@inproceedings{wang-etal-2026-arca,
title = "{ARCA}: Attribution-Regularized Causal Alignment for Task-Aware Bias Mitigation",
author = "Wang, Botao and
Li, Lin and
Shi, Kaize and
Liu, Jiahao and
Lyu, Chong and
Zhong, Xian",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.420/",
pages = "9616--9633",
ISBN = "979-8-89176-459-0",
abstract = "Pretrained masked language models (MLMs) remain important backbones for natural language understanding (NLU) tasks, as they often provide a more favorable balance between task performance and computational efficiency than large decoder-only models. However, existing debiasing methods often decouple bias mitigation from downstream fine-tuning, making it difficult to preserve the mitigation effects after task adaptation. To address this, we propose ARCA (Attribution-Regularized Causal Alignment), an end-to-end framework that unifies debiasing quality and task utility. ARCA uses differential gradient attribution to identify task-specific bias words and regularize the intervention for causal alignment on counterfactual samples. This single-pipeline approach enables joint optimization of task utility and debiasing. Experiments on three mainstream NLU tasks show that ARCA achieves a strong trade-off between bias mitigation and task utility compared with representative debiasing baselines."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-arca">
<titleInfo>
<title>ARCA: Attribution-Regularized Causal Alignment for Task-Aware Bias Mitigation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Botao</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lin</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kaize</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiahao</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Chong</namePart>
<namePart type="family">Lyu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xian</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Pretrained masked language models (MLMs) remain important backbones for natural language understanding (NLU) tasks, as they often provide a more favorable balance between task performance and computational efficiency than large decoder-only models. However, existing debiasing methods often decouple bias mitigation from downstream fine-tuning, making it difficult to preserve the mitigation effects after task adaptation. To address this, we propose ARCA (Attribution-Regularized Causal Alignment), an end-to-end framework that unifies debiasing quality and task utility. ARCA uses differential gradient attribution to identify task-specific bias words and regularize the intervention for causal alignment on counterfactual samples. This single-pipeline approach enables joint optimization of task utility and debiasing. Experiments on three mainstream NLU tasks show that ARCA achieves a strong trade-off between bias mitigation and task utility compared with representative debiasing baselines.</abstract>
<identifier type="citekey">wang-etal-2026-arca</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.420/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>9616</start>
<end>9633</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ARCA: Attribution-Regularized Causal Alignment for Task-Aware Bias Mitigation
%A Wang, Botao
%A Li, Lin
%A Shi, Kaize
%A Liu, Jiahao
%A Lyu, Chong
%A Zhong, Xian
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F wang-etal-2026-arca
%X Pretrained masked language models (MLMs) remain important backbones for natural language understanding (NLU) tasks, as they often provide a more favorable balance between task performance and computational efficiency than large decoder-only models. However, existing debiasing methods often decouple bias mitigation from downstream fine-tuning, making it difficult to preserve the mitigation effects after task adaptation. To address this, we propose ARCA (Attribution-Regularized Causal Alignment), an end-to-end framework that unifies debiasing quality and task utility. ARCA uses differential gradient attribution to identify task-specific bias words and regularize the intervention for causal alignment on counterfactual samples. This single-pipeline approach enables joint optimization of task utility and debiasing. Experiments on three mainstream NLU tasks show that ARCA achieves a strong trade-off between bias mitigation and task utility compared with representative debiasing baselines.
%U https://aclanthology.org/2026.emnlp-main.420/
%P 9616-9633
Markdown (Informal)
[ARCA: Attribution-Regularized Causal Alignment for Task-Aware Bias Mitigation](https://aclanthology.org/2026.emnlp-main.420/) (Wang et al., EMNLP 2026)
ACL