@inproceedings{xu-etal-2026-managing,
title = "Managing Evidence at Document Scale: A Survey of Multi-Page Visually Rich Document Understanding",
author = "Xu, Lewei and
Ding, Yihao and
Xu, Zihan and
Luo, Siwen and
Liu, Daochang and
Liu, Wei",
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.872/",
pages = "20159--20187",
ISBN = "979-8-89176-459-0",
abstract = "Documents spanning tens to hundreds of pages are pervasive in real-world workflows, from scientific papers to financial reports and technical manuals. In this setting, answer-relevant evidence is sparse and distributed across pages, and the document itself often exceeds a backbone MLLM{'}s context window, making MP-VRDU a distinct problem from single-page VRDU. We survey MP-VRDU through the lens of evidence management, the organising principle that distinguishes multi-page from the single-page setting: the problem of deciding how document-scale evidence is selected and reasoned over when it cannot all be held at once. We introduce a taxonomy spanning four architectural families, multimodal representation, inference-time adaptation, and training strategies for cross-page evidence. We further consolidate MP-VRDU datasets and outline open challenges in efficiency, faithfulness, evidence pipelines, supervision integrity, and reproducible evaluation that define the frontier."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="xu-etal-2026-managing">
<titleInfo>
<title>Managing Evidence at Document Scale: A Survey of Multi-Page Visually Rich Document Understanding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Lewei</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yihao</namePart>
<namePart type="family">Ding</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zihan</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Siwen</namePart>
<namePart type="family">Luo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daochang</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wei</namePart>
<namePart type="family">Liu</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>Documents spanning tens to hundreds of pages are pervasive in real-world workflows, from scientific papers to financial reports and technical manuals. In this setting, answer-relevant evidence is sparse and distributed across pages, and the document itself often exceeds a backbone MLLM’s context window, making MP-VRDU a distinct problem from single-page VRDU. We survey MP-VRDU through the lens of evidence management, the organising principle that distinguishes multi-page from the single-page setting: the problem of deciding how document-scale evidence is selected and reasoned over when it cannot all be held at once. We introduce a taxonomy spanning four architectural families, multimodal representation, inference-time adaptation, and training strategies for cross-page evidence. We further consolidate MP-VRDU datasets and outline open challenges in efficiency, faithfulness, evidence pipelines, supervision integrity, and reproducible evaluation that define the frontier.</abstract>
<identifier type="citekey">xu-etal-2026-managing</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.872/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>20159</start>
<end>20187</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Managing Evidence at Document Scale: A Survey of Multi-Page Visually Rich Document Understanding
%A Xu, Lewei
%A Ding, Yihao
%A Xu, Zihan
%A Luo, Siwen
%A Liu, Daochang
%A Liu, Wei
%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 xu-etal-2026-managing
%X Documents spanning tens to hundreds of pages are pervasive in real-world workflows, from scientific papers to financial reports and technical manuals. In this setting, answer-relevant evidence is sparse and distributed across pages, and the document itself often exceeds a backbone MLLM’s context window, making MP-VRDU a distinct problem from single-page VRDU. We survey MP-VRDU through the lens of evidence management, the organising principle that distinguishes multi-page from the single-page setting: the problem of deciding how document-scale evidence is selected and reasoned over when it cannot all be held at once. We introduce a taxonomy spanning four architectural families, multimodal representation, inference-time adaptation, and training strategies for cross-page evidence. We further consolidate MP-VRDU datasets and outline open challenges in efficiency, faithfulness, evidence pipelines, supervision integrity, and reproducible evaluation that define the frontier.
%U https://aclanthology.org/2026.emnlp-main.872/
%P 20159-20187
Markdown (Informal)
[Managing Evidence at Document Scale: A Survey of Multi-Page Visually Rich Document Understanding](https://aclanthology.org/2026.emnlp-main.872/) (Xu et al., EMNLP 2026)
ACL