<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="he-etal-2025-seeing">
    <titleInfo>
        <title>Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Zoe</namePart>
        <namePart type="given">Wanying</namePart>
        <namePart type="family">He</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Sean</namePart>
        <namePart type="family">Trott</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Meenakshi</namePart>
        <namePart type="family">Khosla</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2025-11</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Christos</namePart>
            <namePart type="family">Christodoulopoulos</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Tanmoy</namePart>
            <namePart type="family">Chakraborty</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Carolyn</namePart>
            <namePart type="family">Rose</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Violet</namePart>
            <namePart type="family">Peng</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Suzhou, China</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-332-6</identifier>
    </relatedItem>
    <abstract>Recent studies show that deep vision-only and language-only models—trained on disjoint modalities—nonetheless project their inputs into a partially aligned representational space. Yet we still lack a clear picture of _where_ in each network this convergence emerges, _what_ visual or linguistic cues support it, _whether_ it captures human preferences in many-to-many image-text scenarios, and _how_ aggregating exemplars of the same concept affects alignment. Here, we systematically investigate these questions. We find that alignment peaks in mid-to-late layers of both model types, reflecting a shift from modality-specific to conceptually shared representations. This alignment is robust to appearance-only changes but collapses when semantics are altered (e.g., object removal or word-order scrambling), highlighting that the shared code is truly semantic. Moving beyond the one-to-one image-caption paradigm, a forced-choice “Pick-a-Pic” task shows that human preferences for image-caption matches are mirrored in the embedding spaces across all vision-language model pairs. This pattern holds bidirectionally when multiple captions correspond to a single image, demonstrating that models capture fine-grained semantic distinctions akin to human judgments. Surprisingly, averaging embeddings across exemplars amplifies alignment rather than blurring detail. Together, our results demonstrate that unimodal networks converge on a shared semantic code that aligns with human judgments and strengthens with exemplar aggregation.</abstract>
    <identifier type="citekey">he-etal-2025-seeing</identifier>
    <location>
        <url>https://aclanthology.org/2025.emnlp-main.1806/</url>
    </location>
    <part>
        <date>2025-11</date>
        <extent unit="page">
            <start>35645</start>
            <end>35660</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models
%A He, Zoe Wanying
%A Trott, Sean
%A Khosla, Meenakshi
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-332-6
%F he-etal-2025-seeing
%X Recent studies show that deep vision-only and language-only models—trained on disjoint modalities—nonetheless project their inputs into a partially aligned representational space. Yet we still lack a clear picture of _where_ in each network this convergence emerges, _what_ visual or linguistic cues support it, _whether_ it captures human preferences in many-to-many image-text scenarios, and _how_ aggregating exemplars of the same concept affects alignment. Here, we systematically investigate these questions. We find that alignment peaks in mid-to-late layers of both model types, reflecting a shift from modality-specific to conceptually shared representations. This alignment is robust to appearance-only changes but collapses when semantics are altered (e.g., object removal or word-order scrambling), highlighting that the shared code is truly semantic. Moving beyond the one-to-one image-caption paradigm, a forced-choice “Pick-a-Pic” task shows that human preferences for image-caption matches are mirrored in the embedding spaces across all vision-language model pairs. This pattern holds bidirectionally when multiple captions correspond to a single image, demonstrating that models capture fine-grained semantic distinctions akin to human judgments. Surprisingly, averaging embeddings across exemplars amplifies alignment rather than blurring detail. Together, our results demonstrate that unimodal networks converge on a shared semantic code that aligns with human judgments and strengthens with exemplar aggregation.
%U https://aclanthology.org/2025.emnlp-main.1806/
%P 35645-35660

Markdown (Informal)

[Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models](https://aclanthology.org/2025.emnlp-main.1806/) (He et al., EMNLP 2025)

• Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models (He et al., EMNLP 2025)

ACL

• Zoe Wanying He, Sean Trott, and Meenakshi Khosla. 2025. Seeing Through Words, Speaking Through Pixels: Deep Representational Alignment Between Vision and Language Models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 35645–35660, Suzhou, China. Association for Computational Linguistics.