',Berger,'nsubj_belong_These prep_belong_to num_categories_two amod_categories_main pobj_to_categories partmod_categories_based prep_based_on nn__machine nn__learning pobj_on_ cc__and conj__language cc__or nn__domain nn__specific nn__rules conj__ ' +o,7917,'The research of opinion mining began in 1997 , the early research results mainly focused on the polarity of opinion words and treated the text-level opinion mining as a classification of either positive or negative on the number of positive or negative opinion words in one text ',Turney,'det_research_The nsubj_began_research prep_research_of nn_mining_opinion pobj_of_mining prep_focused_in pobj_in_1997 det_results_the amod_results_early nn_results_research nsubj_focused_results advmod_focused_mainly ccomp_began_focused prep_focused_on det_polarity_the pobj_on_polarity prep_polarity_of nn_words_opinion pobj_of_words partmod_words_ cc__and conj__treated det_mining_the amod_mining_text-level nn_mining_opinion dobj__mining prep_mining_as det_classification_a pobj_as_classification prep_classification_of preconj_positive_either amod_words_positive cc_positive_or advmod_positive_negative conj_positive_on det_number_the pobj_on_number prep_number_of amod_opinion_positive cc_positive_or conj_positive_negative pobj_of_opinion pobj_of_words prep__in num_text_one pobj_in_text ' +o,7918,'Related Works Generally speaking , approaches to MWE extraction proposed so far can be divided into three categories : a -RRB- statistical approaches based on frequency and co-occurrence affinity , b -RRB- knowledgebased or symbolic approaches using parsers , lexicons and language filters , and c -RRB- hybrid approaches combining different methods ',Smadja,'nn_Works_Related nsubjpass_divided_Works advmod_speaking_Generally partmod_Works_speaking appos_Works_approaches prep_Works_to nn_extraction_MWE pobj_to_extraction partmod_extraction_proposed advmod_far_so advmod_proposed_far aux_divided_can auxpass_divided_be prep_divided_into num_categories_three pobj_into_categories dep_approaches_a amod_approaches_statistical dep_categories_approaches prep_approaches_based dep_based_on pobj_on_frequency cc_frequency_and amod_affinity_co-occurrence conj_frequency_affinity dep_filters_b amod_approaches_knowledgebased cc_knowledgebased_or conj_knowledgebased_symbolic nsubj_filters_approaches partmod_approaches_using dobj_using_parsers conj_parsers_lexicons cc_parsers_and conj_parsers_language ccomp_divided_filters cc_divided_and dep_approaches_c nsubj_approaches_hybrid conj_divided_approaches xcomp_approaches_combining amod_methods_different dobj_combining_methods ' +n,7919,'While these approaches have had som e success to date , their usability as parsers in systems for natural language understanding is suspect ',Collins,'mark_had_While det_approaches_these nsubj_had_approaches aux_had_have advcl_suspect_had nn_success_som nn_success_e dobj_had_success prep_success_to nn__date pobj_to_ poss_usability_their nsubj_suspect_usability prep_usability_as pobj_as_parsers prep_parsers_in pobj_in_systems prep_systems_for amod_understanding_natural nn_understanding_language pobj_for_understanding cop_suspect_is ' +o,7920,'All our MT systems were trained using a variant of the alignment template model described in ',Och,'poss_systems_our nn_systems_MT nsubjpass_trained_systems auxpass_trained_were dep_All_trained xcomp_trained_using det_variant_a dobj_using_variant prep_variant_of det_model_the amod_model_alignment nn_model_template pobj_of_model partmod_model_described prep_described_in ' +o,7921,'Many approaches have been proposed for semisupervised learning in the past , including : generative models , self-learning , cotraining , informationtheoretic regularization , and graphbased transductive methods ',Yarowsky,'amod_approaches_Many nsubjpass_proposed_approaches aux_proposed_have auxpass_proposed_been prep_proposed_for amod_learning_semisupervised pobj_for_learning prep_proposed_in det_past_the pobj_in_past prep_proposed_including amod__generative nn__models pobj_including_ amod__self-learning conj__ amod__cotraining conj__ amod__informationtheoretic nn__regularization conj__ cc__and amod__graphbased amod__transductive nn__methods conj__ ' +o,7922,' , this model is symmetric , because both word bags are generated together from a joint probability distribution ',Dunning,'advmod_symmetric_ det_model_this nsubj_symmetric_model cop_symmetric_is mark_generated_because det_bags_both nn_bags_word nsubjpass_generated_bags auxpass_generated_are advcl_symmetric_generated advmod_generated_together prep_generated_from det_distribution_a amod_distribution_joint nn_distribution_probability pobj_from_distribution ' +o,7923,'Learning in this context consisted of estimating the parameters of the model with simple likelihood based techniques , but incorporating various smoothing and back-off estimation tricks to cope with the sparse data problems ',Collins,'nsubj_consisted_Learning prep_Learning_in det_context_this pobj_in_context prep_consisted_of pcomp_of_estimating det_parameters_the dobj_estimating_parameters prep_parameters_of det_model_the pobj_of_model prep_estimating_with amod_likelihood_simple pobj_with_likelihood partmod_likelihood_based dobj_based_techniques cc_estimating_but conj_estimating_incorporating amod_smoothing_various dobj_incorporating_smoothing cc_smoothing_and amod_tricks_back-off nn_tricks_estimation conj_smoothing_tricks aux_cope_to xcomp_incorporating_cope prep_cope_with det_problems_the amod_problems_sparse nn_problems_data pobj_with_problems ' +o,7924,'757 hbps strong tendency to overestimate the probability of rare bi-phrases ; it is computed as in equation -LRB- 2 -RRB- , except that bi-phrase probabilities are computed based on individual word translation probabilities , somewhat as in IBM model 1 : Pr -LRB- t s -RRB- = 1 s t productdisplay tt summationdisplay ss Pr -LRB- t s -RRB- The target language feature function htl : this is based on a N-gram language model of the target language ',Brown,'num_tendency_757 amod_tendency_hbps amod_tendency_strong aux_overestimate_to dep_tendency_overestimate det_probability_the dobj_overestimate_probability prep_probability_of amod_bi-phrases_rare pobj_of_bi-phrases nsubjpass_computed_it auxpass_computed_is parataxis_overestimate_computed prep_computed_as dep_as_in pobj_in_equation tmod_computed_2 mark_computed_except det_probabilities_that amod_probabilities_bi-phrase nsubjpass_computed_probabilities auxpass_computed_are dep_2_computed