<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="lin-etal-2026-scroll">
    <titleInfo>
        <title>Scroll-Net: Geometry-Aware Constrained Decoding for Multi-Page Document Hierarchy Parsing</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Wanting</namePart>
        <namePart type="family">Lin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hiuyi</namePart>
        <namePart type="family">Cheng</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Wei</namePart>
        <namePart type="family">Pan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Lianwen</namePart>
        <namePart type="family">Jin</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>Hierarchical Document Structure Analysis (HDSA) aims to recover the logical tree and reading order of a document, providing a structural foundation for downstream tasks such as document question answering, information extraction, and long-form document understanding. Extending HDSA to multi-page documents remains challenging, as pagination introduces discontinuities between page-local coordinate systems, making it difficult to link structurally related elements across pages. We present Scroll-Net, a geometry-aware constrained decoding framework for multi-page document hierarchy parsing. Scroll-Net maps page-local coordinates into a unified scroll-space coordinate frame, enabling direct comparison of cross-page elements. It employs a dual-stream encoder to disentangle geometric and semantic cues, injecting geometric inductive bias before multimodal fusion. Under the pre-order convention used by the evaluated benchmarks, Scroll-Net constrains autoregressive parent decoding with a hard reading-order causal mask. This backward-pointer constraint makes decoded parent graphs acyclic under the imposed order while retaining O(K) per-step decoding, reducing inference latency by 36.5% and peak memory by 46.5%. Across four benchmarks, Scroll-Net remains competitive on simple layouts while achieving state-of-the-art results on structurally complex documents, improving TEDS by 4.8 points on DocHieNet.</abstract>
    <identifier type="citekey">lin-etal-2026-scroll</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.1483/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>33908</start>
            <end>33922</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Scroll-Net: Geometry-Aware Constrained Decoding for Multi-Page Document Hierarchy Parsing
%A Lin, Wanting
%A Cheng, Hiuyi
%A Pan, Wei
%A Jin, Lianwen
%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 lin-etal-2026-scroll
%X Hierarchical Document Structure Analysis (HDSA) aims to recover the logical tree and reading order of a document, providing a structural foundation for downstream tasks such as document question answering, information extraction, and long-form document understanding. Extending HDSA to multi-page documents remains challenging, as pagination introduces discontinuities between page-local coordinate systems, making it difficult to link structurally related elements across pages. We present Scroll-Net, a geometry-aware constrained decoding framework for multi-page document hierarchy parsing. Scroll-Net maps page-local coordinates into a unified scroll-space coordinate frame, enabling direct comparison of cross-page elements. It employs a dual-stream encoder to disentangle geometric and semantic cues, injecting geometric inductive bias before multimodal fusion. Under the pre-order convention used by the evaluated benchmarks, Scroll-Net constrains autoregressive parent decoding with a hard reading-order causal mask. This backward-pointer constraint makes decoded parent graphs acyclic under the imposed order while retaining O(K) per-step decoding, reducing inference latency by 36.5% and peak memory by 46.5%. Across four benchmarks, Scroll-Net remains competitive on simple layouts while achieving state-of-the-art results on structurally complex documents, improving TEDS by 4.8 points on DocHieNet.
%U https://aclanthology.org/2026.emnlp-main.1483/
%P 33908-33922

Markdown (Informal)

[Scroll-Net: Geometry-Aware Constrained Decoding for Multi-Page Document Hierarchy Parsing](https://aclanthology.org/2026.emnlp-main.1483/) (Lin et al., EMNLP 2026)

• Scroll-Net: Geometry-Aware Constrained Decoding for Multi-Page Document Hierarchy Parsing (Lin et al., EMNLP 2026)

ACL

• Wanting Lin, Hiuyi Cheng, Wei Pan, and Lianwen Jin. 2026. Scroll-Net: Geometry-Aware Constrained Decoding for Multi-Page Document Hierarchy Parsing. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 33908–33922, Budapest, Hungary. Association for Computational Linguistics.