@inproceedings{zhu-etal-2026-holodp,
title = "{H}olo{DP}: A Multilingual Benchmark Suite and Script-Aware Framework for Document Parsing",
author = "Zhu, Mengxiao and
Ying, Wencan and
Cao, Yingjie and
Liu, Jie and
Sha, Jiu",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.1525/",
doi = "10.18653/2026.findings-emnlp.1525",
pages = "32453--32470",
ISBN = "979-8-89176-462-0",
abstract = "Millions of multilingual documents remain locked in image formats, but current document parsers are still largely evaluated and optimized for a narrow range of high-resource languages and regular writing layouts. As a result, their reliability drops sharply when facing underrepresented scripts, vertical or right-to-left writing, and mixed-script document structures. We present HoloDP, a multilingual benchmark suite and script-aware framework for document parsing. HoloDP first introduces a standardized evaluation suite covering 13 languages across seven script families, with tasks spanning page-level layout analysis, region-level parsing, and mixed-script structural composition. Built on this resource, HoloDP further develops a script-aware parser that combines a Swin Transformer encoder, a four-branch direction-adaptive vision-to-text module, and a multilingual decoder. The adaptive module explicitly captures horizontal, vertical, diagonal, and global spatial priors, enabling the model to better handle heterogeneous writing systems and layouts. Extensive experiments show that HoloDP delivers consistent gains on underrepresented and mixed-script benchmarks while remaining competitive on high-resource Chinese and English documents. Ablation studies confirm that the benchmark resources, direction-aware adaptation, and script-conditioned modeling provide complementary benefits. We will release the code, models, and benchmarks for further research."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhu-etal-2026-holodp">
<titleInfo>
<title>HoloDP: A Multilingual Benchmark Suite and Script-Aware Framework for Document Parsing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Mengxiao</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wencan</namePart>
<namePart type="family">Ying</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yingjie</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jie</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiu</namePart>
<namePart type="family">Sha</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Millions of multilingual documents remain locked in image formats, but current document parsers are still largely evaluated and optimized for a narrow range of high-resource languages and regular writing layouts. As a result, their reliability drops sharply when facing underrepresented scripts, vertical or right-to-left writing, and mixed-script document structures. We present HoloDP, a multilingual benchmark suite and script-aware framework for document parsing. HoloDP first introduces a standardized evaluation suite covering 13 languages across seven script families, with tasks spanning page-level layout analysis, region-level parsing, and mixed-script structural composition. Built on this resource, HoloDP further develops a script-aware parser that combines a Swin Transformer encoder, a four-branch direction-adaptive vision-to-text module, and a multilingual decoder. The adaptive module explicitly captures horizontal, vertical, diagonal, and global spatial priors, enabling the model to better handle heterogeneous writing systems and layouts. Extensive experiments show that HoloDP delivers consistent gains on underrepresented and mixed-script benchmarks while remaining competitive on high-resource Chinese and English documents. Ablation studies confirm that the benchmark resources, direction-aware adaptation, and script-conditioned modeling provide complementary benefits. We will release the code, models, and benchmarks for further research.</abstract>
<identifier type="citekey">zhu-etal-2026-holodp</identifier>
<identifier type="doi">10.18653/2026.findings-emnlp.1525</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1525/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>32453</start>
<end>32470</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HoloDP: A Multilingual Benchmark Suite and Script-Aware Framework for Document Parsing
%A Zhu, Mengxiao
%A Ying, Wencan
%A Cao, Yingjie
%A Liu, Jie
%A Sha, Jiu
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F zhu-etal-2026-holodp
%X Millions of multilingual documents remain locked in image formats, but current document parsers are still largely evaluated and optimized for a narrow range of high-resource languages and regular writing layouts. As a result, their reliability drops sharply when facing underrepresented scripts, vertical or right-to-left writing, and mixed-script document structures. We present HoloDP, a multilingual benchmark suite and script-aware framework for document parsing. HoloDP first introduces a standardized evaluation suite covering 13 languages across seven script families, with tasks spanning page-level layout analysis, region-level parsing, and mixed-script structural composition. Built on this resource, HoloDP further develops a script-aware parser that combines a Swin Transformer encoder, a four-branch direction-adaptive vision-to-text module, and a multilingual decoder. The adaptive module explicitly captures horizontal, vertical, diagonal, and global spatial priors, enabling the model to better handle heterogeneous writing systems and layouts. Extensive experiments show that HoloDP delivers consistent gains on underrepresented and mixed-script benchmarks while remaining competitive on high-resource Chinese and English documents. Ablation studies confirm that the benchmark resources, direction-aware adaptation, and script-conditioned modeling provide complementary benefits. We will release the code, models, and benchmarks for further research.
%R 10.18653/2026.findings-emnlp.1525
%U https://aclanthology.org/2026.findings-emnlp.1525/
%U https://doi.org/10.18653/2026.findings-emnlp.1525
%P 32453-32470
Markdown (Informal)
[HoloDP: A Multilingual Benchmark Suite and Script-Aware Framework for Document Parsing](https://aclanthology.org/2026.findings-emnlp.1525/) (Zhu et al., Findings 2026)
ACL