prep_computed_based dep_based_on amod_probabilities_individual nn_probabilities_word nn_probabilities_translation pobj_on_probabilities advmod_as_somewhat advmod_2_as dep_as_in nn_model_IBM pobj_in_model num__1 dep_2_ parataxis_overestimate_Pr nn_s_t nsubj_1_s dep_1_= ccomp_Pr_1 nsubj_t_s dep_tendency_t amod_Pr_productdisplay nn_Pr_tt nn_Pr_summationdisplay nn_Pr_ss dobj_t_Pr nn_s_t appos_Pr_s det_language_The nn_language_target nsubj_feature_language rcmod_Pr_feature nn_htl_function dobj_feature_htl nsubjpass_based_this auxpass_based_is parataxis_t_based prep_based_on det_model_a amod_model_N-gram nn_model_language pobj_on_model prep_model_of det_language_the nn_language_target pobj_of_language ' +o,7925,'Thus , the WSJ+NANC model has better oracle rates than the WSJ model for both the WSJ and BROWN domains ',McClosky,'advmod_has_Thus det_model_the nn_model_WSJ+NANC nsubj_has_model amod_rates_better nn_rates_oracle dobj_has_rates prep_rates_than det__the nn__WSJ nn__model pobj_than_ prep__for preconj_WSJ_both det_WSJ_the pobj_for_WSJ cc_WSJ_and nn_domains_BROWN conj_WSJ_domains ' +o,7926,'However , the approach raises two major challenges : 7In practice , MERT training will be used to train relative weights for the different model components ',Och,'advmod_raises_However det_approach_the nsubj_raises_approach num_challenges_two amod_challenges_major dobj_raises_challenges amod__7In nn__practice amod__MERT nn__training nsubjpass_used_ aux_used_will auxpass_used_be parataxis_raises_used aux_train_to purpcl_used_train amod_weights_relative dobj_train_weights prep_train_for det_components_the amod_components_different nn_components_model pobj_for_components ' +o,7927,'73 122 Baseline System and Experimental Setup We take BBNs HierDec , a string-to-dependency decoder as described in , as our baseline for the following two reasons : It provides a strong baseline , which ensures the validity of the improvement we would obtain ',Shen,'num_System_73 num_System_122 nn_System_Baseline cc_System_and nn_Setup_Experimental conj_System_Setup nsubj_take_We dep_System_take nn_HierDec_BBNs dobj_take_HierDec det_decoder_a amod_decoder_string-to-dependency appos_HierDec_decoder advmod_described_as amod_decoder_described prep_take_in pobj_in_ prep_take_as poss_baseline_our pobj_as_baseline prep_baseline_for det_reasons_the amod_reasons_following num_reasons_two pobj_for_reasons nsubj_provides_It parataxis_take_provides det_baseline_a amod_baseline_strong dobj_provides_baseline nsubj_ensures_which rcmod_baseline_ensures det_validity_the dobj_ensures_validity prep_validity_of det_improvement_the pobj_of_improvement nsubj_obtain_we aux_obtain_would dep_System_obtain ' +o,7928,'using the BLEU metric ',Papineni,'det_metric_the amod_metric_BLEU dobj_using_metric ' +p,7929,'The most notable of these include the trigram HMM tagger , maximum entropy tagger , transformation-based tagger , and cyclic dependency networks ',Ratnaparkhi,'det_notable_The advmod_notable_most nsubj_include_notable prep_notable_of pobj_of_these det__the amod__trigram nn__HMM nn__tagger dobj_include_ amod__maximum amod__entropy nn__tagger conj__ amod__transformation-based nn__tagger conj__ cc__and amod__cyclic amod__dependency nn__networks conj__ ' +o,7930,'To combine the many differently-conditioned features into a single model , we provide them as features to the linear model -LRB- Equation 2 -RRB- and use minimum error-rate training to obtain interpolation weights m This is similar to an interpolation of backed-off estimates , if we imagine that all of the different contextsaredifferently-backedoffestimatesofthe complete context ',Och,'aux_combine_To dep_provide_combine det_features_the amod_features_many amod_features_differently-conditioned dobj_combine_features prep_combine_into det_model_a amod_model_single pobj_into_model nsubj_provide_we dobj_provide_them prep_provide_as pobj_as_features prep_features_to det_model_the amod_model_linear pobj_to_model appos_model_Equation num_Equation_2 cc_provide_and conj_provide_use amod__minimum amod__error-rate nn__training dobj_use_ aux_obtain_to xcomp_use_obtain nn_weights_interpolation nsubj_m_weights ccomp_obtain_m nsubj_similar_This cop_similar_is ccomp_m_similar prep_similar_to det_interpolation_an pobj_to_interpolation prep_interpolation_of amod_estimates_backed-off pobj_of_estimates mark_imagine_if nsubj_imagine_we advcl_similar_imagine complm_complete_that nsubj_complete_all prep_all_of det_contextsaredifferently-backedoffestimatesofthe_the amod_contextsaredifferently-backedoffestimatesofthe_different pobj_of_contextsaredifferently-backedoffestimatesofthe ccomp_imagine_complete dobj_complete_context ' +o,7931,'In the following section we show how this drawback can be overcome using statistical alignments ',Brown,'prep_show_In det_section_the amod_section_following pobj_In_section nsubj_show_we advmod_overcome_how det_drawback_this nsubjpass_overcome_drawback aux_overcome_can auxpass_overcome_be ccomp_show_overcome purpcl_overcome_using amod_alignments_statistical dobj_using_alignments ' +o,7932,'The third function is an original variant of the second ; the fourth is original ; and the fifth is prompted by the arguments of ',Dunning,'det_function_The amod_function_third nsubj_variant_function cop_variant_is det_variant_an amod_variant_original prep_variant_of det_second_the pobj_of_second det_fourth_the nsubj_original_fourth cop_original_is parataxis_variant_original cc_variant_and det_fifth_the nsubjpass_prompted_fifth auxpass_prompted_is conj_variant_prompted prep_prompted_by det_arguments_the pobj_by_arguments prep_arguments_of ' +o,7933,'2005 ; Choi et al , 2006 ; Ku et al , 2006 ; and McDonald , 2008 -RRB- ',Titov,'dep_2005_Choi cc_Choi_et conj_Choi_al appos_Choi_2006 dep_Choi_Ku cc_Ku_et conj_Ku_al appos_Ku_2006 dep_Choi_ cc__and conj__McDonald appos_Choi_2008 ' +o,7934,'3A hypergraph is analogous to a parse forest ',Huang,'num_hypergraph_3A nsubj_analogous_hypergraph cop_analogous_is prep_analogous_to det_forest_a amod_forest_parse pobj_to_forest ' +o,7935,'Such methods can achieve better performance , reaching tagging accuracy of up to 85 \% on unknown words for English ',Marcus,'amod_methods_Such nsubj_achieve_methods aux_achieve_can amod_performance_better dobj_achieve_performance xcomp_achieve_reaching amod_accuracy_tagging dobj_reaching_accuracy prep_accuracy_of quantmod_85_up dep_85_to num_\%_85 pobj_of_\% prep_\%_on amod_words_unknown pobj_on_words prep_words_for nn__English pobj_for_ ' +o,7936,'4 Comparison to Related Work Previous work has compared generative and discriminative models having the same structure , such as the Naive Bayes and Logistic regression models and other models ',Klein,'num_Comparison_4 nsubj_compared_Comparison prep_Comparison_to nn_work_Related nn_work_Work amod_work_Previous pobj_to_work aux_compared_has amod_models_generative cc_generative_and conj_generative_discriminative dobj_compared_models partmod_models_having det_structure_the amod_structure_same dobj_having_structure dep_as_such prep_structure_as det_Bayes_the nn_Bayes_Naive pobj_as_Bayes cc_Bayes_and nn__Logistic nn__regression nn__models conj_Bayes_ cc_Bayes_and amod_models_other conj_Bayes_models ' +o,7937,'Many methods for calculating the similarity have been proposed ',Papineni,'amod_methods_Many nsubjpass_proposed_methods prep_methods_for pcomp_for_calculating det_similarity_the dobj_calculating_similarity aux_proposed_have auxpass_proposed_been ' +o,7938,'Our story makes use of a weighted formalism known as quasi-synchronous grammar -LRB- hereafter , QG -RRB- , originally developed by D for machine translation ',Smith,'poss_story_Our nsubj_makes_story dobj_makes_use prep_use_of det_formalism_a amod_formalism_weighted pobj_of_formalism partmod_formalism_known prep_known_as amod_grammar_quasi-synchronous pobj_as_grammar appos_grammar_hereafter dep_hereafter_QG advmod_developed_originally partmod_grammar_developed prep_developed_by nn__D pobj_by_ prep__for nn_translation_machine pobj_for_translation ' +o,7939,' also selects patterns based on the number of pairs that generate them , but the number of selected patterns is a constant , independent of the number of input word pairs ',Turney,'advmod_selects_ advmod_selects_also dobj_selects_patterns prep_selects_based dep_based_on det_number_the pobj_on_number prep_number_of pobj_of_pairs nsubj_generate_that rcmod_pairs_generate dobj_generate_them cc_pairs_but det_number_the conj_pairs_number prep_number_of amod_patterns_selected pobj_of_patterns dep_patterns_is det__a amod__constant nsubj_selects_ appos__independent prep_independent_of det_number_the pobj_of_number prep_number_of nn_pairs_input nn_pairs_word pobj_of_pairs ' +o,7940,'Of particular relevance is other work on parsing the Penn WSJ Treebank ',Ratnaparkhi,'prep_is_Of amod_relevance_particular pobj_Of_relevance amod_work_other nsubj_is_work prep_work_on pcomp_on_parsing det_Treebank_the nn_Treebank_Penn nn_Treebank_WSJ dobj_parsing_Treebank ' +o,7941,'Previous studies called the class of algorithms illustrated in Figure 2 cautious or sequential because in each iteration they acquire 1 or a small set of rules ',Abney,'amod_studies_Previous nsubj_called_studies det_class_the dobj_called_class prep_class_of pobj_of_algorithms partmod_algorithms_illustrated prep_illustrated_in pobj_in_Figure num_Figure_2 amod_class_cautious cc_cautious_or conj_cautious_sequential dep_in_because prep_called_in det_iteration_each pobj_in_iteration nsubj_acquire_they rcmod_iteration_acquire dobj_acquire_1 cc_1_or det_set_a amod_set_small conj_1_set prep_set_of pobj_of_rules ' +o,7942,'This alignment representation is a generalization of the baseline alignments described in and allows for many-to-many alignments ',Brown,'det_representation_This nn_representation_alignment nsubjpass_described_representation auxpass_described_is det_generalization_a attr_described_generalization prep_generalization_of det_alignments_the nn_alignments_baseline pobj_of_alignments prep_described_in pobj_in_ cc_described_and conj_described_allows prep_allows_for amod_alignments_many-to-many pobj_for_alignments ' +o,7943,'1 Introduction In the multilingual track of the CoNLL 2007 shared task on dependency parsing, a single parser must be trained to handle data from ten different languages: Arabic (Hajic et al. , 2004), Basque (Aduriz et al. , 2003), Catalan, (Mart et al. , 2007), Chinese (Chen et al. , 2003), Czech (Bohmova et al. , 2003), English (Marcus et al. , 1993; Johansson and Nugues, 2007), Greek (Prokopidis et al. , 2005), Hungarian (Csendes et al. , 2005), Italian (Montemagni et al. , 2003), and Turkish (Oflazer et al. , 2003).1 Our contribution is a study in multilingual parser optimization using the freely available MaltParser system, which performs 1For more information about the task and the data sets, see Nivre et al.',Marcus,'' +o,7944,'For example , Weeds -LRB- see below -RRB- took verbs as contexts for nouns in object position : so they regarded two nouns to be similar to the extent that they occur as direct objects of the same set of verbs ',Weeds,'prep_took_For pobj_For_example nn__Weeds nsubj_took_ dep__see advmod_see_below dobj_took_verbs prep_verbs_as pobj_as_contexts prep_took_for pobj_for_nouns prep_nouns_in nn_position_object pobj_in_position mark_regarded_so nsubj_regarded_they advcl_took_regarded num_nouns_two dobj_regarded_nouns aux_similar_to cop_similar_be xcomp_regarded_similar prep_similar_to det_extent_the pobj_to_extent complm_occur_that nsubj_occur_they ccomp_similar_occur prep_occur_as amod_objects_direct pobj_as_objects prep_objects_of det_set_the amod_set_same pobj_of_set prep_set_of pobj_of_verbs ' +o,7945,'Statistical and information theoretic approaches , , , Using lexical collocations to determine PPA with statistical techniques was first proposed by ',Berger,'nn_approaches_Statistical cc_Statistical_and conj_Statistical_information nn_approaches_theoretic nn__ appos__ appos__ nsubjpass_proposed_ partmod__Using amod_collocations_lexical dobj_Using_collocations aux_determine_to xcomp_Using_determine dobj_determine_PPA prep_PPA_with amod_techniques_statistical pobj_with_techniques auxpass_proposed_was advmod_proposed_first dep_approaches_proposed prep_proposed_by pobj_by_ ' +o,7946,'So far , most of the statistical machine translation systems are based on the single-word alignment models as described in as well as the Hidden Markov alignment model ',Brown,'dep_far_So advmod_based_far nsubjpass_based_most prep_most_of det_systems_the amod_systems_statistical nn_systems_machine nn_systems_translation pobj_of_systems auxpass_based_are prep_based_on det_models_the amod_models_single-word amod_models_alignment pobj_on_models prep_based_as amod__described prep_described_in pobj_in_ dep_well_as cc__well dep_well_as det_model_the nn_model_Hidden nn_model_Markov nn_model_alignment conj__model pobj_as_ ' +o,7947,'This formula follows the convention of in letting so designate the null state ',Brown,'det_formula_This