<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="le-etal-2026-audit">
    <titleInfo>
        <title>When Audit Quality Fails to Predict Downstream Utility: Synthetic-Data Selection for Low-Resource Classification</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Xuan-Bach</namePart>
        <namePart type="family">Le</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Phat</namePart>
        <namePart type="given">T</namePart>
        <namePart type="family">Tran-Truong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Son</namePart>
        <namePart type="given">Ha</namePart>
        <namePart type="family">Xuan</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
    </relatedItem>
    <abstract>Quality-aware synthetic-data selection assumes that examples rated highly by an LLM judge will help downstream learning. We test this assumption in a controlled replay across two low-resource African-language classification tasks, four languages, and seven selectors. Six selectors retain 64 examples per task-language cell. CoSDA-V2 yields the highest judged label correctness (0.904 versus 0.767 for naive) and the lowest measured shortcut score, but AlpaGasus leads CoSDA-V2 on downstream Macro-F1 in both the main replay (0.202 versus 0.163) and the three-seed refit (0.193 versus 0.157).\par Adding the audit’s strict gate to AlpaGasus reduces Macro-F1 by 0.029, driven by one cell where only 31 of 64 candidates survive. Because the gate changes pool composition and training-set size, this comparison measures their joint effect. LESS also falls below naive when unconstrained top-k collapses label balance, but recovers with quotas. At this budget, retention constraints account for more of the observed differences than a consistent ranking advantage. In this replay, audit scores do not reliably predict downstream utility. We therefore recommend reporting both metrics for the same retained sets.</abstract>
    <identifier type="citekey">le-etal-2026-audit</identifier>
    <location>
        <url>https://aclanthology.org/2026.findings-emnlp.1184/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>25027</start>
            <end>25045</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T When Audit Quality Fails to Predict Downstream Utility: Synthetic-Data Selection for Low-Resource Classification
%A Le, Xuan-Bach
%A Tran-Truong, Phat T.
%A Xuan, Son Ha
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F le-etal-2026-audit
%X Quality-aware synthetic-data selection assumes that examples rated highly by an LLM judge will help downstream learning. We test this assumption in a controlled replay across two low-resource African-language classification tasks, four languages, and seven selectors. Six selectors retain 64 examples per task-language cell. CoSDA-V2 yields the highest judged label correctness (0.904 versus 0.767 for naive) and the lowest measured shortcut score, but AlpaGasus leads CoSDA-V2 on downstream Macro-F1 in both the main replay (0.202 versus 0.163) and the three-seed refit (0.193 versus 0.157).\par Adding the audit’s strict gate to AlpaGasus reduces Macro-F1 by 0.029, driven by one cell where only 31 of 64 candidates survive. Because the gate changes pool composition and training-set size, this comparison measures their joint effect. LESS also falls below naive when unconstrained top-k collapses label balance, but recovers with quotas. At this budget, retention constraints account for more of the observed differences than a consistent ranking advantage. In this replay, audit scores do not reliably predict downstream utility. We therefore recommend reporting both metrics for the same retained sets.
%U https://aclanthology.org/2026.findings-emnlp.1184/
%P 25027-25045

Markdown (Informal)

[When Audit Quality Fails to Predict Downstream Utility: Synthetic-Data Selection for Low-Resource Classification](https://aclanthology.org/2026.findings-emnlp.1184/) (Le et al., Findings 2026)

• When Audit Quality Fails to Predict Downstream Utility: Synthetic-Data Selection for Low-Resource Classification (Le et al., Findings 2026)

ACL

• Xuan-Bach Le, Phat T. Tran-Truong, and Son Ha Xuan. 2026. When Audit Quality Fails to Predict Downstream Utility: Synthetic-Data Selection for Low-Resource Classification. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 25027–25045, Budapest, Hungary. Association for Computational Linguistics.