<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wu-etal-2026-vision">
    <titleInfo>
        <title>Vision-Braille: A Curriculum Learning Toolkit and Braille-Chinese Corpus for Braille Translation</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Alan</namePart>
        <namePart type="family">Wu</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ye</namePart>
        <namePart type="family">Yuan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Zhiping</namePart>
        <namePart type="family">Xiao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Ming</namePart>
        <namePart type="family">Zhang</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>We present Vision-Braille, the first publicly available end-to-end system for translating Chinese Braille extracted from images into written Chinese. This system addresses the unique challenges of limited annotated resources and tone omission. The task is out of reach for general-purpose LLMs: given the complete Chinese Braille encoding specification and a worked example in context, Qwen3-235B-A22B reaches only 9.46 BLEU, and the only prior Braille-to-Chinese model with released checkpoints scores 5.36 BLEU on passage-level input. Braille is effectively absent from pretraining corpora, decoding it demands exact multi-step symbolic reasoning, and writers omit up to 90% of tone markers, leaving homophones that must be recovered from context. Vision-Braille integrates a Braille OCR pipeline with an open-weights LLM fine-tuned on a synthetic Braille-Chinese corpus that reproduces authentic tone-omission habits, using a four-stage curriculum that moves from short, fully toned input to long passages with tone markers almost entirely stripped. On passage-level translation with 10% tone retention, Curriculum-Braille achieves 83.28 BLEU at a 9.94% character error rate, and 76.66 BLEU on passages hand-transcribed by an expert from the Chinese Braille Library, with no reliance on any proprietary API. For the 150,000 school-age blind children in China, Vision-Braille offers an inclusive NLP solution that supports participation in mainstream education by enabling teachers to grade Braille homework without extensive training.</abstract>
    <identifier type="citekey">wu-etal-2026-vision</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2624/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>59658</start>
            <end>59672</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Vision-Braille: A Curriculum Learning Toolkit and Braille-Chinese Corpus for Braille Translation
%A Wu, Alan
%A Yuan, Ye
%A Xiao, Zhiping
%A Zhang, Ming
%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 wu-etal-2026-vision
%X We present Vision-Braille, the first publicly available end-to-end system for translating Chinese Braille extracted from images into written Chinese. This system addresses the unique challenges of limited annotated resources and tone omission. The task is out of reach for general-purpose LLMs: given the complete Chinese Braille encoding specification and a worked example in context, Qwen3-235B-A22B reaches only 9.46 BLEU, and the only prior Braille-to-Chinese model with released checkpoints scores 5.36 BLEU on passage-level input. Braille is effectively absent from pretraining corpora, decoding it demands exact multi-step symbolic reasoning, and writers omit up to 90% of tone markers, leaving homophones that must be recovered from context. Vision-Braille integrates a Braille OCR pipeline with an open-weights LLM fine-tuned on a synthetic Braille-Chinese corpus that reproduces authentic tone-omission habits, using a four-stage curriculum that moves from short, fully toned input to long passages with tone markers almost entirely stripped. On passage-level translation with 10% tone retention, Curriculum-Braille achieves 83.28 BLEU at a 9.94% character error rate, and 76.66 BLEU on passages hand-transcribed by an expert from the Chinese Braille Library, with no reliance on any proprietary API. For the 150,000 school-age blind children in China, Vision-Braille offers an inclusive NLP solution that supports participation in mainstream education by enabling teachers to grade Braille homework without extensive training.
%U https://aclanthology.org/2026.emnlp-main.2624/
%P 59658-59672

Markdown (Informal)

[Vision-Braille: A Curriculum Learning Toolkit and Braille-Chinese Corpus for Braille Translation](https://aclanthology.org/2026.emnlp-main.2624/) (Wu et al., EMNLP 2026)

• Vision-Braille: A Curriculum Learning Toolkit and Braille-Chinese Corpus for Braille Translation (Wu et al., EMNLP 2026)

ACL

• Alan Wu, Ye Yuan, Zhiping Xiao, and Ming Zhang. 2026. Vision-Braille: A Curriculum Learning Toolkit and Braille-Chinese Corpus for Braille Translation. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 59658–59672, Budapest, Hungary. Association for Computational Linguistics.