<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="han-etal-2026-multilingual">
    <titleInfo>
        <title>Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages?</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Donghoon</namePart>
        <namePart type="family">Han</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">SungHyun</namePart>
        <namePart type="family">Moon</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Aidyn</namePart>
        <namePart type="family">Zhakatayev</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Junghun</namePart>
        <namePart type="family">Cha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">SeungJae</namePart>
        <namePart type="family">Lee</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>Recent multilingual vision–language encoders cover hundreds of languages in a single model, yet on two state-of-the-art instances retrieval on low-resource languages (LRL; e.g. Swahili) trails high-resource ones (HRL; e.g. English) by 30⁺ pp. We ask where in the trained encoder this gap is located. Prior modality-gap and cross-lingual subspace work suggests a linear language direction at the output crowds out alignment-relevant geometry. We falsify this: LEACE drives the linear language classifier from &gt;99% to near chance and iterated INLP to 37–50% while LRL retrieval moves within \pm 1.5 pp and all tier means within 2.2 pp, tracking random controls. The linear bias is a symptom, not the cause. Instead, the alignment-causal factor lies along the encoder’s forward path: the EOS (end-of-sequence) hidden state’s per-language trajectory diverges with depth. Substituting the EOS with its parallel English value three blocks before the projector lifts Swahili from 22.1% to 69.1% on one encoder (and reproduces on the other); three controls rule out pooled-position tautology and English specificity. A front-layer trunk that pulls each language’s projection toward the parallel-content centroid corroborates the diagnosis at training time, recovering +9.6 / +17.1 pp on LRL XM3600 retrieval (1,000-image subset), with consistent gains across three further benchmarks while preserving HRL performance.</abstract>
    <identifier type="citekey">han-etal-2026-multilingual</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.840/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>19474</start>
            <end>19505</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages?
%A Han, Donghoon
%A Moon, SungHyun
%A Zhakatayev, Aidyn
%A Cha, Junghun
%A Lee, SeungJae
%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 han-etal-2026-multilingual
%X Recent multilingual vision–language encoders cover hundreds of languages in a single model, yet on two state-of-the-art instances retrieval on low-resource languages (LRL; e.g. Swahili) trails high-resource ones (HRL; e.g. English) by 30⁺ pp. We ask where in the trained encoder this gap is located. Prior modality-gap and cross-lingual subspace work suggests a linear language direction at the output crowds out alignment-relevant geometry. We falsify this: LEACE drives the linear language classifier from >99% to near chance and iterated INLP to 37–50% while LRL retrieval moves within \pm 1.5 pp and all tier means within 2.2 pp, tracking random controls. The linear bias is a symptom, not the cause. Instead, the alignment-causal factor lies along the encoder’s forward path: the EOS (end-of-sequence) hidden state’s per-language trajectory diverges with depth. Substituting the EOS with its parallel English value three blocks before the projector lifts Swahili from 22.1% to 69.1% on one encoder (and reproduces on the other); three controls rule out pooled-position tautology and English specificity. A front-layer trunk that pulls each language’s projection toward the parallel-content centroid corroborates the diagnosis at training time, recovering +9.6 / +17.1 pp on LRL XM3600 retrieval (1,000-image subset), with consistent gains across three further benchmarks while preserving HRL performance.
%U https://aclanthology.org/2026.emnlp-main.840/
%P 19474-19505

Markdown (Informal)

[Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages?](https://aclanthology.org/2026.emnlp-main.840/) (Han et al., EMNLP 2026)

• Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages? (Han et al., EMNLP 2026)

ACL

• Donghoon Han, SungHyun Moon, Aidyn Zhakatayev, Junghun Cha, and SeungJae Lee. 2026. Where Do Multilingual Vision-Language Encoders Fail on Low-Resource Languages?. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 19474–19505, Budapest, Hungary. Association for Computational Linguistics.