<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="chen-etal-2025-realign">
    <titleInfo>
        <title>ReAlign: Structured Revision for Small Language Model Alignment</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Ruijun</namePart>
        <namePart type="family">Chen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jiajian</namePart>
        <namePart type="family">Guo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hongzhan</namePart>
        <namePart type="family">Chen</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Fanqi</namePart>
        <namePart type="family">Wan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Qifan</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xiaojun</namePart>
        <namePart type="family">Quan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2025-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
        </titleInfo>
        <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">Tanmoy</namePart>
            <namePart type="family">Chakraborty</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Carolyn</namePart>
            <namePart type="family">Rose</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Violet</namePart>
            <namePart type="family">Peng</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Suzhou, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-335-7</identifier>
    </relatedItem>
    <abstract>Aligning small language models with human preferences is challenging, as weak policies struggle to generate informative on-policy samples and suffer from unstable gradients when trained on off-policy signals from stronger models. In this work, we propose ReAlign, a training framework that combines the stability of on-policy learning with the guidance of reviser-assisted supervision. In the ReAlign, we first train a lightweight reviser to improve policy-generated responses using preference-based supervision, conditioned on both the prompt and the initial output. And then, the policy is optimized using a combination of standard on-policy preference pairs and reviser-enhanced pairs constructed as a structured revision task, where the latter provide richer, more learnable feedback. Experimental results on AlpacaEval-2 and Arena-Hard demonstrate that ReAlign significantly boosts alignment performance for weak policies, outperforming strong preference optimization baselines.</abstract>
    <identifier type="citekey">chen-etal-2025-realign</identifier>
    <identifier type="doi">10.18653/v1/2025.findings-emnlp.642</identifier>
    <location>
        <url>https://aclanthology.org/2025.findings-emnlp.642/</url>
    </location>
    <part>
        <date>2025-11</date>
        <extent unit="page">
            <start>12005</start>
            <end>12020</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T ReAlign: Structured Revision for Small Language Model Alignment
%A Chen, Ruijun
%A Guo, Jiajian
%A Chen, Hongzhan
%A Wan, Fanqi
%A Wang, Qifan
%A Quan, Xiaojun
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F chen-etal-2025-realign
%X Aligning small language models with human preferences is challenging, as weak policies struggle to generate informative on-policy samples and suffer from unstable gradients when trained on off-policy signals from stronger models. In this work, we propose ReAlign, a training framework that combines the stability of on-policy learning with the guidance of reviser-assisted supervision. In the ReAlign, we first train a lightweight reviser to improve policy-generated responses using preference-based supervision, conditioned on both the prompt and the initial output. And then, the policy is optimized using a combination of standard on-policy preference pairs and reviser-enhanced pairs constructed as a structured revision task, where the latter provide richer, more learnable feedback. Experimental results on AlpacaEval-2 and Arena-Hard demonstrate that ReAlign significantly boosts alignment performance for weak policies, outperforming strong preference optimization baselines.
%R 10.18653/v1/2025.findings-emnlp.642
%U https://aclanthology.org/2025.findings-emnlp.642/
%U https://doi.org/10.18653/v1/2025.findings-emnlp.642
%P 12005-12020

Markdown (Informal)

[ReAlign: Structured Revision for Small Language Model Alignment](https://aclanthology.org/2025.findings-emnlp.642/) (Chen et al., Findings 2025)

• ReAlign: Structured Revision for Small Language Model Alignment (Chen et al., Findings 2025)

ACL

• Ruijun Chen, Jiajian Guo, Hongzhan Chen, Fanqi Wan, Qifan Wang, and Xiaojun Quan. 2025. ReAlign: Structured Revision for Small Language Model Alignment. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 12005–12020, Suzhou, China. Association for Computational Linguistics.