<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yeo-etal-2026-diffusion">
    <titleInfo>
        <title>Diffusion Large Language Models for Visual Speech Recognition</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Jeong</namePart>
        <namePart type="given">Hun</namePart>
        <namePart type="family">Yeo</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Chae</namePart>
        <namePart type="given">Won</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hyeongseop</namePart>
        <namePart type="family">Rha</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yong</namePart>
        <namePart type="given">Man</namePart>
        <namePart type="family">Ro</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>Existing Visual Speech Recognition (VSR) systems commonly rely on left-to-right autoregressive decoding, which can force premature decisions on visually ambiguous tokens before sufficient context is available. We propose DLLM-VSR, to the best of our knowledge, the first Diffusion Large Language Model (DLLM)-based VSR framework, formulating transcription as iterative masked denoising with flexible-order decoding. With confidence-based unmasking, DLLM-VSR commits high-confidence positions early and uses the committed tokens as bidirectional context to refine ambiguous ones. To adapt DLLMs to VSR, we introduce a two-stage masked-denoising training strategy that separates visual-to-text content alignment from length modeling. We further observe a performance gap compared with an upper-bound setting where the ground-truth transcript length is provided at inference, allowing the model to focus on transcript content decoding. To reduce this gap, we develop length-guided candidate decoding, which uses video duration to construct plausible transcript-length hypotheses and reranks the decoded candidates using length plausibility and decoding confidence. The proposed method achieves a 19.4% word error rate on LRS3, establishing state-of-the-art performance among methods using only LRS3 as labeled training data.</abstract>
    <identifier type="citekey">yeo-etal-2026-diffusion</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1826/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>41596</start>
            <end>41610</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Diffusion Large Language Models for Visual Speech Recognition
%A Yeo, Jeong Hun
%A Kim, Chae Won
%A Rha, Hyeongseop
%A Ro, Yong Man
%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 yeo-etal-2026-diffusion
%X Existing Visual Speech Recognition (VSR) systems commonly rely on left-to-right autoregressive decoding, which can force premature decisions on visually ambiguous tokens before sufficient context is available. We propose DLLM-VSR, to the best of our knowledge, the first Diffusion Large Language Model (DLLM)-based VSR framework, formulating transcription as iterative masked denoising with flexible-order decoding. With confidence-based unmasking, DLLM-VSR commits high-confidence positions early and uses the committed tokens as bidirectional context to refine ambiguous ones. To adapt DLLMs to VSR, we introduce a two-stage masked-denoising training strategy that separates visual-to-text content alignment from length modeling. We further observe a performance gap compared with an upper-bound setting where the ground-truth transcript length is provided at inference, allowing the model to focus on transcript content decoding. To reduce this gap, we develop length-guided candidate decoding, which uses video duration to construct plausible transcript-length hypotheses and reranks the decoded candidates using length plausibility and decoding confidence. The proposed method achieves a 19.4% word error rate on LRS3, establishing state-of-the-art performance among methods using only LRS3 as labeled training data.
%U https://aclanthology.org/2026.emnlp-main.1826/
%P 41596-41610

Markdown (Informal)

[Diffusion Large Language Models for Visual Speech Recognition](https://aclanthology.org/2026.emnlp-main.1826/) (Yeo et al., EMNLP 2026)

• Diffusion Large Language Models for Visual Speech Recognition (Yeo et al., EMNLP 2026)

ACL

• Jeong Hun Yeo, Chae Won Kim, Hyeongseop Rha, and Yong Man Ro. 2026. Diffusion Large Language Models for Visual Speech Recognition. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 41596–41610, Budapest, Hungary. Association for Computational Linguistics.