@inproceedings{park-zubiaga-2026-better,
    title = "Better Scores, Worse Grounding: Hidden Regressions after Fine-Tuning in Dialogue Fact Verification",
    author = "Park, Hyunkyung  and
      Zubiaga, Arkaitz",
    editor = "Choi, Jinho D.  and
      Chen, Yun-Nung  and
      Funakoshi, Kotaro  and
      Emami, Ali",
    booktitle = "Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue",
    month = aug,
    year = "2026",
    address = "Atlanta, Georgia, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.sigdial-1.51/",
    pages = "720--737",
    abstract = "In dialogue fact verification (DFV), responses often depend on prior turns for correct interpretation, yet systems are still judged mainly by aggregate benchmark scores. We study a hidden grounding regression: aggregate Macro-F1 improves after fine-tuning while previously correct, context-dependent pronoun cases become newly wrong and show stronger premise-side sensitivity than cases that remain correct. On three referent-annotated audit sets constructed from DialFact and FaithDial, we audit six encoder-only verifiers before and after matched source-specific fine-tuning through prediction-transition analysis and the Premise-Preference Score (PPS), a control-adjusted masking diagnostic. Fine-tuning improves Macro-F1 across the six-model/three-evaluation-set panel, yet newly regressed cases show stronger premise-side sensitivity than stable-correct cases under PPS in 17 of 18 evaluated comparisons, indicating that aggregate gains can conceal regressions on a controlled dialogue-grounding audit."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="park-zubiaga-2026-better">
    <titleInfo>
        <title>Better Scores, Worse Grounding: Hidden Regressions after Fine-Tuning in Dialogue Fact Verification</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Hyunkyung</namePart>
        <namePart type="family">Park</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Arkaitz</namePart>
        <namePart type="family">Zubiaga</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-08</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Jinho</namePart>
            <namePart type="given">D</namePart>
            <namePart type="family">Choi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Yun-Nung</namePart>
            <namePart type="family">Chen</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Kotaro</namePart>
            <namePart type="family">Funakoshi</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Ali</namePart>
            <namePart type="family">Emami</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Atlanta, Georgia, USA</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>In dialogue fact verification (DFV), responses often depend on prior turns for correct interpretation, yet systems are still judged mainly by aggregate benchmark scores. We study a hidden grounding regression: aggregate Macro-F1 improves after fine-tuning while previously correct, context-dependent pronoun cases become newly wrong and show stronger premise-side sensitivity than cases that remain correct. On three referent-annotated audit sets constructed from DialFact and FaithDial, we audit six encoder-only verifiers before and after matched source-specific fine-tuning through prediction-transition analysis and the Premise-Preference Score (PPS), a control-adjusted masking diagnostic. Fine-tuning improves Macro-F1 across the six-model/three-evaluation-set panel, yet newly regressed cases show stronger premise-side sensitivity than stable-correct cases under PPS in 17 of 18 evaluated comparisons, indicating that aggregate gains can conceal regressions on a controlled dialogue-grounding audit.</abstract>
    <identifier type="citekey">park-zubiaga-2026-better</identifier>
    <location>
        <url>https://aclanthology.org/2026.sigdial-1.51/</url>
    </location>
    <part>
        <date>2026-08</date>
        <extent unit="page">
            <start>720</start>
            <end>737</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Better Scores, Worse Grounding: Hidden Regressions after Fine-Tuning in Dialogue Fact Verification
%A Park, Hyunkyung
%A Zubiaga, Arkaitz
%Y Choi, Jinho D.
%Y Chen, Yun-Nung
%Y Funakoshi, Kotaro
%Y Emami, Ali
%S Proceedings of the 27th Annual Meeting of the Special Interest Group on Discourse and Dialogue
%D 2026
%8 August
%I Association for Computational Linguistics
%C Atlanta, Georgia, USA
%F park-zubiaga-2026-better
%X In dialogue fact verification (DFV), responses often depend on prior turns for correct interpretation, yet systems are still judged mainly by aggregate benchmark scores. We study a hidden grounding regression: aggregate Macro-F1 improves after fine-tuning while previously correct, context-dependent pronoun cases become newly wrong and show stronger premise-side sensitivity than cases that remain correct. On three referent-annotated audit sets constructed from DialFact and FaithDial, we audit six encoder-only verifiers before and after matched source-specific fine-tuning through prediction-transition analysis and the Premise-Preference Score (PPS), a control-adjusted masking diagnostic. Fine-tuning improves Macro-F1 across the six-model/three-evaluation-set panel, yet newly regressed cases show stronger premise-side sensitivity than stable-correct cases under PPS in 17 of 18 evaluated comparisons, indicating that aggregate gains can conceal regressions on a controlled dialogue-grounding audit.
%U https://aclanthology.org/2026.sigdial-1.51/
%P 720-737
Markdown (Informal)

[Better Scores, Worse Grounding: Hidden Regressions after Fine-Tuning in Dialogue Fact Verification](https://aclanthology.org/2026.sigdial-1.51/) (Park & Zubiaga, SIGDIAL 2026)

ACL