@inproceedings{zhong-etal-2026-describe,
title = "{D}e{S}cribe: Discourse-Driven Extraction and Structuring of Claim Verification for Better Faithfulness Evaluation",
author = "Zhong, Yang and
Liu, Yang Janet and
Litman, Diane",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.141/",
pages = "2737--2771",
ISBN = "979-8-89176-444-6",
abstract = "Evaluating the faithfulness of multi-sentence summaries generated by language models (LMs) remains challenging due to their length and the subtlety of factual errors. While claim decomposition methods have proven useful for assessing short summaries (1-3 sentences), their effectiveness for longer, coherent summaries across diverse domains is not well understood. To better understand this issue, we first conduct an empirical study analyzing the quality of claims produced by existing decomposition approaches. Motivated by our findings, we propose a new discourse-informed method, Describe, which first extracts claims from the full summary using discourse structure, and then evaluates them through a structure-guided process that progresses from intra- to inter-sentential claim validation, enabling more fine-grained faithfulness evaluation. Experimental results across benchmarks reveal the benefits of discourse-guided decomposition and evaluation. Further analysis of model-generated error rationales and ablation studies shows our method effectively identifies nuanced errors within sentences, particularly those arising from complex discourse structures."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhong-etal-2026-describe">
<titleInfo>
<title>DeScribe: Discourse-Driven Extraction and Structuring of Claim Verification for Better Faithfulness Evaluation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yang</namePart>
<namePart type="given">Janet</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Diane</namePart>
<namePart type="family">Litman</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Evaluating the faithfulness of multi-sentence summaries generated by language models (LMs) remains challenging due to their length and the subtlety of factual errors. While claim decomposition methods have proven useful for assessing short summaries (1-3 sentences), their effectiveness for longer, coherent summaries across diverse domains is not well understood. To better understand this issue, we first conduct an empirical study analyzing the quality of claims produced by existing decomposition approaches. Motivated by our findings, we propose a new discourse-informed method, Describe, which first extracts claims from the full summary using discourse structure, and then evaluates them through a structure-guided process that progresses from intra- to inter-sentential claim validation, enabling more fine-grained faithfulness evaluation. Experimental results across benchmarks reveal the benefits of discourse-guided decomposition and evaluation. Further analysis of model-generated error rationales and ablation studies shows our method effectively identifies nuanced errors within sentences, particularly those arising from complex discourse structures.</abstract>
<identifier type="citekey">zhong-etal-2026-describe</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.141/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>2737</start>
<end>2771</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T DeScribe: Discourse-Driven Extraction and Structuring of Claim Verification for Better Faithfulness Evaluation
%A Zhong, Yang
%A Liu, Yang Janet
%A Litman, Diane
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F zhong-etal-2026-describe
%X Evaluating the faithfulness of multi-sentence summaries generated by language models (LMs) remains challenging due to their length and the subtlety of factual errors. While claim decomposition methods have proven useful for assessing short summaries (1-3 sentences), their effectiveness for longer, coherent summaries across diverse domains is not well understood. To better understand this issue, we first conduct an empirical study analyzing the quality of claims produced by existing decomposition approaches. Motivated by our findings, we propose a new discourse-informed method, Describe, which first extracts claims from the full summary using discourse structure, and then evaluates them through a structure-guided process that progresses from intra- to inter-sentential claim validation, enabling more fine-grained faithfulness evaluation. Experimental results across benchmarks reveal the benefits of discourse-guided decomposition and evaluation. Further analysis of model-generated error rationales and ablation studies shows our method effectively identifies nuanced errors within sentences, particularly those arising from complex discourse structures.
%U https://aclanthology.org/2026.findings-aacl.141/
%P 2737-2771
Markdown (Informal)
[DeScribe: Discourse-Driven Extraction and Structuring of Claim Verification for Better Faithfulness Evaluation](https://aclanthology.org/2026.findings-aacl.141/) (Zhong et al., Findings 2026)
ACL