To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation

Research output: Chapter in book/report/conference proceedingConference contributionResearchpeer review

View graph of relations

Details

Original languageEnglish
Title of host publicationProceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science
Place of PublicationAbu Dhabi, United Arab Emirates
Pages39-51
Number of pages13
ISBN (electronic)9781959429203
Publication statusPublished - Nov 2022
EventThe 2022 Conference on Empirical Methods in Natural Language Processing - Abu Dhabi, United Arab Emirates
Duration: 7 Dec 202211 Dec 2022

Publication series

NameNLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022

Abstract

Gender bias may emerge from an unequal representation of agency and power, for example, by portraying women frequently as passive and powerless (“She accepted her future”) and men as proactive and powerful (“He chose his future”). When language models learn from respective texts, they may reproduce or even amplify the bias. An effective way to mitigate bias is to generate counterfactual sentences with opposite agency and power to the training. Recent work targeted agency-specific verbs from a lexicon to this end. We argue that this is insufficient, due to the interaction of agency and power and their dependence on context. In this paper, we thus develop a new rewriting model that identifies verbs with the desired agency and power in the context of the given sentence. The verbs’ probability is then boosted to encourage the model to rewrite both connotations jointly. According to automatic metrics, our model effectively controls for power while being competitive in agency to the state of the art. In our main evaluation, human annotators favored its counterfactuals in terms of both connotations, also deeming its meaning preservation better.

ASJC Scopus subject areas

Sustainable Development Goals

Cite this

To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation. / Stahl, Maja; Spliethöver, Maximilian; Wachsmuth, Henning.
Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science. Abu Dhabi, United Arab Emirates, 2022. p. 39-51 (NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022).

Research output: Chapter in book/report/conference proceedingConference contributionResearchpeer review

Stahl, M, Spliethöver, M & Wachsmuth, H 2022, To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation. in Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science. NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022, Abu Dhabi, United Arab Emirates, pp. 39-51, The 2022 Conference on Empirical Methods in Natural Language Processing, Abu Dhabi, United Arab Emirates, 7 Dec 2022. <https://aclanthology.org/2022.nlpcss-1.6/>
Stahl, M., Spliethöver, M., & Wachsmuth, H. (2022). To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation. In Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science (pp. 39-51). (NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022).. https://aclanthology.org/2022.nlpcss-1.6/
Stahl M, Spliethöver M, Wachsmuth H. To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation. In Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science. Abu Dhabi, United Arab Emirates. 2022. p. 39-51. (NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022).
Stahl, Maja ; Spliethöver, Maximilian ; Wachsmuth, Henning. / To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation. Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science. Abu Dhabi, United Arab Emirates, 2022. pp. 39-51 (NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022).
Download
@inproceedings{48399e8886b443998c12f7af9c092dda,
title = "To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation",
abstract = "Gender bias may emerge from an unequal representation of agency and power, for example, by portraying women frequently as passive and powerless (“She accepted her future”) and men as proactive and powerful (“He chose his future”). When language models learn from respective texts, they may reproduce or even amplify the bias. An effective way to mitigate bias is to generate counterfactual sentences with opposite agency and power to the training. Recent work targeted agency-specific verbs from a lexicon to this end. We argue that this is insufficient, due to the interaction of agency and power and their dependence on context. In this paper, we thus develop a new rewriting model that identifies verbs with the desired agency and power in the context of the given sentence. The verbs{\textquoteright} probability is then boosted to encourage the model to rewrite both connotations jointly. According to automatic metrics, our model effectively controls for power while being competitive in agency to the state of the art. In our main evaluation, human annotators favored its counterfactuals in terms of both connotations, also deeming its meaning preservation better.",
author = "Maja Stahl and Maximilian Splieth{\"o}ver and Henning Wachsmuth",
note = "Publisher Copyright: {\textcopyright} 2022 Association for Computational Linguistics.; The 2022 Conference on Empirical Methods in Natural Language Processing ; Conference date: 07-12-2022 Through 11-12-2022",
year = "2022",
month = nov,
language = "English",
series = "NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022",
pages = "39--51",
booktitle = "Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science",

}

Download

TY - GEN

T1 - To Prefer or to Choose? Generating Agency and Power Counterfactuals Jointly for Gender Bias Mitigation

AU - Stahl, Maja

AU - Spliethöver, Maximilian

AU - Wachsmuth, Henning

N1 - Publisher Copyright: © 2022 Association for Computational Linguistics.

PY - 2022/11

Y1 - 2022/11

N2 - Gender bias may emerge from an unequal representation of agency and power, for example, by portraying women frequently as passive and powerless (“She accepted her future”) and men as proactive and powerful (“He chose his future”). When language models learn from respective texts, they may reproduce or even amplify the bias. An effective way to mitigate bias is to generate counterfactual sentences with opposite agency and power to the training. Recent work targeted agency-specific verbs from a lexicon to this end. We argue that this is insufficient, due to the interaction of agency and power and their dependence on context. In this paper, we thus develop a new rewriting model that identifies verbs with the desired agency and power in the context of the given sentence. The verbs’ probability is then boosted to encourage the model to rewrite both connotations jointly. According to automatic metrics, our model effectively controls for power while being competitive in agency to the state of the art. In our main evaluation, human annotators favored its counterfactuals in terms of both connotations, also deeming its meaning preservation better.

AB - Gender bias may emerge from an unequal representation of agency and power, for example, by portraying women frequently as passive and powerless (“She accepted her future”) and men as proactive and powerful (“He chose his future”). When language models learn from respective texts, they may reproduce or even amplify the bias. An effective way to mitigate bias is to generate counterfactual sentences with opposite agency and power to the training. Recent work targeted agency-specific verbs from a lexicon to this end. We argue that this is insufficient, due to the interaction of agency and power and their dependence on context. In this paper, we thus develop a new rewriting model that identifies verbs with the desired agency and power in the context of the given sentence. The verbs’ probability is then boosted to encourage the model to rewrite both connotations jointly. According to automatic metrics, our model effectively controls for power while being competitive in agency to the state of the art. In our main evaluation, human annotators favored its counterfactuals in terms of both connotations, also deeming its meaning preservation better.

UR - http://www.scopus.com/inward/record.url?scp=85154595235&partnerID=8YFLogxK

M3 - Conference contribution

T3 - NLPCSS 2022 - 5th Workshop on Natural Language Processing and Computational Social Science ,NLP+CSS, Held at the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022

SP - 39

EP - 51

BT - Proceedings of the Fifth Workshop on Natural Language Processing and Computational Social Science

CY - Abu Dhabi, United Arab Emirates

T2 - The 2022 Conference on Empirical Methods in Natural Language Processing

Y2 - 7 December 2022 through 11 December 2022

ER -

By the same author(s)