<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="goyal-etal-2026-vastu">
    <titleInfo>
        <title>VASTU: Language Models Struggle to Recognize Online Community Values</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Agam</namePart>
        <namePart type="family">Goyal</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Xianyang</namePart>
        <namePart type="family">Zhan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Charlotte</namePart>
        <namePart type="family">Lambert</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Koustuv</namePart>
        <namePart type="family">Saha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Eshwar</namePart>
        <namePart type="family">Chandrasekharan</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>Online communities develop distinct norms for content they collectively value, yet it remains unclear whether current language models can recognize locally valued contributions in context. We formalize this as community-conditioned preference prediction and introduce Vastu (\underlineValue-\underlineAware \underlineSocial \underlineTuning), a benchmark of 75,000 Reddit comments from 15 communities spanning Gaming, Science, Q&amp;A, Advice, and Politics. We evaluate four model families—prompted LLMs, LoRA-adapted SLMs, supervised encoders, and feature-based classifiers—across global, local, and context-conditioned settings. Our central finding is that parametric adaptation consistently outperforms prompting: supervised encoders reach 0.74 AUROC and fine-tuned SLMs 0.64–0.71, while the best prompted result is only 0.62. This gap is not merely quantitative—vanilla prompting yields over 80% false-negative rates, systematically discarding content communities actually value. Conversational context narrows but does not close this divide. Together, these results suggest that local preference recognition requires community-specific training signal, not just better prompting. Our work supports future research on community-aware reward modeling, feed curation, and positive moderation.</abstract>
    <identifier type="citekey">goyal-etal-2026-vastu</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.231/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>5311</start>
            <end>5327</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T VASTU: Language Models Struggle to Recognize Online Community Values
%A Goyal, Agam
%A Zhan, Xianyang
%A Lambert, Charlotte
%A Saha, Koustuv
%A Chandrasekharan, Eshwar
%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 goyal-etal-2026-vastu
%X Online communities develop distinct norms for content they collectively value, yet it remains unclear whether current language models can recognize locally valued contributions in context. We formalize this as community-conditioned preference prediction and introduce Vastu (\underlineValue-\underlineAware \underlineSocial \underlineTuning), a benchmark of 75,000 Reddit comments from 15 communities spanning Gaming, Science, Q&A, Advice, and Politics. We evaluate four model families—prompted LLMs, LoRA-adapted SLMs, supervised encoders, and feature-based classifiers—across global, local, and context-conditioned settings. Our central finding is that parametric adaptation consistently outperforms prompting: supervised encoders reach 0.74 AUROC and fine-tuned SLMs 0.64–0.71, while the best prompted result is only 0.62. This gap is not merely quantitative—vanilla prompting yields over 80% false-negative rates, systematically discarding content communities actually value. Conversational context narrows but does not close this divide. Together, these results suggest that local preference recognition requires community-specific training signal, not just better prompting. Our work supports future research on community-aware reward modeling, feed curation, and positive moderation.
%U https://aclanthology.org/2026.emnlp-main.231/
%P 5311-5327

Markdown (Informal)

[VASTU: Language Models Struggle to Recognize Online Community Values](https://aclanthology.org/2026.emnlp-main.231/) (Goyal et al., EMNLP 2026)

• VASTU: Language Models Struggle to Recognize Online Community Values (Goyal et al., EMNLP 2026)

ACL

• Agam Goyal, Xianyang Zhan, Charlotte Lambert, Koustuv Saha, and Eshwar Chandrasekharan. 2026. VASTU: Language Models Struggle to Recognize Online Community Values. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 5311–5327, Budapest, Hungary. Association for Computational Linguistics.