@inproceedings{arif-etal-2026-grain,
title = "With a Grain of {SALT}: Are {LLM}s Fair Across Social Dimensions?",
author = "Arif, Samee and
Khan, Zohaib and
Butt, Maaidah Kaleem and
Rashid, Muhammad Suhaib and
Raza, Agha Ali and
Athar, Awais",
editor = "Chang, Kai-Wei and
Mehrabi, Ninareh and
Krishna, Satyapriya and
Das, Anubrata and
Dhamala, Jwala and
Cao, Yang Trista and
Kumarage, Tharindu and
Ramakrishna, Anil and
Christodoulopoulos, Christos and
Wan, Yixin and
Galystan, Aram and
Kumar, Anoop and
Gupta, Rahul",
booktitle = "Proceedings of the 6th Workshop on Trustworthy {NLP} ({T}rust{NLP} 2026)",
month = jul,
year = "2026",
address = "San Diego, California",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.trustnlp-main.48/",
pages = "618--636",
ISBN = "979-8-89176-418-7",
abstract = "In this paper we present a systematic study of social bias in small- to mid-scale Large Language Models (LLMs), focusing on gender, religion, and race. Using our SALT (\textbf{S}ocial \textbf{A}ppropriateness in \textbf{L}LM \textbf{T}ext) dataset, we explore two bias categories{---}Theoretical and Practical. Theoretical bias covers General Debate and Positioned Debate while practical bias includes Career Advice, Personal Advice, and Resume Generation. We quantify bias using win-rate gaps in general debate, and negative-role assignments in positioned debate. For Practical bias, we anonymize model outputs to remove explicit demographic cues and use DeepSeek-R1 as an automated evaluator, measuring outcome disparities across groups. We also examine systemic issues in LLM-based evaluation including evaluation bias, positional bias, and length bias and validate our findings through human annotation. Our results show consistent disadvantages for White, Christian, and male-associated outputs across multiple tasks. Larger models often amplify these disparities, highlighting that scale does not guarantee fairness."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="arif-etal-2026-grain">
<titleInfo>
<title>With a Grain of SALT: Are LLMs Fair Across Social Dimensions?</title>
</titleInfo>
<name type="personal">
<namePart type="given">Samee</namePart>
<namePart type="family">Arif</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zohaib</namePart>
<namePart type="family">Khan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maaidah</namePart>
<namePart type="given">Kaleem</namePart>
<namePart type="family">Butt</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Muhammad</namePart>
<namePart type="given">Suhaib</namePart>
<namePart type="family">Rashid</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Agha</namePart>
<namePart type="given">Ali</namePart>
<namePart type="family">Raza</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Awais</namePart>
<namePart type="family">Athar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 6th Workshop on Trustworthy NLP (TrustNLP 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Kai-Wei</namePart>
<namePart type="family">Chang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ninareh</namePart>
<namePart type="family">Mehrabi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Satyapriya</namePart>
<namePart type="family">Krishna</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anubrata</namePart>
<namePart type="family">Das</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jwala</namePart>
<namePart type="family">Dhamala</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="given">Trista</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tharindu</namePart>
<namePart type="family">Kumarage</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anil</namePart>
<namePart type="family">Ramakrishna</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yixin</namePart>
<namePart type="family">Wan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aram</namePart>
<namePart type="family">Galystan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Anoop</namePart>
<namePart type="family">Kumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rahul</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-418-7</identifier>
</relatedItem>
<abstract>In this paper we present a systematic study of social bias in small- to mid-scale Large Language Models (LLMs), focusing on gender, religion, and race. Using our SALT (Social Appropriateness in LLM Text) dataset, we explore two bias categories—Theoretical and Practical. Theoretical bias covers General Debate and Positioned Debate while practical bias includes Career Advice, Personal Advice, and Resume Generation. We quantify bias using win-rate gaps in general debate, and negative-role assignments in positioned debate. For Practical bias, we anonymize model outputs to remove explicit demographic cues and use DeepSeek-R1 as an automated evaluator, measuring outcome disparities across groups. We also examine systemic issues in LLM-based evaluation including evaluation bias, positional bias, and length bias and validate our findings through human annotation. Our results show consistent disadvantages for White, Christian, and male-associated outputs across multiple tasks. Larger models often amplify these disparities, highlighting that scale does not guarantee fairness.</abstract>
<identifier type="citekey">arif-etal-2026-grain</identifier>
<location>
<url>https://aclanthology.org/2026.trustnlp-main.48/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>618</start>
<end>636</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T With a Grain of SALT: Are LLMs Fair Across Social Dimensions?
%A Arif, Samee
%A Khan, Zohaib
%A Butt, Maaidah Kaleem
%A Rashid, Muhammad Suhaib
%A Raza, Agha Ali
%A Athar, Awais
%Y Chang, Kai-Wei
%Y Mehrabi, Ninareh
%Y Krishna, Satyapriya
%Y Das, Anubrata
%Y Dhamala, Jwala
%Y Cao, Yang Trista
%Y Kumarage, Tharindu
%Y Ramakrishna, Anil
%Y Christodoulopoulos, Christos
%Y Wan, Yixin
%Y Galystan, Aram
%Y Kumar, Anoop
%Y Gupta, Rahul
%S Proceedings of the 6th Workshop on Trustworthy NLP (TrustNLP 2026)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California
%@ 979-8-89176-418-7
%F arif-etal-2026-grain
%X In this paper we present a systematic study of social bias in small- to mid-scale Large Language Models (LLMs), focusing on gender, religion, and race. Using our SALT (Social Appropriateness in LLM Text) dataset, we explore two bias categories—Theoretical and Practical. Theoretical bias covers General Debate and Positioned Debate while practical bias includes Career Advice, Personal Advice, and Resume Generation. We quantify bias using win-rate gaps in general debate, and negative-role assignments in positioned debate. For Practical bias, we anonymize model outputs to remove explicit demographic cues and use DeepSeek-R1 as an automated evaluator, measuring outcome disparities across groups. We also examine systemic issues in LLM-based evaluation including evaluation bias, positional bias, and length bias and validate our findings through human annotation. Our results show consistent disadvantages for White, Christian, and male-associated outputs across multiple tasks. Larger models often amplify these disparities, highlighting that scale does not guarantee fairness.
%U https://aclanthology.org/2026.trustnlp-main.48/
%P 618-636
Markdown (Informal)
[With a Grain of SALT: Are LLMs Fair Across Social Dimensions?](https://aclanthology.org/2026.trustnlp-main.48/) (Arif et al., TrustNLP 2026)
ACL
- Samee Arif, Zohaib Khan, Maaidah Kaleem Butt, Muhammad Suhaib Rashid, Agha Ali Raza, and Awais Athar. 2026. With a Grain of SALT: Are LLMs Fair Across Social Dimensions?. In Proceedings of the 6th Workshop on Trustworthy NLP (TrustNLP 2026), pages 618–636, San Diego, California. Association for Computational Linguistics.