@inproceedings{joshi-etal-2022-er,
title = "{ER}-Test: Evaluating Explanation Regularization Methods for Language Models",
author = "Joshi, Brihi and
Chan, Aaron and
Liu, Ziyi and
Nie, Shaoliang and
Sanjabi, Maziar and
Firooz, Hamed and
Ren, Xiang",
editor = "Goldberg, Yoav and
Kozareva, Zornitsa and
Zhang, Yue",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2022",
month = dec,
year = "2022",
address = "Abu Dhabi, United Arab Emirates",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.findings-emnlp.242",
doi = "10.18653/v1/2022.findings-emnlp.242",
pages = "3315--3336",
abstract = "By explaining how humans would solve a given task, human rationales can provide strong learning signal for neural language models (NLMs). Explanation regularization (ER) aims to improve NLM generalization by pushing the NLM{'}s machine rationales (Which input tokens did the NLM focus on?) to align with human rationales (Which input tokens would humans focus on). Though prior works primarily study ER via in-distribution (ID) evaluation, out-of-distribution (OOD) generalization is often more critical in real-world scenarios, yet ER{'}s effect on OOD generalization has been underexplored.In this paper, we introduce ER-Test, a framework for evaluating ER models{'} OOD generalization along three dimensions: unseen datasets, contrast set tests, and functional tests. Using ER-Test, we comprehensively analyze how ER models{'} OOD generalization varies with the rationale alignment criterion (loss function), human rationale type (instance-level v/s task-level), number and choice of rationale-annotated instances, and time budget for rationale annotation. Across two tasks and six datasets, we show that ER has little impact on ID performance but yields large OOD performance gains, with the best ER criterion being task-dependent. Also, ER can improve OOD performance even with task-level or few human rationales. Finally, we find that rationale annotation is more time-efficient than label annotation for improving OOD performance. Our results with ER-Test help demonstrate ER{'}s utility and establish best practices for using ER effectively.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="joshi-etal-2022-er">
<titleInfo>
<title>ER-Test: Evaluating Explanation Regularization Methods for Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Brihi</namePart>
<namePart type="family">Joshi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aaron</namePart>
<namePart type="family">Chan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ziyi</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaoliang</namePart>
<namePart type="family">Nie</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maziar</namePart>
<namePart type="family">Sanjabi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hamed</namePart>
<namePart type="family">Firooz</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiang</namePart>
<namePart type="family">Ren</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-12</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2022</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yoav</namePart>
<namePart type="family">Goldberg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zornitsa</namePart>
<namePart type="family">Kozareva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Abu Dhabi, United Arab Emirates</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>By explaining how humans would solve a given task, human rationales can provide strong learning signal for neural language models (NLMs). Explanation regularization (ER) aims to improve NLM generalization by pushing the NLM’s machine rationales (Which input tokens did the NLM focus on?) to align with human rationales (Which input tokens would humans focus on). Though prior works primarily study ER via in-distribution (ID) evaluation, out-of-distribution (OOD) generalization is often more critical in real-world scenarios, yet ER’s effect on OOD generalization has been underexplored.In this paper, we introduce ER-Test, a framework for evaluating ER models’ OOD generalization along three dimensions: unseen datasets, contrast set tests, and functional tests. Using ER-Test, we comprehensively analyze how ER models’ OOD generalization varies with the rationale alignment criterion (loss function), human rationale type (instance-level v/s task-level), number and choice of rationale-annotated instances, and time budget for rationale annotation. Across two tasks and six datasets, we show that ER has little impact on ID performance but yields large OOD performance gains, with the best ER criterion being task-dependent. Also, ER can improve OOD performance even with task-level or few human rationales. Finally, we find that rationale annotation is more time-efficient than label annotation for improving OOD performance. Our results with ER-Test help demonstrate ER’s utility and establish best practices for using ER effectively.</abstract>
<identifier type="citekey">joshi-etal-2022-er</identifier>
<identifier type="doi">10.18653/v1/2022.findings-emnlp.242</identifier>
<location>
<url>https://aclanthology.org/2022.findings-emnlp.242</url>
</location>
<part>
<date>2022-12</date>
<extent unit="page">
<start>3315</start>
<end>3336</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ER-Test: Evaluating Explanation Regularization Methods for Language Models
%A Joshi, Brihi
%A Chan, Aaron
%A Liu, Ziyi
%A Nie, Shaoliang
%A Sanjabi, Maziar
%A Firooz, Hamed
%A Ren, Xiang
%Y Goldberg, Yoav
%Y Kozareva, Zornitsa
%Y Zhang, Yue
%S Findings of the Association for Computational Linguistics: EMNLP 2022
%D 2022
%8 December
%I Association for Computational Linguistics
%C Abu Dhabi, United Arab Emirates
%F joshi-etal-2022-er
%X By explaining how humans would solve a given task, human rationales can provide strong learning signal for neural language models (NLMs). Explanation regularization (ER) aims to improve NLM generalization by pushing the NLM’s machine rationales (Which input tokens did the NLM focus on?) to align with human rationales (Which input tokens would humans focus on). Though prior works primarily study ER via in-distribution (ID) evaluation, out-of-distribution (OOD) generalization is often more critical in real-world scenarios, yet ER’s effect on OOD generalization has been underexplored.In this paper, we introduce ER-Test, a framework for evaluating ER models’ OOD generalization along three dimensions: unseen datasets, contrast set tests, and functional tests. Using ER-Test, we comprehensively analyze how ER models’ OOD generalization varies with the rationale alignment criterion (loss function), human rationale type (instance-level v/s task-level), number and choice of rationale-annotated instances, and time budget for rationale annotation. Across two tasks and six datasets, we show that ER has little impact on ID performance but yields large OOD performance gains, with the best ER criterion being task-dependent. Also, ER can improve OOD performance even with task-level or few human rationales. Finally, we find that rationale annotation is more time-efficient than label annotation for improving OOD performance. Our results with ER-Test help demonstrate ER’s utility and establish best practices for using ER effectively.
%R 10.18653/v1/2022.findings-emnlp.242
%U https://aclanthology.org/2022.findings-emnlp.242
%U https://doi.org/10.18653/v1/2022.findings-emnlp.242
%P 3315-3336
Markdown (Informal)
[ER-Test: Evaluating Explanation Regularization Methods for Language Models](https://aclanthology.org/2022.findings-emnlp.242) (Joshi et al., Findings 2022)
ACL