@inproceedings{wu-etal-2021-evaluating,
title = "Evaluating Neural Model Robustness for Machine Comprehension",
author = "Wu, Winston and
Arendt, Dustin and
Volkova, Svitlana",
editor = "Merlo, Paola and
Tiedemann, Jorg and
Tsarfaty, Reut",
booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume",
month = apr,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.eacl-main.210",
doi = "10.18653/v1/2021.eacl-main.210",
pages = "2470--2481",
abstract = "We evaluate neural model robustness to adversarial attacks using different types of linguistic unit perturbations {--} character and word, and propose a new method for strategic sentence-level perturbations. We experiment with different amounts of perturbations to examine model confidence and misclassification rate, and contrast model performance with different embeddings BERT and ELMo on two benchmark datasets SQuAD and TriviaQA. We demonstrate how to improve model performance during an adversarial attack by using ensembles. Finally, we analyze factors that effect model behavior under adversarial attack, and develop a new model to predict errors during attacks. Our novel findings reveal that (a) unlike BERT, models that use ELMo embeddings are more susceptible to adversarial attacks, (b) unlike word and paraphrase, character perturbations affect the model the most but are most easily compensated for by adversarial training, (c) word perturbations lead to more high-confidence misclassifications compared to sentence- and character-level perturbations, (d) the type of question and model answer length (the longer the answer the more likely it is to be incorrect) is the most predictive of model errors in adversarial setting, and (e) conclusions about model behavior are dataset-specific.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2021-evaluating">
<titleInfo>
<title>Evaluating Neural Model Robustness for Machine Comprehension</title>
</titleInfo>
<name type="personal">
<namePart type="given">Winston</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dustin</namePart>
<namePart type="family">Arendt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Svitlana</namePart>
<namePart type="family">Volkova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2021-04</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume</title>
</titleInfo>
<name type="personal">
<namePart type="given">Paola</namePart>
<namePart type="family">Merlo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jorg</namePart>
<namePart type="family">Tiedemann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Reut</namePart>
<namePart type="family">Tsarfaty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Online</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>We evaluate neural model robustness to adversarial attacks using different types of linguistic unit perturbations – character and word, and propose a new method for strategic sentence-level perturbations. We experiment with different amounts of perturbations to examine model confidence and misclassification rate, and contrast model performance with different embeddings BERT and ELMo on two benchmark datasets SQuAD and TriviaQA. We demonstrate how to improve model performance during an adversarial attack by using ensembles. Finally, we analyze factors that effect model behavior under adversarial attack, and develop a new model to predict errors during attacks. Our novel findings reveal that (a) unlike BERT, models that use ELMo embeddings are more susceptible to adversarial attacks, (b) unlike word and paraphrase, character perturbations affect the model the most but are most easily compensated for by adversarial training, (c) word perturbations lead to more high-confidence misclassifications compared to sentence- and character-level perturbations, (d) the type of question and model answer length (the longer the answer the more likely it is to be incorrect) is the most predictive of model errors in adversarial setting, and (e) conclusions about model behavior are dataset-specific.</abstract>
<identifier type="citekey">wu-etal-2021-evaluating</identifier>
<identifier type="doi">10.18653/v1/2021.eacl-main.210</identifier>
<location>
<url>https://aclanthology.org/2021.eacl-main.210</url>
</location>
<part>
<date>2021-04</date>
<extent unit="page">
<start>2470</start>
<end>2481</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Evaluating Neural Model Robustness for Machine Comprehension
%A Wu, Winston
%A Arendt, Dustin
%A Volkova, Svitlana
%Y Merlo, Paola
%Y Tiedemann, Jorg
%Y Tsarfaty, Reut
%S Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume
%D 2021
%8 April
%I Association for Computational Linguistics
%C Online
%F wu-etal-2021-evaluating
%X We evaluate neural model robustness to adversarial attacks using different types of linguistic unit perturbations – character and word, and propose a new method for strategic sentence-level perturbations. We experiment with different amounts of perturbations to examine model confidence and misclassification rate, and contrast model performance with different embeddings BERT and ELMo on two benchmark datasets SQuAD and TriviaQA. We demonstrate how to improve model performance during an adversarial attack by using ensembles. Finally, we analyze factors that effect model behavior under adversarial attack, and develop a new model to predict errors during attacks. Our novel findings reveal that (a) unlike BERT, models that use ELMo embeddings are more susceptible to adversarial attacks, (b) unlike word and paraphrase, character perturbations affect the model the most but are most easily compensated for by adversarial training, (c) word perturbations lead to more high-confidence misclassifications compared to sentence- and character-level perturbations, (d) the type of question and model answer length (the longer the answer the more likely it is to be incorrect) is the most predictive of model errors in adversarial setting, and (e) conclusions about model behavior are dataset-specific.
%R 10.18653/v1/2021.eacl-main.210
%U https://aclanthology.org/2021.eacl-main.210
%U https://doi.org/10.18653/v1/2021.eacl-main.210
%P 2470-2481
Markdown (Informal)
[Evaluating Neural Model Robustness for Machine Comprehension](https://aclanthology.org/2021.eacl-main.210) (Wu et al., EACL 2021)
ACL
- Winston Wu, Dustin Arendt, and Svitlana Volkova. 2021. Evaluating Neural Model Robustness for Machine Comprehension. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 2470–2481, Online. Association for Computational Linguistics.