lint

doc/bibliography.bib

@@ -10,31 +10,31 @@ @Book{jm3
   }
 
 @misc{chen2022blasertextfreespeechtospeechtranslation,
-      title={BLASER: A Text-Free Speech-to-Speech Translation Evaluation Metric}, 
+      title={BLASER: A Text-Free Speech-to-Speech Translation Evaluation Metric},
       author={Mingda Chen and Paul-Ambroise Duquenne and Pierre Andrews and Justine Kao and Alexandre Mourachko and Holger Schwenk and Marta R. Costa-jussà},
       year={2022},
       eprint={2212.08486},
       archivePrefix={arXiv},
       primaryClass={cs.CL},
-      url={https://arxiv.org/abs/2212.08486}, 
+      url={https://arxiv.org/abs/2212.08486},
 }
 
 @misc{duquenne2022speechmatrixlargescaleminedcorpus,
-      title={SpeechMatrix: A Large-Scale Mined Corpus of Multilingual Speech-to-Speech Translations}, 
+      title={SpeechMatrix: A Large-Scale Mined Corpus of Multilingual Speech-to-Speech Translations},
       author={Paul-Ambroise Duquenne and Hongyu Gong and Ning Dong and Jingfei Du and Ann Lee and Vedanuj Goswani and Changhan Wang and Juan Pino and Benoît Sagot and Holger Schwenk},
       year={2022},
       eprint={2211.04508},
       archivePrefix={arXiv},
       primaryClass={cs.CL},
-      url={https://arxiv.org/abs/2211.04508}, 
+      url={https://arxiv.org/abs/2211.04508},
 }
 
 @misc{hendrycks2019benchmarkingneuralnetworkrobustness,
-      title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations}, 
+      title={Benchmarking Neural Network Robustness to Common Corruptions and Perturbations},
       author={Dan Hendrycks and Thomas Dietterich},
       year={2019},
       eprint={1903.12261},
       archivePrefix={arXiv},
       primaryClass={cs.LG},
-      url={https://arxiv.org/abs/1903.12261}, 
+      url={https://arxiv.org/abs/1903.12261},
 }

doc/notes.typ

@@ -19,7 +19,7 @@ We want to study the behaviour of $cal(M)$ when the source text is "corrupted" b
 If this assumption is not true, then we might have to think harder about separating metric variance due to semantic change vs due to change in style.
 Either way, the translation will be influenced by the domain shift due to filler words, and we can sample metric evaluations $M$, as shown below.
 
-//In our case, they want to know which metric is the most robust against filler words. 
+//In our case, they want to know which metric is the most robust against filler words.
 //This is not an adversarial case, since filler words occur naturally.
 //We don't know the real world distribution of filler words, but we could use a LLM to sample from $bb(P)(hat(x) | x)$, where $x$ is the clean input, and $hat(x)$ is the filler-word-corrupted input.