nsubj_follows_formula det_convention_the dobj_follows_convention prep_convention_of pobj_of_ prep_follows_in pcomp_in_letting advmod_state_so amod_state_designate det_state_the amod_state_null dobj_letting_state ' +o,7948,'As in , the parameter C8 D0 B4C4 CX B4D0D8 CX BND0DB CX B5CYC8BNC0BNDBBND8BNA1BNC4BVB5 is further smoothed as follows : C8 D0BD B4C4 CX CYC8BNC0BNDBBND8BNA1BNC4BVB5 A2 C8 D0BE B4D0D8 CX CYC8BNC0BNDBBND8BNA1BNC4BVBNC4 CX B5A2 C8 D0BF B4D0DB CX CYC8BNC0BNDBBND8BNA1BNC4BVBNC4 CX B4D0D8 CX B5B5 Note this smoothing is different from the syntactic counterpart ',Collins,'prep_smoothed_As dep_As_in pobj_in_ det_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_the amod_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_parameter nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_C8 nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_D0 nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_B4C4 nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_CX nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_B4D0D8 nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_CX nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_BND0DB nn_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5_CX nsubjpass_smoothed_B5CYC8BNC0BNDBBND8BNA1BNC4BVB5 auxpass_smoothed_is advmod_smoothed_further mark_follows_as advcl_smoothed_follows nn_B5B5_C8 nn_B5B5_D0BD nn_B5B5_B4C4 nn_B5B5_CX nn_B5B5_CYC8BNC0BNDBBND8BNA1BNC4BVB5 nn_B5B5_A2 nn_B5B5_C8 nn_B5B5_D0BE nn_B5B5_B4D0D8 nn_B5B5_CX nn_B5B5_CYC8BNC0BNDBBND8BNA1BNC4BVBNC4 nn_B5B5_CX nn_B5B5_B5A2 nn_B5B5_C8 nn_B5B5_D0BF nn_B5B5_B4D0DB nn_B5B5_CX nn_B5B5_CYC8BNC0BNDBBND8BNA1BNC4BVBNC4 nn_B5B5_CX nn_B5B5_B4D0D8 nn_B5B5_CX nsubj_Note_B5B5 ccomp_follows_Note det_smoothing_this nsubj_different_smoothing cop_different_is ccomp_Note_different prep_different_from det_counterpart_the amod_counterpart_syntactic pobj_from_counterpart ' +p,7949,'Averaging has been shown to reduce overfitting as well as reliance on the order of the examples during training ',Collins,'nsubjpass_shown_Averaging aux_shown_has auxpass_shown_been aux_reduce_to xcomp_shown_reduce amod__overfitting dobj_reduce_ dep_well_as cc__well dep_well_as conj__reliance prep_reduce_on det_order_the pobj_on_order prep_order_of det_examples_the pobj_of_examples prep_reduce_during pobj_during_training ' +o,7950,'The experiment used all 578 sentences in the ATIS corpus with a parse tree , in the Penn Treebank ',Marcus,'det_experiment_The nsubj_used_experiment det_sentences_all num_sentences_578 dobj_used_sentences prep_sentences_in det_corpus_the nn_corpus_ATIS pobj_in_corpus prep_used_with det_tree_a amod_tree_parse pobj_with_tree prep_used_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank ' +o,7951,'Therefore in grammar rules are already factorized into a set of probabilities ',Collins,'mark_factorized_in amod_rules_ nn_rules_grammar nsubjpass_factorized_rules auxpass_factorized_are advmod_factorized_already dep_Therefore_factorized prep_factorized_into det_set_a pobj_into_set prep_set_of pobj_of_probabilities ' +o,7952,'Following , Rapp -LRB- 2004 -RRB- , and Wettler et al ',Church,'nn_Rapp_ pobj_Following_Rapp appos_Rapp_2004 cc_Rapp_and conj_Rapp_Wettler cc_Wettler_et conj_Wettler_al ' +o,7953,'-LRB- 1 -RRB- 1We follow the notations in for English-French , ie , e f , although our models are tested , in this paper , for English-Chinese ',Brown,'dep_follow_1 nsubj_follow_1We det_notations_the dobj_follow_notations prep_notations_in pobj_in_ prep_f_for pobj_for_English-French nsubj_f_ie dep_ie_e dep_follow_f mark_tested_although poss_models_our nsubjpass_tested_models auxpass_tested_are advcl_f_tested prep_tested_in det_paper_this pobj_in_paper prep_tested_for pobj_for_English-Chinese ' +o,7954,'Much research has been carried out recently in this area ',Brown,'amod_research_Much nsubjpass_carried_research aux_carried_has auxpass_carried_been prt_carried_out advmod_carried_recently prep_carried_in det__this nn__area pobj_in_ ' +o,7955,'Finally , Section 4 reports the results of parsing experiments using our exhaustive k-best CYK parser with the concise PCFGs induced from the Penn WSJ treebank ',Marcus,'advmod_reports_Finally nsubj_reports_Section num_Section_4 det_results_the dobj_reports_results prep_results_of amod_experiments_parsing pobj_of_experiments partmod_experiments_using poss_parser_our amod_parser_exhaustive amod_parser_k-best nn_parser_CYK dobj_using_parser prep_using_with det_PCFGs_the amod_PCFGs_concise pobj_with_PCFGs partmod_PCFGs_induced prep_induced_from det_treebank_the nn_treebank_Penn nn_treebank_WSJ pobj_from_treebank ' +o,7956,'~ lr ~ l-l -LRB- ; is a part of the TiMBL software package which is available from http://ilkkubnl 3 Results We have used the baseNP data presented in 2 ',Ramshaw,'dep_lr_~ punct_l-l_~ pobj_lr_l-l cop_part_is det_part_a parataxis_used_part prep_part_of det_package_the nn_package_TiMBL nn_package_software pobj_of_package nsubj_available_which cop_available_is rcmod_part_available prep_available_from punct_part_http://ilkkubnl num_Results_3 nsubj_used_Results dep_Results_We aux_used_have dep_l-l_used dep_presented_the nn_data_baseNP nsubj_presented_data dep_used_presented prep_presented_in pobj_in_ num__2 ' +o,7957,'Like the data used by , this data was retagged by the Brill tagger in order to obtain realistic part-of speech -LRB- POS -RRB- tags 5 ',Ramshaw,'prep_retagged_Like det_data_the pobj_Like_data partmod_data_used prep_used_by pobj_by_ det_data_this nsubjpass_retagged_data auxpass_retagged_was prep_retagged_by det_tagger_the nn_tagger_Brill pobj_by_tagger mark_obtain_in dep_obtain_order aux_obtain_to purpcl_retagged_obtain amod_tags_realistic dep_realistic_part-of nn_tags_speech appos_tags_POS dobj_obtain_tags amod_tags_5 ' +o,7958,'The Powells algorithm used in this work is similar as the one from but we modi ed the line optimization codes , a subroutine of Powells algorithm , with reference to ',Och,'det_algorithm_The nn_algorithm_Powells dobj_used_algorithm prep_used_in det_work_this pobj_in_work cop_similar_is dep_used_similar prep_similar_as det_ed_the dep_ed_one prep_one_from pobj_from_ prep_one_but dep_but_we nn_ed_modi pobj_as_ed det_codes_the nn_codes_line nn_codes_optimization nsubj_used_codes det_subroutine_a appos_codes_subroutine prep_subroutine_of nn_algorithm_Powells pobj_of_algorithm prep_used_with pobj_with_reference prep_used_to ' +o,7959,'We employ the phrase-based SMT framework , and use the Moses toolkit , and the SRILM language modelling toolkit , and evaluate our decoded translations using the BLEU measure , using a single reference translation ',Koehn,'nsubj_employ_We det__the amod__phrase-based nn__SMT nn__framework dobj_employ_ cc_employ_and conj_employ_use det__the nn__Moses nn__toolkit dobj_use_ cc_employ_and det_language_the nn_language_SRILM nsubj_modelling_language conj_employ_modelling amod__toolkit dobj_modelling_ cc_employ_and conj_employ_evaluate poss_translations_our amod_translations_decoded dobj_evaluate_translations xcomp_evaluate_using det__the amod__BLEU nn__measure dobj_using_ dep_evaluate_using det_translation_a amod_translation_single nn_translation_reference dobj_using_translation ' +o,7960,'Unfortunately , this is not always the case , and the above methodology suffers from the weaknesses pointed out by concerning parse-parse-match procedures ',Wu,'advmod_case_Unfortunately nsubj_case_this cop_case_is neg_case_not advmod_case_always det_case_the cc_case_and det_methodology_the amod_methodology_above nsubj_suffers_methodology conj_case_suffers prep_suffers_from det_weaknesses_the pobj_from_weaknesses partmod_weaknesses_pointed prt_pointed_out prep_pointed_by pobj_by_ prep__concerning amod_procedures_parse-parse-match pobj_concerning_procedures ' +o,7961,'(Ng and Low 2004, Toutanova et al, 2003, Brants 2000, Ratnaparkhi 1996, Samuelsson 1993).',Ratnaparkhi,'' +o,7962,'For our baseline , we have selected the method based on binomial loglikelihood ratio test -LRB- BLRT -RRB- described in ',Dunning,'prep_selected_For poss_baseline_our pobj_For_baseline nsubj_selected_we aux_selected_have det_method_the dobj_selected_method prep_selected_based dep_based_on amod_test_binomial amod_test_loglikelihood nn_test_ratio pobj_on_test abbrev_test_BLRT partmod_test_described prep_described_in ' +o,7963,'Regarding error detection in corpora , discusses inconsistencies in the Penn Treebank and relates them to interannotator differences in tagging style ',Ratnaparkhi,'prep_discusses_Regarding nn_detection_error pobj_Regarding_detection prep_detection_in pobj_in_corpora nsubj_discusses_ dobj_discusses_inconsistencies prep_discusses_in det_Treebank_the nn_Treebank_Penn pobj_in_Treebank cc_discusses_and conj_discusses_relates nsubj_interannotator_them aux_interannotator_to xcomp_relates_interannotator dobj_interannotator_differences prep_interannotator_in pcomp_in_tagging dobj_tagging_style ' +p,7964,'Synchronous parsing models have been explored with moderate success ',Wu,'amod_models_Synchronous nn_models_parsing nsubjpass_explored_models aux_explored_have auxpass_explored_been prep_explored_with amod_success_moderate pobj_with_success ' +o,7965,'The diversity function rewards summaries that cover many important aspects and plays the redundancy reducing role that is common in most extractive summarization frameworks ',Goldstein,'det_function_The nn_function_diversity nsubj_rewards_function dobj_rewards_summaries nsubj_cover_that rcmod_summaries_cover amod_aspects_many amod_aspects_important dobj_cover_aspects cc_cover_and conj_cover_plays det_redundancy_the dobj_plays_redundancy partmod_redundancy_reducing dobj_reducing_role nsubj_common_that cop_common_is rcmod_role_common prep_common_in advmod_extractive_most amod_frameworks_extractive nn_frameworks_summarization pobj_in_frameworks ' +o,7966,'But because we want the insertion state a1a16a20 to model digressions or unseen topics, we take the novel step of forcing its language model to be complementary to those of the other states by setting a2 a3a27a38 a21 a8 a8 a4 a8 a24 a26a11a28a30a29a6 a39a41a40a43a42a45a44a16a46 a1a48a47a1a50a49 a20 a2 a3 a26a17a21 a8a9a8 a4 a8 a24 a51a53a52a55a54a57a56 a21 a39a58a40a43a42a45a44a16a46 a1a59a47a1a50a49 a20 a2 a3a27a26a11a21a50a60 a4 a8 a24a30a24 a17 4Following Barzilay and Lee (2003), proper names, numbers and dates are (temporarily) replaced with generic tokens to help ensure that clusters contain sentences describing the same event type, rather than same actual event.',Barzilay,'' +o,7967,'The self-training protocol is the same as in : we parse the entire unlabeled corpus in one iteration ',McClosky,'det_protocol_The amod_protocol_self-training nsubj_same_protocol cop_same_is det_same_the dep_in_as prep_same_in pobj_in_ nsubj_parse_we parataxis_same_parse det_corpus_the amod_corpus_entire amod_corpus_unlabeled dobj_parse_corpus prep_corpus_in num_iteration_one pobj_in_iteration ' +p,7968,'Systems based on perceptron have been shown to be competitive in NER and text chunking We specify the model and the features with the LBJ modeling language ',Kazama,'nsubjpass_shown_Systems prep_Systems_based dep_based_on pobj_on_perceptron aux_shown_have auxpass_shown_been aux_competitive_to cop_competitive_be xcomp_shown_competitive prep_competitive_in pobj_in_NER cc_NER_and conj_NER_text partmod_NER_chunking xcomp_chunking_ nsubj_specify_We ccomp__specify det_model_the dobj_specify_model cc_model_and det_features_the conj_model_features prep_specify_with det_language_the nn_language_LBJ nn_language_ nn_language_modeling pobj_with_language ' +o,7969,'The final SMT system performance is evaluated on a uncased test set of 3071 sentences using the BLEU , NIST and METEOR scores ',Papineni,'det_performance_The amod_performance_final nn_performance_SMT nn_performance_system nsubjpass_evaluated_performance auxpass_evaluated_is prep_evaluated_on det_set_a amod_set_uncased nn_set_test pobj_on_set prep_set_of num_sentences_3071 pobj_of_sentences partmod_sentences_using det__the amod__BLEU dobj_using_ nn__NIST conj__ cc__and amod_scores_METEOR amod_scores_ conj__scores ' +o,7970,'The POS data set and the CTS data set have previously been used for testing other adaptation methods , though the setup there is different from ours ',Blitzer,'det_set_The dep_set_POS nn_set_data nsubjpass_used_set cc_set_and det_set_the nn_set_CTS nn_set_data conj_set_set aux_used_have advmod_used_previously auxpass_used_been prep_used_for dep_methods_testing amod_methods_other nn_methods_adaptation pcomp_for_methods advmod_methods_ mark_different_though det_setup_the nsubj_different_setup advmod_different_there cop_different_is advcl_methods_different prep_different_from pobj_from_ours ' +o,7971,'63 Unsupervised sentiment classification Turney proposed the unsupervised method for sentiment classification , and similar method is utilized by many other researchers ',Turney,'num_classification_63 nn_classification_Unsupervised nn_classification_sentiment