<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="nguyen-etal-2026-trialign">
    <titleInfo>
        <title>TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Thi-Nhung</namePart>
        <namePart type="family">Nguyen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Linhao</namePart>
        <namePart type="family">Luo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Rollin</namePart>
        <namePart type="family">Omari</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Junae</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Thuy-Trang</namePart>
        <namePart type="family">Vu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Dinh</namePart>
        <namePart type="family">Phung</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>Personalized large language models adapt responses to users’ preferences and social attributes, but can introduce substantial universal truth inconsistencies across social groups, where some groups systematically receive less accurate responses on objective tasks. Existing alignment methods either ignore personalization or mainly focus on subjective preference alignment, largely overlooking fairness and consistency in universal truths. To address this gap, we study Truth-Invariant Alignment (TIA), an alignment problem for personalized LLMs that aims to ensure universal truths remain consistent across social groups while preserving personalization. We propose TriAlign, the first offline multi-agent reinforcement learning (MARL) framework for TIA, where each social group is modeled as an agent interacting. TriAlign jointly optimizes universal truth accuracy, cross-group truth consistency, and personalization through a fairness-aware objective and an explicit inconsistency penalty. Experiments across diverse benchmarks demonstrate that TriAlign achieves a stronger balance among these three objectives than strong baselines, reducing universal truth disparities across social groups while improving both objective task performance and personalization quality.</abstract>
    <identifier type="citekey">nguyen-etal-2026-trialign</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2640/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>60056</start>
            <end>60079</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment
%A Nguyen, Thi-Nhung
%A Luo, Linhao
%A Omari, Rollin
%A Kim, Junae
%A Vu, Thuy-Trang
%A Phung, Dinh
%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 nguyen-etal-2026-trialign
%X Personalized large language models adapt responses to users’ preferences and social attributes, but can introduce substantial universal truth inconsistencies across social groups, where some groups systematically receive less accurate responses on objective tasks. Existing alignment methods either ignore personalization or mainly focus on subjective preference alignment, largely overlooking fairness and consistency in universal truths. To address this gap, we study Truth-Invariant Alignment (TIA), an alignment problem for personalized LLMs that aims to ensure universal truths remain consistent across social groups while preserving personalization. We propose TriAlign, the first offline multi-agent reinforcement learning (MARL) framework for TIA, where each social group is modeled as an agent interacting. TriAlign jointly optimizes universal truth accuracy, cross-group truth consistency, and personalization through a fairness-aware objective and an explicit inconsistency penalty. Experiments across diverse benchmarks demonstrate that TriAlign achieves a stronger balance among these three objectives than strong baselines, reducing universal truth disparities across social groups while improving both objective task performance and personalization quality.
%U https://aclanthology.org/2026.emnlp-main.2640/
%P 60056-60079

Markdown (Informal)

[TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment](https://aclanthology.org/2026.emnlp-main.2640/) (Nguyen et al., EMNLP 2026)

• TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment (Nguyen et al., EMNLP 2026)

ACL

• Thi-Nhung Nguyen, Linhao Luo, Rollin Omari, Junae Kim, Thuy-Trang Vu, and Dinh Phung. 2026. TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 60056–60079, Budapest, Hungary. Association for Computational Linguistics.