Machine Reading, Fast and Slow: When Do Models “Understand” Language?

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Standard

Machine Reading, Fast and Slow: When Do Models “Understand” Language? / Ray Choudhury, Sagnik; Rogers, Anna; Augenstein, Isabelle.

Proceedings of the 29th International Conference on Computational Linguistics. Association for Computational Linguistics (ACL), 2022. p. 78–93.

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Harvard

Ray Choudhury, S, Rogers, A & Augenstein, I 2022, Machine Reading, Fast and Slow: When Do Models “Understand” Language? in Proceedings of the 29th International Conference on Computational Linguistics. Association for Computational Linguistics (ACL), pp. 78–93, THE 29TH
INTERNATIONAL CONFERENCE ON
COMPUTATIONAL LINGUISTICS, 12/10/2022. <https://aclanthology.org/2022.coling-1.8/>

APA

Ray Choudhury, S., Rogers, A., & Augenstein, I. (2022). Machine Reading, Fast and Slow: When Do Models “Understand” Language? In Proceedings of the 29th International Conference on Computational Linguistics (pp. 78–93). Association for Computational Linguistics (ACL). https://aclanthology.org/2022.coling-1.8/

Vancouver

Ray Choudhury S, Rogers A, Augenstein I. Machine Reading, Fast and Slow: When Do Models “Understand” Language? In Proceedings of the 29th International Conference on Computational Linguistics. Association for Computational Linguistics (ACL). 2022. p. 78–93

Author

Ray Choudhury, Sagnik ; Rogers, Anna ; Augenstein, Isabelle. / Machine Reading, Fast and Slow: When Do Models “Understand” Language?. Proceedings of the 29th International Conference on Computational Linguistics. Association for Computational Linguistics (ACL), 2022. pp. 78–93

Bibtex

@inproceedings{8fc5f176b9694f01a7f5fcf8fe8154ee,
title = "Machine Reading, Fast and Slow: When Do Models “Understand” Language?",
abstract = "Two of the most fundamental issues in Natural Language Understanding (NLU) at present are: (a) how it can established whether deep learning-based models score highly on NLU benchmarks for the ”right” reasons; and (b) what those reasons would even be. We investigate the behavior of reading comprehension models with respect to two linguistic ”skills”: coreference resolution and comparison. We propose a definition for the reasoning steps expected from a system that would be ”reading slowly”, and compare that with the behavior of five models of the BERT family of various sizes, observed through saliency scores and counterfactual explanations. We find that for comparison (but not coreference) the systems based on larger encoders are more likely to rely on the ”right” information, but even they struggle with generalization, suggesting that they still learn specific lexical patterns rather than the general principles of comparison.",
author = "{Ray Choudhury}, Sagnik and Anna Rogers and Isabelle Augenstein",
year = "2022",
language = "English",
pages = "78–93",
booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
publisher = "Association for Computational Linguistics (ACL)",
address = "United States",
note = "THE 29TH<br/>INTERNATIONAL CONFERENCE ON<br/>COMPUTATIONAL LINGUISTICS, COLIN 2022 ; Conference date: 12-10-2022 Through 17-10-2022",

}

RIS

TY - GEN

T1 - Machine Reading, Fast and Slow: When Do Models “Understand” Language?

AU - Ray Choudhury, Sagnik

AU - Rogers, Anna

AU - Augenstein, Isabelle

PY - 2022

Y1 - 2022

N2 - Two of the most fundamental issues in Natural Language Understanding (NLU) at present are: (a) how it can established whether deep learning-based models score highly on NLU benchmarks for the ”right” reasons; and (b) what those reasons would even be. We investigate the behavior of reading comprehension models with respect to two linguistic ”skills”: coreference resolution and comparison. We propose a definition for the reasoning steps expected from a system that would be ”reading slowly”, and compare that with the behavior of five models of the BERT family of various sizes, observed through saliency scores and counterfactual explanations. We find that for comparison (but not coreference) the systems based on larger encoders are more likely to rely on the ”right” information, but even they struggle with generalization, suggesting that they still learn specific lexical patterns rather than the general principles of comparison.

AB - Two of the most fundamental issues in Natural Language Understanding (NLU) at present are: (a) how it can established whether deep learning-based models score highly on NLU benchmarks for the ”right” reasons; and (b) what those reasons would even be. We investigate the behavior of reading comprehension models with respect to two linguistic ”skills”: coreference resolution and comparison. We propose a definition for the reasoning steps expected from a system that would be ”reading slowly”, and compare that with the behavior of five models of the BERT family of various sizes, observed through saliency scores and counterfactual explanations. We find that for comparison (but not coreference) the systems based on larger encoders are more likely to rely on the ”right” information, but even they struggle with generalization, suggesting that they still learn specific lexical patterns rather than the general principles of comparison.

M3 - Article in proceedings

SP - 78

EP - 93

BT - Proceedings of the 29th International Conference on Computational Linguistics

PB - Association for Computational Linguistics (ACL)

T2 - THE 29TH<br/>INTERNATIONAL CONFERENCE ON<br/>COMPUTATIONAL LINGUISTICS

Y2 - 12 October 2022 through 17 October 2022

ER -

ID: 341057090