<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yu-etal-2026-smart">
    <titleInfo>
        <title>SMART: Sink-based Modality-Aware Redistribution of Transformer Attention</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Dian</namePart>
        <namePart type="family">Yu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yao</namePart>
        <namePart type="family">Yao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hai</namePart>
        <namePart type="family">Zhao</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>Multimodal Large Language Models (MLLMs) often exhibit modality bias, where predictions over-rely on one modality while underutilizing others. However, the mechanism underlying modality bias remains poorly understood. In this work, we first introduce Banana-Counting, a controlled diagnostic dataset for measuring modality preference under cross-modal conflicts. Evaluations across diverse MLLMs reveal pervasive modality imbalance. Then, through attention analysis, we show that modality bias is strongly associated with imbalanced cross-modal attention allocation and modality-specific attention sinks that absorb redundant attention mass. Motivated by these findings, we propose SMART, an inference-time method that mitigates modality bias by redistributing redundant sink attention toward under-attended modalities. SMART consistently improves multimodal balance across model families. On Banana-Counting, it reduces the modality gap from 27.73 to 0.66 on LLaVA-v1.5-7B while improving balanced multimodal accuracy. Moreover, SMART generalizes effectively to downstream multimodal reasoning tasks without retraining or parameter updates.</abstract>
    <identifier type="citekey">yu-etal-2026-smart</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2358/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>53547</start>
            <end>53560</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T SMART: Sink-based Modality-Aware Redistribution of Transformer Attention
%A Yu, Dian
%A Yao, Yao
%A Zhao, Hai
%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 yu-etal-2026-smart
%X Multimodal Large Language Models (MLLMs) often exhibit modality bias, where predictions over-rely on one modality while underutilizing others. However, the mechanism underlying modality bias remains poorly understood. In this work, we first introduce Banana-Counting, a controlled diagnostic dataset for measuring modality preference under cross-modal conflicts. Evaluations across diverse MLLMs reveal pervasive modality imbalance. Then, through attention analysis, we show that modality bias is strongly associated with imbalanced cross-modal attention allocation and modality-specific attention sinks that absorb redundant attention mass. Motivated by these findings, we propose SMART, an inference-time method that mitigates modality bias by redistributing redundant sink attention toward under-attended modalities. SMART consistently improves multimodal balance across model families. On Banana-Counting, it reduces the modality gap from 27.73 to 0.66 on LLaVA-v1.5-7B while improving balanced multimodal accuracy. Moreover, SMART generalizes effectively to downstream multimodal reasoning tasks without retraining or parameter updates.
%U https://aclanthology.org/2026.emnlp-main.2358/
%P 53547-53560

Markdown (Informal)

[SMART: Sink-based Modality-Aware Redistribution of Transformer Attention](https://aclanthology.org/2026.emnlp-main.2358/) (Yu et al., EMNLP 2026)

• SMART: Sink-based Modality-Aware Redistribution of Transformer Attention (Yu et al., EMNLP 2026)

ACL

• Dian Yu, Yao Yao, and Hai Zhao. 2026. SMART: Sink-based Modality-Aware Redistribution of Transformer Attention. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 53547–53560, Budapest, Hungary. Association for Computational Linguistics.