@inproceedings{nnamdi-etal-2026-bias,
title = "The bias is in the generator: locating medical chain-of-thought subgroup gaps upstream of the verifier",
author = "Nnamdi, Micky C. and
Marteau, Benoit Louis and
Tan, Shaun Qien Yeau and
Tamo, J. Ben and
Yuan, Yining and
Wang, May Dongmei",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the {A}sia-Pacific Chapter of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.aacl-long.224/",
pages = "5027--5050",
ISBN = "979-8-89176-439-2",
abstract = "Verifier-guided best-of-$N$ decoding has two stages: a generator samples candidate chains of thought, and a verifier selects among them. A subgroup disparity may already be present in the generator{'}s candidates or be amplified during selection. We separate the two stages using adversarial demographic counterfactuals of medical exam questions. On HuatuoGPT-o1-8B, a substantial race accuracy gap is already present in a single sampled response, before verifier selection. Across seven generators, only Gemma-3-27B-it under Skywork-RM shows significant verifier-induced amplification after multiple-testing correction. Even an oracle selector leaves a residual gap on every medical generator analyzed, showing that correct answers are less often available for one subgroup. Counterfactually paired low-rank adaptation (LoRA) of the generator reduces the single-sample gap on two medical models, although on HuatuoGPT-o1-8B the reduction does not persist after self-judge selection. Single-variant controls indicate that the single-sample reduction is not explained by additional training steps alone. A short fairness instruction modestly reduces the single-sample gap, and a direct preference optimization (DPO) variant of the paired training also reduces it and, on HuatuoGPT-o1-8B, keeps race mention near its baseline rate. For the medical generators studied, candidate generation rather than inference-time selection is the primary locus for mitigating this gap."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nnamdi-etal-2026-bias">
<titleInfo>
<title>The bias is in the generator: locating medical chain-of-thought subgroup gaps upstream of the verifier</title>
</titleInfo>
<name type="personal">
<namePart type="given">Micky</namePart>
<namePart type="given">C</namePart>
<namePart type="family">Nnamdi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Benoit</namePart>
<namePart type="given">Louis</namePart>
<namePart type="family">Marteau</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shaun</namePart>
<namePart type="given">Qien</namePart>
<namePart type="given">Yeau</namePart>
<namePart type="family">Tan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">J</namePart>
<namePart type="given">Ben</namePart>
<namePart type="family">Tamo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yining</namePart>
<namePart type="family">Yuan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">May</namePart>
<namePart type="given">Dongmei</namePart>
<namePart type="family">Wang</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 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-439-2</identifier>
</relatedItem>
<abstract>Verifier-guided best-of-N decoding has two stages: a generator samples candidate chains of thought, and a verifier selects among them. A subgroup disparity may already be present in the generator’s candidates or be amplified during selection. We separate the two stages using adversarial demographic counterfactuals of medical exam questions. On HuatuoGPT-o1-8B, a substantial race accuracy gap is already present in a single sampled response, before verifier selection. Across seven generators, only Gemma-3-27B-it under Skywork-RM shows significant verifier-induced amplification after multiple-testing correction. Even an oracle selector leaves a residual gap on every medical generator analyzed, showing that correct answers are less often available for one subgroup. Counterfactually paired low-rank adaptation (LoRA) of the generator reduces the single-sample gap on two medical models, although on HuatuoGPT-o1-8B the reduction does not persist after self-judge selection. Single-variant controls indicate that the single-sample reduction is not explained by additional training steps alone. A short fairness instruction modestly reduces the single-sample gap, and a direct preference optimization (DPO) variant of the paired training also reduces it and, on HuatuoGPT-o1-8B, keeps race mention near its baseline rate. For the medical generators studied, candidate generation rather than inference-time selection is the primary locus for mitigating this gap.</abstract>
<identifier type="citekey">nnamdi-etal-2026-bias</identifier>
<location>
<url>https://aclanthology.org/2026.aacl-long.224/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>5027</start>
<end>5050</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The bias is in the generator: locating medical chain-of-thought subgroup gaps upstream of the verifier
%A Nnamdi, Micky C.
%A Marteau, Benoit Louis
%A Tan, Shaun Qien Yeau
%A Tamo, J. Ben
%A Yuan, Yining
%A Wang, May Dongmei
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Proceedings of the 15th International Joint Conference on Natural Language Processing and the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-439-2
%F nnamdi-etal-2026-bias
%X Verifier-guided best-of-N decoding has two stages: a generator samples candidate chains of thought, and a verifier selects among them. A subgroup disparity may already be present in the generator’s candidates or be amplified during selection. We separate the two stages using adversarial demographic counterfactuals of medical exam questions. On HuatuoGPT-o1-8B, a substantial race accuracy gap is already present in a single sampled response, before verifier selection. Across seven generators, only Gemma-3-27B-it under Skywork-RM shows significant verifier-induced amplification after multiple-testing correction. Even an oracle selector leaves a residual gap on every medical generator analyzed, showing that correct answers are less often available for one subgroup. Counterfactually paired low-rank adaptation (LoRA) of the generator reduces the single-sample gap on two medical models, although on HuatuoGPT-o1-8B the reduction does not persist after self-judge selection. Single-variant controls indicate that the single-sample reduction is not explained by additional training steps alone. A short fairness instruction modestly reduces the single-sample gap, and a direct preference optimization (DPO) variant of the paired training also reduces it and, on HuatuoGPT-o1-8B, keeps race mention near its baseline rate. For the medical generators studied, candidate generation rather than inference-time selection is the primary locus for mitigating this gap.
%U https://aclanthology.org/2026.aacl-long.224/
%P 5027-5050
Markdown (Informal)
[The bias is in the generator: locating medical chain-of-thought subgroup gaps upstream of the verifier](https://aclanthology.org/2026.aacl-long.224/) (Nnamdi et al., AACL-IJCNLP 2026)
ACL