@inproceedings{broomfield-etal-2026-thousand,
title = "A Thousand Words or An Image: Studying the Influence of Persona Modality in Multimodal {LLM}s",
author = "Broomfield, Julius and
Sharma, Kartik and
Kumar, Srijan",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1564/",
doi = "10.18653/2026.findings-emnlp.1564",
pages = "33309--33328",
ISBN = "979-8-89176-462-0",
abstract = "Language models have demonstrated remarkable advancements in embodying diverse personas, enhancing their effectiveness as conversational agents. Simultaneously, there have been significant strides in these models in processing and integrating visual information. However, despite images being a natural carrier of persona attributes (e.g. age, occupation, or location), it remains largely unexplored whether multimodal language models (MLLMs) can faithfully ground visual persona information. In this work, we introduce persona embodiment as a novel diagnostic for visual grounding in MLLMs. Unlike simple visual recognition, this task requires models to extract multiple persona attributes from an image, integrate them into a coherent representation, and apply them to downstream language tasks. To this end, we introduce a modality-parallel dataset of personas, containing 2,400 prompts each represented in four modalities. Across our experiments, we find that models frequently underuse visual information relative to text, and that this neglect persists even when richer visual information is clearly available, highlighting underlying limitations and paving the way for future research to bridge this gap."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="broomfield-etal-2026-thousand">
<titleInfo>
<title>A Thousand Words or An Image: Studying the Influence of Persona Modality in Multimodal LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Julius</namePart>
<namePart type="family">Broomfield</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kartik</namePart>
<namePart type="family">Sharma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Srijan</namePart>
<namePart type="family">Kumar</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>Language models have demonstrated remarkable advancements in embodying diverse personas, enhancing their effectiveness as conversational agents. Simultaneously, there have been significant strides in these models in processing and integrating visual information. However, despite images being a natural carrier of persona attributes (e.g. age, occupation, or location), it remains largely unexplored whether multimodal language models (MLLMs) can faithfully ground visual persona information. In this work, we introduce persona embodiment as a novel diagnostic for visual grounding in MLLMs. Unlike simple visual recognition, this task requires models to extract multiple persona attributes from an image, integrate them into a coherent representation, and apply them to downstream language tasks. To this end, we introduce a modality-parallel dataset of personas, containing 2,400 prompts each represented in four modalities. Across our experiments, we find that models frequently underuse visual information relative to text, and that this neglect persists even when richer visual information is clearly available, highlighting underlying limitations and paving the way for future research to bridge this gap.</abstract>
<identifier type="citekey">broomfield-etal-2026-thousand</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1564</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1564/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>33309</start>
<end>33328</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T A Thousand Words or An Image: Studying the Influence of Persona Modality in Multimodal LLMs
%A Broomfield, Julius
%A Sharma, Kartik
%A Kumar, Srijan
%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 broomfield-etal-2026-thousand
%X Language models have demonstrated remarkable advancements in embodying diverse personas, enhancing their effectiveness as conversational agents. Simultaneously, there have been significant strides in these models in processing and integrating visual information. However, despite images being a natural carrier of persona attributes (e.g. age, occupation, or location), it remains largely unexplored whether multimodal language models (MLLMs) can faithfully ground visual persona information. In this work, we introduce persona embodiment as a novel diagnostic for visual grounding in MLLMs. Unlike simple visual recognition, this task requires models to extract multiple persona attributes from an image, integrate them into a coherent representation, and apply them to downstream language tasks. To this end, we introduce a modality-parallel dataset of personas, containing 2,400 prompts each represented in four modalities. Across our experiments, we find that models frequently underuse visual information relative to text, and that this neglect persists even when richer visual information is clearly available, highlighting underlying limitations and paving the way for future research to bridge this gap.
%R 10.18653/2026.findings-emnlp.1564
%U https://aclanthology.org/2026.findings-emnlp.1564/
%U https://doi.org/10.18653/2026.findings-emnlp.1564
%P 33309-33328
Markdown (Informal)
[A Thousand Words or An Image: Studying the Influence of Persona Modality in Multimodal LLMs](https://aclanthology.org/2026.findings-emnlp.1564/) (Broomfield et al., Findings 2026)
ACL