@inproceedings{zhao-bethard-2020-berts,
title = "How does {BERT}`s attention change when you fine-tune? An analysis methodology and a case study in negation scope",
author = "Zhao, Yiyun and
Bethard, Steven",
editor = "Jurafsky, Dan and
Chai, Joyce and
Schluter, Natalie and
Tetreault, Joel",
booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
month = jul,
year = "2020",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2020.acl-main.429/",
doi = "10.18653/v1/2020.acl-main.429",
pages = "4729--4747",
abstract = "Large pretrained language models like BERT, after fine-tuning to a downstream task, have achieved high performance on a variety of NLP problems. Yet explaining their decisions is difficult despite recent work probing their internal representations. We propose a procedure and analysis methods that take a hypothesis of how a transformer-based model might encode a linguistic phenomenon, and test the validity of that hypothesis based on a comparison between knowledge-related downstream tasks with downstream control tasks, and measurement of cross-dataset consistency. We apply this methodology to test BERT and RoBERTa on a hypothesis that some attention heads will consistently attend from a word in negation scope to the negation cue. We find that after fine-tuning BERT and RoBERTa on a negation scope task, the average attention head improves its sensitivity to negation and its attention consistency across negation datasets compared to the pre-trained models. However, only the base models (not the large models) improve compared to a control task, indicating there is evidence for a shallow encoding of negation only in the base models."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-bethard-2020-berts">
<titleInfo>
<title>How does BERT‘s attention change when you fine-tune? An analysis methodology and a case study in negation scope</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yiyun</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Steven</namePart>
<namePart type="family">Bethard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2020-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics</title>
</titleInfo>
<name type="personal">
<namePart type="given">Dan</namePart>
<namePart type="family">Jurafsky</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Chai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Natalie</namePart>
<namePart type="family">Schluter</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joel</namePart>
<namePart type="family">Tetreault</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>Large pretrained language models like BERT, after fine-tuning to a downstream task, have achieved high performance on a variety of NLP problems. Yet explaining their decisions is difficult despite recent work probing their internal representations. We propose a procedure and analysis methods that take a hypothesis of how a transformer-based model might encode a linguistic phenomenon, and test the validity of that hypothesis based on a comparison between knowledge-related downstream tasks with downstream control tasks, and measurement of cross-dataset consistency. We apply this methodology to test BERT and RoBERTa on a hypothesis that some attention heads will consistently attend from a word in negation scope to the negation cue. We find that after fine-tuning BERT and RoBERTa on a negation scope task, the average attention head improves its sensitivity to negation and its attention consistency across negation datasets compared to the pre-trained models. However, only the base models (not the large models) improve compared to a control task, indicating there is evidence for a shallow encoding of negation only in the base models.</abstract>
<identifier type="citekey">zhao-bethard-2020-berts</identifier>
<identifier type="doi">10.18653/v1/2020.acl-main.429</identifier>
<location>
<url>https://aclanthology.org/2020.acl-main.429/</url>
</location>
<part>
<date>2020-07</date>
<extent unit="page">
<start>4729</start>
<end>4747</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T How does BERT‘s attention change when you fine-tune? An analysis methodology and a case study in negation scope
%A Zhao, Yiyun
%A Bethard, Steven
%Y Jurafsky, Dan
%Y Chai, Joyce
%Y Schluter, Natalie
%Y Tetreault, Joel
%S Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics
%D 2020
%8 July
%I Association for Computational Linguistics
%C Online
%F zhao-bethard-2020-berts
%X Large pretrained language models like BERT, after fine-tuning to a downstream task, have achieved high performance on a variety of NLP problems. Yet explaining their decisions is difficult despite recent work probing their internal representations. We propose a procedure and analysis methods that take a hypothesis of how a transformer-based model might encode a linguistic phenomenon, and test the validity of that hypothesis based on a comparison between knowledge-related downstream tasks with downstream control tasks, and measurement of cross-dataset consistency. We apply this methodology to test BERT and RoBERTa on a hypothesis that some attention heads will consistently attend from a word in negation scope to the negation cue. We find that after fine-tuning BERT and RoBERTa on a negation scope task, the average attention head improves its sensitivity to negation and its attention consistency across negation datasets compared to the pre-trained models. However, only the base models (not the large models) improve compared to a control task, indicating there is evidence for a shallow encoding of negation only in the base models.
%R 10.18653/v1/2020.acl-main.429
%U https://aclanthology.org/2020.acl-main.429/
%U https://doi.org/10.18653/v1/2020.acl-main.429
%P 4729-4747
Markdown (Informal)
[How does BERT’s attention change when you fine-tune? An analysis methodology and a case study in negation scope](https://aclanthology.org/2020.acl-main.429/) (Zhao & Bethard, ACL 2020)
ACL