<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="seong-etal-2026-vlms">
    <titleInfo>
        <title>When VLMs ‘Fix’ Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Jin</namePart>
        <namePart type="family">Seong</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wencke</namePart>
        <namePart type="family">Liermann</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Minho</namePart>
        <namePart type="family">Kim</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Jong-hun</namePart>
        <namePart type="family">Shin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Soojong</namePart>
        <namePart type="family">Lim</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>Accurate transcription of handwritten mathematics is crucial for educational AI systems, yet current benchmarks fail to evaluate this capability properly. Most prior studies focus on single-line expressions and rely on lexical metrics such as BLEU, which fail to assess the semantic reasoning across multi-line student solutions. In this paper, we present the first systematic study of multi-line handwritten math Optical Character Recognition (OCR), revealing a critical failure mode of Vision-Language Models (VLMs): over-correction. Instead of faithfully transcribing a student’s work, these models often “fix” errors, thereby hiding the very mistakes an educational assessment aims to detect. To address this, we propose PINK (Penalized INK-based score), a semantic evaluation metric that leverages a Large Language Model (LLM) for rubric-based grading and explicitly penalizes over-correction. Our comprehensive evaluation of 15 state-of-the-art VLMs on the FERMAT dataset reveals substantial ranking reversals compared to BLEU: models like GPT-4o are heavily penalized for aggressive over-correction, whereas Gemini 2.5 Flash emerges as the most faithful transcriber. Furthermore, human expert studies show that PINK aligns significantly better with human judgment (55.0% preference over BLEU’s 39.5%), providing a more reliable evaluation framework for handwritten math OCR in educational settings.</abstract>
    <identifier type="citekey">seong-etal-2026-vlms</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.476/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>10932</start>
            <end>10953</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T When VLMs ‘Fix’ Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR
%A Seong, Jin
%A Liermann, Wencke
%A Kim, Minho
%A Shin, Jong-hun
%A Lim, Soojong
%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 seong-etal-2026-vlms
%X Accurate transcription of handwritten mathematics is crucial for educational AI systems, yet current benchmarks fail to evaluate this capability properly. Most prior studies focus on single-line expressions and rely on lexical metrics such as BLEU, which fail to assess the semantic reasoning across multi-line student solutions. In this paper, we present the first systematic study of multi-line handwritten math Optical Character Recognition (OCR), revealing a critical failure mode of Vision-Language Models (VLMs): over-correction. Instead of faithfully transcribing a student’s work, these models often “fix” errors, thereby hiding the very mistakes an educational assessment aims to detect. To address this, we propose PINK (Penalized INK-based score), a semantic evaluation metric that leverages a Large Language Model (LLM) for rubric-based grading and explicitly penalizes over-correction. Our comprehensive evaluation of 15 state-of-the-art VLMs on the FERMAT dataset reveals substantial ranking reversals compared to BLEU: models like GPT-4o are heavily penalized for aggressive over-correction, whereas Gemini 2.5 Flash emerges as the most faithful transcriber. Furthermore, human expert studies show that PINK aligns significantly better with human judgment (55.0% preference over BLEU’s 39.5%), providing a more reliable evaluation framework for handwritten math OCR in educational settings.
%U https://aclanthology.org/2026.emnlp-main.476/
%P 10932-10953

Markdown (Informal)

[When VLMs ’Fix’ Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR](https://aclanthology.org/2026.emnlp-main.476/) (Seong et al., EMNLP 2026)

• When VLMs ’Fix’ Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR (Seong et al., EMNLP 2026)

ACL

• Jin Seong, Wencke Liermann, Minho Kim, Jong-hun Shin, and Soojong Lim. 2026. When VLMs ’Fix’ Students: Identifying and Penalizing Over-Correction in the Evaluation of Multi-line Handwritten Math OCR. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 10932–10953, Budapest, Hungary. Association for Computational Linguistics.