nsubj_proposed_classification advmod_proposed_Turney det_method_the amod_method_unsupervised dobj_proposed_method prep_proposed_for nn__sentiment nn__classification pobj_for_ cc_proposed_and amod_method_similar nsubjpass_utilized_method auxpass_utilized_is conj_proposed_utilized prep_utilized_by amod__many amod__other nn__researchers pobj_by_ ' +p,7972,'Weusemaximumentropy models , which are particularly well-suited for tasks -LRB- like ours -RRB- with many overlapping features , to harness these linguistic insights by using features in our models which encode , directly or indirectly , the linguistic correlates to SE types ',Berger,'amod_models_Weusemaximumentropy nsubj_correlates_models partmod_models_ nsubj_well-suited_which cop_well-suited_are advmod_well-suited_particularly parataxis__well-suited prep_well-suited_for pobj_for_tasks dep_tasks_like pobj_like_ours prep_tasks_with amod_features_many amod_features_overlapping pobj_with_features aux_harness_to xcomp__harness det_insights_these amod_insights_linguistic dobj_harness_insights prep_harness_by pcomp_by_using dobj_using_features prep_features_in poss_models_our pobj_in_models nsubj_encode_which rcmod_features_encode advmod_correlates_directly cc_directly_or conj_directly_indirectly det_linguistic_the nsubj_correlates_linguistic aux_SE_to xcomp_correlates_SE dobj_SE_types ' +o,7973,'This feature , which is based on the lexical parameters of the IBM Model 1 , provides a complementary probability for each tuple in the translation table ',Brown,'det_feature_This nsubj_provides_feature nsubjpass_based_which auxpass_based_is rcmod_feature_based prep_based_on det_parameters_the amod_parameters_lexical pobj_on_parameters prep_parameters_of det__the nn__IBM nn__Model num__1 pobj_of_ det_probability_a amod_probability_complementary dobj_provides_probability prep_probability_for det_tuple_each pobj_for_tuple prep_tuple_in det_table_the nn_table_translation pobj_in_table ' +o,7974,'The results were evaluated using the CoNLL shared task evaluation tools 5 The approaches tested were Error Driven Pruning -LRB- EDP -RRB- and Transformational Based Learning of IOB tagging -LRB- TBL -RRB- ',Ramshaw,'det_results_The nsubjpass_evaluated_results auxpass_evaluated_were xcomp_evaluated_using det_evaluation_the amod_evaluation_CoNLL amod_evaluation_shared nn_evaluation_task dobj_using_evaluation nsubj__tools num_tools_5 det_approaches_The nsubj_tested_approaches rcmod_tools_tested cop__were nn__Error nn__Driven nn__Pruning abbrev__EDP dep_evaluated_ cc__and nn_Learning_Transformational nn_Learning_Based conj__Learning prep_Learning_of nn_tagging_IOB pobj_of_tagging abbrev_tagging_TBL ' +o,7975,'Class-based methods cluster words into classes of similar words , so that one can base the estimate of a word pair \'s probability on the averaged cooccurrence probability of the classes to which the two words belong ',Brown,'amod_methods_Class-based nsubj__methods nn_words_cluster dobj__words prep__into pobj_into_classes prep_classes_of amod_words_similar pobj_of_words advmod_base_so dep_base_that nsubj_base_one aux_base_can dep__base det_estimate_the dobj_base_estimate prep_estimate_of det_pair_a nn_pair_word poss_probability_pair possessive_pair_\'s pobj_of_probability prep_base_on det_probability_the amod_probability_averaged nn_probability_cooccurrence pobj_on_probability prep_probability_of det_classes_the pobj_of_classes rel_belong_to pobj_to_which det_words_the num_words_two nsubj_belong_words rcmod_classes_belong ' +o,7976,'The second approach takes triples -LRB- verb , prep , noun2 -RRB- and -LRB- nounl , prep , noun2 -RRB- , like those in Table 10 , as training data for acquiring semantic knowledge and performs PP-attachment disambiguation on quadruples ',Brown,'det__The amod__second nn__approach nsubj_takes_ dobj_takes_triples nn_noun2_verb dep_noun2_prep appos_triples_noun2 cc_triples_and conj_triples_nounl dep_nounl_prep dep_nounl_noun2 prep_nounl_like pobj_like_those prep_nounl_in pobj_in_Table num_Table_10 prep_triples_as nn_data_training pobj_as_data prep_data_for pcomp_for_acquiring amod_knowledge_semantic dobj_acquiring_knowledge cc_takes_and conj_takes_performs amod_disambiguation_PP-attachment dobj_performs_disambiguation prep_performs_on pobj_on_quadruples ' +o,7977,'CIT -RRB- ',Wu,'nn_-RRB-_CIT ' +o,7978,'We do not consider mixed features between words and POS tags as in , that is , a single feature consists of either words or tags ',Ramshaw,'nsubj_consider_We aux_consider_do neg_consider_not amod_features_mixed dobj_consider_features prep_features_between pobj_between_words cc_words_and dep_tags_POS conj_words_tags dep_in_as prep_consider_in pobj_in_ nsubj_is_that rcmod__is det_feature_a amod_feature_single nsubj_consists_feature ccomp_is_consists prep_consists_of preconj_words_either pobj_of_words cc_words_or conj_words_tags ' +o,7979,'For example , we can use automatically extracted hyponymy relations , or automatically induced MN clusters ',Brown,'prep_use_For pobj_For_example nsubj_use_we aux_use_can advmod_extracted_automatically dep_use_extracted amod_relations_hyponymy nsubj__relations xcomp_extracted_ cc_extracted_or advmod_induced_automatically conj_extracted_induced nn_clusters_MN nsubj__clusters xcomp_induced_ ' +o,7980,'The techniques used to solve this problem can be roughly classified into two main categories : those relying on pre-existing knowledge resources -LRB- thesauri , semantic networks , taxonomies or encyclopedias -RRB- and those inducing distributional properties of words from corpora ',Hughes,'det_techniques_The nsubj_used_techniques aux_solve_to xcomp_used_solve det_problem_this dobj_solve_problem aux_classified_can auxpass_classified_be advmod_classified_roughly rcmod_problem_classified prep_classified_into num_categories_two amod_categories_main pobj_into_categories dep_problem_those partmod_those_relying prep_relying_on amod_resources_pre-existing nn_resources_knowledge pobj_on_resources dep_those_thesauri amod_networks_semantic appos_thesauri_networks conj_networks_taxonomies cc_networks_or conj_networks_encyclopedias dep_problem_ cc_problem_and det_properties_those amod_properties_inducing amod_properties_distributional conj_problem_properties prep_properties_of pobj_of_words prep_words_from nn__corpora pobj_from_ ' +o,7981,' , or in more recent implementation , the MOSES MT system1 ',Koehn,'cc__or conj__in amod_implementation_more amod_implementation_recent pobj_in_implementation det_system1_the nn_system1_MOSES