-
Notifications
You must be signed in to change notification settings - Fork 4
ErgSemantics_Design
Design principles for DELPH-IN/MRS semantic representations identified during the Hankø meeting:
- Interface representation should include all information that is constrained by the grammar
- Interface representation should be highly normalized, abstracting away from details of surface syntax
- Close paraphrases should lead to comparable or identical structures
in the interface representation
- Ex: Predicative copula does not contribute an EP (The fierce dog barks ~ The dog that is fierce barks)
- Ex: Dative alternation, passive (modulo discourse rel)
- Minimize ambiguity
- Corollary: Differentiate lexical predicates only if that distinction corresponds to morphosyntactic differences
- Decomposition is desirable, but not a goal in itself. If the predicates required for decomposition exist, then it is good.
cf. Copestake et al 2005 and "Slacker Semantics" (Copestake 2009)
While the assumption of an intrinsic argument should be non-controversial for the predications associated with verbal and nominal phrases, more explanation is in order for extending this assumption to the predications introduced by adjectives and even adverbs.
One design decision for the ERG is that the same predicative copula be is used both for present participles as in the cat is sleeping and for predicative adjectives as in the cat is happy. Since we only want one EP for the verb phrase is sleeping, the copula be does not introduce an EP of its own, but rather constrains variable properties (tense and aspect) of the semantics introduced by its verbal complement. Hence predicative adjectives need to supply the eventuality that the copula will so constrain, and this variable is unsurprisingly the intrinsic argument of the EP introduced by the adjective.
A second design decision is that the same lexical entry is used both for predicative and attributive uses of adjectives, rather than either making distinct lexical entry pairs for most adjectives, or deriving the predicative entry from the attributive one via lexical rule. As a consequence, the EP introduced by this lexical entry has an intrinsic argument present even when the entry is used as an attributive adjective, where nothing else in the structure will constrain its variable properties. The grammar makes use of this ARG0 when the adjective itself is modified by a non-scopal word or phrase, as in extremely happy, with extremely introducing an EP which takes the ARG0 of the adjective it modifies as its ARG1. Without an intrinsic argument present in the EPs of attributive adjectives, all modifiers of adjectives would have to be treated as scopal, but this would create an unwanted asymmetry for consistently won and consistently happy, where we expect manner adverbs to be non-scopal modifiers of verbs.
A third design decision is that morphologically related adjective-adverb pairs introduce the same EP, with the syntactic category difference reflected in the semantics only in that the ARG1 of this predication will turn out to be an instance for adjectives but an eventuality for adverbs. Given the prior design decisions, even adverbs therefore have an intrinsic eventuality argument in their EPs, and this ARG0 is again employed by the ERG in composing the semantics of adverbial phrases containing a non-scopal modifier of the adverb, as in consistently victoriously. Here again, if victoriously lacked an intrinsic argument, then consistently would have to be treated as a scopal modifier taking the label of victoriously as its argument, leading to a similarly undesirable asymmetry between consistently won and consistently victoriously, assuming that the adverb modifying the verb here is non-scopal.
The presence of these intrinsic arguments in the predications for adverbs and attribute adjectives is thus motivated by the desires to minimize proliferation of lexical entries, to capture the morphological regularity of most adjective-adverb pairs in English, and to enable a consistent representation of non-scopal modifiers when they themselves modify modifiers.
Home | Forum | Discussions | Events