<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zeng-etal-2026-illusion">
    <titleInfo>
        <title>The Illusion of Expertise: Narrow Finetuning Induces Persistent Domain Priors in Vision-Language Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Qinglin</namePart>
        <namePart type="family">Zeng</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jusheng</namePart>
        <namePart type="family">Zhang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jiangqi</namePart>
        <namePart type="family">Huang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ningyuan</namePart>
        <namePart type="family">Liu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Luozhang</namePart>
        <namePart type="family">Xu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wenhao</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Keze</namePart>
        <namePart type="family">Wang</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>Narrow-domain finetuning makes vision-language models (VLMs) sound like domain experts while quietly making them less attentive to the image. We attribute this gap to a persistent domain prior: an input-agnostic shift in hidden state, installed by narrow finetuning, that biases the model toward domain-style outputs regardless of visual evidence. To test the claim we introduce ActDiff, a white-box diagnostic that isolates the finetuning trace by differencing base and finetuned hidden states under progressively degraded images. The trace is compact and persists under blank and Gaussian-noise inputs; the finetuned model is correspondingly less, not more, disrupted when the image is destroyed (gap 0.323 \ensuremath\pm 0.036 nats on BIOMED, generic eval). Subtracting the trace at inference recovers 54% of the MMStar accuracy collapse on BIOMED, with a peak-and-collapse dose response that random and wrong-domain controls do not produce. A single first-moment regulariser added at training time shrinks the trace 34x and improves generic NLL. The prior thus has a clean diagnostic, a causal handle at inference, and a one-line mitigation at training.</abstract>
    <identifier type="citekey">zeng-etal-2026-illusion</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1668/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>37926</start>
            <end>37947</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T The Illusion of Expertise: Narrow Finetuning Induces Persistent Domain Priors in Vision-Language Models
%A Zeng, Qinglin
%A Zhang, Jusheng
%A Huang, Jiangqi
%A Liu, Ningyuan
%A Xu, Luozhang
%A Wang, Wenhao
%A Wang, Keze
%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 zeng-etal-2026-illusion
%X Narrow-domain finetuning makes vision-language models (VLMs) sound like domain experts while quietly making them less attentive to the image. We attribute this gap to a persistent domain prior: an input-agnostic shift in hidden state, installed by narrow finetuning, that biases the model toward domain-style outputs regardless of visual evidence. To test the claim we introduce ActDiff, a white-box diagnostic that isolates the finetuning trace by differencing base and finetuned hidden states under progressively degraded images. The trace is compact and persists under blank and Gaussian-noise inputs; the finetuned model is correspondingly less, not more, disrupted when the image is destroyed (gap 0.323 \ensuremath\pm 0.036 nats on BIOMED, generic eval). Subtracting the trace at inference recovers 54% of the MMStar accuracy collapse on BIOMED, with a peak-and-collapse dose response that random and wrong-domain controls do not produce. A single first-moment regulariser added at training time shrinks the trace 34x and improves generic NLL. The prior thus has a clean diagnostic, a causal handle at inference, and a one-line mitigation at training.
%U https://aclanthology.org/2026.emnlp-main.1668/
%P 37926-37947

Markdown (Informal)

[The Illusion of Expertise: Narrow Finetuning Induces Persistent Domain Priors in Vision-Language Models](https://aclanthology.org/2026.emnlp-main.1668/) (Zeng et al., EMNLP 2026)

• The Illusion of Expertise: Narrow Finetuning Induces Persistent Domain Priors in Vision-Language Models (Zeng et al., EMNLP 2026)

ACL

• Qinglin Zeng, Jusheng Zhang, Jiangqi Huang, Ningyuan Liu, Luozhang Xu, Wenhao Wang, and Keze Wang. 2026. The Illusion of Expertise: Narrow Finetuning Induces Persistent Domain Priors in Vision-Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 37926–37947, Budapest, Hungary. Association for Computational Linguistics.