nn_system1_MT appos_implementation_system1 ' +o,7982,'Recent comparisons of approaches that can be trained on corpora have shown that in most cases statistical aproaches yield better results than finite-state , rule-based , or memory-based taggers ',Marcus,'amod_comparisons_Recent nsubj_shown_comparisons prep_comparisons_of pobj_of_approaches nsubjpass_trained_that aux_trained_can auxpass_trained_be rcmod_approaches_trained prep_trained_on nn__corpora pobj_on_ aux_shown_have complm__that prep__in amod_cases_most pobj_in_cases amod_aproaches_statistical nsubj__aproaches ccomp_shown_ xcomp__yield amod_results_better dobj_yield_results prep_yield_than amod__finite-state conj_finite-state_rule-based cc_finite-state_or conj_finite-state_memory-based dep_finite-state_taggers pobj_than_ ' +o,7983,'Thus , some research has been focused on deriving different word-sense groupings to overcome the finegrained distinctions of WN , , , , and ',Mihalcea,'advmod_focused_Thus det_research_some nsubjpass_focused_research aux_focused_has auxpass_focused_been prep_focused_on pcomp_on_deriving amod_groupings_different amod_groupings_word-sense dobj_deriving_groupings aux_overcome_to xcomp_deriving_overcome det_distinctions_the amod_distinctions_finegrained dobj_overcome_distinctions prep_distinctions_of nn__WN pobj_of_ conj__ conj__ conj__ conj__ cc__and conj__ ' +o,7984,'33 Voted Perceptron Unlike other methods discussed so far , voted perceptron training attempts to minimize the difference between the global feature vector for a training instance and the same feature vector for the best-scoring labeling of that instance according to the current model ',Collins,'nsubj_voted_33 nn_Perceptron_Voted nsubj_discussed_Perceptron prep_discussed_Unlike amod_methods_other pobj_Unlike_methods rcmod_33_discussed advmod_far_so advmod_discussed_far nn__perceptron nn__training nsubj_attempts_ ccomp_voted_attempts aux_minimize_to xcomp_attempts_minimize det_difference_the dobj_minimize_difference prep_difference_between det_vector_the amod_vector_global nn_vector_feature pobj_between_vector prep_vector_for det_instance_a nn_instance_training pobj_for_instance cc_vector_and det_vector_the amod_vector_same nn_vector_feature conj_vector_vector prep_vector_for det_labeling_the amod_labeling_best-scoring pobj_for_labeling prep_labeling_of det_instance_that pobj_of_instance prep_minimize_according dep_according_to det_model_the amod_model_current pobj_to_model ' +o,7985,'These records are also known as field books and reference sets in literature ',Canisius,'det_records_These nsubjpass_known_records auxpass_known_are advmod_known_also prep_known_as nn_books_field pobj_as_books cc_books_and nn_sets_reference conj_books_sets prep_known_in pobj_in_literature ' +o,7986,'We used Pharoah as a baseline system for comparison ; the s-phrases used in our system include all phrases , with the same scores , as those used by Pharoah , allowing a direct comparison ',Koehn,'nsubj_used_We nn__Pharoah dobj_used_ prep_used_as det_system_a nn_system_baseline pobj_as_system prep_used_for pobj_for_comparison det_s-phrases_the nsubj_include_s-phrases partmod_s-phrases_used prep_used_in poss_system_our pobj_in_system parataxis_used_include det_phrases_all dobj_include_phrases prep_include_with det_scores_the amod_scores_same pobj_with_scores mark_used_as nsubj_used_those advcl_include_used prep_used_by pobj_by_Pharoah xcomp_used_allowing det_comparison_a amod_comparison_direct dobj_allowing_comparison ' +o,7987,'(Berger 1996, Ratnaparkhi 1996, 1998, Mikheev 1998, 2000).',Ratnaparkhi,'' +o,7988,'We also used the following resources : the Charniak parser to carry out the syntactic analysis ; the wn : : similaritypackage to compute the Jiang & Conrath -LRB- J&C -RRB- distance needed to implement the lexical similarity siml -LRB- T , H -RRB- as defined in ; SVM-lightTK to encode the basic tree kernel function , KT , in SVM-light ',Pedersen,'nsubj_used_We advmod_used_also det_resources_the amod_resources_following nsubj_similaritypackage_resources det_parser_the nn_parser_Charniak dep_resources_parser partmod_parser_ aux_carry_to xcomp__carry prt_carry_out det_analysis_the amod_analysis_syntactic dobj_carry_analysis det_wn_the dep_analysis_wn ccomp_used_similaritypackage acomp_similaritypackage_ aux_compute_to xcomp__compute det_Jiang_the dobj_compute_Jiang cc_Jiang_& nn__Conrath appos__J&C nn__distance conj_Jiang_ partmod_Jiang_needed aux_implement_to xcomp_needed_implement det_siml_the amod_siml_lexical nn_siml_similarity dobj_implement_siml appos_siml_T dep_T_H advmod_defined_as amod_siml_defined prep_similaritypackage_in pobj_in_ nsubj__SVM-lightTK parataxis_used_ aux_encode_to xcomp__encode det_function_the amod_function_basic nn_function_tree nn_function_kernel dobj_encode_function appos_function_KT prep_function_in nn__SVM-light pobj_in_ ' +n,7989,'2 Motivation and Prior Work While several authors have looked at the supervised adaptation case , there are less -LRB- and especially less successful -RRB- studies on semi-supervised domain adaptation ',Blitzer,'dep_Motivation_2 cc_Motivation_and amod_Work_Prior conj_Motivation_Work mark_looked_While amod_authors_several nsubj_looked_authors aux_looked_have advcl_are_looked prep_looked_at det_case_the amod_case_supervised nn_case_adaptation pobj_at_case expl_are_there dep_Motivation_are acomp_are_less dep_-LRB-_and advmod_less_especially dep_-LRB-_less dep_less_successful dep_on_studies prep_are_on amod_adaptation_semi-supervised nn_adaptation_domain pobj_on_adaptation ' +o,7990,'Only one word is labeled with the concept ; the syntactic head word is preferred ',Collins,'advmod_word_Only num_word_one nsubjpass_labeled_word auxpass_labeled_is prep_labeled_with det_concept_the pobj_with_concept det__the amod__syntactic nn__head nn__word nsubj_preferred_ cop_preferred_is parataxis_labeled_preferred ' +o,7991,'0 10 20 30 40 5097.20 97.25 97.30 97.35 97.40 Iteration Accuracy on development data Every iterationEvery 4th iterationEvery 8th iteration Every 16th iterationOnce at the beginning No supervised data Figure 2: Dependence on the inclusion of the supervised training data (English) English Czech No supervised data 97.37 95.88 Once at the beginning 97.40 96.00 Every training iteration 97.44 96.21 Table 6: Dependence on the inclusion of the supervised training data 5.4 The morphological analyzers and the perceptron feature templates The whole experiment can be performed with the original perceptron feature set described in (Collins, 2002) instead of the feature set described in this article.',Collins,'' +o,7992,'10Our experiments have shown that using averaging helps tremendously , confirming both the theoretical and practical results of ',Collins,'amod_experiments_10Our nsubj_shown_experiments aux_shown_have complm_helps_that csubj_helps_using dobj_using_averaging ccomp_shown_helps advmod_helps_tremendously xcomp_helps_confirming preconj_theoretical_both dep_theoretical_the dobj_confirming_theoretical cc_theoretical_and amod_results_practical conj_theoretical_results prep_theoretical_of ' +o,7993,'For instance , implementing an efficient version of the MXPOST POS tagger will simply involve composing and configuring the appropriate text file reading component , with the sequential tagging component , the collection of feature extraction components and the maximum entropy model component ',Ratnaparkhi,'prep_involve_For pobj_For_instance partmod_instance_implementing det_version_an amod_version_efficient dobj_implementing_version prep_version_of det__the nn__MXPOST dep__POS nn__tagger pobj_of_ aux_involve_will advmod_involve_simply xcomp_involve_composing cc_composing_and conj_composing_configuring det_component_the amod_component_appropriate nn_component_text nn_component_file nn_component_reading dobj_composing_component prep_involve_with det_tagging_the amod_tagging_sequential pobj_with_tagging nsubj_involve_component det_collection_the appos_component_collection prep_collection_of nn_components_feature nn_components_extraction pobj_of_components cc_components_and det_component_the amod_component_maximum nn_component_entropy nn_component_model conj_components_component ' +o,7994,'SCL for Discriminative Parse Selection So far , pivot features on the word level were used ',Blitzer,'nsubjpass_used_SCL prep_SCL_for nn_Selection_Discriminative nn_Selection_Parse pobj_for_Selection advmod_far_So advmod_Selection_far amod_features_pivot nsubjpass_used_features prep_features_on det_level_the nn_level_word pobj_on_level auxpass_used_were ' +o,7995,'The resolution of alignment can vat3 , from low to high : section , paragraph , sentence , phrase , and word ',Brown,'det_resolution_The prep_resolution_of amod_vat3_alignment nn_vat3_can pobj_of_vat3 prep_resolution_from amod__low prep_low_to pobj_to_high nn_word_section conj_word_paragraph conj_word_sentence conj_word_phrase cc_word_and dep_high_word pobj_from_ ' +o,7996,'-LRB- p 18 -RRB- Whether this is a useful perspective for machine translation is debatable -- however , it is a dead-on description of transliteration ',Brown,'number_18_p complm_perspective_Whether nsubj_perspective_this cop_perspective_is det_perspective_a amod_perspective_useful ccomp_description_perspective prep_perspective_for nn_translation_machine pobj_for_translation cop__is amod__debatable dep_translation_ dep__however nsubj_description_it cop_description_is det_description_a amod_description_dead-on dep_18_description prep_description_of pobj_of_transliteration ' +o,7997,'In order to determine interannotator agreement for step 2 of the coding procedure for the database of annotated texts , we calculated kappa statistics ',Carletta,'mark_determine_In dep_determine_order aux_determine_to dep_calculated_determine nn_agreement_interannotator dobj_determine_agreement prep_determine_for pobj_for_step num_step_2 prep_step_of det_procedure_the amod_procedure_coding pobj_of_procedure prep_procedure_for det_database_the pobj_for_database prep_database_of amod_texts_annotated pobj_of_texts nsubj_calculated_we nn_statistics_kappa dobj_calculated_statistics ' +o,7998,'First , the addition of each modification improves the F-score for both true and system mentions 9The H&K results shown here are not directly comparable with those reported in , since H&K evaluated their system on the ACE 2004 coreference corpus ',Haghighi,'advmod_improves_First det_addition_the nsubj_improves_addition prep_addition_of det_modification_each pobj_of_modification dep_comparable_improves det_F-score_the dobj_improves_F-score prep_F-score_for det_true_both pobj_for_true cc_improves_and nsubj_mentions_system conj_improves_mentions amod_results_9The nn_results_H&K dobj_mentions_results partmod_results_shown advmod_comparable_here cop_comparable_are neg_comparable_not advmod_comparable_directly prep_comparable_with pobj_with_those partmod_those_reported prep_reported_in pobj_in_ prep_reported_since pobj_since_H&K partmod_H&K_evaluated poss_system_their dobj_evaluated_system prep_evaluated_on det_ACE_the pobj_on_ACE num_corpus_2004 nn_corpus_coreference nsubj_comparable_corpus ' +o,7999,'1 Introduction The statistical machine translation framework -LRB- SMT -RRB- formulates the problem of translating a sentence from a source language S into a target language T as the maximization problem of the conditional probability : TM LM = argmaxT p -LRB- SjT -RRB- p -LRB- T -RRB- , -LRB- 1 -RRB- where p -LRB- SjT -RRB- is called a translation model -LRB- TM -RRB- , representing the generation probability from T into S , p -LRB- T -RRB- is called a language model -LRB- LM -RRB- and represents the likelihood of the target language ',Brown,'num_Introduction_1 det_framework_The amod_framework_statistical nn_framework_machine nn_framework_translation nsubj_formulates_framework abbrev_framework_SMT dep_Introduction_formulates det_problem_the dobj_formulates_problem prep_problem_of pcomp_of_translating det_sentence_a dobj_translating_sentence prep_translating_from det_S_a nn_S_source nn_S_language pobj_from_S prep_translating_into det_T_a nn_T_target nn_T_language pobj_into_T prep_T_as det_problem_the nn_problem_maximization pobj_as_problem prep_problem_of det_probability_the amod_probability_conditional pobj_of_probability nn_LM_TM nsubj_=_LM parataxis_formulates_= amod_p_argmaxT dobj_=_p appos_p_SjT partmod_p_p appos_p_T dep_called_1 advmod_called_where nsubjpass_called_p appos_p_SjT auxpass_called_is dep_called_called det_model_a nn_model_translation dobj_called_model abbrev_model_TM xcomp_called_representing det_probability_the nn_probability_generation dobj_representing_probability prep_representing_from pobj_from_T prep_representing_into pobj_into_S nsubjpass_called_p appos_p_T auxpass_called_is dep_=_called det_model_a nn_model_language dobj_called_model abbrev_model_LM cc_called_and conj_called_represents det_likelihood_the dobj_represents_likelihood prep_likelihood_of det_language_the nn_language_target pobj_of_language ' \ No newline at end of file