LEXPLAIN: Improving Model Explanations via Lexicon Supervision

Orevaoghene Ahia; Hila Gonen; Vidhisha Balachandran; Yulia Tsvetkov; Noah A. Smith

LEXPLAIN: Improving Model Explanations via Lexicon Supervision

Orevaoghene Ahia, Hila Gonen, Vidhisha Balachandran, Yulia Tsvetkov, Noah A. Smith

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

Abstract

Model explanations that shed light on the model's predictions are becoming a desired additional output of NLP models, alongside their predictions. Challenges in creating these explanations include making them trustworthy and faithful to the model's predictions. In this work, we propose a novel framework for guiding model explanations by supervising them explicitly. To this end, our method, LEXPLAIN, uses task-related lexicons to directly supervise model explanations. This approach consistently improves the plausibility of model's explanations without sacrificing performance on the task, as we demonstrate on sentiment analysis and toxicity detection. Our analyses show that our method also demotes spurious correlations (i.e., with respect to African American English dialect) on toxicity detection, improving fairness.

Original language	English
Title of host publication	StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference
Editors	Alexis Palmer, Jose Camacho-Collados
Publisher	Association for Computational Linguistics (ACL)
Pages	207-216
Number of pages	10
ISBN (Electronic)	9781959429760
State	Published - 2023
Externally published	Yes
Event	12th Joint Conference on Lexical and Computational Semantics, StarSEM 2023, co-located with ACL 2023 - Toronto, Canada Duration: 13 Jul 2023 → 14 Jul 2023

Publication series

Name	Proceedings of the Annual Meeting of the Association for Computational Linguistics
ISSN (Print)	0736-587X

Conference

Conference	12th Joint Conference on Lexical and Computational Semantics, StarSEM 2023, co-located with ACL 2023
Country/Territory	Canada
City	Toronto
Period	13/07/23 → 14/07/23

Bibliographical note

Publisher Copyright:
© 2023 Association for Computational Linguistics.

Funding

This research is supported in part by by the National Science Foundation (NSF) under grants IIS2203097 and IIS2125201. This research is supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), via the HIATUS Program contract #2022-22072200004. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for governmental purposes notwithstanding any copyright annotation therein.

Funders	Funder number
National Science Foundation	IIS2203097, IIS2125201
Office of the Director of National Intelligence
Intelligence Advanced Research Projects Activity	2022-22072200004

Cite this

Ahia, O., Gonen, H., Balachandran, V., Tsvetkov, Y., & Smith, N. A. (2023). LEXPLAIN: Improving Model Explanations via Lexicon Supervision. In A. Palmer, & J. Camacho-Collados (Eds.), StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference (pp. 207-216). (Proceedings of the Annual Meeting of the Association for Computational Linguistics). Association for Computational Linguistics (ACL).

Ahia, Orevaoghene ; Gonen, Hila ; Balachandran, Vidhisha et al. / LEXPLAIN : Improving Model Explanations via Lexicon Supervision. StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference. editor / Alexis Palmer ; Jose Camacho-Collados. Association for Computational Linguistics (ACL), 2023. pp. 207-216 (Proceedings of the Annual Meeting of the Association for Computational Linguistics).

@inproceedings{640a756a72b04d339aa0ba5f7c55a94e,

title = "LEXPLAIN: Improving Model Explanations via Lexicon Supervision",

abstract = "Model explanations that shed light on the model's predictions are becoming a desired additional output of NLP models, alongside their predictions. Challenges in creating these explanations include making them trustworthy and faithful to the model's predictions. In this work, we propose a novel framework for guiding model explanations by supervising them explicitly. To this end, our method, LEXPLAIN, uses task-related lexicons to directly supervise model explanations. This approach consistently improves the plausibility of model's explanations without sacrificing performance on the task, as we demonstrate on sentiment analysis and toxicity detection. Our analyses show that our method also demotes spurious correlations (i.e., with respect to African American English dialect) on toxicity detection, improving fairness.",

author = "Orevaoghene Ahia and Hila Gonen and Vidhisha Balachandran and Yulia Tsvetkov and Smith, {Noah A.}",

