@inproceedings{zhao-etal-2026-hg,
title = "{HG}-Bench: A Benchmark for Multi-Page Handwritten Answer-Region Grounding in Automated Homework Assessment",
author = "Zhao, Chuangxin and
Shi, Boyan and
Wang, Yanling and
LU, Yijian and
Xiao, Canran and
Chen, Jiali and
Xia, Jun and
Wang, Yan and
Qi, Ji and
Li, Juanzi",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2417/",
pages = "54888--54904",
ISBN = "979-8-89176-459-0",
abstract = "Automated homework assessment requires not only recognizing student answers but accurately locating complete answers and intermediate reasoning steps in noisy, multi-page handwritten work. We address the missing evaluation setting of *page-aware, two-level answer-region grounding*: given homework page images, a model must localize complete answer regions and their ordered step-level subregions. We introduce **HG-Bench**, comprising 500 human-annotated K{--}12 homework samples curated from a 1,489,278-image source pool, with hierarchically linked question- and step-level boxes. Its page-aware protocol separately measures complete-answer localization ($\mathcal{F}_A$) and step-level decomposition ($\mathcal{F}_S^{\mu}$), distinguishing structured spatial grounding from visible-text parsing. Across frontier closed-source and open-weight VLMs, no zero-shot system exceeds 55.22{\%} on $\mathcal{F}_A$ or 48.22{\%} on $\mathcal{F}_S^{\mu}$, whereas a GLM-4.6V 9B reference fine-tuned on approximately 10k in-domain examples reaches 74.97/72.26. These results expose step-level handwritten grounding as a concrete capability gap and provide a reproducible benchmark, evaluation protocol, and trained reference for future work on automated homework assessment."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-etal-2026-hg">
<titleInfo>
<title>HG-Bench: A Benchmark for Multi-Page Handwritten Answer-Region Grounding in Automated Homework Assessment</title>
</titleInfo>
<name type="personal">
<namePart type="given">Chuangxin</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Boyan</namePart>
<namePart type="family">Shi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yanling</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yijian</namePart>
<namePart type="family">LU</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Canran</namePart>
<namePart type="family">Xiao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiali</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Xia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ji</namePart>
<namePart type="family">Qi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Juanzi</namePart>
<namePart type="family">Li</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>Automated homework assessment requires not only recognizing student answers but accurately locating complete answers and intermediate reasoning steps in noisy, multi-page handwritten work. We address the missing evaluation setting of *page-aware, two-level answer-region grounding*: given homework page images, a model must localize complete answer regions and their ordered step-level subregions. We introduce **HG-Bench**, comprising 500 human-annotated K–12 homework samples curated from a 1,489,278-image source pool, with hierarchically linked question- and step-level boxes. Its page-aware protocol separately measures complete-answer localization (\mathcalF_A) and step-level decomposition (\mathcalF_S^μ), distinguishing structured spatial grounding from visible-text parsing. Across frontier closed-source and open-weight VLMs, no zero-shot system exceeds 55.22% on \mathcalF_A or 48.22% on \mathcalF_S^μ, whereas a GLM-4.6V 9B reference fine-tuned on approximately 10k in-domain examples reaches 74.97/72.26. These results expose step-level handwritten grounding as a concrete capability gap and provide a reproducible benchmark, evaluation protocol, and trained reference for future work on automated homework assessment.</abstract>
<identifier type="citekey">zhao-etal-2026-hg</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2417/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>54888</start>
<end>54904</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HG-Bench: A Benchmark for Multi-Page Handwritten Answer-Region Grounding in Automated Homework Assessment
%A Zhao, Chuangxin
%A Shi, Boyan
%A Wang, Yanling
%A LU, Yijian
%A Xiao, Canran
%A Chen, Jiali
%A Xia, Jun
%A Wang, Yan
%A Qi, Ji
%A Li, Juanzi
%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 zhao-etal-2026-hg
%X Automated homework assessment requires not only recognizing student answers but accurately locating complete answers and intermediate reasoning steps in noisy, multi-page handwritten work. We address the missing evaluation setting of *page-aware, two-level answer-region grounding*: given homework page images, a model must localize complete answer regions and their ordered step-level subregions. We introduce **HG-Bench**, comprising 500 human-annotated K–12 homework samples curated from a 1,489,278-image source pool, with hierarchically linked question- and step-level boxes. Its page-aware protocol separately measures complete-answer localization (\mathcalF_A) and step-level decomposition (\mathcalF_S^μ), distinguishing structured spatial grounding from visible-text parsing. Across frontier closed-source and open-weight VLMs, no zero-shot system exceeds 55.22% on \mathcalF_A or 48.22% on \mathcalF_S^μ, whereas a GLM-4.6V 9B reference fine-tuned on approximately 10k in-domain examples reaches 74.97/72.26. These results expose step-level handwritten grounding as a concrete capability gap and provide a reproducible benchmark, evaluation protocol, and trained reference for future work on automated homework assessment.
%U https://aclanthology.org/2026.emnlp-main.2417/
%P 54888-54904
Markdown (Informal)
[HG-Bench: A Benchmark for Multi-Page Handwritten Answer-Region Grounding in Automated Homework Assessment](https://aclanthology.org/2026.emnlp-main.2417/) (Zhao et al., EMNLP 2026)
ACL
- Chuangxin Zhao, Boyan Shi, Yanling Wang, Yijian LU, Canran Xiao, Jiali Chen, Jun Xia, Yan Wang, Ji Qi, and Juanzi Li. 2026. HG-Bench: A Benchmark for Multi-Page Handwritten Answer-Region Grounding in Automated Homework Assessment. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 54888–54904, Budapest, Hungary. Association for Computational Linguistics.