Counterfactual Interventions Reveal the Causal Effect of Relative Clause Representations on Agreement Prediction

Shauli Ravfogel, Grusha Prasad, Tal Linzen, Yoav Goldberg

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

36 Scopus citations

Abstract

When language models process syntactically complex sentences, do they use their representations of syntax in a manner that is consistent with the grammar of the language? We propose AlterRep, an intervention-based method to address this question. For any linguistic feature of a given sentence, AlterRep generates counterfactual representations by altering how the feature is encoded, while leaving intact all other aspects of the original representation. By measuring the change in a model’s word prediction behavior when these counterfactual representations are substituted for the original ones, we can draw conclusions about the causal effect of the linguistic feature in question on the model’s behavior. We apply this method to study how BERT models of different sizes process relative clauses (RCs). We find that BERT variants use RC boundary information during word prediction in a manner that is consistent with the rules of English grammar; this RC boundary information generalizes to a considerable extent across different RC types, suggesting that BERT represents RCs as an abstract linguistic category.

Original languageEnglish
Title of host publicationCoNLL 2021 - 25th Conference on Computational Natural Language Learning, Proceedings
EditorsArianna Bisazza, Omri Abend
PublisherAssociation for Computational Linguistics (ACL)
Pages194-209
Number of pages16
ISBN (Electronic)9781955917056
StatePublished - 2021
Event25th Conference on Computational Natural Language Learning, CoNLL 2021 - Virtual, Online
Duration: 10 Nov 202111 Nov 2021

Publication series

NameCoNLL 2021 - 25th Conference on Computational Natural Language Learning, Proceedings

Conference

Conference25th Conference on Computational Natural Language Learning, CoNLL 2021
CityVirtual, Online
Period10/11/2111/11/21

Bibliographical note

Publisher Copyright:
© 2021 Association for Computational Linguistics.

Funding

This work was supported by United States–Israel Binational Science Foundation award 2018284, and has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme, grant agreement No. 802774 (iEXTRACT). We thank Robert Frank for fruitful discussion of an early version of this work, and Marius Mosbach for his helpful comments.

FundersFunder number
Horizon 2020 Framework Programme802774
European Commission
United States-Israel Binational Science Foundation2018284

    Fingerprint

    Dive into the research topics of 'Counterfactual Interventions Reveal the Causal Effect of Relative Clause Representations on Agreement Prediction'. Together they form a unique fingerprint.

    Cite this