@inproceedings{samu-islam-2026-bap,
title = "{B}a{P}-Bench: A Bias-Aware {B}angla Paraphrase Benchmark for Measuring Instruction-Guided Gender Debiasing and Evaluating Transformer Models",
author = "Samu, Most. Sharmin Sultana and
Islam, Md. Rakibul",
editor = "Fale{\'n}ska, Agnieszka and
Basta, Christine and
Costa-juss{\`a}, Marta and
Sta{\'n}czak, Karolina and
Nozza, Debora and
Attanasio, Giuseppe and
Neplenbroek, Vera and
Hardmeier, Christian",
booktitle = "Proceedings of the 7th Workshop on Gender Bias in Natural Language Processing ({G}e{BNLP})",
month = nov,
year = "2026",
address = "Hengqin, China (remote)",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.gebnlp-1.9/",
pages = "126--144",
ISBN = "979-8-89176-447-7",
abstract = "Large language models (LLMs) are widely used for paraphrase generation, making it important to evaluate their ability to produce fair outputs under explicit debiasing instructions, particularly for under-resourced languages such as Bangla. Existing work has focused on paraphrase generation and detection but lacks benchmarks for evaluating instruction-guided debiasing. We introduce BaP-Bench, a bias-aware Bangla paraphrase benchmark constructed using four LLMs to evaluate instruction-guided mitigation of stance/sentiment toward women (as a proxy for gender bias). To enable large-scale evaluation, we employ automatic classifier-based annotation to assess the effectiveness of LLM-generated debiased paraphrases. We further demonstrate the utility of BaP-Bench as a downstream benchmark by evaluating transformer-based models on paraphrase detection under debiasing instructions. Results show that LLMs vary considerably in their ability to follow debiasing instructions while preserving paraphrase quality, and that transformer-based models achieve strong performance on the downstream task. BaP-Bench provides a foundation for studying instruction-guided fairness in Bangla paraphrasing and supports future research on bias-aware paraphrase generation and evaluation."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="samu-islam-2026-bap">
<titleInfo>
<title>BaP-Bench: A Bias-Aware Bangla Paraphrase Benchmark for Measuring Instruction-Guided Gender Debiasing and Evaluating Transformer Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Most.</namePart>
<namePart type="given">Sharmin</namePart>
<namePart type="given">Sultana</namePart>
<namePart type="family">Samu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Md.</namePart>
<namePart type="given">Rakibul</namePart>
<namePart type="family">Islam</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 7th Workshop on Gender Bias in Natural Language Processing (GeBNLP)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Agnieszka</namePart>
<namePart type="family">Faleńska</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christine</namePart>
<namePart type="family">Basta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Marta</namePart>
<namePart type="family">Costa-jussà</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Karolina</namePart>
<namePart type="family">Stańczak</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Debora</namePart>
<namePart type="family">Nozza</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Giuseppe</namePart>
<namePart type="family">Attanasio</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vera</namePart>
<namePart type="family">Neplenbroek</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christian</namePart>
<namePart type="family">Hardmeier</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China (remote)</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-447-7</identifier>
</relatedItem>
<abstract>Large language models (LLMs) are widely used for paraphrase generation, making it important to evaluate their ability to produce fair outputs under explicit debiasing instructions, particularly for under-resourced languages such as Bangla. Existing work has focused on paraphrase generation and detection but lacks benchmarks for evaluating instruction-guided debiasing. We introduce BaP-Bench, a bias-aware Bangla paraphrase benchmark constructed using four LLMs to evaluate instruction-guided mitigation of stance/sentiment toward women (as a proxy for gender bias). To enable large-scale evaluation, we employ automatic classifier-based annotation to assess the effectiveness of LLM-generated debiased paraphrases. We further demonstrate the utility of BaP-Bench as a downstream benchmark by evaluating transformer-based models on paraphrase detection under debiasing instructions. Results show that LLMs vary considerably in their ability to follow debiasing instructions while preserving paraphrase quality, and that transformer-based models achieve strong performance on the downstream task. BaP-Bench provides a foundation for studying instruction-guided fairness in Bangla paraphrasing and supports future research on bias-aware paraphrase generation and evaluation.</abstract>
<identifier type="citekey">samu-islam-2026-bap</identifier>
<location>
<url>https://aclanthology.org/2026.gebnlp-1.9/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>126</start>
<end>144</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T BaP-Bench: A Bias-Aware Bangla Paraphrase Benchmark for Measuring Instruction-Guided Gender Debiasing and Evaluating Transformer Models
%A Samu, Most. Sharmin Sultana
%A Islam, Md. Rakibul
%Y Faleńska, Agnieszka
%Y Basta, Christine
%Y Costa-jussà, Marta
%Y Stańczak, Karolina
%Y Nozza, Debora
%Y Attanasio, Giuseppe
%Y Neplenbroek, Vera
%Y Hardmeier, Christian
%S Proceedings of the 7th Workshop on Gender Bias in Natural Language Processing (GeBNLP)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China (remote)
%@ 979-8-89176-447-7
%F samu-islam-2026-bap
%X Large language models (LLMs) are widely used for paraphrase generation, making it important to evaluate their ability to produce fair outputs under explicit debiasing instructions, particularly for under-resourced languages such as Bangla. Existing work has focused on paraphrase generation and detection but lacks benchmarks for evaluating instruction-guided debiasing. We introduce BaP-Bench, a bias-aware Bangla paraphrase benchmark constructed using four LLMs to evaluate instruction-guided mitigation of stance/sentiment toward women (as a proxy for gender bias). To enable large-scale evaluation, we employ automatic classifier-based annotation to assess the effectiveness of LLM-generated debiased paraphrases. We further demonstrate the utility of BaP-Bench as a downstream benchmark by evaluating transformer-based models on paraphrase detection under debiasing instructions. Results show that LLMs vary considerably in their ability to follow debiasing instructions while preserving paraphrase quality, and that transformer-based models achieve strong performance on the downstream task. BaP-Bench provides a foundation for studying instruction-guided fairness in Bangla paraphrasing and supports future research on bias-aware paraphrase generation and evaluation.
%U https://aclanthology.org/2026.gebnlp-1.9/
%P 126-144
Markdown (Informal)
[BaP-Bench: A Bias-Aware Bangla Paraphrase Benchmark for Measuring Instruction-Guided Gender Debiasing and Evaluating Transformer Models](https://aclanthology.org/2026.gebnlp-1.9/) (Samu & Islam, GeBNLP 2026)
ACL