note = "Publisher Copyright: {\textcopyright} 2023 Association for Computational Linguistics.; 12th Joint Conference on Lexical and Computational Semantics, StarSEM 2023, co-located with ACL 2023 ; Conference date: 13-07-2023 Through 14-07-2023",

year = "2023",

language = "אנגלית",

series = "Proceedings of the Annual Meeting of the Association for Computational Linguistics",

publisher = "Association for Computational Linguistics (ACL)",

pages = "207--216",

editor = "Alexis Palmer and Jose Camacho-Collados",

booktitle = "StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference",

address = "ארצות הברית",

}

Ahia, O, Gonen, H, Balachandran, V, Tsvetkov, Y & Smith, NA 2023, LEXPLAIN: Improving Model Explanations via Lexicon Supervision. in A Palmer & J Camacho-Collados (eds), StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference. Proceedings of the Annual Meeting of the Association for Computational Linguistics, Association for Computational Linguistics (ACL), pp. 207-216, 12th Joint Conference on Lexical and Computational Semantics, StarSEM 2023, co-located with ACL 2023, Toronto, Canada, 13/07/23.

LEXPLAIN: Improving Model Explanations via Lexicon Supervision. / Ahia, Orevaoghene; Gonen, Hila; Balachandran, Vidhisha et al.
StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference. ed. / Alexis Palmer; Jose Camacho-Collados. Association for Computational Linguistics (ACL), 2023. p. 207-216 (Proceedings of the Annual Meeting of the Association for Computational Linguistics).

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

TY - GEN

T1 - LEXPLAIN

T2 - 12th Joint Conference on Lexical and Computational Semantics, StarSEM 2023, co-located with ACL 2023

AU - Ahia, Orevaoghene

AU - Gonen, Hila

AU - Balachandran, Vidhisha

AU - Tsvetkov, Yulia

AU - Smith, Noah A.

PY - 2023

Y1 - 2023

N2 - Model explanations that shed light on the model's predictions are becoming a desired additional output of NLP models, alongside their predictions. Challenges in creating these explanations include making them trustworthy and faithful to the model's predictions. In this work, we propose a novel framework for guiding model explanations by supervising them explicitly. To this end, our method, LEXPLAIN, uses task-related lexicons to directly supervise model explanations. This approach consistently improves the plausibility of model's explanations without sacrificing performance on the task, as we demonstrate on sentiment analysis and toxicity detection. Our analyses show that our method also demotes spurious correlations (i.e., with respect to African American English dialect) on toxicity detection, improving fairness.

AB - Model explanations that shed light on the model's predictions are becoming a desired additional output of NLP models, alongside their predictions. Challenges in creating these explanations include making them trustworthy and faithful to the model's predictions. In this work, we propose a novel framework for guiding model explanations by supervising them explicitly. To this end, our method, LEXPLAIN, uses task-related lexicons to directly supervise model explanations. This approach consistently improves the plausibility of model's explanations without sacrificing performance on the task, as we demonstrate on sentiment analysis and toxicity detection. Our analyses show that our method also demotes spurious correlations (i.e., with respect to African American English dialect) on toxicity detection, improving fairness.

UR - http://www.scopus.com/inward/record.url?scp=85175400418&partnerID=8YFLogxK

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:85175400418

T3 - Proceedings of the Annual Meeting of the Association for Computational Linguistics

SP - 207

EP - 216

BT - StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference

A2 - Palmer, Alexis

A2 - Camacho-Collados, Jose

PB - Association for Computational Linguistics (ACL)

Y2 - 13 July 2023 through 14 July 2023

ER -

Ahia O, Gonen H, Balachandran V, Tsvetkov Y, Smith NA. LEXPLAIN: Improving Model Explanations via Lexicon Supervision. In Palmer A, Camacho-Collados J, editors, StarSEM 2023 - 12th Joint Conference on Lexical and Computational Semantics, Proceedings of the Conference. Association for Computational Linguistics (ACL). 2023. p. 207-216. (Proceedings of the Annual Meeting of the Association for Computational Linguistics).

LEXPLAIN: Improving Model Explanations via Lexicon Supervision

Abstract

Publication series

Conference

Bibliographical note

Funding

Other files and links

Fingerprint

Cite this