@inproceedings{satheesh-etal-2026-neutrality,
title = "On the Neutrality of Model Confidence in {G}erman and {E}nglish Gender-Bias Benchmarks",
author = "Satheesh, Shalaka and
Jacob, Nika and
Beckh, Katharina",
editor = "Atwell, Katherine and
Borah, Angana and
Dementieva, Daryna and
Kreiss, Elisa and
Kotonya, Neema and
Liu, Jiarui and
Olson, Liz and
Wan, Ruyuan and
Zhao, Jieyu",
booktitle = "Proceedings of the Fifth Workshop on {NLP} for Positive Impact ({NLP}4{PI})",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.nlp4pi-1.29/",
pages = "339--348",
ISBN = "979-8-89176-495-8",
abstract = "Confidence, the probability a model assigns to its prediction, is commonly treated as a signal for correctness, and is central to quality monitoring and selective question answering. We ask whether this signal is neutral with respect to social stereotypes. A model{'}s confidence may reflect correctness well overall and yet be higher where its answers align with a stereotype and detecting this requires comparing confidence across stereotype directions. We investigate this on the gender subset of Bias Benchmark for Question Answering (BBQ) in English and in German (GG-BBQ), evaluating 22 base and instruction-tuned models from five providers. For some models, we find small but directionally consistent effects for both ambiguous and disambiguated questions, suggesting that confidence-based selection can favour stereotype-aligned content and that confidence is worth examining alongside answer-based scores in bias evaluations on multiple-choice questions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="satheesh-etal-2026-neutrality">
<titleInfo>
<title>On the Neutrality of Model Confidence in German and English Gender-Bias Benchmarks</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shalaka</namePart>
<namePart type="family">Satheesh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nika</namePart>
<namePart type="family">Jacob</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Katharina</namePart>
<namePart type="family">Beckh</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 Fifth Workshop on NLP for Positive Impact (NLP4PI)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Katherine</namePart>
<namePart type="family">Atwell</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Angana</namePart>
<namePart type="family">Borah</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daryna</namePart>
<namePart type="family">Dementieva</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Elisa</namePart>
<namePart type="family">Kreiss</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Neema</namePart>
<namePart type="family">Kotonya</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiarui</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Liz</namePart>
<namePart type="family">Olson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruyuan</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jieyu</namePart>
<namePart type="family">Zhao</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">979-8-89176-495-8</identifier>
</relatedItem>
<abstract>Confidence, the probability a model assigns to its prediction, is commonly treated as a signal for correctness, and is central to quality monitoring and selective question answering. We ask whether this signal is neutral with respect to social stereotypes. A model’s confidence may reflect correctness well overall and yet be higher where its answers align with a stereotype and detecting this requires comparing confidence across stereotype directions. We investigate this on the gender subset of Bias Benchmark for Question Answering (BBQ) in English and in German (GG-BBQ), evaluating 22 base and instruction-tuned models from five providers. For some models, we find small but directionally consistent effects for both ambiguous and disambiguated questions, suggesting that confidence-based selection can favour stereotype-aligned content and that confidence is worth examining alongside answer-based scores in bias evaluations on multiple-choice questions.</abstract>
<identifier type="citekey">satheesh-etal-2026-neutrality</identifier>
<location>
<url>https://aclanthology.org/2026.nlp4pi-1.29/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>339</start>
<end>348</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T On the Neutrality of Model Confidence in German and English Gender-Bias Benchmarks
%A Satheesh, Shalaka
%A Jacob, Nika
%A Beckh, Katharina
%Y Atwell, Katherine
%Y Borah, Angana
%Y Dementieva, Daryna
%Y Kreiss, Elisa
%Y Kotonya, Neema
%Y Liu, Jiarui
%Y Olson, Liz
%Y Wan, Ruyuan
%Y Zhao, Jieyu
%S Proceedings of the Fifth Workshop on NLP for Positive Impact (NLP4PI)
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-495-8
%F satheesh-etal-2026-neutrality
%X Confidence, the probability a model assigns to its prediction, is commonly treated as a signal for correctness, and is central to quality monitoring and selective question answering. We ask whether this signal is neutral with respect to social stereotypes. A model’s confidence may reflect correctness well overall and yet be higher where its answers align with a stereotype and detecting this requires comparing confidence across stereotype directions. We investigate this on the gender subset of Bias Benchmark for Question Answering (BBQ) in English and in German (GG-BBQ), evaluating 22 base and instruction-tuned models from five providers. For some models, we find small but directionally consistent effects for both ambiguous and disambiguated questions, suggesting that confidence-based selection can favour stereotype-aligned content and that confidence is worth examining alongside answer-based scores in bias evaluations on multiple-choice questions.
%U https://aclanthology.org/2026.nlp4pi-1.29/
%P 339-348
Markdown (Informal)
[On the Neutrality of Model Confidence in German and English Gender-Bias Benchmarks](https://aclanthology.org/2026.nlp4pi-1.29/) (Satheesh et al., NLP4PI 2026)
ACL