@inproceedings{mu-etal-2026-acre,
title = "{ACRE}: An Acoustic-Cue and Contextual-Knowledge Reasoning Enhancement Framework for Speech Relation Extraction",
author = "Mu, Wenxuan and
Ning, Jinzhong and
Zhao, Di and
Zhang, Yijia",
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.1244/",
pages = "26343--26352",
ISBN = "979-8-89176-462-0",
abstract = "Speech relation extraction (SpeechRE) aims to extract structured relational triples directly from speech. However, existing studies still face three major challenges: **the lack of Chinese SpeechRE benchmarks**, **insufficient modeling of speech-specific cues**, and **evaluation bias introduced by character-level exact matching** in Chinese scenarios. To address these challenges, we first construct **CSRTE**, a real-world Chinese SpeechRE benchmark built from human speech, providing unified data and evaluation settings. We then propose **ACRE**, a reasoning enhancement framework for SpeechRE, which integrates acoustic-cue and contextual-knowledge into a structured acoustic-semantic reasoning chain, enabling the model to go beyond literal semantic content **(what-to-say)** and capture speech-specific features **(how-to-say)** for relation reasoning. We further combine supervised fine-tuning with reinforcement learning and design composite rewards to improve the quality of the model{'}s reasoning process. In addition, we propose pronunciation-aware evaluation **(PAE)**, which considers pronunciation and tone consistency during entity matching to reduce the evaluation bias. Experiments on both Chinese and English SpeechRE benchmarks demonstrate the effectiveness of ACRE and the usefulness of PAE. Our source code and datasets are available at https://github.com/MWXGOD/CSRTE."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="mu-etal-2026-acre">
<titleInfo>
<title>ACRE: An Acoustic-Cue and Contextual-Knowledge Reasoning Enhancement Framework for Speech Relation Extraction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wenxuan</namePart>
<namePart type="family">Mu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jinzhong</namePart>
<namePart type="family">Ning</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Di</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yijia</namePart>
<namePart type="family">Zhang</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>Speech relation extraction (SpeechRE) aims to extract structured relational triples directly from speech. However, existing studies still face three major challenges: **the lack of Chinese SpeechRE benchmarks**, **insufficient modeling of speech-specific cues**, and **evaluation bias introduced by character-level exact matching** in Chinese scenarios. To address these challenges, we first construct **CSRTE**, a real-world Chinese SpeechRE benchmark built from human speech, providing unified data and evaluation settings. We then propose **ACRE**, a reasoning enhancement framework for SpeechRE, which integrates acoustic-cue and contextual-knowledge into a structured acoustic-semantic reasoning chain, enabling the model to go beyond literal semantic content **(what-to-say)** and capture speech-specific features **(how-to-say)** for relation reasoning. We further combine supervised fine-tuning with reinforcement learning and design composite rewards to improve the quality of the model’s reasoning process. In addition, we propose pronunciation-aware evaluation **(PAE)**, which considers pronunciation and tone consistency during entity matching to reduce the evaluation bias. Experiments on both Chinese and English SpeechRE benchmarks demonstrate the effectiveness of ACRE and the usefulness of PAE. Our source code and datasets are available at https://github.com/MWXGOD/CSRTE.</abstract>
<identifier type="citekey">mu-etal-2026-acre</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.1244/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>26343</start>
<end>26352</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T ACRE: An Acoustic-Cue and Contextual-Knowledge Reasoning Enhancement Framework for Speech Relation Extraction
%A Mu, Wenxuan
%A Ning, Jinzhong
%A Zhao, Di
%A Zhang, Yijia
%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 mu-etal-2026-acre
%X Speech relation extraction (SpeechRE) aims to extract structured relational triples directly from speech. However, existing studies still face three major challenges: **the lack of Chinese SpeechRE benchmarks**, **insufficient modeling of speech-specific cues**, and **evaluation bias introduced by character-level exact matching** in Chinese scenarios. To address these challenges, we first construct **CSRTE**, a real-world Chinese SpeechRE benchmark built from human speech, providing unified data and evaluation settings. We then propose **ACRE**, a reasoning enhancement framework for SpeechRE, which integrates acoustic-cue and contextual-knowledge into a structured acoustic-semantic reasoning chain, enabling the model to go beyond literal semantic content **(what-to-say)** and capture speech-specific features **(how-to-say)** for relation reasoning. We further combine supervised fine-tuning with reinforcement learning and design composite rewards to improve the quality of the model’s reasoning process. In addition, we propose pronunciation-aware evaluation **(PAE)**, which considers pronunciation and tone consistency during entity matching to reduce the evaluation bias. Experiments on both Chinese and English SpeechRE benchmarks demonstrate the effectiveness of ACRE and the usefulness of PAE. Our source code and datasets are available at https://github.com/MWXGOD/CSRTE.
%U https://aclanthology.org/2026.findings-emnlp.1244/
%P 26343-26352
Markdown (Informal)
[ACRE: An Acoustic-Cue and Contextual-Knowledge Reasoning Enhancement Framework for Speech Relation Extraction](https://aclanthology.org/2026.findings-emnlp.1244/) (Mu et al., Findings 2026)
ACL